Industrializing Additive Manufacturing: Mirko Meboldt Christoph Klahn Editors

Mirko Meboldt
Christoph Klahn Editors
Industrializing
Additive
Manufacturing
Proceedings of AMPA2020
Industrializing Additive Manufacturing
Mirko Meboldt Christoph Klahn
•
Editors
Industrializing Additive
Manufacturing
Proceedings of AMPA2020
123
Editors
Mirko Meboldt Christoph Klahn
Product Development Group Zurich pd|z Design for New Technologies ipdz
ETH Zürich Inspire AG
Zürich, Switzerland Zürich, Switzerland
ISBN 978-3-030-54333-4 ISBN 978-3-030-54334-1 (eBook)

https://doi.org/10.1007/978-3-030-54334-1
© Springer Nature Switzerland AG 2021
This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part
of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations,
recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission
or information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar
methodology now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this
publication does not imply, even in the absence of a specific statement, that such names are exempt from
the relevant protective laws and regulations and therefore free for general use.
The publisher, the authors and the editors are safe to assume that the advice and information in this
book are believed to be true and accurate at the date of publication. Neither the publisher nor the
authors or the editors give a warranty, expressed or implied, with respect to the material contained
herein or for any errors or omissions that may have been made. The publisher remains neutral with regard
to jurisdictional claims in published maps and institutional affiliations.
This Springer imprint is published by the registered company Springer Nature Switzerland AG
The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland
Preface
The 2nd AMPA Conference (Additive Manufacturing in Products and

Applications) held at ETH Zurich from September 1 to 3, 2020, is again focusing
on industrial series product applications and value chains enabled by additive
manufacturing. Three years after the first AMPA Conference, additive manufac-
turing is leaving the hype behind and is growing into established industrial seg-
ments. Additive manufacturing is gaining ground and problems relating to value
chains, design and production, quality assurance and cost models are becoming
increasingly important.
In recent months of the COVID-19 pandemic, additive manufacturing impres-
sively demonstrated what the technology can do when supply chains break down.
Worldwide, spare parts, consumables and new products were manufactured locally
by additive manufacturing.
Additive manufacturing is taking the next step on its journey toward a broad
range of series production, and we are facing many new scientific challenges.
Transforming the potential benefits of additive manufacturing into a successful
industrial or end user product is a challenge to all disciplines along the product
development process.
The topics of Additive Manufacturing in Products and Applications cover all
fields necessary to develop and produce successful products. Design Tools &
Methods: Identifying and designing AM parts, Process Chain Integration: Setting
up a safe and efficient production infrastructure, Business Cases of AM
Applications: Quantifying the benefits of AM, Unique Customer Benefits: Learning
from good and unusual examples and Teaching and Training: Bringing knowledge
and experience to new users.
v
vi Preface
All scientific contributions are double-blind peer-reviewed by industrial com-

mittee members for industrial relevance and by members of the scientific committee
for scientific quality. In a two-staged process, 35 contributions were selected out of
55 submitted abstracts.
We thank everyone who contributed to the success of the Additive
Manufacturing in Products and Applications conference: Thanks to the authors for
their valuable papers and talks, to the members of the industrial and scientific
committees for their hard but fair reviews and for chairing sessions, to the partic-
ipants of the sessions for the fruitful discussions and last but not least to all those
who supported the conference in the background.
Christoph Klahn
Mirko Meboldt
Organization
Organizing Committee
Christoph Klahn inspire AG, ipdz, Switzerland

Mirko Meboldt ETH Zürich, pd|z, Switzerland
Martin Stöckli inspire AG, Switzerland
Petra Kahl inspire AG, Switzerland
Daniel Omidvarkarjan inspire AG, ipdz, Switzerland
Urs Hofmann inspire AG, ipdz, Switzerland
Industrial Committee
Dominique Beuchat 3D Precision SA, Switzerland
Jeannette Clifford Sika Automotive AG, Switzerland
Michael Cloots IRPD AG, Switzerland
Hans Gut Güdel Group AG, Switzerland
Arno Held AM Ventures Holding GmbH, Germany
Lorenz Herrmann ABB Schweiz AG, Switzerland
Steffen Jung Nova Werke AG, Switzerland
Christoph Kiener Siemens AG, Germany
Harald Kissel Sandvik Machining Solutions AB, Sweden
Martin May Schunk GmbH & Co. KG, Germany
Klaus Müller-Lohmeier Festo SE & Co. KG, Germany
Maximilian Munsch AMPOWER GmbH & Co. KG, Germany
Patrizia Richner Sonova AG, Switzerland
Ralph Rosenbauer ALPA Capaul & Weber, Switzerland
Marco Salvisberg GF Precicast SA, Switzerland
Thomas Scheiwiller Bühler AG, Switzerland
Oliver Schlatter Injex AG, Switzerland
Martin Schöpf Robert Bosch GmbH, Germany
Ralf Schumacher Medartis AG, Switzerland
vii
viii Organization
Tobias Weber TRUMPF Laser- und Systemtechnik GmbH,

Germany
Eric Wycisk AMPOWER GmbH & Co. KG, Germany
Scientific Committee
Rosa Ballardini University of Lapland, Faculty of Law, Finland
Klas Boivie SINTEF Manufacturing AS, Norway
David Butler University of Strathclyde, UK
Olaf Diegel Lund University, Product Development, Sweden
Jens Ekengren Örebro University, Mechanical Engineering,
Sweden
Richard Hague University of Nottingham, EPSRC Centre
for Additive Manufacturing, UK
Russel Harris University of Leeds, Future Manufacturing
Processes Research Group, UK
Andreas Kirchheim ZHAW, Zentrum für Produkt- und
Prozessentwicklung (ZPP), Switzerland
Christoph Klahn inspire AG, ipdz, Switzerland
Christian Lindemann University Paderborn, DMRC, Germany
Bingjian Liu University of Nottingham Ningbo, Product
Design and Manufacture, China
Kaspar Löffel FHNW, Institut für Produkt- und
Produktionsengineering, Switzerland
Mirko Meboldt ETH Zürich, pd|z, Switzerland
Dimitris Mourtzis University of Patras, Laboratory for
Manufacturing Systems and Automation,
Greece
Alberto Ortona SUPSI, Hybrid Materials Laboratory,
Switzerland
Eujin Pei Brunel University London, UK
Maren Petersen University of Bremen, Institute Technology
and Education, Germany
Manfred Schmid inspire AG, icams, Switzerland
Marianne Schmid Daners ETH Zürich, pd|z, Switzerland
Jon Spangenberg DTU, Section of Manufacturing Engineering,
Denmark
Adriaan Spierings inspire AG, icams, Switzerland
Panagiotis Stavropoulos University of Patras, Laboratory for
Manufacturing Systems and Automation,
Greece
Tao Sun University of Virginia, USA
Klaus-Dieter Thoben University of Bremen, BIK, Germany
Organization ix
Anna Valente SUPSI, Automation, Robots and Machines

Laboratory, Switzerland
Oğuzhan Yilmaz Gazi University, Additive Manufacturing
Technology Application and Research Center
(EKTAM), Turkey
Michael Zäh TUM, Werkzeugmaschinen und
Fertigungstechnik, Germany
Markus Zimmermann TUM, Produktentwicklung und Leichtbau,
Germany
Contents
Design for AM
Generative Design Optimization and Characterization of Triple
Periodic Lattice Structures in AlSi10Mg . . . . . . . . . . . . . . . . . . . . . . . . 3
Patrik Karlsson, Lars Pejryd, and Niclas Strömberg
Structured Approach for Changing Designer’s Mindset Towards
Additive Manufacturing: From Theory to Practice . . . . . . . . . . . . . . . . 17
Gustavo Menezes de Souza Melo, Gerret Lukas, Johannes Willkomm,
Stephan Ziegler, Günther Schuh, and Johannes Henrich Schleifenbaum
An Interactive, Fully Digital Design Workflow for a Custom 3D
Printed Facial Protection Orthosis (Face Mask) . . . . . . . . . . . . . . . . . . . 26
Neha Sharma, Dennis Welker, Shuaishuai Cao, Barbara von Netzer,
Philipp Honigmann, and Florian Thieringer
Opportunities of 3D Machine Learning for Manufacturability
Analysis and Component Recognition in the Additive Manufacturing
Process Chain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
Tobias Nickchen, Gregor Engels, and Johannes Lohn
Review on the Design Approaches of Cellular Architectures Produced
by Additive Manufacturing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
Marco Pelanconi and Alberto Ortona
Process Chain
Multi-material 3D Printing of Thermoplastic Elastomers
for Development of Soft Robotic Structures with Integrated
Sensor Elements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
Antonia Georgopoulou, Bram Vanderborght, and Frank Clemens
xi
xii Contents
Solution Approaches and Process Concepts for Powder Bed-Based

Melting of Glass . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
Susanne Kasch, Thomas Schmidt, Fabian Eichler, Laura Katharina Thurn,
Simon Jahn, and Sebastian Bremen
Additive Manufacturing of Ti-Nb Dissimilar Metals
by Laser Metal Deposition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
Di Cui, Briac Lanfant, Marc Leparoux, and Sébastian Favre
Investigation of Plastic Freeformed, Open-Pored Structures
with Regard to Producibility, Reproducibility
and Liquid Permeability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
Andre Hirsch, Christian Dalmer, and Elmar Moritzer
Novel 4-Axis 3D Printing Process to Print Overhangs Without
Support Material . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
Michael Wüthrich, Wilfried J. Elspass, Philip Bos, and Simon Holdener
Hybrid Manufacturing: A New Additive Manufacturing Approach
for Closed Pump Impellers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146
Robin Rettberg and Thomas Kraenzler
Adaptive Slicing and Process Optimization for Direct Metal
Deposition to Fabricate Exhaust Manifolds . . . . . . . . . . . . . . . . . . . . . . 160
Daniel Eisenbarth, Alessandro Menichelli, Fabian Soffel,
and Konrad Wegener
Quality
Drift Detection in Selective Laser Melting (SLM) Using a Machine
Learning Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177
Pinku Yadav, Olivier Rigo, Corinne Arvieu, Emilie Le Guen,
and Eric Lacoste
Influence of the Inert Gas Flow on the Laser Powder Bed Fusion
(LPBF) Process . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 192
Florian Wirth, Alex Frauchiger, Kai Gutknecht, and Michael Cloots
Artificial Intelligence for Monitoring and Control of Metal
Additive Manufacturing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205
Giulio Masinelli, Sergey A. Shevchik, Vigneashwara Pandiyan,
Tri Quang-Le, and Kilian Wasmer
Simulation
Development of a Process Model for Bead Deposition Rates
and Cooling Behavior of Large Scale Additive Manufacturing Parts . . . 223
Michel Layher, Lukas Eckhardt, Andreas Hopf, and Jens Bliedtner
Contents xiii
Estimations of Interlayer Contacts in Extrusion Additive

Manufacturing Using a CFD Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . 241
Raphaël Comminal, Sina Jafarzadeh, Marcin Serdeczny,
and Jon Spangenberg
Influence of Fibers on the Flow Through the Hot-End in Material
Extrusion Additive Manufacturing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 251
Marcin Serdeczny, Raphaël Comminal, David Bue Pedersen,
and Jon Spangenberg
Deploying Artificial Intelligence for Component-Scale Multi-physical
Field Simulation of Metal Additive Manufacturing . . . . . . . . . . . . . . . . 268
Ehsan Hosseini, P. Gh. Ghanbari, F. Keller, S. Marelli, and Edoardo Mazza
Prototyping and Testing

Experimental Investigation of Filament Behaviour in Material
Extrusion Additive Manufacturing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 279
Mark Golab, Sam Massey, and James Moultrie
Debinding and Sintering of Dense Ceramic Structures Made
with Fused Deposition Modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 293
Frank Clemens, Josef Schulz, Lovro Gorjan, Antje Liersch,
Tutu Sebastian, and Fateme Sarraf
Feasibility Investigation of Gears Manufactured by Fused
Filament Fabrication . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 304
Hans-Jörg Dennig, Livia Zumofen, and Andreas Kirchheim
Qualification of Additively Manufactured Blood Vessel Models
for the Evaluation of Braided Stent Designs . . . . . . . . . . . . . . . . . . . . . . 321
Juliane Kuhl, Ngoc Tuan Ngo, Jan-Hendrik Buhk, Andreas Ding,
Andrés Braschkat, Jens Fiehler, and Dieter Krause
Innovative Use Case

Additive Manufactured and Topology Optimized Flexpin
for Planetary Gears . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 337
Anton Höller, Frank Huber, Livia Zumofen, Andreas Kirchheim,
Hanspeter Dinner, and Hans-Jörg Dennig
A Review of Optimised Additively Manufactured Steel Connections
for Modular Building Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 357
Zhengyao Li, Konstantinos Daniel Tsavdaridis, and Leroy Gardner
Application of Topology Optimisation to Steel Node-Connections
and Additive Manufacturing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 374
Moustafa Mahmoud Abdelwahab and Konstantinos Daniel Tsavdaridis
xiv Contents
The AM Dowel – A Novel Insert for the Integration of Threads

into Additive Manufactured Polymer Components . . . . . . . . . . . . . . . . . 391
Daniel Omidvarkarjan, Peter Balicki, Harry Baumgartner,
Ralph Rosenbauer, Filippo Fontana, and Mirko Meboldt
Novel Pressure Swirl Nozzle Design Enabled
by Additive Manufacturing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 399
Michael Umbricht, Kaspar Löffel, Marc Huber, Patrick Lüscher,
Janine Bochsler, Daniel Weiss, and Tom Duda
Design of an Additively Manufactured Customized Gripper System
for Human Robot Collaboration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 415
Nikolai Hangst, Stefan Junk, and Thomas Wendt
Enhanced Cooling Design in Wire Drawing Tooling Using
Additive Manufacturing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 426
Joakim Larsson, Patrik Karlsson, Jens Ekengren, and Lars Pejryd
Aortic Model in a Neurointerventional Training Model – Modular
Design and Additive Manufacturing . . . . . . . . . . . . . . . . . . . . . . . . . . . . 437
Nadine Wortmann, Andreas M. Frölich, Anna A. Kyselyova,
Helena I. De Sousa Guerreiro, Jens Fiehler, and Dieter Krause
Integration of Additive Manufacturing into Process Chain
of Porcelain Preservation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 455
Bingjian Liu, Fangjin Zhang, Xu Sun, and Adam Rushworth
Business Cases
Decision Support System for a Metal Additive Manufacturing
Process Chain Design for the Automotive Industry . . . . . . . . . . . . . . . . 469
Markus Johannes Kratzer, Julian Mayer, Florian Höfler,
and Nikolaus Urban
Business Models for Additive Manufacturing: A Strategic View
from a Procurement Perspective . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 483
Andreas H. Glas, Matthias M. Meyer, and Michael Eßig
A Performance Upgrade of an Industrial Gas Turbine Based
on Additive Manufactured Components . . . . . . . . . . . . . . . . . . . . . . . . . 500
Pankaj Bajaj, Fulvio Magni, and Peter Flohr
Author Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 511

Design for AM
Generative Design Optimization
and Characterization of Triple Periodic
Lattice Structures in AlSi10Mg
Patrik Karlsson, Lars Pejryd, and Niclas Strömberg(B)
Örebro University, 701 82 Örebro, Sweden

niclas.stromberg@oru.se,
http://www.oru.se
Abstract. In this work, generative design optimization and characteriza-

tion of triple periodic lattice structures in AlSi10Mg are considered. Struc-
tures with Gyroid, Schwarz-D and G-prime lattices are designed optimally
by utilizing a generative design optimization approach. The approach is
based on topology optimization, support vector machines (SVM), radial
basis function networks (RBFN), morphing operations, design of experi-
ments and metamodels. Firstly, topology optimization solutions are gener-
ated which are represented using SVM, secondly, sizing solutions obtained
by setting the SIMP parameter equal to one are represented with RBFN.
Thirdly, graded lattice structures using the RBFN are morphed together
with the SVM to final conceptual designs. Fourthly, design of experiments
of the conceptual designs are performed using non-linear finite element
analyses (FEA) and, finally, metamodel-based design optimization is con-
ducted using convex combinations of Kriging, RBFN, polynomial chaos
expansion and support vector regression models. In order to validate the
optimal designs, new tensile test specimens that include the periodic lat-
tice structures are suggested. The specimens with all three lattices are
manufactured in AlSi10Mg using direct metal laser sintering with an EOS
M290 machine. Tensile tests of these specimens are then performed and
validated using nonlinear FEA. The test specimens are also characterized
with respect to geometry and defects by means of computed tomography,
optical microscopy and scanning electron microscopy. The study demon-
strates the high potential of using the proposed generative design optimiza-
tion approach with triple periodic lattice structures for producing robust
lightweight designs using additive manufacturing. In order to demonstrate
the industrial relevance the established GE engine bracket is studied in the
paper and discussed at the conference.
Keywords: Generative design · Lattice structures · AlSi10Mg
1 Introduction
In the last ten years or so, additive manufacturing (AM) has become a topic
of intensive research and development work in both industry and academia.
c Springer Nature Switzerland AG 2021
M. Meboldt and C. Klahn (Eds.): AMPA 2020,
Industrializing Additive Manufacturing, pp. 3–16, 2021.
https://doi.org/10.1007/978-3-030-54334-1_1
4 P. Karlsson et al.
The high interest stems from e.g. short lead time and the possibilities that the
technology may offer to generate highly complex structures with “no or lim-
ited extra cost of manufacturing”. The development of several processes and
equipment dedicated to the additive manufacturing of metal based components
is of course also one of the major reasons for this increased interest, especially
from industry. This have resulted in expectations that this technology may bring
upon a “complete shift in the manufacturing industry”. The high expectations
that is quite often seen, may result in significant disappointments (“just another
fancy and expensive technology”) if the full potential of the possibilities that AM
offers in the design of components with additional functionalities is not utilized.
In order to do so, the design process also needs to be developed in order to be
able to handle new design elements and process steps in an efficient way. One
of the potential show stoppers for increased use of AM in industry, identified
by Swedish industry in their Strategic Research Agenda (SRA), is the lack of
engineers educated and trained in design for additive manufacturing (DfAM).
The need for new and/or improved design methodology and the education of
engineers on these methods has therefore been identified as important. One type
of design elements that is now possible to use in the manufacture of light weight
components is lattices. Lattice structures may provide a high stiffness to weight
ratio and can be included anywhere in parts, due to the flexibility of the AM
methods. This, together with topology optimisation (TO), may provide signifi-
cant flexibility in design options, but may also be a challenge for the efficiency
of the design process. Especially in the steps from TO-based concept genera-
tion to CAD description that is necessary for the generation of machine code
for AM equipment. Much of lattices that have been used so far in components
are based on struts and vary from simple cubic lattices to topology optimized
lattices that alter their shape to increase the stiffness to weight ratio [1]. A new
contender for the strut lattices has in recent years appeared in the form of peri-
odic surface based lattices, even though the concept has been around for a long
time [2]. These triply periodic minimal surfaces (TPMS) are made up from a
single surface that wraps around itself to create a cubic surface lattice structure.
The TPMS lattices appears to be stiffer than the strut lattices and they could
also provide a higher redundancy to fabrication errors [3,4].
In this work, a two-steps pragmatic approach for combining topology opti-
mization (TO) concepts with optimal graded triply periodic minimal surfaces
(TPMS) to final design concepts is presented. The approach is based on a two-
steps procedure according to Fig. 1. In the first step, topology optimization is
performed by setting the SIMP factor to four. The TO-solution is classified
using support vector machines as proposed in [5]. In such manner, an implicit
surface representation of the TO-solution is obtained. In the second step, size
optimization is performed by setting the SIMP factor equal to one. The optimal
thickness distribution is represented by radial basis function networks which in
turn is utilized to functionally grade our implicit representations of the TPMS:
Gyroid, G-prime and Schwarz-D. After these two steps, the implicit SVM-based
TO-surfaces and the optimal graded implicit TPMS are combined by computer
Generative Design Optimization 5
Fig. 1. The basic idea of the generative design optimization approach.
graphics operations, such as e.g. boolean, blending and morphing operations,

in order to set up design of experiments of final design concepts of TO-based
layouts and optimal graded TPMS. In order to evaluate material properties
utilized in the design optimization method, tensile test specimens were man-
ufactured by AM, validated by tensile testing and characterized by means of
optical microscopy, scanning electron microscopy and computed tomography.
Functional graded TPMS have become most popular in tissue engineering.
Already in 2008, Gabrielli et al. [6] developed methods for functionally graded
TPMS designs to be used as bone substitutes. Functionally graded Schwarz-
D titanium structures for bone implants were manufactured by selective laser
melting in [7]. A review of functionally graded lattice structures in additive man-
ufacturing of orthopedic implants can be found in [8]. A study on the mechan-
ical properties of functionally graded Gyroid structures fabricated by selective
laser melting was performed by Yang et al. [9]. Compressive tests of function-
ally graded TPMS were conducted in [10]. Most recently, functionally graded
hybrid TPMS lattices, by blending using a Sigmoid operation, were simulated
and tested by Al-Keta et al. [11]. Studies on optimal graded lattice structures
appear less frequently. Most of the work so far on optimal graded lattice struc-
tures is performed on strut-based designs, see e.g. [1,12,13]. However, last year,
an interesting approach for optimal graded TPMS was proposed by Li et al. [14].
Their work is indeed most relevant for comparison of the presented generative
design optimization approach in this paper.
2 The Generative Design Optimization Approach

In this section, a generative design optimization approach based on implicit sur-
faces of topology optimization concepts and graded TPMS-based lattice struc-
tures is presented, see Fig. 1.
Gyroid f1 , G-prime f2 and Schwarz-D f3 lattices are considered by the fol-

lowing implicit surfaces (see Fig. 2):
f1 = sin(γx) cos(γy) + sin(γy) cos(γz) + sin(γz) cos(γx) ± κ,

f2 = [sin(2γx) sin(γz) cos(γy) + sin(2γy) sin(γx) cos(γz)
+ sin(2γz) sin(γy) cos(γx)] + 0.2 [cos(2γx) cos(2γy)
(1)
+ cos(2γy) cos(2γz) + cos(2γz) cos(2γx)] ± κ,
f3 = sin(γx) sin(γy) sin(γz) + sin(γx) cos(γy)cos(γz)
+ cos(γx) sin(γy) cos(γz) + cos(γx) cos(γy) sin(γz) ± κ,
where γ and κ are parameters controlling the period and the thickness.
Fig. 2. The triple periodic lattice structures.
The implicit surfaces in (1) representing our TPMS-based lattice structures

are graded by setting the SIMP-factor n equal to one1 in the SIMP-based topol-
ogy optimization formulation reading:
⎧
⎪
⎪ min c(d) = F T d
⎪
⎨ ρ ,d ) ⎧
(
⎨ K(ρ)d = F , (2)
⎪
⎪ s.t. V (ρ) ≤ V̂ ,
⎪
⎩ ⎩
≤ ρ ≤ 1,
where the vector ρ contains density variables ≤ ρe ≤ 1, where is a small

positive number representing zero density in order to avoid difficulties with sin-
gular stiffness matrices. The value ρe = 1 represents of course a completely filled
element with material and ρe = represents no material. Furthermore, d is the
displacement vector, F is the external force vector, where

K(ρ) = ρne ke (3)
e
1
A SIMP factor close to two might be a better choice in order to represent the stiffness
as function of the density of the TPMS lattice. In a near future, we will implement
proper material interpolation laws derived by numerical homogenization.

is the SIMP-based stiffness matrix, where is an assembly operator. The total
volume of the design is obtained as

V (ρ) = ρe Ve , (4)
e
where Ve represents the volume of element e and it is constrained by the upper

limit V̂ . The meaning of setting n = 1 is thickness optimization is 2D. The
topology optimization is performed by setting n = 4.
Fig. 3. The basic idea of the proposed SVM-based postprocessing approach [5].
The topology optimization solution obtained by solving (2) using n = 4 is

transformed to an implicit surface by using the support vector formulation of
Mangasarian [15,16]. That is, the topology optimization solution of zeros and
ones is classified using the following linear programming problem:
⎧ N N N
⎪
⎪ min
⎪ qi + pi + C vi
⎪
⎪ (q ,p ,b,v )
⎨ ⎧ i=1 i=1
N
i=1
⎪
⎨ (5)
i
⎪
⎪ 1 − v i − y ( k(xi , xj )y j (qj − pj ) + b) ≤ 0, i = 1, . . . N,
⎪ s.t.
⎪
⎪
⎩ ⎪
⎩ j=1
qi ≥ 0, pi ≥ 0, vi ≥ 0, i = 1, . . . N,
where k(xi , xj ) represents a kernel function, the sampling point xi is the center
point of a finite element e, y i = 1 corresponds to a density ρe = 1 and y i is set to
−1 for densities ρe = 0. A geometric model of a topology optimization solution
is then defined by the following implicit surface:
⎧
N ⎨ < 0, outside the solid,
Φ(x) = y j (qj − pj )k(x, xj ) + b = = 0, boundary of the solid, (6)
⎩
j=1 > 0, inside the solid,
where q, p and b are given by the optimal solution of (5). Thus, (6) serves
the purpose to be the “CAD” model of the TO concept and it is established
completely automatically without any interaction with CAD engineers. The basic
idea of the support vector machine (SVM) approach is illustrated in Fig. 3.
A major benefit of having the topology optimization concept represented as
an implicit surface in (6) is that standard computer graphics operations, such as
e.g. boolean, blending and morphing operations, are applicable. In such manner,
the triple periodic lattice structures given in (1) can easily be integrated with
the topology concept as presented in Fig. 1. Letting g be the implicit surface of
the topology optimization solution and h be the graded lattice structure, then
we first generate a cover by g̃ = min(g + α, −g + α) and, secondly, merge this
cover with the graded lattice structure using max(g̃, h) in order to generate the
final concept. By changing the scalar α we of course change the thickness of
the cover and this can also be done locally by morphing operations. In such
manner, detailed design optimization of our TO concepts with TMPS-based
lattice structures can be done by setting up design of experiments, executing
non-linear finite element analyses and establishing metamodels. The latter is
obtained by letting the computer experiments from the finite element analyses
be represented by optimal convex combinations of metamodels. The taxicab
norm of the leave-one-out cross validation errors is minimized for an ensemble
of metamodels consisting of Kriging models, radial basis function networks and
support vector regression models. This is performed by solving the following
linear programming problem [17]:
⎧
⎪
⎪
N

⎪ (wmin
⎪ pi + qi
⎨ ,p ,q )
⎧i=1
⎨ Y w − f̂ = p − q, (7)
⎪
⎪ T
⎪
⎪ s.t. w 1 = 1,
⎩ ⎩
wi , pj , qj ≥ 0, i = 1, . . . , M, j = 1, . . . , N,
where Y consists of the leave-one-out cross validation errors and wi are the
weights in the convex combination of metamodels, which is given by
M

yen = yen (x) = wi yi (x), (8)
i=1
where yi (x) represents a Kriging model, a radial basis function network or a

support vector regression model.
Another useful property of the SVM-based implicit surfaces is the treatment
of multiple load cases. The standard approach for handling multiple loads is
to minimize the weighted compliance. But by using the SVM-based implicit
surfaces, we can simply blend together the solutions from the load cases taken
independently. This is illustrated for the GE bracket in Fig. 4. Letting g1 , g2
and g3 represent the implicit surface for the three load cases presented in the
figure, then the presented blended design is given by max(max(g1 , g2 ), g3 ). The
critical load for the structure is then studied by J2-plasticity analysis using
Abaqus/Standard.
Fig. 4. Topology optimization of the GE bracket for multiple load cases by blending
support vector machines.
3 Validation of Tensile Tests
In order to perform design optimization using TPMS-based lattice structures as

presented in the previous section, reliable material data and constitutive material
models are needed. The design optimization process would be easier if standard
bulk properties and models could be used instead of finding new material param-
eters and constitutive equations. It is well-known that structural properties drop
Fig. 5. Tensile tests of lattice structures.

for thin members. So, the question is if or when an assumption of bulk behaviour
is appropriate for the TPMS-based lattice structures. In this work this is investi-
gated by designing, printing, testing and analyzing tensile test specimen of triple
periodic lattice structures according to Fig. 5.
The purpose of the study is to validate if J2-plasticity with linear isotropic
hardening using bulk properties is a proper model for the TMPS-based lattice
structures when performing design optimization. The plasticity model is first
set up using material data for AlSi10Mg from EOS and calibrate the model for
these data using the tensile tests for the test specimen without lattice structures.
It is clear as usual that defining the yield stress using the Rp0.2 value implies
a non-conservative model. Therefore, the yield stress is adjusted to match the
tensile test curves. Next, the calibrated J2-plasticity model is used to simulate
the TMPS-based lattice structures and the simulation is compared to the tensile
test results for the corresponding lattice structure. The validation procedure is
presented for the Schwarz-D structure in Fig. 6. In general, the tensile tests of
the lattice structures and the FEA predictions match very well. However, larger
standard deviations of the tests of the lattice structures compared to bulk data
are observed.
Fig. 6. Non-linear finite element analyses of tensile tests.
4 Characterization of Tensile Tests

Test specimens were, subsequently to tensile testing, characterized by means of
scanning electron microscopy (SEM), optical microscopy (OM) and computed
tomography (CT). Material characterizations by means of microscopy techniques
were done by using an Zeiss FEG-SEM Zigma 300 VP and a Zeiss Imager M2m
stereo microscope. CT analysis of test samples was done using a Bruker Skyscan
1272 scanner. Reconstruction of CT data was performed using InstaRecon soft-
ware and utilizing an optimization for beam hardening, ring artefacts and post-
alignment correction. The postprocessing were performed using VG Studio Max
from Volume Graphics and the custom method VGDefX for porosity analysis.
In Fig. 7, the results from fractography of bulk specimens by means of SEM
are shown. The results generally indicated ductile fracture and dimples in size
range of about 0.2–1 µm were observed, Fig. 7 a). Typical defects detected in
SEM were pores with size of approximately 20–50 µm, as seen in regions marked
by dashed red circles in Fig. 7 b). Some pores contained particles, exemplified in
the region marked with dashed red circle in Fig. 7 c) and magnified in Fig. 7 d),
with particle size of about 1–10 µm.
The results from fractography of samples with lattice structures by means of
OM and SEM are seen in Figs. 8 and 9. In order to get an overview of the fractured
samples surfaces, surface analysis by means of OM, utilizing focus stacking, was
performed. Downskin were observed in OM and a typical surface region containing
downskin is marked by dashed lined square in Fig. 8 a). The same fracture surface
region, marked in Fig. 8 a), was analyzed by SEM and it was found that the down-
skin regions contained accumulation of powder particles, Fig. 8 b)–d). In the right
lower part of Fig. 8 b), the fracture surface is seen and in the right upper part of
the figure, accumulation of powder particles is evident. The interface between the
fracture surface and the as printed surface contained partly melted powder, Fig. 8
Fig. 7. Typical fracture surfaces of the bulk specimens. Dimples a) and pores, marked
by dashed red circles in b)–c) and magnified in d), were observed in SEM.
Fig. 8. Fracture surfaces of samples with lattice structures observed in OM a) and

SEM b)–d). Arrows indicates build direction.
c), and larger pores with size of approximately 50–100 µm in regions of particle
clusters, as seen in Fig. 8 d).
Similar to bulk specimens, Fig. 7 b)–d), pores were also observed in the frac-
ture surface analysis of the samples with lattice structures in SEM, Fig. 9 a)–b).
The pores varied in size, from nano- to micro sized pores, as seen in the marked
regions in Fig. 9 b), and some pores contained particles with sizes of approxi-
mately 10–50 µm, Fig. 9 a). It has been shown that pores in parts produced by
AM may originate from oxide inclusions and entrapped gas [18–20]. Thus, the
particles found in pores in this study, Fig. 1 c)–d) and Fig. 9 a), may be inclu-
sions. However, chemical analysis of the particles in SEM did not indicate foreign
elements and, thus, the particles are probably partly melted powder particles,
resulting in lack of fusion porosity. Additional pores observed in the present
study may though be a result of entrapped gas during manufacturing of samples
by AM.
The defects identified in specimens with lattice structures by means of SEM
were generally pores. However, the gyro samples contained additional defects.
Cracks with clusters of micron-sized particles, marked with red arrows in Fig. 9
c), were detected in SEM. Some of these particles were also found in cluster
formations on open fracture surface regions. The particles observed in SEM are
magnified in Fig. 9 d) and as can be seen in the figure, the particles appears to
be partly melted particles. However, more research on this is needed in order to
fully explain the particle formation found in Gyro samples.
Fig. 9. Typical defects in specimens with lattice structures in general a)–b) and Gyro
samples specifically c)–d).
In Fig. 10, the results from 3D analysis of specimens by means of CT are

shown for the Schwarz D sample. The CT analysis of specimens enabled 3D-view,
Fig. 10 a), and 3D porosity analysis, Fig. 10 b), of the samples. Downskin areas
observed in OM and SEM were confirmed by CT, which additionally enabled an
easy overview of the downskin positions. Clusters of particles were observed on
both curved regions and on the outer shell of the tensile test specimens with lat-
tice structures. Downskin may be a result of lattice design challenging the limits
of AM design [21]. The results from porosity analysis performed by CT indicated
pore sizes of approximately 20–300 µm. This is seen for a fractured Schwarz D
specimen in Fig. 10 b), which is a transparent view of the sample in Fig. 10 a).
The larger pores may be a result of the large voids between particles downskin
areas, where accumulation of particles was observed in SEM, Fig. 8 d). These
pores may act as crack initiation sites, causing material failure. Defects such as
pores are generally found in AM produced parts and it is commonly known that
the defects influence mechanical properties of the material negatively [22]. Addi-
tionally, size, amount and position of defects influence mechanical properties of
the material and is probably the cause of the tensile test data scattering as seen
for the Schwarz-D samples in Fig. 6. Thus, the material in the present study may
be optimized by reducing the defects observed in bulk and lattice samples.
Fig. 10. 3D-view a) and porosity analysis results b) of the Schwarz-D sample, investi-
gated by means of CT.
5 Concluding Remarks
In this work, a generative design optimization approach for functional grading
of triply periodic minimal surface lattice structures by using topology optimiza-
tion, support vector machines and radial basis function networks is proposed
and implemented. In a very near future, the approach will be improved by using
a proper material interpolation law for the lattice derived by using numerical
homogenization instead of using SIMP or RAMP. Established bulk data is the
starting point, but this will be needed to be further tested and validated in order
to improve the design optimization approach. AM processing, especially of thin
structures and/or structures with thin elements included in the component, is
still in need for further development and investigations in order to understand
and possibly remove flaws and porosity that may possess weaknesses and thereby
reducing strength and life of components. In this work, test specimens were char-
acterized by means of CT, OM and SEM. Porosity, possibly caused by entrapped
gas during manufacturing of samples by AM, and lack of fusion porosity were
observed for both bulk samples and specimens with lattice structures. Size of
the pores ranged from nano- to micro sized pores and pores were mainly found
in fracture surfaces and in downskin regions, where accumulation of powder par-
ticles was evident and larger pores were observed. Additionally, fractography by
means of SEM of samples with Gyro lattice structure reviled other defects such
as partly melted particles.
References
1. Daynes, S., Feih, S., Lu, W.F., Wei, J.: Optimisation of functionally graded lattice
structures using isostatic lines. Mater. Des. 127, 215–223 (2017)
2. Schwarz, H.A.: Gesammelte Mathematische Abhandlungen. Springer, Berlin (1890)
3. Jansson, A., Ekegren, J., Pejryd, L.: Numerical analysis of compression strength in
network structures based on trusses, and periodic surfaces, aimed for additive man-
ufacturing. In: The Proceedings of Progression in Additive Manufacturing (2018)
4. Jansson, A., Pejryd, L.: In-situ computed tomography investigation of the com-
pression behaviour of strut, and periodic surface lattices. In: The Proceeding of
the ICT2019, Paper id: 2366 (2019)
5. Strömberg, N.: Efficient detailed design optimization of topology optimization con-
cepts by using support vector machines and metamodels. Eng. Optim. 52, 1136–
1148 (2019)
6. Gabbrielli, R., Turner, I.G., Bowen, C.R.: Development of modelling methods for
materials to be used as bone substitutes. Key Eng. Mater. 361–363, 903–906
(2008)
7. Han, C., Li, Y., Wang, Q., Wen, S., Wei, Q., Yan, C., Hao, L., Liu, J., Shi, Y.:
Continuous functionally graded porous titanium scaffolds manufactured by selec-
tive laser melting for bone implants. J. Mech. Behav. Biomed. Mater. 80, 119–127
(2018)
8. Mahmoud, D., Elbestawi, M.A.: Lattices structures and functionally graded mate-
rials applications in additive manufacturing of orthopedic implants: a review. J.
Manuf. Mater. Process. 1(2), 1–19 (2017)
9. Yang, L., Mertens, R., Ferrucci, M., Yan, C., Shi, Y., Yang, S.: Continuous graded
gyroid cellular structures fabricated by selective laser melting: design. Manuf.
Mech. Prop. Mater. Des. 162, 394–404 (2019)
10. Afshar, M., Anaraki, A.P., Montazerian, H.: Compressive characteristic of radially
graded porosity scaffolds archiectured with minimal surfaces. Mater. Sci. Eng. 92,
254–267 (2018)
11. Al-Ketan, O., Lee, D.W., Rowshan, R., Al-Rub, R.K.A.: Functionally graded and
multi-morphology sheet TPMS lattices: design, manufacturing, and mechanical
properties. J. Mech. Behav. Biomed. Mater. 102, 1–17 (2020)
12. Cheng, L., Bai, J., To, A.C.: Functionally graded lattice structure topology opti-
mization for the design of additive manufactured components with stress con-
straints. Comput. Methods Appl. Mech. Eng. 344, 334–359 (2019)
13. Jin, X., Li, G.X., Zhang, M.: Optimal design of three-dimensional non-uniform
nylon lattice structures for selective laser sintering manufacturing. Adv. Mech.
Eng. 10(7), 1–19 (2018)
14. Li, D., Dai, N., Tang, Y., Dong, G., Zhao, Y.F.: Design and optimization of graded
cellular structures with triply periodic level surface-based topological shapes. J.
Mech. Des. 141, 071402-1-13 (2019)
15. Mangasarian, O.L.: Generalized support vector machines. In: Advances in Large
Margin Classifiers, pp. 135–146. Cambridge, MA (2000)
16. Mangasarian, O.L.: Exact 1-norm support vector machines via unconstrained con-
vex differentiate minimization. J. Mach. Learn. Res. 7, 1517–1530 (2006)
17. Strömberg, N.: Comparison of optimal linear, affine and convex combinations of
metamodels. Eng. Optim. 52, 1–17 (2020)
18. Thijs, L., Van Humbeeck, J., Kempen, K., Yasa, E., Kruth, J.-P.: Investigation
on the inclusions in maraging steel produced by selective laser melting. In: 5th
International Conference on Advanced Research in Virtual Rapid Prototyping, pp.
297–304 (2011). https://doi.org/10.1201/b11341-48
19. Surreddi, K.B., Oikonomou, C., Karlsson, P., Olsson, M., Pejryd, L.: In-situ micro-
tensile testing of additive manufactured maraging steels in the SEM: influence of
build orientation, thickness and roughness on the resulting mechanical properties.
La Metallurgia Italiana 3, 27–33 (2018)
20. Tang, M., Chris, P.: Pistorius, oxides, porosity and fatigue performance of
AlSi10Mg parts produced by selective laser melting. Int. J. Fatigue 94, 192–201
(2017)
21. Adam, G.A.O., Zimmer, D.: On design for additive manufacturing: evaluating geo-
metrical limitations. Rapid Prototyp. J. 21, 662–670 (2015). https://doi.org/10.
1108/RPJ-06-2013-0060
22. Yasa, E., Kempen, K., Kruth, J.: Microstructure and mechanical properties of
maraging steel 300 after selective laser melting. In: Proceedings of the 21st Inter-
national Solid Freeform Fabrication Symposium, pp. 383–396 (2010)
Structured Approach for Changing Designer’s
Mindset Towards Additive Manufacturing:
From Theory to Practice
Gustavo Menezes de Souza Melo1(&), Gerret Lukas2,

Johannes Willkomm1, Stephan Ziegler1, Günther Schuh2,
and Johannes Henrich Schleifenbaum1
1
Chair for Digital Additive Production (DAP), RWTH Aachen University,
52074 Aachen, Germany
gustavo.melo@dap.rwth-aachen.de
2
Laboratory for Machine Tools and Production Engineering (WZL), RWTH
Aachen University, 52074 Aachen, Germany
Abstract. Additive Manufacturing (AM) has a great potential of disrupting

product design and supply-chain in many industries by means of its unique
capabilities. Regarding the product design, the potential benefits comprise
functional integration, reduced assembly efforts, reduced weight and increased
performance. Although AM has been around for decades, designers still think in
the restrictions imposed by conventional manufacturing. The awareness of the
potentials of AM has not yet been pushed in the minds of designers and the
adoption of AM in design process often fails due to a status quo in design or
limited knowledge of the employees. Against this background, this paper pro-
poses a framework to change Designer’s mindset towards AM. By means of in-
depth interviews with designers and design engineers from different industries,
the common challenges and implemented solutions were investigated. From
these expert interviews, the following key challenges were identified: AM-
adjusted design methodology, standards implementation and software support.
Based on those, a wide literature review of possible solutions was carried out
and its result was combined with the already implemented solutions in industry.
The proposed framework not only takes advantage of currently available human
capital in the organization but also paves a sustainable way to train new per-
sonnel and create momentum towards AM adoption. By means of a structured
learning path and a knowledge management platform integrated into design
software, the proposed framework effectively extracts tangible and part-specific
design rules and assures optimal knowledge transfer among employees. This
framework was subsequently validated in a workshop with industry experts.
Keywords: Additive Manufacturing Design methods DfAM Product

development Change management

https://doi.org/10.1007/978-3-030-54334-1_2
18 G. M. d. S. Melo et al.
1 Introduction
AM has been emerging strongly in recent years. Growth in machine sales and increased
numbers of equipment manufacturers show how the AM market has been expanding [1].
AM enables the fabrication of products with high complex design with various func-
tionalities [2–4]. However, design engineers often think in the restrictions imposed by
conventional manufacturing or link AM to unrealistic expectations [2]. The awareness of
the potentials and restrictions of AM has not been effectively pushed in the minds of
design engineers. Moreover, the adoption of AM in design process often fails due to a
status quo in design or limited knowledge of the employees. A sustainable adoption of
AM is only possible by means of complete mindset shift of designers and design engineers
from conventional manufacturing towards AM [3]. In other words, today’s professionals
need to change the way they approach design problems. Against this background, some
companies have already started their journey to train their employees in AM.
Academic literature has dealt with education in the field of AM for almost a decade.
Since AM has been of growing interest, Geraedts et al. [4] investigated the role of AM
in the light of design engineering in three domains: business, research, and education.
At the same time Williams and Seepersad [5] developed a concept combining project-
based and problem-based learning for a university course. In both papers, the dominant
topic was education for future designers in AM. Then, Ford and Dean [6] discussed the
general necessity of teaching conventional manufacturing in comparison to AM. They
conclude that designer should not ignore conventional design and AM should be added
to the curriculum. A sole focus on AM could result in diminishing conventional
technologies. Loy [7] puts this conclusion into a different perspective, by stating that
design educators face a number of different challenges in terms of AM in design
education. Minetola et al. [8] use a survey to investigate the impacts of early exposure
with AM in engineering education and find that a “think-additive” approach early on
leads to a full facilitation of the benefits of AM. Simpson et al. [9] and Prabhu et al.
[10] conclude in a similar way. Yet, only Watschke et al. [11] propose a methodical
approach for design education, however they focus on the ideation process.
In the light of previous and current research, the prevailing need of companies for
designer with an AM mindset has not yet been addressed. As researcher focused on the
secondary and tertiary education to train future talent, the education for professionals,
also referred as continuing education, has been neglected. Of course, educating pro-
fessionals in AM is core to a number of certificate courses and workshops, but literature
does not provide a systematic approach that addresses the needs of companies.
Therefore, this paper aims to develop a systematic framework to educate design
engineer professionals and provide insight into the development of an AM mindset in
industrial companies.
Against this background, this paper presents a survey among industry participants
for a deeper understanding of challenges, goals and current implemented solutions in
companies. Subsequently, a broad literature study is carried out in order to collect
further best practices among the academia and industry beyond our focus
group. Finally, based on the two steps before mentioned, a systematic framework is
developed and validated through a workshop with the interview participants.
Structured Approach for Changing Designer’s Mindset 19
2 Method
The presented research utilizes qualitative research. AM mindset cannot be described

by a defined set of variables, it rather emerges from a dynamic model based on
qualitative data: In order to generate such data, we used two types of methods. We base
on the concept for grounded theory [12] as we obtain data by interviewing a group of
representatives of companies that facilitate AM in their organization. In addition to the
interviews, we conducted a systematic review of existing literature based on codes from
the interviews. Table 1 gives an overview of the addressed industries. The participants
of each company are in charge of the AM activities and are ranked in middle
management.
Table 1. Overview of interview participants.

Industry Number of companies Number of participants
Machine manufacturing 3 3
Automotive 3 3
Materials & process 4 4
The interviews were semi-structured, conducted by one of the authors and recorded
for documentation purpose. Before the interview, the participants received a guideline
containing seven open questions to prepare the interview. During the interview, the
interviewer could alter the question, if needed, to enlarge on topics of interest for the
study. In order to keep track of such changes to the guidelines, interview reviews were
conducted and if necessary the guidelines were adjusted. However, every participant
was only interviewed once. After all interviews were conducted, we coded the tran-
scripts and categorized findings. Our three main categories were “common challenges”,
“implemented solutions” and “shared goals”. As our participant group was small
compared to other qualitative studies, we ensured iteration between initial coding and
categorizing for an objective analysis of the interview data.
Subsequently to the interviews, a systematic literature review based on the pro-
cedure by Kitchenham [13] was applied. This systematic is divided into three phases:
planning, conducting and reporting. Within the planning phase the objectives of the
literature review was defined. From the interviews (“common challenges” and “shared
goals”), three key areas were identified as vital for mindset shift towards AM:
• AM-adjusted design methodology, which raises awareness and increase know-how
• Implementation of standards, which uses a structured and accessible approach
• Provision of better software, which supports expert knowledge exchange
During the conductive phase, the literature was collected and analysed. For this
purpose, the literature was divided in two categories according to their scientific value:
primary source (e.g. paper, standards and technical books) and second source (e.g.
magazine, internal knowledge and online guidelines). By means of the systematic
proposed by Kitchenham, numerous literature studies were screened, with a focus on

the last decade. The identified literature was subsequently presented to in-house AM
design experts who selected the most relevant and comprehensive ones. In the reporting
phase, the literature is summarized and reported for further investigation.
3 Result
The goal of the interviews in our study group was obtaining a description of the status
quo of continuing education for AM at the respective companies. During the coding of
the interviews, we identified emerging themes via our in-vivo codes, as exemplified in
Table 2. We avoided any early categorization for an objective analysis of the data [14].
At the end of our interview transcription and coding, we found that three categories
were fitting our themes and codes best: “common challenges”, “implemented solu-
tions”, and “shared goals”.
Table 2. Examples from in-vivo codes, themes, and categories.

In-vivo code Theme Category
“However, we are still in the learning phase. You get the Design Common
standard design rules, but often it’s still try and fail and methodology challenges
then redesign.”
“The very first fruitful way was that we sent designers to Design Implemented
user training.” training solutions
“Certification of components: This is a K.O. criterion for Certification Shared goals
us. If there is not something clearer there, I do not know
whether it [AM] goes on here.”
Among the challenges, we identified in our study, AM know-how and methods for
continuing education in particular were the most relevant topics for the majority of the
partners. While software and norms were the second and third largest concerns.
“Implemented solutions” summarize indication for approaches to solve the challenges
along the implementation of an AM mindset. Unsuccessful solutions have led to
remaining issues in the category of “Common challenges”. Across our interviewee
group we found a number of different approaches. Very common was to establish an
in-house expert team responsible to develop and hold workshops in different aspects of
AM. Those teams also managed internal databases for design guidelines and best-
practice projects. In summary, three categories of implemented solutions were identi-
fied: Collecting information through learning by doing, creation of own guidelines, and
discussion with experts in user groups. However, there is agreement among the par-
ticipant that the challenges still remain and the shared goals have not yet been reached.
A new type of design methodology has to be established. This methodology must be
adapted to AM. Furthermore, the new methodology needs support by international
standards and procedure for certification. Lastly, due to the complexity, software tool
must improve and be able to support decisions faced by designers.
Throughout the literature review study, it is elaborated that an AM-adjusted design

methodology requires not only AM design workflow, but also the key design guide-
lines and how effectively learn DfAM. Among the most important workflows are the
one from ASTM 52910 (Additive manufacturing—Design—Requirements, guidelines
and recommendations) [15] and workflow of approaches typically enabled by AM
(e.g.: topology optimization [16], cellular materials – lattice structures [17], monolithic
design – part consolidation [2], and function integration [18]). In the topic of design
guidelines, besides the major reference ASTM 52910, online available guidelines were
suggested along recent and updated AM design books (e.g.: A Practical Guide to
Design for Additive Manufacturing) [19]. Lastly, the most suitable learning approaches
for DfAM were presented, from lecture, through problem until project-based learning
[5, 20].
Regarding the implementation of standards, key norms, handbooks and guidelines
were covered. Among those are the already mentioned ASTM 52910 and the VDI 3405
– part 3 [21]. The most valuable contribution is the method which describes how to
implement those guidelines. In general, the AM industry currently lacks fundamental
principles for establishing derivative rules based on guidelines and best practices. To be
useful to designers, design guidance needs to consist of rules with numeric values
capturing the limitations of AM technologies, processes, and machines. The Guide-to-
Principle-to-Rule Approach offers a structured implementation framework from the
abstractness of design guidelines, through design principles, until the concreteness of
design rules [22].
Concerning the provision of better software, three phases throughout the product
design process were delimited: before, during and after design. Before design, some
solutions based on Artificial Intelligence (AI)/Augmented Intelligence were presented
as promising, e.g.: AI sketch-based design tool [23]. During design, the most estab-
lished solution was generative design, along with new coming approaches as real-time
generative-design (e.g. Autodesk and Desktop Metal Live Parts). Lastly after design,
printability checker and build simulation analysis tools are the most common used ones
(e.g. ANSYS).
4 Development and Validation of Framework
The broad literature review demonstrated that continuing education is key actor to raise
awareness, increase know-how and knowledge exchange. Moreover, a structured
approach was presented to extract, from high level guidelines presented in standards,
consistent and tangible rules for their wider adoption. Lastly, software works as a right
support to enhance human capabilities, acting as soon as possible in the design process.
Based on those findings, a holistic solution was proposed.
The solution takes into account two main important aspects: the product and the
individual. The first, represented here by the product design process, ranges from
conceptual to final design. The second, represented here by the design knowledge,
ranges from novice to expert. On one hand, the final design requires an expert level of
knowledge, on the other, novices perform better and more innovative than experts
during the conceptual phase of product design due to lack of fixation [24, 25]. The
solution is presented through the framework in Fig. 1.
Fig. 1. Framework with structured learning path, objectives and final output used in assistant for
design software.
This path from conceptual to final design optimally linking novice and expert
knowledge is assured by means of different learning methods, from lecture, via
problem and lastly project-based. Firstly, lecture-based learning explores what is
possible through AM, presenting some approaches and features to be used in this sense.
Problem-based learning aims the correlation design-material-process and their trade-off
to meet requirements (from a client, for example). Lastly, project-based learning
addresses how to quantify restrictions, being therefore able to optimize regarding
quality, cost and time.
Each learning phase covers, in different time frames – respectively days, weeks and
months – specific topics in order to guarantee a transition from potential of additive
manufacturing until its restrictions. The covered topics are design heuristics, guides,
principles and rules. Design Heuristics are cognitive shortcuts that help designers
explore variations in designs. Design Guides offer feature-based best practices when
using AM in product design to take advantage of AM capabilities. Design Principles
are basic, logical correlations capturing process parameter and control parameters. And
Design Rules are explicit value-based constraints that provide needed insight into
manufacturability. Those rules are subsequently stored in a knowledge repository
which works as database for a software support application.
In conclusion, this integrated solution uses continuing education to raise awareness
and improve knowledge exchange among professionals and uses a structured and
optimal path to get concrete design rules from the abstractness of norm and standards.
This proposed framework works therefore as a bridge between theory and practice in
how to design for AM. Lastly, a software support application makes the knowledge
developed during this learning process available for new design engineers and coming
products.
By means of a workshop, the above mentioned solution was presented to repre-

sentatives from interviewed companies in order to collect their industrial and business
perspectives. At the end of the workshop, a round table discussion took place and the
perspective of all participants was individually expressed and clarified. The proposed
approach was positively evaluated by all participants of the workshop. Numerous
valuable points came up from the discussion which are summarized below:
• Design for AM should cover not only the relations between material, properties and
3D printing process, but also post-processing;
• The proposed solution suits the niche of businesses with low product diversity and
mid-series production due to the highly needed internal efforts and costs;
• In order to be scalable, as shareable approach should be merged into the solution for
cost and risk sharing;
• OEMs play an important role in order to make this scalable solution possible. IP of
design is the main point of attention;
• The main objective is to identify the right moment, financially speaking, to quit the
learning track and use the shared knowledge base for desired design rules.
5 Conclusion and Outlook
The paper revealed that the interviewees have similar experiences regarding the con-
tinuing education of designers and other employees in AM. The design methodology,
norms & standards and design software have been identified as key areas to improve
AM adoption. The proposed solution consists of a structured framework to optimally
take advantage of company’s human capital and to extract tangible rules for 3D
printing an optimal part, fostering alongside knowledge transfer and creating aware-
ness. Lastly, a software application makes expert know-how more easily available with
fewer resources. This integrated solution successfully tackle the three key areas iden-
tified as vital for mindset shift towards AM: AM-adjusted design methodology,
implementation of standards and provision of better software.
However, our theory is grounded on a qualitative approach, thus cannot provide
any statistical evidence. Nonetheless, the researched phenomena of continuing edu-
cation for AM is no topic for quantitative research only. Therefore, we conclude that
our model contributes to the understanding of professional continuing education for
AM design, but can be extended by evaluating a larger group of employees on different
hierarchical levels and professional tenure.
As outlook, the future work comprises the development of an open innovation
platform for sharing of design rules. A first solution proposed is based on the auto-
matically extraction of design rules via on-premise software, subsequently encryption
and lastly upload to the cloud only after prior authorization of the respective IP owner.
The main issue to be investigated is how refractory will be the industry to share their
development in exchange of others.
References
1. Campbell, I., Diegel, O., Huff, R., Kowen, J., Wohlers, T.: Wohlers Report 2019: 3D
Printing and Additive Manufacturing State of the Industry. Wohlers Associates, Fort Collins
(2019)
2. Schmelzle, J., Kline, E.V., Dickman, C.J., Reutzel, E.W., Jones, G., Simpson, T.W.:
(Re) designing for part consolidation: understanding the challenges of metal additive
manufacturing. J. Mech. Des. 137(11), 111404 (2015)
3. Ahuja, B., Karg, M., Schmidt, M.: Additive manufacturing in production: challenges and
opportunities. In: Laser 3D Manufacturing II, California, vol. 9353. International Society for
Optics and Photonics (2015)
4. Geraedts, J., Doubrovski, E., Verlinden, J., Stellingwerff, M.: Three views on additive
manufacturing: business, research and education. In: Horváth, I., Albers, A., Behrendt, M.,
Rusák, Z. (eds.) Ninth International Symposium on Tools and Methods of Competitive
Engineering, Karlsruhe, pp. 1–15 (2012).
5. Williams, C.B., Seepersad, C.C.: Design for additive manufacturing curriculum: a problem-
and project-based approach. In: International Solid Freeform Fabrication Symposium,
Virginia, pp. 81–92 (2012)
6. Ford, P., Dean, L.: Additive manufacturing in product design education: out with the old and
in with the new? In: DS 76: Proceedings of E&PDE 2013, the 15th International Conference
on Engineering and Product Design Education, Dublin, pp. 611–616 (2013).
7. Loy, J.: The future for design education: preparing the design workforce for additive
manufacturing. Int. J. Rapid Manuf. 5, 199–212 (2015)
8. Pei, E., Minetola, P., Iuliano, L., Bassoli, E., Gatto, A.: Impact of additive manufacturing on
engineering education–evidence from Italy. Rapid Prototyp. J. 21(5), 535–555 (2015)
9. Simpson, T.W., Williams, C.B., Hripko, M.: Preparing industry for additive manufacturing
and its applications: summary & recommendations from a national science foundation
workshop. Addit. Manuf. 13, 166–178 (2017)
10. Prabhu, R., Miller, S.R., Simpson, T.W., Meisel, N.: The earlier the better? Investigating the
importance of timing on effectiveness of design for additive manufacturing education. In:
ASME 2018 International Design Engineering Technical Conferences and Computers and
Information in Engineering Conference, Quebec City, Canada (2018)
11. Watschke, H., Bavendiek, A.K., Giannakos, A., Vietor, T.: A methodical approach to
support ideation for additive manufacturing in design education. In: DS 87-5 Proceedings of
the 21st International Conference on Engineering Design (ICED 17): Vol 5: Design for X,
Design to X, Vancouver, Canada, pp. 41–50 (2017).
12. Tweed, A., Charmaz, K.: Grounded theory methods for mental health practitioners. In:
Harper, D., Thompson, A.R. (eds.) Qualitative research methods in mental health and
psychotherapy, pp. 131–146. Wiley-Blackwell, Oxford (2012)
13. Kitchenham, B.: Procedures for performing systematic reviews, Keele, UK (2004)
14. Gummesson, E.: All research is interpretive! J. Bus. Ind. Mark. 18, 482–492 (2003)
15. ISO/ASTM52910-18, Additive manufacturing—Design—Requirements, guidelines and
recommendations. ASTM International, West Conshohocken, PA (2018)
16. Gebisa, A.W., Lemu, H.G.: A case study on topology optimized design for additive
manufacturing. Mater. Sci. Eng. Conf. Ser. 276(1), 012026 (2017)
17. Merkt, S., Klocke, F.: Qualifizierung von generativ gefertigten Gitterstrukturen für
maßgeschneiderte Bauteilfunktionen, Aachen, Germany (2015)
18. Klaiber, D., Fröhlich, T., Vietor, T.: Strategies for function integration in engineering design:
from differential design to function adoption. Procedia CIRP 84, 599–604 (2019)
19. Diegel, O., Nordin, A., Motte, D.: A Practical Guide to Design for Additive Manufacturing.
Springer, Singapore (2020)
20. Yang, L.: Introducing the state-of-the-art additive manufacturing research in education. In:
Pei, E., Monzón, M., Bernard, A. (eds.) Additive Manufacturing: Developments in Training
and Education, pp. 53–65. Springer, Cham (2019)
21. VDI 3405 Blatt 3:2015-12: Additive Fertigungsverfahren - Konstruktionsempfehlungen für
die Bauteilfertigung mit Laser-Sintern und Laser-Strahlschmelzen (2015)
22. Mani, M., Witherell, P., Jee, H.: Design rules for additive manufacturing: a categorization.
In: ASME 2017 International Design Engineering Technical Conferences and Computers
and Information in Engineering Conference. American Society of Mechanical Engineers,
Digital Collection, Ohio (2017)
23. Wang, F., Kang, L., Li, Y.: Sketch-based 3D shape retrieval using Convolutional Neural
Networks. In: 2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR),
Boston, MA, pp. 1875–1883 (2015)
24. Kim, J., Ryu, H.: A design thinking rationality framework: framing and solving design
problems in early concept generation. Hum. Comput. Interact. 29, 516–553 (2014)
25. Crilly, N.: Fixation and creativity in concept development: the attitudes and practices of
expert designers. Des. Stud. 38, 54–91 (2015)
An Interactive, Fully Digital Design Workflow
for a Custom 3D Printed Facial Protection
Orthosis (Face Mask)
Neha Sharma1,2 , Dennis Welker2,3, Shuaishuai Cao1,2,

Barbara von Netzer4, Philipp Honigmann5 ,
and Florian Thieringer1,2(&)
1
Department of Oral and Cranio-Maxillofacial Surgery,
University Hospital Basel, Spitalstrasse 21, 4031 Basel, Switzerland
florian.thieringer@usb.ch
2
Medical Additive Manufacturing Research Group,
Department of Biomedical Engineering, University of Basel,
Gewerbestrasse 16, 4123 Allschwil, Switzerland
3
Department of Biomechanics, University of Applied Sciences,
Badstrasse 24, 77652 Offenburg, Germany
4
University Clinic of Oral and Maxillofacial Surgery,
Medical University of Innsbruck, Anichstrasse 35, 6020 Innsbruck, Austria
5
Hand Surgery, Cantonal Hospital Basel-Land,
Rheinstrasse 26, 4410 Liestal, Switzerland
Abstract. Sport-related injuries have an increased prevalence of maxillofacial

fractures among professional (soccer) players. From professional players’ per-
spective, these injuries can have career-detrimental effects when followed with
prolonged recovery periods. Therefore, to facilitate an earlier training and
competition return, and reduce the chances of re-injury, the use of face-
protective orthosis, commonly known as a face mask, in rehabilitative man-
agement is of paramount importance.
To date, the fabrication of a customized face mask has been an entirely manual
and time-consuming process. To mitigate the issues with conventional cus-
tomized face masks, the authors have presented a fully digital “contactless design
and production” workflow for the fabrication of a patient-specific face mask. This
work aimed to integrate the existing tools of medical image processing software,
computer-aided design (CAD), three-dimensional (3D) digitization, and additive
manufacturing (AM) to provide a cost-effective, practitioner/patient-friendly
solution for the design and manufacturing of patient-specific face-protective
orthosis or face masks. Considering the functional and clinical aspects at the
fractured site, a virtually designed face mask was fabricated in-house with carbon-
reinforced polylactic acid composite material via material extrusion or Fused
Deposition Modeling (FDM) technology. The face mask had a comfortable fit,
required no alterations, was lightweight, and shortened the convalescence period
for the patient. The results from the selected design case accurately represent the
clinical scenario and shows the potential of the proposed workflow in similar
facial fracture situations. With greater ease of fabrication and production validity,
this study highlights an alternative approach applicable in clinical practice.

https://doi.org/10.1007/978-3-030-54334-1_3
An Interactive, Fully Digital Design Workflow 27
Keywords: Face-protective orthosis Additive manufacturing Face

scanning Fused Deposition Modeling Composite thermoplastics
1 Introduction
Sport-related injuries result from a variety of different mechanisms and often vary in
location and pattern. These injuries account for 6 to 10% of maxillofacial traumas, with
the most common association to the fractures of nose followed by zygomatic bones [1,
2]. In general, sport-related injuries have an increased prevalence of facial fractures in
soccer and basketball players [3, 4]. From professional players’ perspectives, especially
in high contact sports, these injuries can have career-detrimental effects when followed
with prolonged convalescence periods. Face-protective orthoses, commonly known as
face masks, are patient-specific splints, that primarily protect the face and redistribute
the impact forces during sport activities [5]. Therefore, to facilitate an earlier training
and competition return, and reduce the chances of re-injury, the role of face-protective
orthosis in rehabilitative management is of paramount importance [6].
Face masks can be either prefabricated or customized (also known as patient-
specific). Over-the-counter or prefabricated face masks are commonly available in the
market and fit a range of athletes during sports activities. However, the use of prefab-
ricated face mask in the clinical treatment of athletes sustaining maxillofacial trauma is
limited [7, 8]. These devices do not provide an individualized fit to the patient and,
therefore, are considered inferior to customized face masks during the rehabilitation
phase. On the other hand, customized or patient-specific face masks are solely designed
for the individualized rehabilitative management of athletes with maxillofacial injuries
[9]. Until now, the fabrication of customized face masks has been an arduous task. This
fabrication method is an entirely manual, labor-intensive process comprising face
impression or moulage (negative), plaster mold (positive) fabrication, adaptation of
thermoplastic sheet onto the mold, and last, fine-tuning of the mask with cutting tools.
This conventional approach, although widely used among clinicians, is time-consuming
and unpleasant for the patient. Besides, this approach often requires frequent adjust-
ments compromising both the comfort and function of the face mask [5, 9–11].
To address the challenges mentioned above and increase patient compliance, the
integration of less invasive technologies into the clinical workflow is imperative.
Medical image processing software and additive manufacturing (AM) technology have
been applied in several medical applications, including the design and manufacturing
of medical splints for ankle-foot or wrist [12–14]. Furthermore, advancements in
digitization technologies such as three-dimensional (3D) face-scanning have made it
conceivable to generate digital models of surface topography of human face [15]. The
assimilation of these technologies contributes to patient-specific digital data in a con-
tactless manner, and has the potential to change the conventional workflow for cus-
tomized face mask. This work aimed to integrate the existing tools of medical image
processing software, computer-aided design (CAD), 3D digitization, and AM to pro-
vide a cost-effective solution for the design and manufacturing of patient-specific face-
protective orthosis or face masks. More specifically, a practitioner/patient-friendly
28 N. Sharma et al.
“contactless design and production” approach was devised that enabled the clinicians
with point-of-care manufacturing to fabricate a customized face mask for a patient
operated at a distant hospital.
2 Materials and Methods
In this section, an interactive digital design workflow for customized face-protective

orthosis or face mask is introduced for athletes with sports-related injuries of the
maxillofacial region. The entire workflow has been established using a procedural
methodology, including four phases, each involving several steps. An overview of the
schematic representation of the digital workflow is displayed in Fig. 1.
Fig. 1. An overview of the schematic representation of the digital workflow.
2.1 Clinical Case Fracture Treatment

The workflow started with an appropriate clinical case selection, referred to the
Department of Cranio-Maxillofacial Surgery, University Hospital Basel, to discuss the
treatment option for the fabrication of a patient-specific face mask. The patient was a
professional soccer player of the Austrian First League who suffered an injury to the
face during sport activity. He experienced a fracture of the right zygomaticomaxillary
complex (ZMC) region, which was treated by immediate open reduction and internal
rigid fixation. The digital imaging and communications in medicine (DICOM) dataset
from immediate postoperative cone-beam computed tomography (CBCT) provided by
the operating hospital were imported into Materialise Interactive Medical Image
Control System (MIMICS) medical software (MIMICS Innovation Suite v. 20.0,
Materialise, Leuven, Belgium). Following this, Hounsfield unit (HU), which expresses
the grayscale, was adjusted accordingly using the thresholding method. Subsequently, a
semiautomatic segmentation of the region of interest was performed, and the respective
bony and soft tissues 3D volumetric reconstructions were generated, which were
consequently exported and saved in a standard tessellation language (STL) file format.
The 3D volumetric reconstructions confirmed adequate reduction with three-point
fixation of the ZMC region with titanium miniplates at the right frontozygomatic
suture, right infraorbital, and right zygomatic buttress regions (Fig. 2).
Fig. 2. Postoperative CBCT 3D volumetric reconstructions A: Bony 3D volumetric recon-

struction showing titanium miniplates fixation at right frontozygomatic suture, infraorbital
region, and zygomatic buttress regions. B: 3D volumetric reconstructions of soft and bony
components with noticeable soft tissue swelling (right side).
2.2 Computational Image Data Acquisition and Virtual Model

Registration
An accurate visualization of anatomical bone and soft tissue components is an essential
step for the design of a patient-specific face mask. As the immediate postoperative
CBCT dataset does not corroborate with the soft tissue component because of post-
operative swelling, a 3D optical face scan (Vectra M3 3D Imaging system, Canfield
Scientific, New Jersey, USA) was scheduled after the swelling had subsided. Prior to
data acquisition, the optical scanner was calibrated following the manufacturer’s
instructions. After the scanning procedure was completed, the digitized surface
geometry of the face was transferred as polygonal STL (triangular mesh) containing
over 200,000 points and 474,180 triangle elements. The mesh generated was of high
quality, and no further post-processing procedures were required. To register the face’s
surface topography generated from a 3D optical scan dataset to the native anatomical
bony structures generated from the postoperative CBCT dataset, superimposition via a
best-fit alignment method was executed (3-matic medical v. 13.0, Materialise, Leuven,
Belgium). Using an iterative closest point (ICP) algorithm, surface registration protocol
(n-point and global registration) was accomplished between the 3D volumetric
reconstruction of CBCT soft tissue component and 3D optical face model (Fig. 3A,
3B). The registration protocol was based on the selection of similar anatomical points
on the healthy (non-operated) side of the face, unaffected by postoperative swelling.
Integrating 3D volumetric reconstructions of bony anatomical structures with a 3D
30 N. Sharma et al.
optical face scan allowed an accurate representation of the patient’s maxillofacial

region, which served as a reference for the digital designing of face mask (Fig. 3C).
Fig. 3. Illustration of the subsequent steps of the computational medical image registration
protocol. A: Selection of points on the 3D volumetric reconstruction of soft tissue component
from postoperative CBCT. B: Selection of points on the 3D optical face scan model. C: Profile
view of the patient after registration of a 3D optical face model on the 3D volumetric
reconstruction of the bony component from postoperative CBCT.
2.3 3D Modeling of Patient-Specific Face Mask in CAD Software

Following the computational medical image registration protocol, modeling of the face
mask was accomplished in an open-source CAD software (Meshmixer v.3.5.474,
Autodesk Inc.). Anatomical and functional requirements were taken into consideration
during this design phase. The maxillofacial skeleton has areas of strength (maxillofacial
buttresses or pillars) and areas of weakness (in-between walls). The components of the
buttress system in the exemplary case consist of vertical (zygomaticomaxillary) and
horizontal (frontal bar, infraorbital rim) buttresses. These regions have increased bone
thickness and act as a supporting base for the design of the face mask. The STL
(triangular mesh) file from the 3D optical face model was used for the digital modeling
of the face mask. Using Meshmixer’s select tool feature, the region of interest
(ROI) was highlighted using unwarp brush tool. A specific selection mode, limited to
symmetry, was turned on during this phase, to allow equivalent boundary extensions of
the face mask (Fig. 4A). The smooth boundary of the mask was then defined using the
expand mode filter for geodesic distance (Fig. 4B). Keeping the select tool on, the edit
functionality was used, and the mask surface was extracted. Next, using the offset tool,
a clearance of 2 mm was created. (Fig. 4C). This was necessary to have an adaptation
space between the face and inner surface of the face mask to prevent over-compression
of the mask on the patient’s face and also to allow sufficient space for the padded
lining. The extracted surface of the mask was separated from the 3D optical face model
STL. Next, the thickness of the mask was defined. Using the edit mode feature and
keeping the surface connected functionality on, the offset tool was used to extrude the
mask surface and modeled to a thickness of 3 mm (Fig. 4D). To refine the boundaries
of extruded mask, smooth boundary tool was used, keeping the shape-preserving
functionality on (Fig. 4E). Further refinement of the mask was accomplished using the
sculpt tool, limited to surface only functionality. This fine-tuning of the mask over-
laying the fractured regions was limited to the outer surface with a selective increase in
the thickness. Lastly, using Boolean subtraction tool, four rectangular retentive
grooves (dimensions: 5 mm 2 mm) were bilaterally designed onto the frontal and
zygoma region of the mask to secure the fastening band (Fig. 4F). These grooves are
for assembling and disassembling the face mask, making it adjustable during the
rehabilitation phase of the patient. The modeled mask file was at last exported in STL
file format.
Fig. 4. Illustration of the subsequent steps of 3D modeling workflow. A: Selection of the region
of interest (ROI), B: Defining smoothed boundaries, C: Creation of gap between mask and face,
D: Extraction of the mask’s surface to add thickness, E: Shape-preserving symmetrical boundary
refinement of extruded mask, F: Creation of rectangular pattern retention grooves.
2.4 Material Extrusion Additive Manufacturing and Post-processing

of Patient-Specific Face Mask
The STL file of the virtually designed face mask was imported into the slicing software
(MakerBot Print v. 4.5.0.1729, MakerBot Industries, USA) of a 3D printer. The face
mask was fabricated in PLA filament reinforced with short carbon fibers (in a weight
fraction of 30%) (Patona 1.75 mm Black Carbon Fiber PLA filament, Patona Inter-
national S.L.U, Germany) using a desktop Fused Deposition Modeling (FDM) 3D
printer (MakerBot Replicator+, MakerBot Industries, USA), with the following set-
tings: infill: 40%; layer height: 0.2 mm; shells: 3; nozzle diameter: 0.6 mm, extrusion
temperature: 220 °C. For optimal printing, the generation of both raft and support
structures were selected in the slicing software. After manual removal of support
structures, the surface of the face mask was smoothed using 1000 grit sandpaper, and
two layers of carbon-fiber fabric 200 g/m2 (HP-T200/120C, HP-Textiles, Schapen,
Germany) were adhered on the inner side of the mask for additional reinforcement.
32 N. Sharma et al.
In succession, to achieve a glossy smooth surface finish, the mask was coated with a
layer of an epoxy resin material (Epoxy Resin 4305, DD composite, Germany), and left
to dry. Custom graphics were later added for aesthetic purposes. Finally, a padded foam
lining was adhered to the undersurface of the face mask.
3 Results
Figure 5 shows the result of the described interactive fully digital workflow – from
computational 3D planning/designing, additive manufacturing to the realization of
patient-specific face mask in a professional soccer player. Due to the nature of the
maxillofacial fracture pattern, in this case, the customized face mask was fabricated as a
one-piece structure, which provided optimal protection to the operated site. Results
from subjective assessment by the patient were entirely satisfactory. According to these
results, the face mask had a comfortable snug-fit requiring no alterations, was impact-
resistant, and provided a significant earlier return to his athletic practice sessions. The
digitally contoured ocular apertures gave an unobstructed view during sport activity.
Overall, this in-house fabricated face mask was sturdy, lightweight (50 gm), and
aesthetically pleasing.
Fig. 5. Interactive digital workflow. A. 3D computer-aided design and planning B. FDM printed
carbon-reinforced PLA face mask C. A professional soccer player with a customized face mask
during his sport’s practice session.
4 Discussion
Recently, the use of face-protective orthosis or face masks by professional athletes has
significantly increased. While several over-the-counter options for face masks already
exist, these masks are not as efficient as custom-made masks due to diverse patterns of
maxillofacial injuries [7]. Patient-specific face masks are more comfortable and provide
a better fit. However, the traditional fabrication method for custom-made masks come
with some disadvantages, the most significant being the unpleasant fabrication process.
This conventional method is an entirely manual process, starting with a negative
impression (sometimes called moulage) of the face taken with materials such as plaster,
alginate, or silicon rubber. Next, a positive replica of the face is made from the
impression using plaster or dental stone. Subsequently, a thermoplastic sheet material is
heated, and together with the positive replica is placed in a vacuum former. Once set,
the sheet is removed from the mold and adjusted to the dimensions of the mask using a
heated knife or a grinding/cutting tool. Finally, holes are cut for the eyes, nostrils,
mouth as required, and retention grooves are made for straps. Although this conven-
tional process is widely used among clinicians, the process requires frequent iterations,
resulting in a time-consuming process where much material and prosthetic clinic time is
invested [8, 9]. However, such procedures can be outsourced to an external company,
but this fabrication process is expensive and requires long lead times. Therefore, to
increase efficiency from conventional methods and outsourcing, a novel in-house, fully
digital workflow for the design and manufacture of a patient-specific face mask was
devised.
Digitization technologies such as 3D face scanning and AM contributed to the
contactless production of patient-specific digital data, which correlates to anatomical
features [16, 17]. A device created from a patient’s data makes a bespoke orthosis,
which provides the best fit geometry. For a clinical fit patient-specific face mask, each
clinical case should be assessed individually. One of the essential aspects that need
consideration is adequate fixation of the fracture. A face mask designed over
improperly reduced fracture site can cause more damage with fracture-dislocation when
in contact with external forces. Several design criteria are also relevant to determine,
which anatomical structures such as fractured bones and soft tissue lacerations need
protection and which anatomical structures contribute to maximal support for the face
mask. In the present clinical case, bilateral supraorbital rims, glabella, and contralateral
non-injured zygoma region acted as supporting anatomical structures for the face mask.
An elevated surface that helps in the distribution of forces away from the operated
ZMC region protected the fractured region. Lateral and anterior field of view was
maintained by contouring and widening the ocular region. As an ill-fitting face mask
can cause discomfort to the patient, smooth boundary edges and padded lining were
added to prevent abrasion of the underlying skin from mild movement. This image-
based design workflow is minimally invasive, less stressful for the patient, and facil-
itates quick rehabilitation during the convalescence period by protecting the trauma-
tized anatomic region.
In this study, we chose the option of using a biodegradable composite thermoplastic
material, carbon-reinforced PLA, for the fabrication of a lightweight face mask. The
addition of reinforced materials (such as carbon fibers) to PLA to form a thermoplastic
composite, helps in matrix binding and transfers the load to the reinforcing fibers. This
results in carbon-reinforced composites with high strength-to-weight ratio, excellent
corrosion and wear resistance, and high dimensional stability [18]. In terms of utility
and performance of the face mask, it can be ascertained that the AM manufactured
carbon-reinforced PLA mask achieved a good balance between strength and weight
reduction. The subjective assessment by the patient validates the high durability and
friendly wear of the face mask. The complete workflow from planning to fabrication of
34 N. Sharma et al.
mask was accomplished in less than one day. This shows that using an in-house AM
setup not only results in a shorter overall turnaround time, but is also a relatively cost-
effective production solution. The cost of the proposed in-house digital workflow is
represented in a general scenario, for instance, the clinician already has the required
equipment for the scanning and printing process, and the overall cost is lower due to
less time and material consumption. Nevertheless, these cost-effective benefits should
be further evaluated where clinicians verify the proposed workflow with conventional
fabrication methods.
The medical application of AM is increasing with the potential integration of an
automatic design process. Some novel concepts for substitutes in medical splints for
orthopedic aids are already reported [13]. These digital solution platforms automatically
generate a design, which is digitally validated through finite element simulations. Once a
feasible structure is obtained, the design is manufactured. Such digital platforms can be
exploited and expanded in the field of face masks by establishing a digital process chain.
Although the proposed workflow is based on a patient with a fracture of the
zygoma region, this sequential methodology, with slight design alterations (for
example: the selection of specific ROI), can apply to other fractures of the maxillofacial
region that need protection during the convalescence periods. This digitized workflow
enables design freedom in a virtual environment, and various modifications can be
rapidly integrated before proceeding with manufacturing [13, 19]. The digitized
approach provides an easier means of reproducibility in case of lost/damaged mask.
Finally, incorporating digital technologies in a clinical environment allows the clinician
to customize the patient’s mask with minimal effort, increasing patient satisfaction, and
an improvement in treatment efficacy.
5 Conclusion
In this article, we presented a fully digital workflow that combined state-of-the-art

digitization technology with medical imaging and additive manufacturing to rapidly
fabricate a custom-made face-protective orthosis for rehabilitative management in
patients with sports-related maxillofacial injuries. To sum up, the exemplary case
demonstrates how the unique properties of point-of-care manufacturing can be
exploited in the field of orthoses and prostheses through the establishment of a digital
process chain. This workflow has allowed an improvement of some characteristics of
conventional custom-made masks as follows:
• Minimally invasive, contactless production method
• Decrease in lead times
• Easier reproducibility
• Cost-effectiveness
References
1. Adsett, L., Thomson, W.M., Kieser, J.A., Tong, D.C.: Patterns and trends in facial fractures
in New Zealand between 1999 and 2009. N. Z. Dent J. 109(4), 142–147 (2013)
2. Yamamoto, K., Matsusue, Y., Horta, S., Murakami, K., Sugiura, T., Kirita, T.: Clinical
analysis of midfacial fractures. Mater. Sociomed. 26(1), 21 (2014). https://doi.org/10.5455/
msm.2014.26.21-25
3. Hwang, K., You, S.H., Lee, H.S.: Outcome analysis of sports-related multiple facial
fractures. J. Craniofac. Surg. 20(3), 825–829 (2009). https://doi.org/10.1097/SCS.
0b013e3181a14cda
4. Echlin, P.S., Upshur, R.E., Peck, D.M., Skopelja, E.N.: Craniomaxillofacial injury in sport: a
review of prevention research. Br. J. Sports Med. 39(5), 254–263 (2005). https://doi.org/10.
1136/bjsm.2004.013128
5. Procacci, P., Ferrari, F., Bettini, G., Bissolotti, G., Trevisiol, L., Nocini, P.F.: Soccer-related
facial fractures: postoperative management with facial protective shields. J. Craniofac. Surg.
20(1), 15–20 (2009). https://doi.org/10.1097/SCS.0b013e3181909c55
6. Guyette, R.F.: Facial injuries in basketball players. Clin. Sports Med. 12(2), 247–264 (1993)
7. Gandy, J.R., Fossett, L., Wong, B.J.: Face masks and basketball: NCAA division I consumer
trends and a review of over-the-counter face masks. Laryngoscope 126(5), 1054–1060
(2016). https://doi.org/10.1002/lary.25584
8. Morita, R., Shimada, K., Kawakami, S.: Facial protection masks after fracture treatment of
the nasal bone to prevent re-injury in contact sports. J. Craniofac. Surg. 18(1), 143–145
(2007). https://doi.org/10.1097/01.scs.0000246729.23483.87
9. Kaplan, S., Driscoll, C.F., Singer, M.T.: Fabrication of a facial shield to prevent facial
injuries during sporting events: a clinical report. J. Prosthet. Dent. 84(4), 387–389 (2000).
https://doi.org/10.1067/mpr.2000.109480
10. Ghoseiri, K., Ghoseiri, G., Bavi, A., Ghoseiri, R.: Face-protective orthosis in sport-related
injuries. Prosthet. Orthot. Int. 37(4), 329–331 (2013). https://doi.org/10.1177/
0309364612463929
11. Haug, S.P., Haug, R.H.: Fabrication of a facial orthotic for protection of a fractured nose.
J. Oral Maxillofac. Surg. 50(7), 765–766 (1992). https://doi.org/10.1016/0278-2391(92)
90117-i
12. Chen, R.K., Jin, Y.A., Wensman, J., Shih, A.: Additive manufacturing of custom orthoses
and prostheses—a review. Addit. Manuf. 12, 77–89 (2016). https://doi.org/10.1016/j.addma.
2016.04.002
13. Lin, H., Shi, L., Wang, D.: A rapid and intelligent designing technique for patient-specific
and 3D-printed orthopedic cast. 3D Print. Med. 2(1), 1–10 (2016). https://doi.org/10.1186/
s41205-016-0007-7
14. Zhao, Y.J., Xiong, Y.X., Wang, Y.: Three-dimensional accuracy of facial scan for facial
deformities in clinics: a new evaluation method for facial scanner accuracy. PLoS ONE 12
(1), e0169402 (2017). https://doi.org/10.1371/journal.pone.0169402
15. Lin, J.T., Nagler, W.: Use of surface scanning for creation of transparent facial orthoses: a
report of two cases. Burns 29(6), 599–602 (2003). https://doi.org/10.1016/S0305-4179(03)
00080-9
16. Bibb, R., Freeman, P., Brown, R., Sugar, A., Evans, P., Bocca, A.: An investigation of three-
dimensional scanning of human body surfaces and its use in the design and manufacture of
prostheses. Proc. Inst. Mech. Eng H 214(6), 589–594 (2000). https://doi.org/10.1243/
0954411001535615
36 N. Sharma et al.
17. Berli, C., Thieringer, F.M., Sharma, N., Müller, J.A., Dedem, P., Fischer, J., Rohr, N.:
Comparing the mechanical properties of pressed, milled, and 3D-printed resins for occlusal
devices. J. Prosthet. Dent. (2020). https://doi.org/10.1016/j.prosdent.2019.10.024
18. Ning, F., Cong, W., Hu, Y., Wang, H.: Additive manufacturing of carbon fiber-reinforced
plastic composites using fused deposition modeling: effects of process parameters on tensile
properties. J. Compos. Mater. 51(4), 451–462 (2017). https://doi.org/10.1177/
0021998316646169
19. Cazón, A., Aizpurúa, J.E., Paterson, A., Bibb, R.J., Campbell, R.I.: Customised design and
manufacture of protective face masks combining a practitioner-friendly modelling approach
and low-cost devices for digitising and additive manufacturing. Virtual Phys. Prototyp. 9(4),
251–261 (2014). https://doi.org/10.1080/17452759.2014.958648
Opportunities of 3D Machine Learning
for Manufacturability Analysis
and Component Recognition
in the Additive Manufacturing Process
Chain
Tobias Nickchen1(B) , Gregor Engels1 , and Johannes Lohn2

1
Paderborn University, 33098 Paderborn, Germany
t.nickchen@sicp.uni-paderborn.de
2
Protiq GmbH, 32825 Blomberg, Germany
jlohn@protiq.com,
https://www.uni-paderborn.de/en/home/,
https://www.protiq.com/en/
Abstract. Additive Manufacturing (AM) is one of the manufacturing

processes with the highest potentials in the current transformation of the
industry. To make use of this potential and to achieve consistent product
quality at decreasing costs not only the 3D printers themselves but also
the whole process chain has to be automated. Due to the high degree of
digitalization and the use of 3D Computer Aided Design (CAD) models
within the entire process chain, it is possible to use these information
for automation via intelligent data analysis. In this paper, the poten-
tial of using 3D Machine Learning (ML) approaches for automation and
optimization of sub-processes of the process chain is analyzed. There-
fore, we consider the information flow of the 3D models in the process
chain of an AM service provider. The potential of using state-of-the-art
algorithms from the field of 3D ML for automation of sub-processes like
manufacturability analysis, production cost calculation or 3D-component
recognition is analyzed and feasibility is examined. For the sub-processes
of manufacturability analysis and 3D-component recognition prototype
solutions have been implemented and evaluated. For the production cost
calculation, only preliminary analyses were carried out, on the basis of
which the possible applications of 3D ML algorithms can be estimated.
With our analyses, we demonstrate that it is possible to further auto-
mate the process chain of AM service providers through the use of 3D
ML algorithms.
Keywords: Additive Manufacturing · 3D Machine Learning · Neural

Networks · Process chain automation

https://doi.org/10.1007/978-3-030-54334-1_4
38 T. Nickchen et al.
1 Introduction
Additive Manufacturing (AM) offers enormous potentials for the use of opti-
mized components in many highly technical industries. However, since it is still
a relatively young process on an industrial scale, many steps of the process
chain that go beyond actual production are characterized by manual work. Since
the beginning of the fourth technological revolution, industrial processes have
been iteratively optimized. The use of interconnected sensors and cyber-physical-
systems leads to intelligent self-adaptive production chains. This enables a more
efficient production and a reduction of costs while achieving higher quality [1,2].
Currently, conventional manufacturing processes cover the majority of industrial
production. Only 0.04% of global goods production is additively manufactured
[3]. According to the same study, however, 5% is quite within the realm of what
is possible in the future, if the AM industry takes advantage of its development
opportunities. In order to keep pace with common manufacturing processes such
as casting, forging or milling and to enable series production, production costs
must be further reduced, consistent quality ensured and the advantage of short
time to market further expanded. To achieve this, the potential of process chain
automation must be further exploited.
For the analysis of the process chain we worked together with an AM service
provider. As the service provider works with 3D printers from the Powder Bed
Fusion (PBF) field [4], we also focus on the process chain of this production
method. PBF processes for processing polymers and metals are widely used in
the AM industry. The actual AM process of the PBF 3D printers themselves is
already completely automated.
Fig. 1. Structure of PBF 3D printers [5].

Opportunities of 3D Machine Learning in the AM Process Chain 39
Nevertheless, large parts of the upstream and downstream processes are char-
acterized by manual work steps. In the context of our work, we focus on the
process steps which are directly linked to the information flow of the 3D mod-
els and identify the potentials of using 3D Machine Learning (ML) algorithms
for process automation. 3D ML is one of the evolving fields of ML and offers
enormous potentials for the analysis of 3D-data.
Our main contribution is the proof of concept that parts of the AM process
chain can be automated by using 3D ML techniques. For the steps manufactura-
bility analysis and 3D-component recognition, we already implemented evalua-
tion systems and proved the feasibility of these systems. For the process step of
production cost calculation, only preliminary analyses were carried out which
underline the potentials for this sub-process.
In Sect. 2, we describe the major parts of the process chain and analyze
which sub-processes are suitable for further automation. Afterwards we give an
overview of state-of-the-art 3D ML algorithms which can be used for our pur-
poses. In the following Sect. 4, the results of our evaluation studies are presented,
followed by a conclusion and the future perspectives.
2 AM process Chain Analysis
In this chapter, we take a deeper look on the AM process chain with its upstream
and downstream processes. For getting a detailed insight of all processes, we
worked together with an industrial AM service provider and focused on the
PBF process chain. All analyses in this paper refer to these processes. We first
describe the basic sub-steps of the process chain in Sect. 2.1 and subsequently
analyze the automation potentials of the different steps in Sect. 2.2.
The basic structure of PBF 3D printers can be seen in Fig. 1. The main
characteristic of PBF is the layer-wise application of powder and subsequent
selective melting of the powder using a laser with up to 200 W for polymers and
around 400 W for metals [6]. This specific approach leads to the necessity of
various additional steps, such as the manufacturability analysis, production cost
calculation or 3D-component recognition.
Since we deal with the analysis of the information flow in the process chain,
we only cover sub-processes that are directly related to the digital 3D models.
Of course, the process chain includes further steps besides those shown in Fig. 2.
However, these have been deliberately removed in our graphic, as they have
no direct relation to the digital information flow. Subsequently, we work out
sub-processes with automation potentials in Sect. 2.2.
2.1 The Process Chain of an AM Service Provider
The process chain of an AM service provider includes many up- and downstream
processes besides the actual production. We deal with all steps that take place in
the environment of the service provider and therefore start with the data upload.
The construction of the Computer Aided Design (CAD) models is on the side
Automated
on server
Data conversion
Manual Build job

Data upload 3D model repair preparation
analysis
Manufacturability
analysis
Production
cost calculation
Additive
Manufacturing
Quality Post- 3D-component

Shipping
assurance processing recognition
Fig. 2. Process steps in the AM process chain which are directly linked to 3D models.
of the customer and is therefore not included. An overview of the process steps
related to the 3D models is given in Fig. 2.
Data Upload. As mentioned before, the process chain begins with an upload
of 3D models by a customer. The data can be uploaded in nearly all common 3D
data types. Regardless of the uploaded data type, the 3D models are automati-
cally converted to the Standard Tessellation Language (STL) [7] file format. The
STL file format is the standard data format used for AM. All following steps are
based on the STL models.
Model Repair. The three subsequent steps 3D model repair, manufacturability

analysis and production cost calculation are already performed automatically on
a server. It is necessary to perform these steps automatically in near real time
so that potential customers can receive a quote for their uploaded 3D models
without delay. 3D model repair involves the checking and correction of errors
that may occur during the CAD construction or the conversion to STL. For that
task, software with ready-to-use functions can be used to repair STL models
completely automatically [8].
Manufacturability Analysis. The manufacturability analysis is necessary to

verify if an STL model with a given geometry can be manufactured with a
specific manufacturing technology. The manufacturability of an object is defined
by geometric features like overhangs, bores or channels and process parameters
like layer thickness or build orientation [9].
Only parts of these features can currently be checked automatically with soft-
ware tools. Problematic here is above all that the existing definitions of manu-
facturability can only be applied to existing models to a limited extent. Possible
guidelines for the manufacturability of 3D models were considered by various
researchers in the last years [10,11]. Most of these design guidelines are based
on standard geometries like cylinders or cuboids. To use these guidelines, 3D
models must be approximated by combinations of the standard geometries. The
guidelines can then be applied to the approximated 3D models. A major problem
is that complex 3D models can often only be approximated very imprecisely by
standard geometries. Therefore, only the guideline of minimum wall thickness is
currently automatically tested by a software tool since this feature is not based
on standard geometries. At the AM service provider, a more detailed check of
the 3D models is therefore performed manually by AM experts in the following
step.
Production Cost Calculation. The production cost calculation is the last

step which has to be performed automatically after the upload to enable real-
time pricing. Based on the 3D models geometry and operational parameters, the
production cost of 3D models has to be calculated. The costs of a component
are influenced by all processes in the process chain. From the preliminary anal-
yses to the manual build job preparation, the actual 3D printing process and
the various post-processing steps right through to shipping. Due to the great
variety of individual geometries, however, it is hard to exactly predict the costs
for some of these steps, e.g. the time necessary for the post-processing steps
like sandblasting or support removal. Owing to the complexity, the relationship
between the geometry of the 3D model and the amount of work required for
the post-processing steps cannot easily be described mathematically. Therefore,
the costs of these procedures can only be approximated with current software
solutions.
Build Job Preparation. The build job preparation is split into manual work
steps carried out by AM engineers work steps executed by software tools. Single
CAD models must be combined to build jobs. Depending on the requirements of
the final product, e.g. different build orientations of a 3D model in the construc-
tion space are necessary. Automated functions are in principle available for calcu-
lation of a good orientation and position of the 3D models in the build chamber
[8,12]. Nevertheless, especially the orientation has a strong influence on compo-
nent properties such as surface quality or mechanical characteristics. Therefore,
objects produced in automatically calculated orientations do not always meet the
specifications a customer expects. On the basis of interviews with the experts of
the AM service provider, it has become clear that especially the orientation of
the 3D models is currently still chosen manually by AM engineers in order to
guarantee the optimal quality of the components to be produced.
Additive Manufacturing. After the build job preparation, the physical pro-
cess steps begin. As mentioned in Sect. 1, the AM process itself is already com-
pletely automated. Depending on process and material, production is followed by
various technical finishing steps such as powder removal or sandblasting. These
processes run apart from the digital information flow. The physical production
and the virtual information flow converge again in the 3D-component recognition
step.
3D-Component Recognition. In PBF processes, different components are

manufactured together in one batch. Especially with the PBF 3D printers for
polymer processing, up to 100 different components are often produced in one
build job. After production, they must be assigned to the appropriate digital 3D
model again, in order to be able to continue with the subsequent post-processing
steps. This is still a manual process which is time consuming and expensive.
Post Processing. After the individual components have been separated again,
all digital information is available in order to be able to carry out post-processing
steps such as surface treatment and to subsequently check whether all customer
criteria have been met in the quality control process step. If this is the case,
the process chain can be completed with the dispatch of the components to the
respective customer.
2.2 Automation Potentials in the AM Process Chain
In this section, we analyze the optimization potential of the process steps

described in Sect. 2.1 to decide which process steps are suitable for automation
using 3D ML algorithms. The process steps manufacturability analysis, build
job preparation, 3D-component recognition and post-processing are potentially
of interest for optimization because they are characterized by manual work. Fur-
thermore, the step production cost calculation can be considered. Although this
step is automated, it still offers further potential for optimization to improve the
current approximation of the calculation of real production costs. In our work,
we do not consider build job preparation any further because we believe that the
expertise required in this step is difficult to replace with intelligent algorithms.

Additionally, the step post-processing will neither be considered in the context
of this paper, since intelligent data analysis is not relevant for that process.
Manufacturability Analysis. In Sect. 2.1, we explained that current manu-

facturability criteria are represented by design guidelines which are based on
standard geometries. With increasing complexity of components, the guidelines
are no longer sufficient to represent the real limitations of processability. Since it
is difficult to describe the restrictions by concrete mathematical representations,
we believe it is more goal-oriented to investigate the possibility of generating
conclusions about the manufacturability directly from existing production data.
Intelligent systems are capable of independently learning the features that are
decisive for manufacturability. Therefore, we see the potential of automation
with 3D ML algorithms for this process step.
According to AM experts of the AM service provider, errors occurring in
the production can be triggered by many different causes. Partly it is filigree
details in the 3D models, partly larger connected features that lead to faulty
production. Therefore, a solution for this application has to be able to capture
and process both fine details and global correlations within the 3D models.
Production Cost Calculation. As described in Sect. 2.1, it is difficult to

define the mathematical relationships between the geometry of a 3D model and
the actual costs for 3D printing itself and the pre- and post-processing steps.
The cost of a component depends on many different factors, such as the objects
volume and geometric complexity or the orientation in the build chamber. These
factors in turn influence direct cost factors such as construction time, material
consumption or the amount of work required for the various finishing steps such
as sand blasting and surface preparation or in metal processing the removal of
support structures. Our research and discussions with the AM service provider
have shown that in particular sand blasting, surface preparation and support
removal in metal processing are a major cost factor as they are mostly per-
formed manually. Due to the great geometric individuality of the 3D models, it
is difficult to calculate the exact amount of work required for the manual post-
processing steps. The amount of work and thus the costs are strongly dependent
on the geometric complexity. Complex geometries can contain many different
features like free-form surfaces, small openings or complex internal structures. It
is practically impossible to create a mathematical formula with hand-generated
features to calculate the correlation between these features and the costs for
post-processing.
At this point we see clear opportunities for optimization. Data-driven sys-
tems are able to learn to extract the relevant features. The big advantage of
Deep Learning (DL) models is that they are able to generate so called Deep
Features from raw inputs on their own. This enables them to learn the complex
mathematical relationships that exist here on the basis of their own automati-
cally generated features. By collecting data such as the manual processing time
for post-processing of produced objects during real production, a 3D ML system

can then learn the relationship between the geometry and the processing time
and thus the processing costs.
By determining the costs for individual post-processing steps more precisely,
the total costs for an object can be significantly improved compared to the
estimation currently in use.
3D-Component Recognition. For manually assigning physical components

to digital 3D models, an employee visually compares the properties of the objects
with the 3D models. Due to the increasing number of produced objects, manual
sorting is today already connected with an extremely high effort. Therefore, we
see a high potential for cost savings through an automated solution. Since in
AM completely new components have to be detected in real time every day,
conventional computer vision approaches reach their limits. In contrast, 3D ML
applications can adapt to the daily changing components. During our work on
this topic a first commercial approach recently emerged for this task (AM Flow
[13]). According to their data, they achieve a detection rate of 95%. To the
best of our knowledge, their system has never been evaluated on a standardized,
publicly available data set. In our opinion it is necessary to achieve a detection
rate of almost 100% for complete automation. Therefore, a further optimization
with 3D ML is possible for the step of component recognition.
3 3D Machine Learning
For the process steps of manufacturability analysis, production cost calcula-

tion and 3D-component recognition explained in Sect. 2.2, we have examined
the possibility of using 3D ML approaches to automate that steps. To find the
most suitable solution for our task, we compared several state-of-the-art 3D ML
approaches.
Existing 3D ML algorithms for popular tasks like 3D object recognition,
segmentation or tracking are using either directly 3D data or 2D projections
like images. Both variants contain different positive and negative aspects. By
sensors like 3D scanners, 3D data in the form of point clouds, meshes or Red,
Green, Blue, Depth (RGB-D) data can be generated. 3D representations offer
the advantage of displaying the scanned data in great detail but have the dis-
advantage of being very memory intensive and therefore place high demands on
the processing hard- and software. On the other hand, 2D projections in the
form of images offer the advantage of being less memory intensive but have the
disadvantage of information loss.
In order to find the best solution, we figured out the positive and negative
aspects of the different approaches and decided which approach is most useful
for our purposes.
3.1 Image-Based
Image-based approaches like RotationNet [14] or Group-view Convolutional Neu-
ral Network (GVCNN) [15] have the advantage of less memory consumption,
faster training times and lower hardware requirements compared to approaches
using 3D data. For frequently used benchmark tasks such as the classification of
the Princeton ModelNet40 dataset [16], they achieve classification rates of up to
97% which is state of the art and exceeds 3D-data-based approaches.
Fig. 3. Data representation of RotationNet [14]
In most state-of-the-art image-based approaches the basic drawback of losing

information about the 3D objects by using 2D projections is counteracted by
using multi-view representations of the 3D models. For each 3D object of the
data set, multiple images are generated by using different view points around
the 3D model (see Fig. 3). These groups of images are then used as collective
input data for the model, giving more information to the model what leads
to the state-of-the-art results in classification of ModelNet40 [14,15]. However,
differentiating 3D models from ModelNet40 differs from our issue because filigree
details are relatively irrelevant. Therefore, it must be proven that the image-
based approaches are also capable of differentiating very similar objects.
3.2 3D-Data-Based
Approaches based on 3D data can be assigned to the two sub-classes point-cloud-
based and voxel-based. Voxels are the three-dimensional equivalent of pixels in
two-dimensional space.
Point-Cloud-Based. Point-cloud-based approaches like PointNet++, Relation-

shape Convolutional Neural Network (CNN) [17,18] or Linked Dynamic Graph
CNN [19] directly work on 3D point clouds which can be generated by different
types of sensors (Fig. 4).
Fig. 4. Data representation of relation-shape CNN [18]
Since point clouds are unstructured data sets with possibly varying point
densities in different parts of the cloud, convolutional filters which are often used
for image processing can not be used. For being capable of handling the varying
resolutions, point-cloud-based approaches learn features in multiple hierarchical
scales. However, the possibility to capture filigree and global features at the same
time can only be realized in theory because it is limited by the size of Graphics
Processing Units (GPUs). In practice, a maximum of 5000 points is usually used
which is not sufficient to display all details of complex objects. For the analysis
of fine details of 3D models, point-cloud-based approaches reach their limits.
Voxel-Based. Similar to point clouds, so called voxels capture three dimen-

sional information. The difference to point clouds is that the data is stored
structured in a three-dimensional grid. Therefore, common mathematical oper-
ators like convolution used in Convolutional Neural Networks (CNNs) can be
applied on the 3D data. This enables the same behaviour of voxel-based systems
in the three-dimensional as the conventional CNNs in the two-dimensional space.
The main drawback similar to point-cloud-based approaches is the high usage of
computational memory with growing resolution of the voxel grid. Developers of
the approaches VoxNet, OctNet [20,21] or Spatial-hashing-based Convolutional
Neural Network (HCNN) [22] have thus tried to optimize the data structure
to enable a high resolution. Wang et al. [22] represent the state of the art and
reach a resolution of 5123 voxel using high end GPUs. Therefore, voxel-based
approaches are best suited to capture filigree details in 3D models (Fig. 5).
3.3 Approach Selection
Based on the requirements of the process steps manufacturability analysis, pro-

duction cost calculation and 3D-component recognition and the characteristics
of the different types of approaches, the following potentials for process automa-
tion arise. Since both local and global features must be considered in the sub-
processes of manufacturability analysis and production cost calculation, the use
Fig. 5. Data representation of HCNN [22]
of a voxel-based approach is reasonable for these sub-processes. The longer train-

ing time is negligible, since no recurring training process is necessary. For 3D-
component recognition all described types of algorithms offer positive arguments
which argue for a use. Therefore, we evaluate all described algorithms for that
task.
4 Evaluation
In order to prove that the described algorithms are suitable for solving the tasks
of manufacturability analysis, production cost calculation and 3D-component
recognition, an evaluation must be carried out. To the best of our knowledge
there are currently no benchmark data sets available for the problems we are
investigating. Therefore, we generated our own data sets based on the Thingi10K
data set [23] which contains 10000 3D models from the AM domain and trained
the algorithms with subsets of that data. The subsets were adapted to the respec-
tive problem definition and are oriented towards real production data.
4.1 Manufacturability Analysis

To verify the basic usability of the HCNN approach for manufacturability anal-
ysis, we had to generate a labeled data set. Since manual labeling of thousands
of 3D models is extremely time-consuming, we decided to use the wall thickness
tool which is currently used at the AM service provider. With the help of the
tool and the Thingi10K data set, a labeled data set with about 5000 3D models
has been created where the 3D models are divided into the categories “manufac-
turable” and “non-manufacturable” with respect to the feature “minimum wall
thickness”. Labeling this data set, enables us to verify the ability to recognize
filigree details in 3D models. Some example 3D models can be seen in Fig. 6.
This data set has been used to train the HCNN model which is presented in
Sect. 3.2 with a resolution of 5123 voxels. The HCNN model is able to achieve
an accuracy of 94% correctly classified 3D models in first studies. By applying
Fig. 6. Example 3D models with colorization of thin areas. Red areas are beneath a
threshold value of 0.5 mm.
parameter tuning, this value can be increased even further. The studies have
shown that a voxel-based model has the ability to learn high-resolution features
from 3D models. Especially, it has shown that it is possible to recognize the wall
thickness feature despite strong variance in the exact expression of the feature.
Therefore, the next step of our work will be to back up the generated results
with further data sets. We want to use the data generated during the produc-
tion of the AM service provider to generate additional data sets. With that
information we can create labeled data sets regarding other geometric features
like overhangs, bore holes or channels described by Bikas et al. [9].
4.2 3D-Component Recognition
For 3D-component recognition six different evaluation data sets have been gen-
erated based on the Thingi10K data set in order to evaluate the performance
of the described algorithms for the recognition of physical components. These
data sets contain between 10 and 100 different 3D models and were composed
in a way that they represent typical build jobs of the currently common indus-
trial PBF 3D printers for polymer processing. Three data sets contain random
compositions of 3D models from the AM domain, the other three explicitly very
similar 3D models (Fig. 7). The data sets are available for public download in
order to offer other researchers the possibility of comparison1 .
To prove that the approaches we are using are applicable for a real applica-
tion, we need to adapt it to a possible recognition station setup. After printing,
the objects will be separated on a conveyor belt and passed into a scanning area
consisting of multiple 2D- or 3D-sensors installed in elevated view-points like
1
https://gpstd.s3.eu-central-1.amazonaws.com/Raw datasets.zip.
shown inf Fig. 3. The biggest unknown in this system is the orientation in which
a component is sensed by the sensors. Therefore, the creation of training data
for the 3D ML algorithms is a deciding factor.
Depending on the algorithm, virtual images, point clouds or voxel represen-
tations have been generated from the 3D models, which have then been used
as training data for the systems. For the creation of these input representations
we implemented an algorithm which generates physically sound training data.
This means that only virtual sensor views are used, which can also occur in the
described setup with real sensors. Our evaluation has shown that this method for
training data generation has a major influence compared to randomly generated
training data.
Fig. 7. Example of 3D models with high similarity.
The data sets have been used to train the approaches [14,15,17–19,22] pre-
sented in Sect. 3.1. For the data sets with up to 30 different randomly chosen 3D
models all approaches reached an accuracy of more than 99%. For higher num-
bers of different 3D models the accuracy of the image-based approaches decreases
to values around 93%. The accuracy of the point-cloud-based approaches can still
reach 96% and the voxel-based approach is close to 99%. This confirms the expec-
tations that image-based approaches in particular reach their limits, especially
when there are strong similarities between individual objects.
The evaluation experiments have shown that Deep Neural Network (DNN)-
based approaches are able to recognize physical objects after production with
high accuracy. Especially the generation of physically sound training data brings
us one step closer to the goal of a 100% recognition rate and thus to the possibility
of complete automation. In the next step, we will evaluate the approaches in the
production process of the service provider to move from artificially generated
data sets to real production data.
5 Conclusion
In the context of our work we have shown that the process chain of AM offers
various aspects that can be optimized with the use of 3D ML algorithms. The
studies have shown that state-of-the-art 3D ML algorithms are capable of analyz-
ing 3D models in high detail with regard to various issues. For the sub-processes
of manufacturability analysis and 3D-component recognition we implemented
prototypes which have shown that these two tasks can be optimized using the
proposed 3D ML algorithms. For the production cost calculation we have car-
ried out preliminary studies. These have shown that using 3D ML algorithms
for a more precise calculation of the production costs is promising. In our future
work we are going to verify this assumption by using the algorithms with real
production data.
Our results show that there are different potentials for further automating
the process chain of AM. In our research we will continue to build on the studies
described in this paper and apply the applications in real production.
Acknowledgements. We gratefully acknowledge the funding of this project by com-

puting time provided by the Paderborn Center for Parallel Computing (PC2). A special
thanks belongs to the Phoenix Contact Foundation for supporting this research project.
References
1. Rüßmann, M., Lorenz, M., Gerbert, P., Waldner, M., Justus, J., Engel, P., Har-
nisch, M.: Industry 4.0: the future of productivity and growth in manufacturing
industries. Boston Consult. Group 9(1), 54–89 (2015)
2. Liao, Y., Deschamps, F., Loures, E.D.F.R., Ramos, L.F.P.: Past, present and future
of industry 4.0-a systematic literature review and research agenda proposal. Int.
J. Prod. Res. 55(12), 3609–3629 (2017)
3. Caffrey, T., Wohlers, T., Campbell, R.: Executive summary of the Wohlers report
(2016)
4. King, W.E., Anderson, A.T., Ferencz, R.M., Hodge, N.E., Kamath, C., Khairallah,
S.A., Rubenchik, A.M.: Laser powder bed fusion additive manufacturing of metals;
physics, computational, and materials challenges. Appl. Phys. Rev. 2(4), 041304
(2015)
5. Woesz, A.: Rapid prototyping to produce porous scaffolds with controlled archi-
tecture for possible use in bone tissue engineering. In: Virtual Prototyping & Bio
Manufacturing in Medical Applications, pp. 171–206. Springer (2008)
6. Bikas, H., Stavropoulos, P., Chryssolouris, G.: Additive manufacturing methods
and modelling approaches: a critical review. Int. J. Adv. Manuf. Technol. 83(1–4),
389–405 (2016)
7. Ciobota, N.D.: Standard tessellation language in rapid prototyping technology.
National Institute of Research and Development for Mechatronics and Measure-
ment Technique, Bucuresti, The Scientific Bulletin of Valahia University–Materials
and Mechanics, vol. 7 (2012)
8. Materialise Magics Sinter Module. https://www.materialise.com/en/software/
magics/modules. Accessed 22 Apr 2020
9. Bikas, H., Lianos, A., Stavropoulos, P.: A design framework for additive manufac-
turing. Int. J. Adv. Manuf. Technol. 103(9–12), 3769–3783 (2019)
10. Adam, G.A., Zimmer, D.: Design for additive manufacturing–element transitions
and aggregated structures. CIRP J. Manuf. Sci. Technol. 7(1), 20–28 (2014)
11. Rudolph, J.P., Emmelmann, C.: Analysis of design guidelines for automated order
acceptance in additive manufacturing. Procedia CIRP 60, 187–192 (2017)
12. Zwier, M.P., Wits, W.W.: Design for additive manufacturing: automated build
orientation selection and optimization. Procedia CIRP 55, 128–133 (2016)
13. Am-Vision: 3D-Part Recognition. https://am-flow.com/vision/. Accessed 19 Feb
2020
14. Kanezaki, A., Matsushita, Y., Nishida, Y.: RotationNet: joint object categorization
and pose estimation using multiviews from unsupervised viewpoints. In: Proceed-
ings of the IEEE Conference on Computer Vision and Pattern Recognition, pp.
5010–5019 (2018)
15. Feng, Y., Zhang, Z., Zhao, X., Ji, R., Gao, Y.: GVCNN: group-view convolutional
neural networks for 3D shape recognition. In: Proceedings of the IEEE Conference
on Computer Vision and Pattern Recognition, pp. 264–272 (2018)
16. Wu, Z., Song, S., Khosla, A., Yu, F., Zhang, L., Tang, X., Xiao, J.: 3D shapenets: a
deep representation for volumetric shapes. In: Proceedings of the IEEE Conference
on Computer Vision and Pattern Recognition, pp. 1912–1920 (2015)
17. Qi, C.R., Yi, L., Su, H., Guibas, L.J.: Pointnet++: deep hierarchical feature learn-
ing on point sets in a metric space. In: Advances in Neural Information Processing
Systems, pp. 5099–5108 (2017)
18. Liu, Y., Fan, B., Xiang, S., Pan, C.: Relation-shape convolutional neural network
for point cloud analysis. In: Proceedings of the IEEE Conference on Computer
Vision and Pattern Recognition, pp. 8895–8904 (2019)
19. Zhang, K., Hao, M., Wang, J., de Silva, C.W., Fu, C.: Linked dynamic graph
CNN: learning on point cloud via linking hierarchical features. arXiv preprint
arXiv:1904.10014 (2019)
20. Maturana, D., Scherer, S.: VoxNet: a 3D convolutional neural network for real-
time object recognition. In: 2015 IEEE/RSJ International Conference on Intelligent
Robots and Systems (IROS), pp. 922–928. IEEE (2015)
21. Riegler, G., Osman Ulusoy, A., Geiger, A.: OctNet: learning deep 3D representa-
tions at high resolutions. In: Proceedings of the IEEE Conference on Computer
Vision and Pattern Recognition, pp. 3577–3586 (2017)
22. Shao, T., Yang, Y., Weng, Y., Hou, Q., Zhou, K.: H-CNN: spatial hashing based
CNN for 3D shape analysis. IEEE Trans. Vis. Comput. Graph. 26, 2403–2416
(2018)
23. Zhou, Q., Jacobson, A.: Thingi10K: a dataset of 10,000 3D-printing models. arXiv
preprint arXiv:1605.04797 (2016)
Review on the Design Approaches of Cellular
Architectures Produced by Additive
Manufacturing
Marco Pelanconi(&) and Alberto Ortona
Mechanical Engineering and Materials Technology Institute (MEMTi),

University of Applied Sciences (SUPSI),
Via Cantonale 2C, 6928 Manno, Switzerland
marco.pelanconi@supsi.ch
Abstract. The advent of additive manufacturing (AM) has allowed conceiving

components by their function and no longer by their manufacture. This benefit
allows improving the components’ performances and the fabrication of geo-
metrically complex parts such as cellular structures. The design method is
fundamental for the layout of these components. In this work, we present the
design approaches of regular and irregular strut-based and triply periodic min-
imal surface-based structures. We propose a novel design method of multi-
functional cellular architectures enabling to generate structures with
morphological variations that provide a component with different features and
functionalities in its own volume, depending on its requirements. The final
chapter present the study of a bio inspired cellular material, from the design to
experimental testing through manufacture and simulation. The structure consists
of an ultra-lightweight body made up of gyroid cells and reinforced with carbon
fiber bars. This concept was used to design an innovative helmet, which aims at
combining the lightness and impact resistance.
Keywords: Additive manufacturing Lattice design Helmet
1 Introduction
Porous cellular materials are used in many different engineering industrial fields, such
as high temperature application, catalyst, protection systems, weight saving applica-
tions, thermal storage, composites [1–5]. They are employed exploiting their properties
basically related to the material (polymeric, metallic or ceramic) and to the morphol-
ogy. Additive Manufacturing (AM) made it possible to produce these structures that
previously could not be fabricated with conventional manufacturing methods. AM
allows to fabricate metal, plastic and ceramic objects starting from a 3D computer-
aided drafting (CAD) file [6]. It has opened the doors to the generation of more and
more complex CAD models, leading to the need for more advanced generation tools.
Cellular porous architectures are very complex in their morphology and they need
special design tools. A proper design of the morphology can result in structures with
optimized properties for specific applications [7–9]. Such structures contain large
numbers of geometrical details, which are impossible to generate with standard CAD
https://doi.org/10.1007/978-3-030-54334-1_5
Review on the Design Approaches of Cellular Architectures Produced by AM 53
packages [10]. In general, “classic” CAD tools are not able to perform quick and
efficient design of lattice structures. Ashby and Gibson [11] introduced a simplified
model of foams. Each cell was assumed as “cube like”. Their idea of using cubic lattice
structures was introduced to explain the behavior of foams, with analytical models, in
terms of pressure drop, heat and mass transfer, and stiffness [12–14]. Compared to
random foams, lattice structures are regular and reproducible. They offer more design
freedom, which results in structures with enhanced properties and novel functionalities
[15]. In this paper, we report computational design approaches in different dedicated
software environments, using a combination of purpose-built algorithms and scripts.
The study has focused on regular and irregular strut-based lattices and triply periodic
minimal surface-based structures. A novel design method of multifunctional cellular
architectures is presented (Fig. 1).
Fig. 1. Design solutions for cellular architectures produced by additive manufacturing: a)

irregular strut-based structures; b) regular strut-based structures; c) multifunctional cellular
architectures; d) triply periodic minimal surface based-structures.
2 Irregular Strut-Based Structures
An irregular strut-based structure consists of a random arrangement of cells composed

by struts, which are connected one to another by nodes. This means that the archi-
tecture is not periodic and the cells are geometrically different one from another.
54 M. Pelanconi and A. Ortona
2.1 Random Foam Design

Foams exhibit scattered properties due to their random and non-periodic structure:
defining their behavior is very difficult. One way is to study experimentally their
properties [16] and the other is to simulate computationally [17], but this requires a
CAD model of the foam. A Matlab (MathWorks. Natick, Massachusetts, USA) code
was developed for the generation of random foams made up of struts elements. The
script uses two purpose-built .TXT files, one with points and the other with connection,
which generate a large random foam domain. Setting the size of the sample as input,
the algorithm randomly chose a region inside this domain, cutting a random foam with
the specified dimensions. The result is an array of point and lines [18] that can be scaled
in order to reach a certain dimension of the pores. The array is subsequently converted
into cylinders and spheres of a specified diameter. The code does not allow any other
type of control on the geometry. Ceramic foams are widely employed in high tem-
perature applications thanks to their outstanding properties. We designed and produced
Si-SiC open cell foams as active zone in porous burners for heat radiation application
[19], catalyst carriers and solar radiation absorber [2] (Fig. 2).
Fig. 2. Random foam design
2.2 Voronoi-Based Design

There are only few parameters that can be varied in order to engineer the properties of a
foam. The need to represent a random foam, instead of periodic arrangement of the
regular lattices, with a random engineerable structure has led to the development of a
new algorithm based on the Voronoi 3D tessellation [20, 21]. This method consists in
the decomposition of the space determined by the distances from a given discrete set of
random points. For each point there is a corresponding region consisting of all points
closer to that point than to any other. These regions are called Voronoi cells [22].
The edges of each cell are subsequently converted into solid struts. The code, realized
into Rhinoceros (McNeel, Seattle, Washington, USA) using the Grasshopper plug-in,
allows the generation of Voronoi-based lattices of any shape (Fig. 3) with a
parametrization of the following geometrical quantities: mean cells size, struts diam-
eter, gradient for struts diameter, sample size. The code was further developed allowing
the implementation of a cells size gradient along one or more axis and more directions:
the cells size changes starting from one value and arriving at another, even with more
variations. In this way, it is possible to generate a lattice with different porosities in
different regions of its volume.
Fig. 3. Voronoi-based design
3 Regular Strut-Based Structures
A regular strut-based structure consists of a periodic arrangement of cells composed by

struts, which are connected one to another by nodes. The architecture construction
takes place from a unit cell that is replicated in the three-dimensional space until
forming a sample of the desired overall dimensions.
3.1 Structured Lattice Design

A parametric algorithm code [23, 24] was developed using Matlab, which generates
structured lattices made up of struts elements. The script allows varying the following
parameters: cells type, cell size, struts diameter, elongation angle [25] and XYZ cell
replication (sample overall size). The code allows generating several types of cells [4,
5, 15]: octet, modified octet, Weaire-Phelan, Tetrakaidecahedron, hexagon, modified
hexagon, diamond, straight cube and rotated cube. The selected cell is replicated in the
space until forming the desired sample dimension. With Boolean tools in then possible
to crop the lattice into the desired shape. The struts, consisting of cylinders and spheres,
are then exported in .STEP file into the commercial computer-aided design software
Siemens NX (Siemens. Munich, Germany) and joined in a single solid body.
This code was implemented for the fabrication of a: (i) Si-SiC periodic architecture
for catalytic supports (based on rotated cube cells) within the FP7 BioRobur European
research project [26, 27]; (ii) Tubular Si-infiltrated SiCf/SiCm composites with a lattice
structure inside were fabricated for concentrated solar receiver applications [28, 29];
(iii) SiSiC porous burners [19]; (iv) ceramic automotive catalyst substrates [30];
(v) Al2O3 tubular high temperature heat exchanger with a rotated cube lattice inside
[31] (Fig. 4).
Fig. 4. Structured lattice design
The numerical tool was further developed by using the visual programming lan-
guage Grasshopper that runs within the Rhinoceros 3D CAD software. This new
algorithm allows the generation of lattices structures of any shape, with a quickly
parametrization of the geometrical quantities: cells size, struts diameter, gradient for
struts diameter, sample size. The following types of cell can be generated: x, star, cross,
tesseract, vintiles, diamond, honeycomb, Tetrakaidecahedron, Weaire-Phelan, octet,
modified octet, hexagon, modified hexagon, straight cube and rotated cube. This code
can be further developed for any type of cell. Figure 5 shows a structured lattice design
with a gradient for the strut’s diameter, created using Grasshopper. The advantage of
this tool is the possibility to convert the body into a 3D triangular mesh using Cocoon
add-on: the output is a STL file that can be processed immediately for 3D printing.
3.2 Unstructured Lattice Design

Unstructured lattices are preferred with respect to structured ones. This is due to the
possibility to have components with heterogeneities such as variable cell size and
orientation. A 3D Matlab tools was developed in order to produce unstructured lattices.
This method starts with a 3D CAD model of the bulk volume to fill with lattice cells.
This code needs a cad-mesh with hexahedron elements as input and converts all the
elements in any type of unit cell, which must fit into the hexahedron. The script allows
varying the cells size and the struts diameter. Moreover, it is possible to perform a
Fig. 5. Structured lattice design with Grasshopper
distortion of the cells that allows filling at best the complex volumes. The code was
implemented to produce an Innovative Thermal Management Concepts for Thermal
Protection of Future Space Vehicles (THOR project) [32] with a particular shape.
Figure 6 shows a simple example of an unstructured lattice design.
Fig. 6. Unstructured lattice design: a) CAD model of a bulk volume; b) hexahedral mesh with a
linear degrowth rate of 0.9; c) CAD model with tetrakaidecahedron unit-cells.
4 Multifunctional Cellular Architectures Design
Additive manufacturing allows producing very complex geometry. This means that
different components, which perform different functions, can be fabricated together in
one piece having the same features of them. In this field, we developed a novel design
method of multifunctional cellular architectures enabling to generate structures with
morphological variations that provide a component with different features and func-
tionalities in its own volume, depending on its requirements. The purpose-built
Grasshopper algorithm allows generating structures with different cell types in the same
volume, therefore there is a gradient between a cell type and another, made up of
Voronoi 3D tessellation. The flexibility of Voronoi tessellation can be employed to join
different structures. Figure 7 shows a 2D representation of a multi-lattice produced with
this approach.
Fig. 7. Multifunctional cellular architectures design: a) quad lattice; b) hexagonal lattice; c)

rotated quad lattice; d) multi-lattice structure consisting of quad, hexagonal and rotated cube
cells.
5 Triply Periodic Minimal Surface Design
Additive Cellular structures made up of struts elements are mainly employed for their
high porosity, flow [33] and thermal properties [34, 35]. In applications where the
objective is to maximize the surface area of a component (for example catalytic sub-
strates), these structures are not the best solution. A design approach, based on triply
periodic minimal surface (TPMS) [36], was performed in order to produce high-surface
area structures. A TPMS consists of a single, continuous, smooth and periodic
arrangement of surfaces. The architecture construction takes place from a single surface
that is replicated in the three-dimensional space until forming a sample of the desired
overall dimensions. The surface divides space into two interwoven domains.
The TPMS is then converted into a lattice using two different approaches [37]: (i) Strut-
based structure: one domain is filled with solid material and the other is left empty
(void domain). The resulting structure is a lattice made up of struts, i.e. as the previous
presented structures but with a TPMS unit cell; (ii) Sheet-based structure: the surface is
thickened of a desired value, forming two separate empty domains, which are infinite
and intertwined, but not interconnected. The resulting structure is a warped sheet with
the constant thickness everywhere.
A 3D numerical tool was developed in Grasshopper. It allows the generation of
TPMS-based structures of any shape, with a parametrization of the following geomet-
rical quantities: cells size, struts diameter or surface thickness, sample size. The algo-
rithm was employed for the realization of novel cylindrical catalytic substrates [38]
produced by directly 3D printing of alumina powders [39]. Five different structures were
evaluated in order understand their mechanical and fluid-dynamic behavior (Fig. 8).
Fig. 8. TPMS design
6 Innovative Bio Inspired Helmet
Biological structures such as butterfly wings are natural hybrid materials that are made
up of multiple components that are combined in specific geometries and scales. Butterfly
wings have widely inspired researchers due to their particular design and multifunction
such as attracting their mates (optical) or escaping predators (aero-mechanical). Indeed,
from a mechanical perspective, their wing can be considered a structure that is optimized
for bending loads. In a cross section of a wing scale, the highly porous central region
separates two outer regions which are supported by a frame in whereby load-bearing
bars are connected to the porous core by perpendicular smaller bars. The topology of the
inner porous region maximizes the structure’s rigidity while simultaneously minimizing
its weight. The architecture of the porous structure can be described as the TPMS of the
gyroid (Sect. 5). The structure was further evolved by reinforcing the external elements
with carbon fiber-reinforced plastic (CFRP) rods [40].
The model was designed and simulated with finite element analysis (FEA) to
optimize its hybrid structure and parameters: CFRP rods diameter and gyroid surface
thickness. The structures were AM manufactured through the stereolithography
(SLA) technique. SLA allows for the fabrication of three-dimensional polymeric parts
with UV radiation that induces the photopolymerization of a reactive monomer. In this
process, the STL model is sliced into two-dimensional cross-sections, allowing for their
projection in sequence and building the part layer by layer. The CFRP rods were
inserted in their casings and bonded to them through a thin layer of a two-component
epoxy glue that was applied to the rods before their insertion. In order to evaluate the
mechanical behavior of the structures, 3-point bending, quasi-static experimental tests
were performed with a universal material electromechanical testing machine at standard
conditions. A 3D FEM-based approach was followed to simulate the experimental 3-
point bending tests on the structures. Ultimately, experimental results were compared
with the FEA output in the linear-elastic regime. The validation of the simulation was
performed comparing the bending deformation of the structure with the experimental
one under the same condition. The structure with CFRP rods had more than twice the
stiffness of the non-reinforced structure (46 N/mm and 20 N/mm, respectively) and
withstood a maximum load of about 280 N (100 N without CFRP). The results of the
FE model confirm these values.
An advantage of this solution over the standard sandwich structures is that it
directly connects the solid part of the porous core to the mating reinforcing element and
further minimizes its mass. The proposed topological approach can be applied to many
materials as long as there is a difference in the elastic modulus between the core and the
ribs. The design can become more demanding from a computational point of view if
you want to generate complex objects.
To demonstrate the feasibility of this concept, thanks to nTop Platform (New York,
NY 10013, USA), an innovative helmet was designed (Fig. 9). The structure consists
of an ultra-lightweight helmet made up of gyroid cells and reinforced with carbon fiber
bars. This concept design aims at combining the lightness and transpiration of gyroid
architectures with the unmatched mechanical properties of carbon fiber reinforced
plastics (CFRP), which are placed (like in the butterfly wing) to enhance the impact
protection right where it is needed.
Fig. 9. An innovative helmet designed with nTopology
7 Conclusions
The strategy of designing a component has always been influenced by its manufac-
turing. The advent of AM has allowed conceiving components by their function and no
longer by their manufacture. This benefit allows improving the components’ perfor-
mances and the fabrication of geometrically complex parts such as cellular structures.
The computational design is performed through the simulation of the component
behavior (mechanical, thermal, fluid dynamic, etc.) based on a trial & error approach.
Therefore, the possibility of quickly modify the geometric parameters of the component
becomes fundamental. This study reviews alternative design methods to create cellular
structures with regular and irregular configurations. The design tools presented in this
work were employed to design components for several engineering applications: active
zone in porous burners, heat radiation application, catalyst carriers, solar radiation
absorber, high performance protectors for space vehicles, complex templates for heat
exchangers, water treatment, lightweight materials, composites.
With the irregular strut-based lattice approach, it is possible to generate random
foams and Voronoi lattice. These structures exhibit scattered properties and they are
used when randomness is needed. The regular strut-based lattice approach is based on
the creation structured and unstructured lattice structure. The 3D numerical tools are
based on the replication of a desired unit-cell in the space followed by Boolean tool to
cut the desired final shape. This is the most used approach to design engineering
components due to the control and parametrization of the numerical model. The multi-
lattice approach, based on the Voronoi tessellation, allows to generate structure with
different morphologies in a single volume. It can be used to join different lattices and
generate very complex components. The TPMS design approach consists in the cre-
ation of surface-based components with high mechanical properties. This numerical
tool was used to design an innovative helmet inspired by the butterfly wings scales. The
structure consists of an ultra-lightweight body made up of gyroid cells and reinforced
with carbon fiber bars. This innovative concept aims at combining the lightness and
impact resistance.
References
1. Fend, T.: High porosity materials as volumetric receivers for solar energetics. Optica
Applicata 40(2), 271–284 (2010)
2. Gianella, S., Gaia, D., Ortona, A.: High temperature applications of Si-SiC cellular ceramics.
Adv. Eng. Mater. 14(12), 1074–1081 (2012)
3. Camacho, Y.M., Bensaid, S., Lorentzou, S., Vlachos, N., Pantoleontos, G., Konstan-
dopoulos, A., Luneau, M., Meunier, F.C., Guilhaume, N., Schuurman, Y., Werzner, E.:
Development of a robust and efficient biogas processor for hydrogen production. Part 2:
experimental campaign. Int. J. Hydrog. Energy 43(1), 161–177 (2018)
4. Szczurek, A., Ortona, A., Ferrari, L., Rezaei, E., Medjahdi, G., Fierro, V., Bychanok, D.,
Kuzhir, P., Celzard, A.: Carbon periodic cellular architectures. Carbon 88, 70–85 (2015)
5. Rezaei, E., Haussener, S., Gianella, S., Ortona, A.: Early-stage oxidation behavior at high
temperatures of SiSiC cellular architectures in a porous burner. Ceram. Int. 42(14), 16255–
16261 (2016)
6. Klahn, C., Leutenecker, B., Meboldt, M.: Design strategies for the process of additive
manufacturing. Procedia CIRP 36, 230–235 (2015)
7. Deshpande, V.S., Ashby, M.F., Fleck, N.A.: Foam topology: bending versus stretching
dominated architectures. Acta Mater. 49(6), 1035–1040 (2001)
8. Murr, L.E., Gaytan, S.M., Medina, F., Lopez, H., Martinez, E., Machado, B.I., Hernandez,
D.H., Martinez, L., Lopez, M.I., Wicker, R.B., Bracke, J.: Next-generation biomedical
implants using additive manufacturing of complex, cellular and functional mesh arrays.
Philos. Trans. R. Soc. A: Math. Phys. Eng. Sci. 368, 1999–2032 (2010)
9. Wang, P., Li, J.B., Vafai, K., Zhao, L., Zhou, L.: Thermo-fluid optimization of a solar porous
absorber with a variable pore structure. J. Solar Energy Eng. 139(5), 51012–51017 (2017)
10. Wang, H., Chen, Y., Rosen, D.W.: A hybrid geometric modeling method for large scale
conformal cellular structures. In: ASME 2005 International Design Engineering Technical
Conferences and Computers and Information in Engineering Confer (September 2005)
11. Gibson, L.J., Ashby, M.F.: Cellular Solids: Structure and Properties. Cambridge University
Press, Cambridge (1999)
12. Evans, A.G., Hutchinson, J.W., Ashby, M.F.: Cellular metals. Curr. Opin. Solid State Mater.
Sci. 3(3), 288–303 (1998)
13. Giani, L., Groppi, G., Tronconi, E.: Mass-transfer characterization of metallic foams as
supports for structured catalysts. Ind. Eng. Chem. Res. 44(14), 4993–5002 (2005)
14. Lacroix, M., Nguyen, P., Schweich, D., Huu, C.P., Savin-Poncet, S., Edouard, D.: Pressure
drop measurements and modeling on SiC foams. Chem. Eng. Sci. 62(12), 3259–3267 (2007)
15. Ortona, A., Rezaei, E.: Modeling the properties of cellular ceramics: from foams to lattices
and back to foams. In: Advances in Science and Technology, vol. 91, pp. 70–78. Trans Tech
Publications (2014)
16. D’Angelo, C., Ortona, A., Colombo, P.: Finite element analysis of reticulated ceramics under
compression. Acta Mater. 60(19), 6692–6702 (2012)
17. Pusterla, S., Ortona, A., D’Angelo, C., Barbato, M.: The influence of cell morphology on the
effective thermal conductivity of reticulated ceramic foams. J. Porous Mater. 19(3), 307–315
(2012)
18. Bianchi, G., Gianella, S., Ortona, A.: Design and additive manufacturing of periodic ceramic
architectures. J. Ceram. Sci. Technol. 8(1), 59–66 (2017)
19. Ortona, A., Pusterla, S., Fino, P., Mach, F.R.A., Delgado, A., Biamino, S.: Aging of
reticulated Si-SiC foams in porous burners. Adv. Appl. Ceram. 109(4), 246–251 (2010)
20. Bowyer, A.: Computing Dirichlet tessellations. Comput. J. 24(2), 162–166 (1981)
21. Watson, D.F.: Computing the n-dimensional Delaunay tessellation with application to
Voronoi polytopes. Comput. J. 24(2), 167–172 (1981)
22. Fortune, S.: Voronoi diagrams and Delaunay triangulations. In: Computing in Euclidean
geometry, pp. 225–265 (1995)
23. Ortona, A., D’Angelo, C., Gianella, S., Gaia, D.: Cellular ceramics produced by rapid
prototyping and replication. Mater. Lett. 80, 95–98 (2012)
24. Santoliquido, O., Bianchi, G., Dimopoulos Eggenschwiler, P., Ortona, A.: Additive
manufacturing of periodic ceramic substrates for automotive catalyst supports. Int. J. Appl.
Ceram. Technol. 14(6), 1164–1173 (2017)
25. Sullivan, R.M., Ghosn, L.J., Lerch, B.A., Baker, E.H.: Elongated tetrakaidecahedron
micromechanics model for space shuttle external tank foams (2009)
26. Herrmann, A., Pönisch, M., Werzner, E., Laurinat, M., Rezaei, E., Ortona, A., Bensaid, S.,
Fino, D., Trimis, D.: Development of an autothermal biogas processor for hydrogen
production. In: Proceedings of the International Gas Union Research Conference (IGRC
2014) (September 2014)
27. Camacho, Y.M., Bensaid, S., Fino, D., Trimis, D., Herrmann, A., Guilhaume, N.,
Schuurman, Y., Konsatndopoulos, A.G., Lorentzou, S., Gianella, S., Antonini, M.: Biogas
robust processing with combined catalytic reformer and trap: BioRobur Project. WIT Trans.
Ecol. Environ. 195, 463–474 (2015)
28. Ortona, A., Fend, T., Yu, H.W., Raju, K., Fitriani, P., Yoon, D.H.: Tubular Si-infiltrated
SiCf/SiC composites for solar receiver application—Part 1: fabrication by replica and
electrophoretic deposition. Solar Energy Mater. Solar Cells 132, 123–130 (2015)
29. Ortona, A., Yoon, D.H., Fend, T., Feckler, G., Smirnova, O.: Tubular Si-infiltrated SiCf/SiC
composites for solar receiver application–Part 2: thermal performance analysis and
prediction. Solar Energy Mater. Solar Cells 140, 382–387 (2015)
30. Papetti, V., Eggenschwiler, P.D., Della Torre, A., Lucci, F., Ortona, A., Montenegro, G.:
Additive manufactured open cell polyhedral structures as substrates for automotive catalysts.
Int. J. Heat Mass Transf. 126, 1035–1047 (2018)
31. Pelanconi, M., et al.: Thermal design, optimization and additive manufacturing of ceramic
regular structures to maximize the radiative heat transfer. Mater. Des. 163, 107539 (2019)
32. Esser, B., Barcena, J., Kuhn, M., Okan, A., Haynes, L., Gianella, S., Ortona, A., Liedtke, V.,
Francesconi, D., Tanno, H.: Innovative thermal management concepts and material solutions
for future space vehicles. J. Spacecr. Rockets 53(6), 1051–1060 (2016)
33. Boomsma, K., Poulikakos, D., Ventikos, Y.: Simulations of flow through open cell metal
foams using an idealized periodic cell structure. Int. J. Heat Fluid Flow 24(6), 825–834
(2003)
34. Boomsma, K., Poulikakos, D.: On the effective thermal conductivity of a three-
dimensionally structured fluid-saturated metal foam. Int. J. Heat Mass Transf. 44(4), 827–
836 (2001)
35. Ferrari, L., Barbato, M., Esser, B., Petkov, I., Kuhn, M., Gianella, S., Barcenad, J., Jimenez,
C., Francesconi, D., Liedtke, V., Ortona, A., Ortona, A.: Sandwich structured ceramic matrix
composites with periodic cellular ceramic cores: an active cooled thermal protection for
space vehicles. Compos. Struct. 154, 61–68 (2016)
36. Schoen, A.H.: Infinite periodic minimal surfaces without self-intersections (1970)
37. Kapfer, S.C., Hyde, S.T., Mecke, K., Arns, C.H., Schröder-Turk, G.E.: Minimal surface
scaffold designs for tissue engineering. Biomaterials 32(29), 6875–6882 (2011)
38. Al-Ketan, O., Pelanconi, M., Ortona, A., Abu Al-Rub, R.K.: Additive manufacturing of
architected catalytic ceramic substrates based on triply periodic minimal surfaces. J. Am.
Ceram. Soc. 102(10), 6176–6193 (2019)
39. Santoliquido, O., Colombo, P., Ortona, A.: Additive manufacturing of ceramic components
by digital light processing: a comparison between the “bottom-up” and the “top-down”
approaches. J. Eur. Ceram. Soc. 39(6), 21 (2019)
40. Pelanconi, M., Ortona, A.: Nature-inspired, ultra-lightweight structures with gyroid cores
produced by additive manufacturing and reinforced by unidirectional carbon fiber ribs.
Materials 12(24), 4134 (2019)
Process Chain
Multi-material 3D Printing of Thermoplastic
Elastomers for Development of Soft Robotic
Structures with Integrated Sensor Elements
Antonia Georgopoulou1,2(&), Bram Vanderborght2,

and Frank Clemens1
1
Department of Functional Materials,
Empa – Swiss Federal Laboratories for Materials Science and Technology,
Überlandstrasse 129, 8600 Dübendorf, Switzerland
antonia.georgopoulou@empa.ch
2
Department of Mechanical Engineering (MECH), Vrije Universiteit Brussel
(VUB), and Flanders Make, Pleinlaan 2, 1050 Brussels, Belgium
Abstract. Embedded sensing can benefit soft robots with the ability to interact
with their environment but producing embedded soft sensors can be challenging.
Multi-material Fused Deposition Modeling (FDM) additive manufacturing
allows producing complex structures, by combining more than one kind of
polymeric material. For multi-material FDM, conductive thermoplastic elas-
tomer filaments have been developed. This allows the printing of flexible
functional structures, based on thermoplastic elastomer structures with con-
ductive paths that are of great interest for stretchable electronics and soft robotic
applications. In this study, stretchable piezoresistive elastomer strain sensor
composites were successfully produced by using multi-material FDM.
A piezoresistive thermoplastic elastomer was printed on the top of a noncon-
ductive, flexible thermoplastic elastomer strip using FDM multi-material 3D
printer. FDM elastomer filaments with different shore hardness as substrate
materials for the gripper structure were used. The hardness of the elastomer
affected the printability and the adhesion to the conductive elastomer material,
which was used as a strain sensor material. The hardness affected the strain
sensor properties too. The piezoresistive response, dynamic behavior, drift,
relaxation and sensitivity of the printed multi-material strips were investigated
by tensile tests. Soft robotic grippers with integrated sensing elements to detect
deformation while touching the objective were selected as a case study. The soft
grippers with the integrated sensors exhibited intelligent response by recog-
nizing when they were griping a small or big object and when an obstacle was
inhibiting their function.
Keywords: Piezoresistive elastomer sensor FDM additive manufacturing

Soft robotic gripper

https://doi.org/10.1007/978-3-030-54334-1_6
68 A. Georgopoulou et al.
1 Introduction
Important applications for additive manufacturing (AM) are the development and
fabrication of products for consumer goods and electronics [1]. 3D printing in com-
bination with functional materials can be used for the production of sensors integrated
in substrates and robotic devices [2–4]. The FDM additive manufacturing is one of
several material extrusion methods, which can be used to develop robotic devices with
integrated sensors. One of its main advantages in comparison to direct energy depo-
sition, material jetting or powder bed fusion AM techniques is the fact that multi-
material printing is easy to implement. In addition, FDM is cost-efficient, has good
resolution and compatibility with many materials and composites [5, 6]. Multi-
mateiral FDM includes a combination of different materials and can be a useful pro-
duction method for robotic systems [7]. In the case of sensing in robotics, multi-
material 3D printing can be used for producing the robotic body with integrated sen-
sors, based on functional materials, in one-step [8]. FDM additive manufacturing is
compatible with thermoplastic elastomer materials like thermoplastic polyurethane
(TPU). By using TPU filaments, it is possible to print soft robotic structures like
pneumatic actuators and grippers.
Soft robotic grippers with integrated sensors are intelligent systems that can alter
the posture, gripping force, and gripping geometry according to the object they are
gripping [9, 10]. These intelligent grippers can find application where a careful
manipulation of objects is needed but also for the handling of complex geometries,
where compliance with the object surface is important [11]. Therefore, soft grippers
have been developed for the food industry to handle sensitive food and perform
complex tasks that involve food like packaging [12, 13]. Furthermore, the soft robotic
gripper can be used for the exploration of unknown environments like space or
underwater and can be particularly useful for acquiring sensitive samples from these
environments [14, 15]. Soft grippers based on pneumatic actuators can be controlled by
pressure [16]. It is also possible to develop soft grippers using servomotors. However,
they must be controlled optically or by piezoresistive sensors to be able to use them for
sensitive objectives. Under harsh conditions, optically monitoring of gripper move-
ments is difficult. Multi-material FDM to develop soft robotic grippers with integrated
sensors has been used before [17]. However, the effect of the stiffness of the soft
gripper structure on the sensing behavior has never been investigated. In this attempt,
soft robotic grippers with integrated strain sensors have been investigated using a
commercial FDM multi-material printer. TPU filament with carbon black filler was
used for printing of the sensing paths on the surface of the gripper structure. The body
of the soft gripper was printed with TPU filaments of two different shore hardness to
investigate the effect of the shore hardness on the sensing behavior of the conductive
sensing structure on the surface of the gripper. Furthermore, the effect of the thickness
of the gripper on the sensing behavior was investigated and the potential of the sensor
to be used for monitoring the function of the robotic gripper was explored.
Multi-material 3D Printing of Thermoplastic Elastomers 69

2.1 Sensor Printing and Robotic Gripper
Filaments based on TPU were supplied from Recreus Industries (Elda, Spain) in two
different shore harnesses (FilaFlex 82A and FilaFlex 95A). The conductive TPU fila-
ment Eel based on thermoplastic polyurethane and carbon black was supplied by
Fenner Drives (Ninjatek Eel, Manheim, USA). For 3D printing, the FDM 3D printer
Pro2 Dual Extruder 3D Printer (Raise 3D, Irvine, USA) was used. The printing of the
multi-material strips and soft grippers was done at a temperature of 230 °C with a
printing speed of 15 mm/s for the perimeters and 20 mm/s for the infill. A layer height
with 0.2 mm was printed with a nozzle size of 0.6 mm. To achieve a dense structure,
infill was set to 100%; the extrusion multiplier was set to 120%. The setup of the multi-
material FDM additive manufacturing process can be seen in Fig. 1.
Fig. 1. Setup of the process of multi-material FDM additive manufacturing with a conductive
and a non-conductive filament.
For the assessment of the mechanical and electrical behavior of the system, the Eel
conductive TPU was printed on the surface of TPU strips with dimensions130 10
0.3 mm.
For the multi-material soft gripper structure, first, the sensor structure was produced
and on top of it, the gripper body was printed. After the printing of the gripper, silver
wires were inserted at the gripper to act as tendons. These tendons were necessary for
the actuation and motion of the robotic gripper. Additional to the wires, a Tower Pro
MG90S micro servo (Adafruit Industries. New York, USA) was used. The control of
the motor was performed with an Arduino microcontroller. Both structures are shown
in Fig. 2.
Fig. 2. a) Sensors integrated on the surface of an elastomer strip from FilaFlex 82A (yellow) and
FilaFlex 95A (blue) b) robotic gripper with integrated sensing elements made with FilaFlex 82
and c) with FilaFlex 95A for the robotic body using multi-material 3D printing.
2.2 Tensile Testing

For minimizing the slipping during the tensile testing, pneumatic clamps were used.
4 bar pressure had to be applied to avoid slipping of the samples during the testing. For
measuring the resistivity during the tensile test, a Keithley 2450 source meter
(Keithley, Solon, USA) was used with the KickStart software from the same company.
For the measurements, a two-terminal sensing mode was used and the change in current
was measured while a constant voltage of 1 V was applied. Two different types of
tensile tests were performed. First, the strips were tested up to the breaking point, and
later strips were tested dynamically with consecutive cycles of loading and releasing,
separated by a dwell time of 30 s at the maximum and minimum strain levels. From the
measurements of the resistance, the relative resistance (Rrel) was calculated according
to the formula where R is the electrical resistance of the sensor and Ro the electrical
resistance of the sensor when no strain is applied to it:
R R0
Rrel ¼ ð1Þ
R0
3 Results and Discussion

3.1 Tensile Test to the Breakpoint
Tensile tests were performed on multi-material printed strips (Fig. 2a). From the results
of the tensile test, the stress-strain was constructed (Fig. 3).
82A
7
95A
6
Stress (MPa)
4
0
0 100 200 300 400 500
Strain (%)
Fig. 3. Stress-strain plot of the TPU strips (with integrated sensor element) with different shore
hardness (82A and 95A). The strip with the lowest shore hardness could endure larger
elongations compared to the substrate of higher shore hardness.
From the stress-strain plot, it can be seen, that the TPU with lower shore hardness
could endure much larger elongation (up to 500%) before the break. The strip with
higher shore hardness broke at 260% strain. As expected, the stiffness was higher for
the strip, which was printed from TPU filament with higher shore hardness. The
elasticity modulus was calculated at 13 MPa for the 82A strip sample and 23 MPa for
the 95A sample. No delamination between the sensor part and the substrate strip was
observed during the tensile testing. The sensor broke before the substrate strip did.
80 82A
60
Rel. Resistance
40
95A
20
0
0 100 200 300 400 500
Strain (%)
Fig. 4. Electrical signal of the piezoresistive sensor printed on the TPU strips of different shore
hardness.
Looking at the sensor response (Fig. 4), up to 250% strain, the strips of the two
different TPU strips show similar behavior. The electrical resistance changed with the
applied deformation, showing that the system exhibited a piezoresistive response.
3.2 Dynamic Tensile Testing

Looking at the dynamic testing (Fig. 5), it was seen that after five cycles of loading and
releasing, the mechanical properties of the strips with integrated sensor elements
responded with good repeatability.
Fig. 5. Mechanical response during dynamic tensile testing between the strain 0 and 30% for
strips of shore hardness a) 82A and b) 95A. Stress relaxation was observed for both sensor
systems but was larger for the strip of lower shore hardness.
At the dwell time, when the strain was held constant for 30 s the mechanical
relaxation was investigated as described by Melnykowycz et al. [18]. Stress relaxation
often occurs in elastomers and can affect the sensor response during the dynamic test.
As proposed by Melnykowycz et al., the drift of the mechanical stress was investigated
between different cycles. In this case, the drift between the second and fifth cycle at
30% strain was evaluated (Table 1).
Table 1. Mechanical relaxation and drift during dynamic tensile testing for strips made with
multi-material 3D printing calculated at 30% strain.
Substrate material Mechanical relaxation Mechanical drift
FilaFlex 82A 24% 20%
FilaFlex 95A 20% 5%
It was observed that at 0% strain, the strips were bucking and therefore the
mechanical relaxation and drift could not be calculated. Based on the measurements
shown in Fig. 4, buckling occurred at strains below 16%. The stress relaxation was
higher in the case of the material with the lower shore hardness, FilaFlex 82A. As for
the mechanical drift, it was relatively low for both systems, except for the FilaFlex 95A
system with a drift of 89%. The presence of drift especially at low strains can be
attributed to the presence of buckling during the tensile testing.
Looking at the piezoresistive behavior of the 3D printed sensor elements on top of
the TPU strips during the dynamic testing it was seen, that the relative resistance could
follow the change in strain for both the loading and the releasing phase of the tensile
test (Fig. 6).
a) 82A Strain Rel.Resistance b) 95A Strain Rel.Resistance
100 100
0.2 0.2
90 90
80 80
0.0 0.0
70 70
Rel.Resistance
Rel.Resistance
-0.2 -0.2 60
Strain (%)
60
Strain (%)
50 50
-0.4 -0.4
40 40
-0.6 -0.6 30
30
20 20
-0.8 -0.8
10 10
-1.0 0 -1.0 0
0 50 100 150 200 250 300 350 0 50 100 150 200 250 300 350
Time (sec) Time (sec)
Fig. 6. Sensor Response during dynamic tensile testing between the strain 0 and 30% for
embedded sensors in substrates with shore hardness a) 82A and b) 95A. Relaxation was also
observed for the electrical signal but in this case, there was not significant dependence on the
shore hardness of the strip observed.
The reverse piezoresistivity that was seen during the tensile test to the breakpoint
for strains lower than 30%, was also seen at the dynamic testing. At low strain higher
conductivity and at high strain lower conductivity can be observed. It is worthwhile to
mention that due to the buckling behavior, the electrical relaxation and drift could not
be determined at 0% strain. However, it can be seen in Fig. 6, that during buckling of
the strips at lower than 16% strain the electrical signal of the 82A show an unexpected
drift. The response of the sensor was linear but the relative resistance decreased with an
increase in strain and increased when the strain decreased. The relaxation and drift that
was observed for the mechanical dynamic behavior of the sensor, was also seen in the
response of the sensor signal (Table 2).
Table 2. Electrical relaxation and relaxation during dynamic tensile testing for the sensors
integrated into an elastomer substrate produced with multi-material 3D printing calculated at 30%
strain.
Substrate material Electrical relaxation Electrical drift
FilaFlex 82A 28% 9%
FilaFlex 95A 25% 8%
In the case of the electrical signal, the relaxation was slightly higher for the system
with a strip of lower shore hardness. This finding agreed with what was also observed
for stress relaxation. However, in the case of the drift of the electrical signal, there was
almost no difference seen when comparing the two systems.
3.3 Application: Robotic Gripper with Integrated Piezoresistive Gauge

Sensor
The aim of this study was the printing of soft robotic grippers with integrated strain
sensing elements. The grippers consisted of a flexible belt and elements called pha-
langes (Fig. 7).
Fig. 7. Sketch of a tentacle, showing the belt, the phalanges and the wires (tendons) which are
connected to the servomotor.
Inside the phalanges structure, a wire (tendon) is connected with a servomotor. If

the servomotor will coil the tendon, the belt will bended until the phalanges will touch
each other. With the assistance of flexible tendons, the belt could bend because of the
reduction in tendon length that is connected to the servomotor. The black lines in Fig. 7
are the piezoresistive sensor parts, which were printed on top of the belt structure.
Varying thickness of the belt from 2 to 6 mm was used to investigate the effect of the
geometrical stiffness, and therefore the bending stress and deformation, on the sensor
behavior. Grippers with three different belt thicknesses (2 mm, 4 mm and 6 mm) were
printed. The grippers were assessed for their sensor performance during consecutive
cycles of opening and closing (Fig. 8).
Fig. 8. Gripper with embedded sensors produced with multi-material FDM additive manufac-
turing operates between positions a) open and b) closed and close up for the gripper of shore
hardness c) 82A and d) 95A.
It is worthwhile to mention that by adding the sensor structure on top of the printed
gripper substrate no significant stiffening effect could be observed and the servomotor
had no problem opening and closing the gripper. Based on mechanics for bending, the
sensor will see larger stress and deformation if the thickness of the gripper will
increase. As it was seen from the sensor response, by increasing the belt thickness, the
change in the electrical resistance of the strain sensor decreased (Fig. 9).
Fig. 9. Signal response of the strain sensor on the surface of the belt structure in a robotic
gripper with three different belt thicknesses during a cyclical test. The gripper moved five times
between positions open and close a) relative resistance during testing and b) resistance during the
testing. The gripper with a belt of the smallest thickness had the largest change in relative
resistance but the sensor response was accompanied by significant noise.
Independent on the belt thickness, all the grippers showed the reverse piezoresis-
tivity that was seen during the tensile testing (Fig. 6). In general, for all belt thicknesses
it could be observed that while the deformation increased, the relative resistance
decreased. For the gripper with the smallest thickness of the belt, it was seen that the
change in the resistance and relative resistance was the highest. However, when
looking at the gripper with the 2 mm belt thickness, observed higher noise at a closed
position and a lager relaxation at an open position can be observed. The appearance of
this noise could be considered undesirable for many applications like soft robotics and
might be an artifact of the servomotor because of small voltage fluctuations. This noise
was only observed in the gripper with the smaller belt thickness and is not noticeable
for the other systems. The relative resistance change changes for the three different
grippers. To verify these differences, some deformations calculations where done
(Table 3).
Table 3. The deformation of the gripper and change in relative resistance when the gripper
moved between positions open and close for grippers with different belt sizes.
Belt Length of the wire at Length of the wire at Relative Relative
thickness open position (mm) close position (mm) deformation change in
(mm) resistance
2 8 9.8 20.2% 60%
4 8 9.2 14.0% 52%
6 8 8.8 9.5% 37%
The wires were used to open and close the tentacles of the soft grippers by a
servomotor and the change in their length, when the gripper moved between positions
open and closed, was used to calculate the relative deformation. The wires were coiled
up and thus the length of the wires gets shorter when the gripper is closed. Looking at
the values in Table 3 it was observed, that the relative deformation decreased by
increasing the belt thickness. Smaller deformation caused a smaller change in resis-
tance, an effect that is expected from a deformation sensor. The results in Table 3 are in
good agreement with optical observations. Due to the design change of the gripper
(thicker band) it was observed, that the phalanges would touch each other at lower
strain, which will block further deformation of the tentacle of the gripper.
The initial value Ro for the resistance was also different for the grippers with
different band thicknesses (Fig. 9b). The Ro was higher for the gripper with the lowest
belt thickness and decreased with the increasing belt thickness. In order to verify if this
difference was caused by the printing procedure, the initial resistance Ro was measured
for all the four tentacles of each gripper (Table 4).
Table 4. The deformation of the gripper and change in relative resistance when the gripper
moved between positions open and close for grippers with different belt sizes.
Gripper Belt Ro at Ro at Ro at Ro at Deviation
material thickness tentacle tentacle tentacle tentacle
(mm) 1 (kX) 2 (kX) 3 (kX) 4 (kX)
FilaFlex 2 36 ± 6 34 ± 8 79 ± 2 102 ± 6 29%
82A
FilaFlex 4 164 ± 4 120 ± 5 48 ± 1 33 ± 3 53%
82A
FilaFlex 4 59 ± 4 67 ± 5 60 ± 4 58 ± 3 4%
95A
FilaFlex 6 116 ± 3 50 ± 1 31 ± 2 33 ± 2 35%
82A
From comparing the values of the resistance in the different tentacles, it was seen
that the values can vary a lot for all the grippers made with the material FilaFlex 82A.
The change in resistance was the same for every tentacle, independent of the initial
value in the resistance as it is a parameter that depends on the deformation, which was
the same for all the tentacles. As a result, there can't be made a conclusion about the
effect of the thickness of the belt on the values of the resistance. However, this was not
the case for the material FilaFlex 95A. In this system, the values of the resistance had
consistency between the different tentacles of the grippe. This large deviation in the
values of the resistance can be traced back to the printing procedure. When observing
optically the produced grippers, for the case of FilaFlex 82A material that there could
be seen traces of carbon black powder all over the first layer. This effect was distin-
guished as the original color of the filament appeared darkened by black particles at
parts of the first layer. This was not the case for the FilaFlex 95A gripper, that the
coloration of the first layer was unaltered. A possible interpretation is that during the
printing, in the case of the FilaFlex 82A, the nozzle that printed the robotic body
carried away some of the particles of the carbon black of the conductive TPU that was
print first. However, this change was not consistent during the printing, the discol-
oration appeared more intense in some tentacles compared to others, and this could be
an explanation for the deviation of the values of the resistance.
Based on those results of the gripper design study, grippers with a belt thickness of
4 mm were made with the FilaFlex 82A the FilaFlex 95A. The grippers were compared
for the electrical signal and stability of the sensor response. Both structures showed a
relative deformation of around 16% for a fully closed position. In order to assess the
sensor signal for the grippers, the soft structures were opened and closed with a dell
time during each position. In addition, an objective (orange) was grabbed by the soft
grippers (Fig. 10 a)). In addition, the electrical signal of the sensor was investigated
when the movement of the tentacles was blocked by an obstacle (Fig. 10 b)).
Fig. 10. a) The robotic gripper with the integrated sensors and a band thickness of 2 mm is
gripping an orange to test object recognition and b) the hand is preventing the gripper from
opening to test obstacle recognition by the integrated sensor.
The sensor response for the grippers made with the materials of different shore
hardness performing different tasks can be seen in Fig. 11
a) 82A b) 95A
8.0x104 8.0x104 open
7.5x104 7.5x104 open
4 4
7.0x10 7.0x10
4 4
6.5x10 6.5x10
Resistance (Ohm)
Resistance (Ohm)
6.0x104 6.0x104
5.5x104
open open 5.5x104
5.0x104 5.0x104
4.5x104 4.5x104
object
4.0x104 4.0x104 closed
4 4
3.5x10 3.5x10
object
3.0x104 closed 3.0x104
4
2.5x10 2.5x104
0 2 4 6 8 10 12 14 16 18 20 0 2 4 6 8 10 12 14 16 18 20
8.0x104 c) 82A 8.0x104 d) 95A
7.5x104 7.5x104
7.0x104 7.0x104
open
6.5x104 open 6.5x104
Resistance (Ohm)
Resistance (Ohm)
6.0x104 6.0x104
5.5x104 5.5x104
5.0x104 5.0x104
4.5x104 4.5x104
obstacle
4.0x104 4.0x104
closed
3.5x104 obstacle 3.5x104
3.0x104 3.0x104
closed
2.5x104 2.5x104
0 2 4 6 8 10 12 14 16 18 20 22 24 26 0 2 4 6 8 10 12 14 16 18 20 22 24 26
Fig. 11. Sensor response for grabbing objective and obstacle during obstacle recognition test for
grippers with a robotic made out of elastomer with shore hardness a) 82A and b) 95A and sensor
response during obstacle recognition test for the robotic body made out of elastomer with shore
hardness c) 82A and d) 95A. Both sensors could indicate when the gripper was gripping an
object, when not and when an obstacle was impairing the function of the system
Looking at Fig. 11 a) and b) it was calculated, that the difference in resistance was
52% for the 82A and 59% for the 95A. The relative change in resistance between the
closed positions and holding an objective was 7.4% and 11.5% for the 82A and 95A,
respectively. Therefore, in both cases, the relative change of the sensor signal was
slightly higher for the TPU gripper with higher shore hardness. An important parameter
for deformation sensors that are targeted for robotic applications is the ability to dis-
tinguish from the sensor signal the different positions of the robot system. The initial
resistance at 6.1 kX was higher for the system with higher shore hardness compared to
the lower shore hardness, with initial resistance of 4.2 kX.
As already mentioned, the maximal deformation from open to the closed position
(touching of the phalanges) was 16.2%. For grabbing the orange, the deformation was
calculated to 6.5%. According to the electrical resistance values, the piezoresistive
sensors integrated into the soft gripper structures can identify if the gripper is open,
closed or if an objective is grabbed. In soft robotics, grippers with integrated
piezoresistive sensor elements that can distinguish between when the gripper is grab-
bing an object, open and closed position lead to the creation of intelligent robotic
systems.
Additionally to the gripping test, an obstacle test was performed. During the
obstacle test, the gripper movement (open-closed-open-closed) was blocked manually
after two cycles. In the first two, the closing and opening of the soft gripper could be
easily detected by the change in resistance. However, at the time point when the
obstacle was imposed, the value for the resistance did not return to the value of the
resistance for the open position. As already explained, wires were used to open and
close the tentacles of the soft grippers by a servomotor. The wires are coiled up and
thus the length of the wires gets shorter when the gripper is closed. The resistance
stayed close to the value for the closed position for the entire time the obstacle was
imposed. After the obstacle was removed, the resistance returned to the previous values
for the position open.
With the ability to distinguish between positions open and close, but also recog-
nizing obstacles and when the gripper is gripping an object, these intelligent soft robots
can be used for applications where effective monitoring of the gripper function is
needed. As a result, these grippers with integrated sensing elements could be used for
more efficient production in many sectors of the industry that requires soft robots, as it
is, for example, the food industry.
In this attempt, soft robotic grippers with integrated sensing elements were pro-
duced in the one-step process using multi-material FDM. These grippers consisted of
two materials, one conductive TPU that can be used to sense the deformation of the
gripper structure and one non-conductive TPU to fabricate the structure of the gripper.
In order to investigate, the behavior of the printed composite systems and the effect of
the shore hardness of the strips, tensile testing was performed. The TPU with lower
shore hardness can be used for applications with larger elongation. However, cycling
experiments showed that the sensor behavior was similar for both types of TPU.
4 Conclusions
For the functional soft gripper application, first, the optimal thickness of the sensor part
of the soft gripper was adjusted to 4 mm thickness. Later, two TPU filaments with
different shore hardness (82A and 95A) were investigated as candidate materials for the
gripper structure. The gripper with the TPU 95A showed a slightly higher difference in
electrical resistance value between open and closed positions. This resulted in a higher
sensitivity. The grippers with the integrated sensing elements exhibited intelligent
function. The deformation of the sensor enriched their function with the ability to
distinguish when the gripper is gripping an object, when not and when an obstacle is
preventing the gripper from functioning properly. This intelligent soft robotic gripper
with the integrated sensors that was produced in one-stem with a simple and low cost
could be potentially found each place in production lines for sensitive objects and lead
to more efficiency and accuracy in production.
Acknowledgement. This project has received funding from the European Union’s Horizon
2020 research and innovation programme under grant agreement No 828818 (SHERO project).
References
1. Ligon, S.C., Liska, R., Stampfl, J., Gurr, M., Mülhaupt, R.: Polymers for 3D printing and
customized additive manufacturing. Chem. Rev. 117, 10212–10290 (2017)
2. Yang, Y., Song, X., Li, X., Chen, Z., Zhou, C., Zhou, Q., Chen, Y.: Recent progress in
biomimetic additive manufacturing technology: from materials to functional structures. Adv.
Mater. 30, 1706539 (2018)
3. Yap, H.K., Ng, H.Y., Yeow, C.H.: High-force soft printable pneumatics for soft robotic
applications. Soft Robot. 3, 144–158 (2016)
4. Mutlu, R., Alici, G., Panhuis, M.I.H., Spinks, G.M.: 3D printed flexure hinges for soft
monolithic prosthetic fingers. Soft Robot. 3, 120–133 (2016)
5. Mohan, N., Senthil, P., Vinodh, S., Jayanth, N.: A review on composite materials and
process parameters optimisation for the fused deposition modelling process. Virtual Phys.
Prototyp. 12, 47–59 (2017)
6. Kumar, S., Kruth, J.P.: Composites by rapid prototyping technology. Mater. Des. 31, 850–
856 (2010)
7. Espalin, D., Muse, D.W., MacDonald, E., Wicker, R.B.: 3D printing multifunctionality:
structures with electronics. Int. J. Adv. Manuf. Technol. 72, 963–978 (2014)
8. Shih, B., Christianson, C., Gillespie, K., Lee, S., Mayeda, J., Huo, Z., Tolley, M.T.: Design
considerations for 3D printed, soft, multimaterial resistive sensors for soft robotics. Front.
Robot. AI. 6, 30 (2019)
9. Zhou, J., Chen, S., Wang, Z.: A soft-robotic gripper with enhanced object adaptation and
grasping reliability. IEEE Robot. Autom. Lett. 2, 2287–2293 (2017)
10. Manti, M., Hassan, T., Passetti, G., D’Elia, N., Laschi, C., Cianchetti, M.: A bioinspired soft
robotic gripper for adaptable and effective grasping. Soft Robot. 2, 107–116 (2015)
11. Shintake, J., Cacucciolo, V., Floreano, D., Shea, H.: Soft robotic grippers. Adv. Mater. 30,
1707035 (2018)
12. Wang, Z., Torigoe, Y., Hirai, S.: A prestressed soft gripper: design, modeling, fabrication,
and tests for food handling. IEEE Robot. Autom. Lett. 2, 1909–1916 (2017)
13. Naghdy, F., Esmaili, M.: Soft fruit grading using a robotics gripper. Int. J. Robot. Autom. 11,
93–101 (1996)
14. Hao, Y., Wang, T., Ren, Z., Gong, Z., Wang, H., Yang, X., Guan, S., Wen, L.: Modeling
and experiments of a soft robotic gripper in amphibious environments. Int. J. Adv. Robot.
Syst. 14, 1729881417707148 (2017)
15. Licht, S., Collins, E., Mendes, M.L., Baxter, C.: Stronger at depth: jamming grippers as deep
sea sampling tools. Soft Robot. 4, 305–316 (2017)
16. Yang, Y., Chen, Y.: Innovative design of embedded pressure and position sensors for soft
actuators. IEEE Robot. Autom. Lett. 3, 656–663 (2018)
17. Schmitt, F., Piccin, O., Barbe, L., Bayle, B.: Soft robots manufacturing: a review. Front.
Robot. AI. 5, 84 (2018)
18. Melnykowycz, M., Koll, B., Scharf, D., Clemens, F.: Comparison of piezoresistive
monofilament polymer sensors. Sensors 14, 1278–1294 (2014)
Solution Approaches and Process Concepts
for Powder Bed-Based Melting of Glass
Susanne Kasch1(&), Thomas Schmidt1, Fabian Eichler2,

Laura Katharina Thurn2, Simon Jahn1, and Sebastian Bremen2
1
ifw Jena– Günter-Köhler-Institut für Fügetechnik
und Werkstoffprüfung GmbH, Jena, Germany
skasch@ifw-jena.de
2
Faculty of Mechanical Engineering and Mechatronics,
University of Applied Sciences Aachen, Aachen, Germany
Abstract. In the study, the process chain of additive manufacturing by means

of powder bed fusion will be presented based on the material glass. In order to
reliably process components additively, new concepts with different solutions
were developed and investigated.
Compared to established metallic materials, the properties of glass materials
differ significantly. Therefore, the process control was adapted to the material
glass in the investigations. With extensive parameter studies based on various
glass powders such as borosilicate glass and quartz glass, scientifically proven
results on powder bed fusion of glass are presented. Based on the determination
of the particle properties with different methods, extensive investigations are
made regarding the melting behavior of glass by means of laser beams. Fur-
thermore, the experimental setup was steadily expanded. In addition to the
integration of coaxial temperature measurement and regulation, preheating of
the building platform is of major importance. This offers the possibility to
perform 3D printing at the transformation temperatures of the glass materials. To
improve the component’s properties, the influence of a subsequent heat treat-
ment was also investigated.
The experience gained was incorporated into a new experimental system,
which allows a much better exploration of the 3D printing of glass. Currently,
studies are being conducted to improve surface texture, building accuracy, and
geometrical capabilities using three-dimensional specimen.
The contribution shows the development of research in the field of 3D
printing of glass, gives an insight into the machine and process engineering as
well as an outlook on the possibilities and applications.
Keywords: Glass powder Laser processing Additive manufacturing

Melting L-PBF Fused silica Borosilicate
1 Introduction
The laser-based powder bed fusion (L-PBF) process offers an alternative to already
established manufacturing processes due to the comparable component properties, the
freedom in geometric design and the production costs in small and medium series [1].

https://doi.org/10.1007/978-3-030-54334-1_7
Solution Approaches and Process Concepts for Powder Bed-Based Melting 83
Laser-based powder-bed fusion is already an industrially established process for the

manufacturing of three-dimensional components from polymers and metallic materials.
Sintering of ceramics and glass is increasingly being investigated [2]. Laser sintering of
glass powders is an effective technology for manufacturing products for micro-
electromechanical systems or objects for medical applications [3]. Glass is a material
with a wide variety of compositions. The characteristics of glasses can be adjusted very
variably, which enables many applications of this material group [4]. The conventional
production of glass powders involves the steps of melting the glass composition from a
batch, glass fritting, grinding and sieving the glass powders into the required particle
size and particle size distribution. This produces non-spherical, irregularly shaped
particles. In addition to the geometric properties (particle size, particle shape) of the
glass powders, the mechanical-physical properties (e.g. flowability, absorption/
transmission versus the laser wavelength) are essential for processing by means of
powder bed fusion with laser beams. Furthermore, the process control has to be adapted
to the thermophysical properties of the glasses such as poor heat conduction or low
thermal shock resistance depending on the thermal expansion coefficients.
The present article examines different glass powders along the L-PBF under these
aspects in order to provide potential users with a systematic approach as well as ma-
terial- and application-specific processing concepts.
2 Fundamentals
The process of laser-based powder bed fusion (L-PBF) is characterized by a multitude

of influencing factors. The influencing parameters are differentiated according to [5]
with regard to the material properties and the laser and process conditions. The material
properties are described in
• Optical properties
• Thermal properties
• Technological properties of the powders
and the laser and process conditions in
• Laser parameters
• Exposure parameters
• Process environment and
• Machine control
The process effectiveness depends on the main influencing factors mentioned,
whereby a significant influence is determined by the properties of the used powders and
the absorption or transmission of the used laser radiation. A targeted optimization of the
overall process must therefore always be considered dependent on the material and
laser wave.
The absorption of the laser radiation in the powder provides the energy required to
melt the powder by means of L-PBF. According to [6], the energy absorption on or in a
powder layer is in general significantly higher compared to absorption on a compact
solid of the same material. For different one-component powders, N. Tolochko carried
84 S. Kasch et al.
out investigations on absorption with two different wavelengths [7]. In the investiga-
tions laser beams from a Nd:YAG and a CO2 laser source were coupled into an
experimental plant. For SiO2 glass powder an absorption for k = 1.06 µm of A = 0.04
and for k = 10.6 µm of A = 0.96 was determined. For an SiO2 glass component,
comparable absorption and transmission values (Fig. 1) are shown for the CO2 laser
wavelength of k = 10.6 µm and the wavelength of a solid-state laser with approxi-
mately 1 µm. On the other hand compact soda-lime or borosilicate glasses, have a
higher absorption rate for wavelengths smaller than 2 µm, due to their material com-
position, which probably also applies to the absorption for these glass powders.
Fig. 1. Transmission spectrum of different compact glasses
Compared to crystalline solids (e.g. metals), solid glass components are amorphous.
The glassy state is classically described as the frozen state of a supercooled liquid,
whose property is characterized by the temperature-dependent viscosity behavior.
Depending on the glass composition, characteristic fix-points [8] must be observed
throughout the entire temperature and viscosity curve during glass production and glass
processing. Methods to analyze this temperature-dependent behavior are differential
scanning calometry (DSC), differential thermal analysis (DTA) or high temperature
microscopy (HTM) of glass powder samples or compacts. The values thus determined
provide information for processing the L-PBF process.
Defined coating of the powders by means of a blade (e.g. rubber lip,) metal blade on
a building platform and their defined lowering after laser irradiation (exposure) perform
the building process in L-PBF. The processability of powders during coating depends
essentially on particle shape, particle size and particle distribution, which can be
determined experimentally by technological parameters bulk density and flow behavior.
Bulk density directly relates to the compaction of a powder, i. e. how strongly it is

compressed (solidification stress). As the solidification stress increases, the bulk density
increases and the void volume between the particles decreases. With fine-grained bulk
solids, the bulk density is usually more strongly influenced by the solidification stress.
The flow properties of powder particles also depend on the particle shape, particle size
distribution, the chemical composition of the powder particles, but also on the tem-
perature as well as on the humidity. The particle shape has a decisive influence on the
flow behavior of a powder. In theory, smooth, round particles larger than 0.5 mm flow
more easily than rough, spherical particles. The adhesive forces between the particles
are also responsible for the flow behavior of powders [9]. Due to their larger specific
surface area, finer powders exhibit higher adhesive forces and change the flowability to
lower values [10].
The particle size distribution, in addition to the compaction and flow of a powder,
also has a significant influence on the sintering or melting behaviour in the L-PBF
process. A homogeneous powder mass distribution ensures a homogeneous energy
input during laser exposure and thus a homogeneous melting of the material. For this
reason, a constant particle size should be achieved. In practice, an average diameter
with a certain deviation is usually achievable [11].
3 Experimental Investigations and Results
3.1 Characterization of the Glass Powders

Various glass powders with different compositions and geometric, thermo- and
mechanical-physical properties were available for the investigations. First, an extensive
analysis of the initial state of the glass powders was carried out to select glass powders
with good processing properties for the experimental investigations in the L-PBF
process. The following methods were used to determine the properties:
• Geometric properties
– Powder geometry (Grain shape microscopically using a scanning electron
microscope (REM) JSM-6300 with energy dispersive X-ray spectroscopy
(EDX) (JEOL, Japan; EDX: NORAN Instruments, USA)
– Particle distribution by means of laser diffraction LS230 (Beckman Coulter)
• Mechanical and physical properties
– Flowability (according to DIN EN ISO 4490) with Hall flowmeter
– Tab density (according to DIN EN ISO 3953)
– Coating quality manually by means of defined doctor blade and
– Coating system SLM 50 by ReaLizer
• Thermo-physical properties
– Temperature-dependent viscosity behavior using a high-temperature microscope
HTM, differential scanning calometry (DSC) and differential thermal analysis
(DTA)
• Optical properties,
– Absorption measurement
86 S. Kasch et al.
Based on these investigations, two SiO2 glass powders and one borosilicate glass
powder were selected for their technological suitability. These are SiO2 powders from
the company QSIL with the designation “NC4A” (GP4) as well as a powder from the
company Heraeus with the designation “Zandosil” (GP6). Both powders consist of
more than 99.999% silicon dioxide and have a low coefficient of expansion of
a = 0,5 * 10–6 K−1. The manufacturer specifies further thermal properties for GP4 as
follows: softening limit 1730 °C, transformation range from 1075 °C to 1210 °C,
processing range from 1700 °C to 2100 °C. The thermal conductivity of this material is
1.38 W/m * K and is significantly lower than that of classic metallic materials. For
powder materials, the thermal conductivities are correspondingly lower up to a factor of
100 [12].
A borosilicate glass powder (GP10) is used from the company Schott AG, which
was specially ground from borosilicate flat glass “BOROFLOAT® 33”. The manu-
facturer specifies the characteristic viscosity dependent fix points as follows: Working
Point (104 dPa * s) with 1270 °C, Softening Point (107.6 dPa * s) with 820 °C,
transformation temperature (Tg) with 525 °C. The specific thermal conductivity of the
material is 1.2 W/m * K and the thermal expansion coefficient of a = 3.25 * 10–6 K−1
is higher compared to SiO2 and thermal stresses can be reduced less easily.
Glass powders are usually present as broken grains due to their production by
crushing and grinding (Fig. 2). Spherical glass powders can be produced by using
complex technologies and are therefore more expensive and not available for all glass
compositions and grain sizes. Powders GP4 and GP10 are characterized by an irregular,
angular broken grain. Powder GP10 also has a high proportion of fines, sticking on a
larger surface, which can be seen in the SEM image by the light particles on the powder
grains. Within the scope of powder characterization, a powder size distribution for GP4
of 137 µm to 340 µm and for GP10 of 115 lm to 250 lm was determined. The SiO2
glass powder GP6 was produced by a special process is spherical and amorphous with
a grain size of 141 µm to 434 µm. The particle distribution of the powders mentioned
corresponds to a Gaussian-like distribution, which is also typical for metallic powders
in the L-PBF process. Based on these results, the minimum layer thickness to be
applied was determined experimentally (see Table 1). It is assumed that the better the
flowability and tapping density of the powder, the smaller the layer thickness that can
be set.
GP 4 – SiO2 glass (QSIL), bro- GP6 – SiO2 glass (Heraeus), GP 10 – Borosilicate glass
ken grain, angular and angular, round grain (SCHOTT), broken grain, vari-
no fines ous angular geometries, high fine
grain content
Fig. 2. REM images of various glass powders

For the technological processing of the glass powders in the powder bed, the
spreading and the layer thickness is of great importance. Therefore, within the scope of
the experimental investigations, the glass powders were first examined with a hand
operated blade and adjustable gap size, homogeneity of the coating application and the
coating thickness. The SLM 50 coating system by ReaLizer was then used to
mechanically process relevant powders and to evaluate the quality of the result,
depending on the layer thickness by means of visual assessment. Figure 3 left shows an
example of the homogeneous coating application for the glass powder GP10 with a
layer thickness of 200 µm, which can be rated as good for this material. In opposite
Fig. 3 right shows exemplarily an inhomogeneous coating result of a borosilicate glass
mainly due to particle size distribution between d10 = 1 µm and d90 = 25 µm and
irregular broken shape. A homogeneous coating thickness is achieved for all three
powders with 1 to 1.5 times the average particle size (see Table 1).
Table 1. Geometrical and technological properties of the used powders

Sample Particle size Flowability Tap Applied
number MW D 10 D 50 D 90 Funnel Funnel density coating
[µm] [µm] [µm] [µm] Ø Ø [g/ml] thickness
2.5 mm 5 mm [µm]
[s] [s]
SiO2 232 137 228 340 95.4 18.42 1.39 300
(GP4)
SiO2 277 141 264 434 194.4 24.90 0.81 400
(GP6)
B3.3 176 115 175 250 148.2 21.30 1.15 200
(GP10)
Fig. 3. Result of single coating: left: Borosilicate glass powder GP10 with SLM 50 by ReaLizer
(substrate platform diameter: 70 mm); right: exemplarily borosilicate glass with smaller particle
size distribution.
88 S. Kasch et al.
Furthermore, the respective processing temperatures are relevant for the thermal
processing. Compared to pure quartz glass, these temperatures are significantly lower
for borosilicate glass. Thermal behavior of pressed borosilicate glass powders is shown
in Fig. 4 at typical processing temperatures. They show the compression or the start of
melting of the glass powder. It can be seen that above the softening point of 820 °C at
900 °C there is a clear shrinkage of the powder compact and above 975 °C melting can
be observed. These results were confirmed by DSC/DTA. In the L-PBF process these
temperatures determine the adjusted quality of the glass component (sintered or
melting) and its later application.
Fig. 4. HTM images – borosilicate glass powder GP10
The optical properties of the glass powders GP4, GP6 and GP10 were determined
with respect to absorption versus the wavelength k = 1.064 µm (Table 2). The deter-
mined values show a good agree with the values in the literature [7]. For SiO2 powders
with broken grains, a slightly higher absorption is determined than for round grains,
which is due to the shape of the grain and the associated scattering and reflection
behavior. Borosilicate glass powder shows an even better absorption than pure SiO2
glass powders, which is probably due to the glass composition. Depending on the
manufacturer, borosilicate glass consists of up to 80% silicon dioxide (SiO2), 13%
boron trioxide (B2O3), 8% alkali oxides (sodium oxide Na2O; potassium oxide K2O),
7% alumina (Al2O3) and 5% alkaline earth oxides (CaO, MgO, …) [8].
Table 2. Absorption characteristics for k = 1,064 µm

Sample number Material Absorption
GP4 SiO2, broken grain 0.0957
GP6 SiO2, round grain 0.0825
GP10 Borosilicate glass, broken grain 0.137
3.2 Experimental Investigation of the L-PBF Process

3.2.1 Systems Engineering
The experimental investigations were carried out on two different systems, a com-
mercial SLM 50 by ReaLizer with a Nd:YAG laser (manufacturer: IPG Laser GmbH)
of the wavelength k = 1.064 µm and an experimental CO2 laser plant with different
sources (SYNRAD 57-1 series Pmax = 100 W, FEHA Pmax = 1200 W) of the wave-
length k = 10.6 µm.
The SLM 50 is an encapsulated system with a heatable copper platform, which can
be preheated up to 190 °C during the process, and a coater, which distributes the
powder on the platform.
The experimental CO2 laser plant was further developed during the investigations to
meet the requirements of powder processing of glass. Figure 5 shows the schematic
principle. To ensure a constant glass temperature during processing, a pyrometric
control system was integrated into the plant. For this purpose, the measuring beam of the
pyrometer is coaxially superimposed on the laser beam so the glass temperature is
recorded directly in the laser action zone. This prevents partial evaporation of the glass
and ensures a constant temperature during the entire construction period. The powder is
applied via a pneumatically movable hopper, through whose slotted opening the powder
can flow to the building platform. This slot is sealed with a temperature-resistant lip all
around.
Fig. 5. Schematic setup of the experimental CO2 laser plant, coater and inductive heating
A challenge with L-PBF processing of glass powders is the required heating of the
building platform to the range of the transformation temperature of the glass. For GP4
approx. 1000 °C, for GP10 approx. 500 °C are required. In the experimental plant
heating is carried out indirectly by induction. A conductive ceramic under the building
platform provides the necessary temperature transfer. This allows high temperatures to
be quickly achieved without contact but leads to an extremely high load on the entire
plant. Therefore, the construction of the plant and the selection of suitable materials is a
great challenge.
90 S. Kasch et al.
3.2.2 Investigations and Results for Melting Glass Powder

The aim of the investigations was to melt the glass powders GP4, GP6 and GP10 using
the laser wavelengths 1.064 µm and 10.6 µm in the L-PBF process and to produce
compact glass components. Laser and process parameters such as laser power, beam
diameter (focus position), scan speed, track pitch and scan strategy were varied
(Table 3). The temperature of the installation space was kept constant according to the
equipment used. In the investigations with the experimental station, the temperature of
the platform was adjusted depending on the transformation temperature of the glass
powders. The temperature-dependent laser power control was adjusted to the melting
temperature of the glass powders that was determined in the HTM investigations.
Based on the results of the coating tests for powders GP4, GP6 and GP10, the layer
thickness was also adjusted to a constant value.
Table 3. Overview of essential laser and process parameters of the test facilities
Heading level SLM 50, ReaLizer Experimental station
SYNRAD 57-1 FEHA
1200
Laser wavelength k [µm] 1.064 10.6 10.6
Laser power Pmax [W] 100 100 1200
Laser power PL [W] 80–100 10–60 200–800
Scanning speed vs [mm/s] 2–50 1–10 800–1200
Temperature Building platform T [°C] 190 500–1000 500–1000
The investigations with SLM 50 by ReaLizer show that both SiO2 powders (GP4,
GP6) cannot be processed regardless of the powder geometry. By increasing volume
energy densities of up to 1 kJ/mm3 and multiple exposures of identical layers, an
attempt was made to get the necessary heat input into the powders. This should allow
the necessary processing temperatures of about 1585 °C to be reached. Due to the
solid-state laser (Nd:YAG) with a wavelength of 1.064 µm, which is only poorly
absorbed in the SiO2 powder, and the low preheating temperature of the substrate plate
(190 °C), it was not possible to produce any solids.
Although the borosilicate glass powder (GP10) also has low absorption at the fixed
laser wavelength used in this system, the powder with its significantly lower softening
temperature (820 °C) could be processed with the SLM 50 [13]. Significant fusions of
the powder are achievable by a volume energy density of about 200 J/mm3 to
250 J/mm3. Evaluations of these experiments showed that there is a conflict of
objectives between achieving a melt line and maintaining geometric accuracy. With the
aid of higher volume energy densities, better fusions can be achieved. However, this
increases the geometric deviations as a result of higher amount of particles sticking on
the surface powder. This is due to the low thermal conductivity of borosilicate glass
powder and the resulting heat accumulation in the powder bed. In addition, tests with
different scanning strategies showed that the best results can be achieved by scanning
the outer contour and the hatch. In the tests, parameters could be determined to produce
test specimens for materials testing and improvement of quality (porosity, surface,
tightness) by a subsequent heat treatment (Fig. 6).
Before
contour
Hatch
Fig. 6. Test specimen of borosilicate glass (GP10), diameter 10 mm, height 10 mm, wall
thickness 0,5 mm
Investigations with different CO2 laser systems at the experimental plant showed
that the two quartz glass powders GP4 and GP6 can be processed due to their almost
100% absorption for the wavelength of a CO2 laser. Compared to metal powders, glass
powders have a lower thermal conductivity by a factor of 100. This makes it more
difficult to bond the upper powder layers to the layers below. For this reason, the glass
powder must be converted to its molten state with high energy densities in order to
achieve fusing. Depending on the exposure strategy, beam diameters of 2 to 3 mm
were used to achieve the necessary processing temperature of 1700 °C to 2100 °C and
to establish the fusing. Two exposure strategies were investigated:
• Slow, progressive exposure due to small beam movement:
– Speed: 1 to 10 mm/s
– Laser power: 10 to 60 W
– Volume energy density: 5 to 7 kJ/mm3
• Number of scans: 1quasi-simultaneous exposure through high beam movement:
– Speed: 800 to 1200 mm/s
– Laser power: 250 to 800 W
– Volume energy density: >30 kJ/mm3
– Number of scans: approx. 1000
The construction rate for circular geometries could be increased considerably by the
quasi-simultaneous exposure. However, this requires an energy density that is 5 times
higher. Furthermore, the experiments showed that SiO2 solids (Fig. 7, Ø 12 mm, height
92 S. Kasch et al.
Fig. 7. Test specimen from SiO2 glass (left: GP6, right: GP4) diameter 12 mm, height 30 mm
30 mm) can be produced, which have a density of approximately 2.2 g/cm3 and are
comparable with literature values. The component itself has a milky appearance,
because during the construction process, in addition to the melting trace, other particles
adhere and do not melt completely. The interior of the wall (approx. 1.5 mm) is fused
compactly (Fig. 8).
Fig. 8. Test specimen from SiO2 glass (GP4), wall thickness 1,5 mm
Fig. 9. Test specimen from SiO2 glass (GP4), left: macroscopic view, right: CT-scanning
Furthermore, the influence of inductive platform/building space heating was inves-

tigated (Fig. 9). At a temperature of 600 °C a homogeneous, pore-free glass body
(10 10 10 mm3) made of SiO2 (GP4) was produced. In comparison to the conven-
tionally produced substrate plate, no differences are optically detectable. The specimen
appears as transparent as the substrate plate. Only in the edge area up to approximately
1.5 mm no complete fusion of the glass particles on the test specimen took place.
4 Summary and Outlook
This paper presents investigations for laser beam powder bed fusion of borosilicate and
quartz glass powders. Different glass powder classes were characterized with respect to
their geometric, thermo- and mechanical-physical properties and the technological
processability in the L-PBF process using different laser wavelengths was investigated.
The plant technology was adapted to the different requirements of glass processing,
especially for quartz glass powders with CO2 laser radiation. In contrast to metallic
materials, the mentioned properties of glass materials, which make glass the relevant
and technically usable material that it is, present challenges to processability by L-PBF.
The poor absorption rate, especially for wavelength of 1.064 µm, in combination with
the high temperature resistance and poor heat conduction are the technical barriers to be
overcome by this basic research.
94 S. Kasch et al.
In order to achieve the goal of a complete fusion of the materials and to provide
potential users with a systematic approach as well as material- and application-specific
processing concepts test specimens were produced and evaluated. For the successful
production of test specimens from the different glass powders, process parameters were
determined, and initial quality evaluations were carried out. In the tests the basic
processability of the glass powders was proven. Glass specimen could be manufactured
successfully from borosilicate and quartz glass powder by laser radiation. As a result of
the experiments, parameter windows could also be diverted, within which processing is
possible. It was found that the parameters strongly depend on the geometric shape of
the specimen due to heat agglomeration. The qualification of process and machine
parameters is therefore much more difficult than with metallic materials.
In further investigations, however, the process and experimental setup must be
further optimized. To improve the component quality, further investigations on post
heat treatment (e.g. stress relief) are to be carried out. Furthermore, the analysis of the
material component quality with regard to porosity, roughness and tightness in relation
to the developed L-PBF parameters is to be continued and possible further fields of
application are to be shown.
Acknowledgements. The investigations were carried out in the project “Application limits in
melting of glass materials” (Aif-IGF 19673 BG), funded by the Federal Ministry of Economics
and Energy on the basis of a resolution of the German Bundestag.
References
1. Küsters, Y., et al.: Robuster Strahlschmelzprozess durch methodische Parameterfindung.
RTejournal 8 (2011)
2. Luo, J., Pan, H., Kinzel, E.: Additive Manufacturing of Glass. J. Manuf. Sci. Eng. Am. Soc.
Mech. Eng. 136, 061024 (2014)
3. Tolochko, N.K., Arshinov, M.K., Arshinov, K.I., Ragulya, A.V.: Laser sintering of SiO 2
powder compacts. Powder Metall. Met. Ceram. 43(1/2), 10–16 (2004)
4. Scholze, H.: Glas - Natur, Struktur und Eigenschaften. Springer, Heidelberg (1988)
5. Bliedtner, J., Müller, H., Barz, A.: Lasermaterialbearbeitung: Grundlagen - Verfahren -
Anwendungen – Beispiele. Hanser, München (2013)
6. Hügel, H.: Laser in der Fertigung - Strahlquellen, Systeme, Fertigungsverfahren, 2. Auflage.
Vieweg+Teubner Verlag, Wiesbaden (2009)
7. Tolochko, N.: Absorptance of powder materials suitable for laser sintering. Rapid
Prototyp. J. 6, 155–160 (2000)
8. Vogel, W.: Glaschemie, 3. Auflage. Springer, Heidelberg (1992)
9. Schulze, D.: Pulver und Schüttgüter: Fließeigenschaften und Handhabung, VDI-Buch, 2.,
bearb. Aufl. Springer, Heidelberg (2009)
10. Schatt, W.: Pulvermetallurgie - Technologien und Werkstoffe, 2. Auflage. Springer,
Heidelberg (2007)
11. Fateri, M., Gebhardt, A., Thuemmler, S., Thurn, L.: Experimental investigation on selective
laser melting of glass. Phys. Procedia 56, 357–364 (2014)
12. Gusarov, A.V., Smurov, I.: Modeling of interaction of laser radiation with powder bed at
selective laser melting. Phys. Procedia 5, 381–394 (2010)
13. Eichler, F., Skupin, M., Thurn, K., Kasch, S., Schmidt, T.: Operating limits for beam melting
of glass materials. In: Vortrag: The 14th International conference on modern technologies in
manufacturing. Department of Manufacturing Engineering (DME) – TU Cluj-Napoca and
DiCoMI, Cluj-Napoca (2019)
Additive Manufacturing of Ti-Nb Dissimilar
Metals by Laser Metal Deposition
Di Cui1(&), Briac Lanfant1, Marc Leparoux1, and Sébastian Favre2

1
Laboratory for Advanced Materials Processing, Empa-Swiss Federal
Laboratories for Materials Science and Technology, 3602 Thun, Switzerland
di.cui@empa.ch
2
Medtronic Europe Sàrl, 1131 Tolochenaz, Switzerland
Abstract. Conventional technologies for joining dissimilar metals have become

insufficient, as the need for designing and fabricating products with complex
shape and integrated composition variation have arisen. Laser metal deposition
(LMD), a powder injecting additive manufacturing (AM) technology, is capable
to build complex geometries and tailor material composition locally within one
single workpiece. In this work, thin walls transitioning from titanium to niobium
were made by LMD with controlled injection of Ti and Nb powders. The mor-
phologies and microstructures were observed with optical microscopy (OM) and
scanning electron microscopy (SEM). The cross-sections showed fully dense
deposition, without cracks in any of the transition area. Deposition of Nb powder
resulted in partially melted Nb particles embedded in the transition area, which
was a result of the significantly higher melting point of Nb. Energy-dispersive
spectroscopy (EDS) confirmed metallurgical bonding at the transition areas and
showed variation of composition along the build direction. In the transition area,
microhardness was 204 ± 5 HV at the Ti-rich side, 155 ± 6 HV in the solid
solution, with an atomic composition of Ti70Nb30, and 120 ± 16 HV at the Nb-
rich side. Electron backscattered diffraction (EBSD) results revealed hcp struc-
ture in the pure Ti region and bcc structure in the transition and pure Nb regions.
Columnar growth was revealed in the pure metal regions and equiaxed growth in
the transition region. X-ray computed tomography (X-CT) showed 3D element
distribution and revealed very small number of pores in the transition area, which
were not observed by previous microscopy on cross-sections.
Keywords: Dissimilar metals Laser metal deposition Metallurgical

bonding Microhardness EBSD
1 Introduction
Laser additive manufacturing (AM) is a solid freeform manufacturing technology. It

enables direct fabrication of detailed work pieces by accurately depositing desired
material at set positions within a pre-determined domain [1]. Laser metal deposition
(LMD) is an advanced powder-injecting laser AM technology capable of directly
producing dense metal parts with complex geometries, and, of special interest, variation

https://doi.org/10.1007/978-3-030-54334-1_8
Additive Manufacturing of Ti-Nb Dissimilar Metals by LMD 97
of composition through the control of the powder type injection. A variety of materials
can be manufactured by LMD for numbers of applications in the fields of medical
device, automobiles and aerospace.
Despite the lower flexibility in geometry complexity compared to powder bedfu-
sion technologies, LMD has the advantage of mixing at least two different powders
with the desired composition to synthesize alloys in-situ [2], metal matrix nanocom-
posites [3] and functionally graded materials (FGM) [4]. Many binary systems have
been studied to produce FGM with LMD. Ti + Ta deposition by LMD was carried out
on Ti6Al4V substrate [5]. An increase of Ta concentration resulted in an increase of
minimum laser power for successful deposition and resulted in good biocompatibility,
Young’s modulus and 0.2% offset yield strength. Ti6Al4V-Inconel 625 FGM was
fabricated but cracks appeared due to formation of brittle intermetallic phases such as in
the Ti-Ni and Ti-Fe material systems [6]. Schneider-Maunoury et al. reported a study
on LMD of Ti-Nb samples with several incremental increases of Nb content [7]. Many
unmelted Nb particles were observed in the produced bimetal part. Microhardness of
the samples decreased with the increase of Nb content. The lowest elastic modulus of
58 ± 8 GPa was found at the composition of Ti40Nb wt%. The adaptation of the
elastic modulus of Ti-Nb parts to the human bone (10–30 GPa) is of prime interest for
orthopedic parts and additive manufacturing offers additionally a customization to the
individual patient.
This paper summarizes the preliminary results on the fabrication of Ti-Nb
assemblies using LMD process with separate injection of Ti and Nb powders. The
focus of this study is the investigation of the interfacial zones between these two
materials and the occurrence of cracks depending of main process parameters.
2.1 Materials
The powders utilized were Cp-Ti grade 1 powder (oxygen < 0.08 wt%, Hall flow 0.1 in.
23 s) with a diameter around 45–106 lm supplied by AP&C Advanced Powders &
Coatings Inc., Canada, and AMPERTEC MAP Nb powder (oxygen = 373 ppm, Hall
flow 0.1 in. 13 s) with a diameter of 63–100 lm, supplied by H.C. Starck Tantalum and
Niobium GmbH, Germany. Powder particles of both titanium and niobium were pri-
marily spherical, with minor satellites (Fig. 1), exhibiting desirable flowability for
powder transportation during the LMD process. Pure niobium and titanium bulk foils
with various thicknesses have been used as substrates.
98 D. Cui et al.
Fig. 1. SEM images of titanium (left) and niobium (right) powder particles.
2.2 Fabrication of Thin Walls

A commercial LMD machine (Mobile 1.0, BeAM, France) has been used to build the 3D
structures. This LMD system uses a continuous wave (CW) fiber laser with a maximum
power of 500 W operating at a wavelength of 1068 nm (YLR-Series, IPG Photonics,
USA) as the heating source. The laser has a focal spot diameter of 800 µm and a Rayleigh
range of 18 mm. The powders are feed through a volumetric powder feeder (Medicoat,
Switzerland) with two powder containers. The conventional microscale powders are then
transported by a carrier gas (argon) through two tubes which are joined by a Y junction
into one single line connected to the processing head. A specific conical nozzle focuses the
powder jet at 3.5 mm below its exit at the same position as the laser focus. The processing
head is mounted on a 3-axis system (x, y, z) and the substrate holder has 2 rotative axes.
The printing process takes place in an airtight chamber that offers possibility to work
under low level of oxygen, which is crucial for processing materials with high affinity of
oxygen like titanium. Images of the system and schematic of the nozzle are presented in
Fig. 2.
Fig. 2. Laser metal deposition (LMD) machine: (a) dual-container powder feeder; (b) interior of
the printing chamber; (c) schematic of the coaxial nozzle.
4 Groups of samples have been fabricated as illustrated in Fig. 3:

G1 - deposition of a Ti wall on a Nb foil (200 lm);
G2 - deposition of a Nb wall on a Ti foil (300 lm);
G3 - deposition of Ti followed by Nb, using a Ti grade 23 plate (4 mm) as substrate;
G4 - deposition of Ti and then a mixing area Ti+Nb followed by a pure Nb structure
using a Ti grade 23 plate (4 mm) as substrate.
All of the samples have been produced under argon (O2 < 30 ppm, H2O < 170
ppm, argon filling overpressure 3 mbar).
Fig. 3. Schematics of the printed structures.
The targeted dimensions of the samples were lengths of 6 mm, and a height of
4 mm for the pure Ti and Nb walls of samples G1, G2 and G3. The pure material
structures were targeted to be 2 mm high with an intermediate mixing zone of 4 mm
for the G4 sample. The z step-increment of the nozzle, i.e. designed layer thickness, has
been fixed at 0.2 mm for all samples.
Beside the laser power, the linear axis moving speed and the powder feedrates were
varied. For comparing the used conditions, a linear energy is defined as power divided
by speed, reflecting input laser energy per unit length. Accordingly, a linear feedrate is
defined as feedrate divided by speed, which reflects input powder mass per unit length.
Finally, energy per feed is defined as power divided by feed rate, reflecting input laser
energy per input powder mass. The printing parameters were selected according to a
previous study performed with Titanium powder [3] and are summarized in Table 1
(samples G1 to G3) and Table 2 (sample G4). For all the G3 samples, a power of
325 W, speed of 2000 mm/min and feed rate of 3.6 g/min have been employed to
deposit the first 4 mm height of Ti on the substrate. A delay of 10 s was then necessary
to switch the powder feeder from pure Ti to Nb and to stabilize the powder flow before
depositing the next 4 mm of Nb on top of the as-deposited Ti using the parameters
presented in Table 1.
100 D. Cui et al.
Table 1. Printing parameters for G1, G2 and G3

Group Sample Power Speed Feedrate Linear Linear Energy
(W) (mm/min) (g/min) energy feedrate per feed
(J/mm) (mg/mm) (J/mg)
G1 1 325 2000 3.9 9.8 2.0 4.9
2 360 2000 3.9 10.8 2.0 5.4
G2 1 150 1000 6.0 9.0 6.0 1.5
2 188 1000 6.0 11.3 6.0 1.9
3 213 1000 6.0 12.8 6.0 2.1
4 248 1000 6.0 14.9 6.0 2.5
5 94 500 6.0 11.3 12.0 0.9
6 281 1500 6.0 11.3 4.0 2.8
7 375 2000 6.0 11.3 3.0 3.8
8 281 1500 7.0 11.3 4.6 2.3
9 375 2000 7.0 11.3 3.5 2.8
G3 1 375 1200 6.0 18.8 5.0 3.8
2 425 1200 6.0 21.3 5.0 4.3
3 475 1200 6.0 23.8 5.0 4.8
4 475 1400 7.0 20.4 5.0 4.1
Table 2. Printing parameters for G4

Sample Powder Height Power Speed Feedrate Linear Linear Energy
(mm) (W) (mm/min) (g/min) energy federate per feed
(J/mm) (mg/mm) (J/mg)
G4_1 Ti 2 325 2000 3.9 9.8 2.0 4.9
Ti+ 4 375 1040 1.9+ 21.6 1.8+ 4.2
Nb 3.5 3.4
Nb 2 375 1500 4.8 15.0 3.2 4.7
It should be noted when analyzing the results that the input laser energy is dis-
tributed in four fractions: (1) absorbed by substrate or previously deposited material;
(2) absorbed by powder particles during flight; (3) scattered by powder particles and
directed away; (4) reflected by substrate or previously deposited material.
2.3 Characterization Methods

The produced samples have been cut perpendicular to the laser scanning directions.
The cross-sections have then been embedded in Demotec 10 resin, ground up to grit
2500 SiC grinding paper, and polished with 6 lm and 3 lm diamond pastes and finally
with an OPS solution (0.04 lm SiO2 with H2O2).
Cross-section morphology and microstructure have been examined with an optical
microscope (OM) (ZEISS Axioplan, Germany) and a scanning electron microscope
(SEM–Hitachi S-4800, Japan). An energy-dispersive spectrometer (EDS–Ametek Edax

Octane plus, USA) has been used to investigate the elemental spatial distribution.
Microhardness HV0.2 has been measured with a load of 200 g and a dwell time of 10 s.
An electron backscattered diffraction camera (EBSD–Ametek, USA) has been used to
investigate the grain morphology and the crystalline structure. X-ray computed
tomography (X-CT-RX Solutions EasyTom XL Ultra, France) has been used on one
sample to visualize the elemental distribution in three dimensions and examine for
potential pores.
3.1 Ti Walls Deposited on Nb Foil

Side views and cross-sections of G1 by OM are presented in Fig. 4. Samples G1_1 and
G1_2 exhibited good adherence to the Nb foil. They both have dimensions close to the
programmed ones (L H = 6 mm 4 mm) with a length of 6.5 mm and height of
3.8 mm. Widths of G1_1 and G1_2 are 800 lm and 940 lm, respectively, similar to
Fig. 4. OM images of G1 samples. (a) and (b) are side view and cross-section of G1_1; (c) and
(d) are side view and cross-section of G1_2. Arrows in the lower left caption indicate direction of
laser movement in the horizontal plane and building direction. Purple broken frames indicate
designed dimension of L H ¼ 6 mm 4 mm. White broken lines indicate positions of the
cross-sections in the as-built thin wall.
102 D. Cui et al.
the provided laser beam diameter of around 800 lm. The 18% wall width difference
between the samples is the result of an increase of 11% in laser power, and therefore
energy at the sample level. At the interface with the Nb foil, both Ti walls are slightly
thinner, as a result of enhanced heat dissipation through the metallic substrate. Both of
the walls appear crack-free and dense in the cross-sections, but thermal deformation of
the substrate can be observed with thin foils.
Several SEM images of the interface between Ti wall and Nb foil are presented in
Fig. 5. SEM imaging of the Ti/Nb interface in G1_1 reveals a thin band of intermediate
contrast between Ti wall and Nb foil (Fig. 5(a–c)). Thickness of this band is usually
around 0.5 lm. Occasionally, the band protrudes into the pure Ti part, reaching a
thickness approximately 30 lm (Fig. 5(c)). EDS line analysis shows variation of rel-
ative concentrations of Ti and Nb across the band, confirming metallurgical bonding
with mixing of the two elements in the band.
Fig. 5. (a) SEM image of the interface between Ti wall and Nb foil. (b) The thin band with a
thickness of 0.5 lm. (c) thin band protruding into pure Ti part, reaching a thickness of 30 lm.
(d) EDS line analysis along the yellow broken line across the protrusion; red and green curves
show variation of intensities of NbL and TiK signals respectively, demonstrating their
correspondence with SEM image contrast.
3.2 Nb Walls Deposited on Ti Foil

Samples G2_1 to G2_9 exhibited good adherence to the Ti foil.
Influence of Power. Samples G2_1 to G2_4 have been fabricated with increasing
power while keeping constant the speed and feed rate. Figure 6 shows bottom parts of
cross-sections of the samples.
All Ti foils appear melted through at the bottom of the Nb wall deposition. A higher
energy density is indeed required for depositing Nb compared to Ti due to the higher
melting temperature of Nb (2742 K instead of 1941 K for Ti). Additionally, many
spherical features are present in this transition area. These features are unmelted or
Fig. 6. OM of cross-sections of G2_1 to G2_4 with processing power indicated. Distances

between bold white lines indicate laser beam diameter. Red arrows show spherical features in the
transition area. The inserted image shows grain boundaries in the wall.
partially melted Nb inlet particles as confirmed by EDS analyses. Grain boundaries in

the main body of the pure Nb walls were revealed by polishing.
In the studied parameters range, a power increase induces an increase of the width of
melted part in the substrate, the volume of the transition area and the width of the wall
as previously observed with Ti.
Influence of Proportionally Changing Power and Speed. Samples G2_5, G2_2,
G2_6 and G2_7 have been produced by proportionally increasing power and speed,
while keeping the feedrate constant. The samples have been fabricated with the same
linear energy, but with a decreasing linear feedrate. Figure 7 shows bottom parts of
cross-sections of the samples. Similar to previous samples, all Ti foils are melted and
partially melted Nb particles are observed.
Fig. 7. OM of cross-sections of G2_5, G2_2, G2_6 and G2_7 with processing power and speed
indicated. They all have an identical linear energy of 11.3 J/mm. Distances between bold white
lines indicate laser beam diameter.
In sample G2_5, the Ti foil has been melted and severely deformed, and traces of
individual deposited layers are observed in the wall body, exhibiting poor homogeneity
of the wall. As the linear energy was sufficient to create a melt pool within the used
parameter range, the lower power for G2_5 could have created a cooler melt pool. The
slower speed then allowed this cooler melt pool to exist for longer time thus increasing
the probability of melted metals to flow sideways before solidifying, resulting in the
protrusions on both sides of G2_5. Compared to other samples where power and speed
104 D. Cui et al.
increased proportionally, the width of melted substrate and the volume of transition
area decreased. The width of wall body decreased dramatically, corresponding to the
greatly decreased linear feedrate.
Influence of Feedrate. Samples G2_6 and G2_8 as well as G2_7 and G2_9 have been
produced respectively with the same power and speed, but different feedrates.
Fig. 8. OM of cross-sections of (a) G2_6, (b) G2_8, (c) G2_7 and (d) G2_9. Distances between
bold white lines indicate laser beam diameter.
Figure 8 shows bottom parts of cross-sections of the samples. Full penetration of the
Ti substrate was observed for the lower feedrate 6 g/min while increasing the feedrate
to 7 g/min prevents the full penetration (white arrow). Large Nb particles were
observed in the mixing zone between Nb and Ti, and their quantity appears to be larger
for lower power and speed. The wall thickness and homogeneity is smaller for higher
power and speed but no significant changes were observed depending on the feedrate in
the investigated range.
Unmelted or Partially Melted Nb Particles. For all samples in G2, meaning where
Nb was injected above Ti, Nb particles have been observed in the transition area, but
not in the main body of the walls. This indicates that conditions with the lowest energy
input in any aspect (power down to 94 W, linear energy down to 9.0 J/mm or energy
per feed down to 0.9 J/mg) have been sufficient to melt Nb inlet particles completely.
To explain the presence of Nb particles only in the transition area and not in the wall
body, different effects could be considered:
1) The heat dissipation at the bottom is higher than the upper in the wall body because
of larger heat transfer through the bulk substrate. This induces the necking observed
at the bottom of the walls built on bulk substrates as seen for instance in Fig. 4.
2) The heat accumulated layer by layer induces a higher melt pool temperature in
layers upper in the wall.
3) The boiling point of Nb (5015 K) is significantly higher than that of Ti (3560 K).
Potential boiling or overheating of a melt pool of Ti+Nb mixture could dissipate
more heat, compared to a melt pool of pure Nb which is hot, but not boiling.
Therefore, melt pool of pure Nb can reach potentially much higher temperature than
that of Ti+Nb mixture. Additionally Ti vapors may absorb the infrared laser
wavelength leading to less energy in the melt pool [8].
These above mentioned effects are all related to heat accumulation and dissipation
due to the presence of the substrate. Obviously, these physical phenomena occur not
only individually but are combined promoting the non-fully melting of the niobium
particles. Higher energy input in the system could however overheat the titanium
substrate leading to perforation, extensive evaporation and enhanced instability of the
melt pool. Smaller Nb particle sizes that would need less energy for melting could be
considered for the transition area.
3.3 Deposition Directly Changing from Ti to Nb

A pure Ti wall is first deposited on a Ti grade 23 substrate, and then Nb is deposited
onto it (scheme G3 – Fig. 3). All samples have similar dimensions close to the target
programmed. Figure 9 presents optical micrographs of the cross-sections of the tran-
sition areas between the deposited Ti and Nb walls for the different processing con-
ditions in Table 1. Negligible cracks nor delamination at the interface between the
substrate and the printed structure have been observed. Futhermore, negligible cracking
nor porosity have been observed within the printed structures as well. In all samples,
there is a transition area between pure Ti part and pure Nb part, with a sharp interface at
the pure Ti side but no well-defined interface at the pure Nb side. Several Nb particles
are observed in this transition area, but not in the pure Nb wall side located on the top
of the manufactured structures.
Fig. 9. OM of cross-sections of samples G3_1 to G3_4 (from left to right). The color and
morphology difference is due to different polarization when taking the images. Black dots on
G3_2 are indents for microhardness measurements.
106 D. Cui et al.
X-CT has been done on a part of the G3_3 sample to visualize the 3D elemental
distribution and check for potential pores. This sample and its reconstructed image are
shown on the left in Fig. 10. Nb with a higher atomic number has a brighter contrast in
the reconstructed image, while Ti appears darker. Six horizontal slices of the recon-
structed image (indicated by the dashed lines) are presented on the right side of Fig. 10.
Slice 1 shows fully dense Ti part with a few Nb particles sticking on the side. An area
with intermediate contrast appears in slice 2, indicating emergence of Ti/Nb solid
solution. Slices 3 to 5 are characteristic of the transition from Ti to Nb. Several bright
spheres in these slices correspond to the Nb particles observed in OM images. A few
pores with sizes of tens of microns are observed in slice 4, which were not observed in
cross-section OM images. The stepped contrast in slice 5 shows traces of a few dif-
ferent scanning tracks, indicating stepped composition gradient in the transition area.
Slice 6 shows the dense pure Nb part.
Fig. 10. Photograph of the G3_3 sample part and its reconstructed image by X-CT. 6 slices of
the reconstructed image are shown on the right. There positions are indicated by the white dashed
lines.
Coupled EDS and EBSD have been performed on sample G3_4. The results are
shown in Fig. 11. It can be noticed that, hcp structure (a-Ti) is only present where no
Nb element is detected. However as soon as Nb appears in the area, a bcc (Nb or b-Ti)
phase is observed, regardless of the presence of Ti. This is in agreement with Nb being
a b-stabilizer for Ti. A sharp interface is observed between the transition area and pure
Ti part, as shown in samples presented in Fig. 9. These interfaces could then be
explained by a mismatch of both crystal structures. This interface would be an inter-
esting subject of testing in future mechanical tests. The inverse pole figure (IPF) maps
in Fig. 11 show a tendency of columnar growth in pure Ti and pure Nb parts, and
equiaxed growth in the transition area.
Fig. 11. Coupled EDS and EBSD mapping of Ti and Nb in transition area of G3_4.
EDS line scan, mapping and microhardness measurements have been done on
sample G3_2. The results are shown in Fig. 12. The line scan (Fig. 12 middle) reveals
a basically 3-step transition in the transition area. Immediately above pure Ti part with
an average microhardness of 148 ± 9 HV is a band with a width of 40–100 lm, where
the atomic composition is Ti90Nb10. Microhardness in this area is highest, with an
average of 204 ± 5 HV. Then in the main part of the transition area, An atomic
composition is Ti70Nb30 was observed in the primary part of the transition area. The
atomic ratio converted to mass ratio is Ti:Nb = 55:45, essentially the same as a Ti
grade 36 alloy composition. This area exhibits an average microhardness of 155 ± 6
HV. The embedded Nb particle has an average microhardness of 81 ± 1 HV which
could indicate weak points in future mechanical tests. Further up in the structure, traces
of a Nb-rich track (Nb > 95 at.%) wrapped in the Ti grade 36 were observed, with an
average microhardness of 120 ± 16 HV. Finally, the pure Nb part has an average
Fig. 12. OM image of G3_2 with microindents corresponding HV0.2 numbers. The EDS line
scan was done along the green arrowed line, and atomic composition of Nb and Ti (yellow and
teal curves, respectively) are shown in the chart. Key composition points are directed by red
arrows to corresponding areas in the OM image, indicating a composition-contrast correlation.
EDS mapping of Nb and Ti on the right shows similar element distribution to that of G3_4, with
slightly more Nb particles.
108 D. Cui et al.
microhardness of 100 ± 12 HV. The highest microhardness in the Ti90Nb10 band

could result from finer grains due to crystal structure mismatch, as it is transitioning
from hcp a-Ti to bcc Ti70Nb30.
Unmelted or Partially Melted Nb Particles. For all G3 samples, Nb has been
deposited on as-deposited Ti walls. Therefore, enhanced heat dissipation through the
bulk substrate should not play a role anymore as compared to the deposition of Nb
directly on a Ti substrate (samples type G2). In this configuration, the heat dissipated in
the Ti-Nb mixing zone due to overheating of Titanium could explain the presence of
non-fully melted Nb particles. Above this transition zone, in the pure Nb wall, no non-
fully melted niobium particles could be observed as for G2 samples.
3.4 Deposition Changing from Ti to Nb with a Mixing Area in Between

In sample G4_1, a pure Ti wall has been first deposited on a Ti grade 23 substrate. Nb
powder feeder has then been activated and a waiting time of 10 s has been imposed
before beginning deposition in the mixing area, in order to guarantee a stable flow of
sufficiently mixed powder in the feeding line. After finishing the mixing area, the Ti
feeder was turned off and Nb feeder was adjusted to the desired feedrate, but no waiting
time has then been imposed at this transition. The parameters listed in Table 2 have
been employed to immediately deposit the Nb part of the structure. Figure 13(a) shows
cross-section of the sample G4_1.
Fig. 13. (a) OM image of cross-section of G4_1. (b) Microhardness profile along the building
direction from Ti through the Ti+Nb mixing zone to Nb. Yellow boxes show transitions between
the pure metal and the mixing zone. (c) EDS line scan across the mixing area.
The pure Ti part has dimensions close to the programmed ones. The mixing area is
longer and pure Nb part is shorter than programmed, which is a result of an unstable
powder flow caused by absence of the 10 s waiting time. Many Nb particles were
observed in the mixing area as seen by the sparks of Nb signal in the EDS mapping and in
Fig. 14, but neither crack nor porosity were present. Microhardness measurements in
Fig. 13(b) and EDS line scan along the sample in Fig. 13(c) show results similar to those
from G3_2: a Ti90Nb10 transition band follows immediately after the pure Ti, exhibiting
an average microhardness of approximately 182 ± 1 HV. The primary part of the mixing
area consists of a Ti70Nb30 solid solution as the matrix, with Nb particles embedded (the
abrupt changes on the Nb curve in Fig. 13(c)). The average microhardness of the solid
solution matrix and the embedded particles were measured to be 154 ± 10 HV and
85 ± 1 HV, respectively. The EDS map of Ti close to the Ti-to-Ti+Nb transition in
Fig. 14 shows a slightly higher concentration of Ti between the first and second
deposited mixing layers, which should be the same case for the layers above. Micro-
hardness of these features ranged from 192 to 259 HV.
Fig. 14. OM image of the mixing area next to the pure Ti part (left) and EDS map of Ti of the
same area (right). White arrows point to the features where Ti is slightly richer.
Unmelted or Partially Melted Nb Particles. As for G2 (Nb on Ti bulk) and G3 (Nb

on Ti wall) samples, Nb particles have been observed throughout the mixing area in
G4_1 where Ti and Nb powders were fed together. Here also the overheating of the
mixture Ti+Nb and the resulting heat dissipation could explain the presence of Nb
particles that seem well distributed within one layer and not confined at the interface
between two subsequent layers as reported by Schneider-Maunoury et al. [7]
From this study and the different configurations investigated, it appears that melting
together materials with different melting points is challenging even for a simple
material system leading to a solid solution. It would be interesting to measure then the
temperature of the melting pool in this mixing zone depending on the Nb content. As
proposed previously, a possibility to solve the issue of having unmelted particles would
be to use finer particles for the higher melting point material, here Nb. However,
working with fine powders induced flowability and safety issues. Preheating Niobium
110 D. Cui et al.
starting particles either in the container, the transporting line or in-flight after exiting
the nozzle could be envisaged but would bring in potential fire hazard and add technical
complexity to the system.
4 Conclusions
This preliminary study demonstrates the ability to deposit Titanium on Niobium and
inversely Niobium on Titanium for building 3D structures using a direct energy
deposition process. Additionally, a mixing zone obtained by the separate feeding of
both powders could also be deposited with the additive manufacturing facility, offering
new opportunities for building functionally graded materials.
The microscopic observations of the interfaces between these two metals as well as
of the mixing zone reveal no cracks and a dense structure. The composition of the built
walls varies from one pure metal to the other one through a solid solution that may
present various compositions. The two metals grow with columnar grains whereas the
solid solution shows an equiaxed structure. The hcp phase is only observed in pure Ti
parts and as soon as Nb is present, the crystal structure becomes bcc. Fine grains due to
the sharp transition between these two crystalline phases induces a high hardness.
Even if the energy input was always sufficient to melt completely the niobium
particles, some unmelted Nb particles have been found at the interfaces with titanium in
the mixing zone also far away from the bulk substrate. Further analyses are required to
explain this phenomenon that could be induced by the evaporation of titanium that
cools down the melt pool and absorbs the laser wavelength. These large particles could
be an issue for the mechanical response of the material.
Future activities will then focus on the correlation between the process parameters
and the Ti-Nb microstructures. This would be necessary for tailoring the performances
such as tensile strength, fatigue strength, corrosion resistance and biocompatibility of
the parts for industrial applications.
Acknowledgements. The authors are thankful to Bernhard von Gunten and Peter Ramseier
(Laboratory for Advanced Materials Processing–EMPA Thun) for EBSD sample preparation
EBSD. The authors would also like to thank Xavier Maeder (Laboratory for Mechanics of
Materials & Nanostructures–EMPA Thun) for the assistance with EBSD analysis and Kai
Zweiacker for assistance with X-CT analysis.
References
1. Loh, G.H., Pei, E., Harrison, D., Monzon, M.D.: An overview of functionally graded additive
manufacturing. Addit. Manuf. 23, 34–44 (2018)
2. Yan, L., Chen, X., Li, W., Newkirk, J., Liou, F.: Direct laser deposition of Ti-6Al-4V from
elemental powder blends. Rapid Prototyp. J. 22(5), 810–816 (2016)
3. Lanfant, B., Bär, F., Mohanta, A., Leparoux, M.: Fabrication of metal matrix composites by
laser metal deposition-a new process approach by direct dry injection of nanopowders.
Materials 12, 3584 (2019)
4. Yan, L., Chen, Y., Liou, F.: Additive manufacturing of functionally graded metallic materials
using laser metal deposition. Addit. Manuf. 31, 100901 (2020)
5. Feurst, J., et al.: LASER powder deposition of titanium-tantalum alloy structured interfaces
for use in orthopedic devices. In: Medical Device Materials VI: Proceedings from the
Materials and Processes for Medical Devices Conference 2011, pp. 159–164. ASM
International, Minneapolis (2013)
6. Pulugurtha, S.R.: Functionally graded Ti6Sl4V and inconel 625 by laser metal deposition. Ph.
D. dissertation, Missouri University of Science and Technology, Rolla (2014)
7. Schneider-Maunoury, C., et al.: An application of differential injection to fabricate
functionally graded Ti-Nb alloys using DED-CLAD® process. J. Mater. Process. Technol.
268, 171–180 (2019)
8. Mohanta, A., et al.: Influence of temporal and spectral profiles of lasers on weld quality of
titanium. Opt. Lasers Eng. 134, 106173 (2020)
Investigation of Plastic Freeformed,
Open-Pored Structures with Regard
to Producibility, Reproducibility
and Liquid Permeability
Andre Hirsch1(&), Christian Dalmer2, and Elmar Moritzer1

1
Kunststofftechnik Paderborn (KTP), Direct Manufacturing Research Center
(DMRC), Paderborn University, 33098 Paderborn, Germany
andre.hirsch@ktp.upb.de
2
Paderborn University, Warburger Str. 100, 33098 Paderborn, Germany
Abstract. The Arburg Plastic Freeforming (APF) is an additive manufacturing

process which allows the production of three-dimensional thermoplastic com-
ponents in layers. The components are produced by depositing fine, molten
plastic droplets. The main advantage of the APF is the open-parameter control of
the associated machine system. Thus, the process parameters can be optimized
for individual applications.
A special and new application of the APF is the production of interconnecting
porous structures. As this is a novel approach with this manufacturing process,
the general producibility and reproducibility must first be proven. Therefore, the
relevant process parameters with an influence on the open-pored structures are
identified. The volume of the individual plastic droplets, the distance between
the droplets and the layer thickness are the three decisive influencing factors.
With the use of analysis methods, the free spaces created in the structure are
described by a uniformly constructed, interconnected pore structure. This means
that the pores are interconnected in three dimensions.
Reproducibility is evaluated by repeated production and thru the changed
conditions during the manufacturing process. In addition, the multiplication and
a change of geometry are evaluated in such a way that there is no influence on
the pore size. Irregularities when depositing the first layer are caused by
unevenness of the building platform. A suitable test arrangement is set up to
determine the liquid permeability. A characteristic value is determined to
describe the permeability to liquids.
Keywords: Arburg Plastic Freeforming Porous plastic structures

Producibility Liquid permeability
1 Introduction
The Arburg Plastic Freeforming (APF) is an additive manufacturing process that allows
three-dimensional, thermoplastic components to be produced layer by layer. The
components are generated by depositing fine, molten plastic droplets. One of the main
advantages of the APF process is the open machine control. Thus, the process
parameters can be adapted and optimized for individual applications. In addition, due to
https://doi.org/10.1007/978-3-030-54334-1_9
Investigation of Plastic Freeformed, Open-Pored Structures 113
the open-parameter control, it is possible to process own materials on the corre-

sponding machine system Freeformer.
The objective of this paper is to investigate the general producibility of open-pored
structures using the APF process. Furthermore, influences on the reproducibility of the
manufacturing process are to be identified and their effects are to be shown. With
regard to possible fields of application, a correlation between the manufactured
structures and their fluid permeability shall be investigated.
2 State of the Art
2.1 Arburg Plastic Freeforming (APF)

The Arburg Plastic Freeforming is characterized in particular by the processing of
standard plastic granules as well as by the production of components out of very fine
molten thermoplastic droplets. The associated machine system for this technology is
the Freeformer from Arburg GmbH & Co KG. Its most important machine components
are shown in Fig. 1. The raw material, a qualified standard thermoplastic granule, is fed
via a hopper. In the material preparation unit, the granulate is molten with a screw as in
the injection molding process. The molten material is then pressed into the material
reservoir. Here, a piezo actuator performs a pulsed nozzle closure. The nozzle moves
up and down, producing almost 250 droplets per second. The movement of the building
platform, for the precise positioning of the discharged droplets in the x- and y-direction,
is realized by two linear motors. After the completion of a layer the platform is lowered
by one-layer thickness in z-direction, using a spindle drive [1–3].
Basis: Qualified standard

Piezo actuator performs granule
pulsed nozzle closure
Material preparation with
screw as in injection molding
process
Nozzle
closure
Material reservoir between
screw and nozzle tip is under
y pressure
Part carrier moves the part

Discharge of z along the x- and y-axes and
tiny droplets downwards layer by layer
from the nozzle tip along the z-axis
x © ARBURG
Fig. 1. Schematic setup of the Freeformer [1]

114 A. Hirsch et al.
In the literature there is a large number of publications by the Arburg company with
process descriptions, the advantages of the technology, the available materials and
application examples [2–5]. In [6–8] the resulting mechanical properties of APF
components were investigated and optimized. Furthermore, new approaches for the
production of two-component (hard-soft) parts with interlocking interfaces as well as
for the suitability of the APF process for the processing of Metal-Injection-Molding
granules are found in the literature [9, 10].
2.2 Porosity
Porosity is a structural characteristic of a component and describes the existence of
pores. Pores are classified into open and closed structures. Closed pores are located
within a component. They are completely enclosed within the component so that there
is no connection to the environment. Open pores are further divided into blind and
continuous pores (see Fig. 2). Blind pores have one connection to the environment,
whereas continuous pores have at least two connections to the environment. When a
medium (liquid or gas) flows through a component, the number and geometry of the
pores contributes to the degree of permeability [11, 12].
Continuous Pore
Closed Pore
Blind Pore
Fig. 2. Division of the pore types into open, closed and blind pores
A further classification is based on the size of the pore size. The following size
ranges are differentiated by diameter:
• Micropores: <2 nm
• Mesopores: 2 to 50 nm
• Macropores: >50 nm (0.05 µm)
The porosity of a material is determined using Eq. (1) and describes the per-
centage of free volume:
qSpecimen
¼1 ð1Þ
qReference
qSpecimen : Density of the porous specimen

qReference : Density of the non-porous material (Reference)
2.3 Applications of Open-Pored Structures

An exemplary application of open-pored structures is filtration, where a filter is able to
serve different purposes. If two media are separated from each other, this is called
filtration or separation. The filter retains particles that are larger than the minimum pore
diameter. Furthermore, it is possible to mix different gases with each other (dispersion)
by using branches and a multi-layer structure within the filter element [11].
2.4 Liquid Permeability

The permeability is a material constant and describes the flow-through capability of a
porous material. Permeability provides information about the volume of a permeant
(gas or liquid) that penetrates a barrier of a given thickness and area per unit of time,
provided that there is a partial pressure difference at the interfaces of the barrier [13].
The following three basic equations are used for mathematical description:
• Continuity equation (conservation of mass)
• Law of Darcy (conservation of momentum)
• Thermodynamic equation of state
The continuity equation implies that the sum of all masses flowing in and out is
equal to the mass change of the sample. Darcy’s law specifies the relationship between
the flow velocity within the pores and the potential gradient, considering the height and
acceleration due to gravity. With the help of the thermodynamic equation of state, the
dependence of the density of the permeant on the pressure, at constant temperature, is
included [14]. By combining the basic equations, Eq. (2) defines the permeability (K)
with the unit m2:
V_ g L kf g
K¼ ¼ ð2Þ
q g A Dh q g
_
V: Volume flow
g: Dynamic viscosity of the permeant
L: Length of the barrier
q: Density of the permeant
g: Acceleration of gravity
A: Cross-sectional area of the barrier through which flow occurs
Dh: Difference in height between inlet and outlet
kf : Permeability coefficient
As can be seen from Eq. (2), permeability can be expressed by a second definition,
the coefficient of permeability (kf), with the unit m/s. The formulaic relationship is
described in DIN 18130-1 and is given in Eq. (3) below:
V_ L q g
kf ¼ ð3Þ
A Dp
Dp: Pressure difference between inlet and outlet
Once the permeability coefficient has been calculated, the permeability of water can
be classified according to DIN 18130-1 (see Table 1).
Table 1. Classification of permeability according to DIN 18130-1

Description Permeability coefficient kf [m/s]
Very highly permeable >10−2
Highly permeable 10−2 bis 10−4
Permeable 10−4 bis 10−6
Low permeability 10−6 bis 10−8
Very low permeability 10−8 bis 10−9
Almost completely impermeable to water <10−9
3 Producibility of Open-Pored Structures
The beginning of this study is based on a randomly produced component (see Fig. 3,
left) from the qualification of the material polypropylene (PP) Moplen HP 500 N for
the APF process. The aim of such material qualification is to determine all relevant
process parameters so that components with the required mechanical and visual
properties can be produced. During this process parameter optimization, a porous,
translucent structure was created by changing the layer thickness (see Fig. 3, right).
2 mm 2 mm
Fig. 3. Initial specimen in full size view (left) and with enlargement (right)
The porous structure looked very homogeneous, which lead to the idea of carrying out
detailed investigations. However, due to the unavoidable warpage of the parts caused
by the semi-crystalline PP, the following tests were not carried out with PP, but with
the already qualified material ABS Terluran GP 35.
3.1 Variation of the Layer Thickness (LT)

Initially, the aim was to produce an open-pored structure using the material ABS
Terluran GP35 by varying the layer thickness (LT). For the following investigations all
parameters, except LT, remain unchanged. The layer thickness is increased from
0.15 mm in steps of one hundredth of a millimeter. The evaluation of the measurement
results is shown below in Fig. 4 and 5.
100
Pore width [μm]
80
60
40 FF 1.48
20
0
0.15 0.16 0.17 0.18 0.19 0.20
Layer thickness [mm]
Fig. 4. Pore widths with different LT
100
Pore length [μm]
80
60
40 FF 1.48
20
0
0.15 0.16 0.17 0.18 0.19 0.20
Layer thickness [mm]
Fig. 5. Pore lengths with different LT
The diagrams in Fig. 4 and 5 show that the size of the pores increases with
increasing layer thicknesses. Pore openings with widths between 22 and 82 µm are
produced. The lengths are in a range of 34 to 71 µm. This corresponds to the order of
magnitude of macropores.
Computer tomography (CT) scans shown in Fig. 6 were taken from the specimen
produced with LT = 0.18 mm and FF = 1.48. The CT images show a rectangular
structure of the cylindrical specimen. This change in geometry is the result of the CT
evaluation, in which a rectangular ROI (“Region of Interest”) was cut out of the overall
image.
500 μm 500 μm
Fig. 6. CT scan of the porous structure (left) and its inversion to show the pore volume (right)
The CT scan of the sample shows the principal composition of the porous structure.
The inversion of the volume allows not only the continuous openings in z-direction to
be detected, but also a connection of the pores in transverse direction. There is an
interconnection in all directions in the volume. Because of this, it is not possible to
differentiate between individual pores within the produced samples, instead there is a
continuous free space. If terms are used in the following to describe the pore size or
geometry, they are related to the pore openings visible on the surface.
3.2 Variation of the Form Factor (FF)

The form factor is used to vary the distance between the single droplets and between
the droplet chains (see Fig. 7) [15]. This process parameter is defined by the ratio of the
width (W) to the height (H) of the individual droplets (W/H-ratio). Consequently, the
FF influences the porosity and filling of the components. For the following investi-
gations, only the parameter FF is varied. According to theory, the distance between the
droplets is reduced as the FF decreases. The basic setting for the FF is 1.48 for the
High form factor Low form factor
Deposited
polymer droplets Movement of
the nozzle
Fig. 7. Schematic description of the process parameter form factor [15]

current material and results a fully filled component. The variation of the parameter is
started with a rounded value of 1.50. The factor is then increased in tenth steps.
Subsequently, the created samples are visually examined and categorized.
The porous structures produced by variation of the FF do not differ visually from
the samples produced by variation of the LT. The decreasing stability of the component
with increasing FF is also noticeable in this case. At an FF of 2.50, individual strands of
the last layer built stand out from the rest of the component.
When evaluating the pore geometry, a correlation between pore length/width and
an increasing form factor can be observed (see Fig. 8 and 9). To describe this corre-
lation, the calculation of the correlation coefficient r according to Bravais-Pearson is
used. The correlation coefficient ranges between −1 and 1 and indicates the direction
and intensity of a linear correlation. If r = 1 (r = −1) there is a strong positive (neg-
ative) linear correlation. If r = 0 there is no linear correlation. Due to the resulting
values of r = 0.999 for the pore width and r = 0.995 for the pore length, a positive
linear correlation is assumed.
250
Pore width [μm]
200
150
100 LT 0.15
Linear (LT 0.15)
50
0
1.5 1.6 1.7 1.8 1.9 2 2.1 2.2 2.3 2.4 2.5
Form Factor
Fig. 8. Pore widths with different FF
250
Pore length [μm]
200
150
100 LT 0.15
Linear (LT 0.15)
50
0
1.5 1.6 1.7 1.8 1.9 2 2.1 2.2 2.3 2.4 2.5
Form factor
Fig. 9. Pore lengths with different FF

The variation of the FF results in porous structures with pore widths in a range of
37 to 192 µm. The pore lengths cover a range between 32 and 194 µm. The pore
geometries can again be assigned to the macropores. In comparison to the variation of
LT, significantly larger pores can be created with the variation of the FF. The FF can be
adjusted in hundredths of a step during part preparation (slicing). In combination with a
linear relationship between the pore size and the FF, it is possible to adjust the pore
sizes almost continuously within the determined ranges.
3.3 Variation of the Discharge Level (DL)

The discharge level describes the mass discharge of material from the dosing volume
through the nozzle. Accordingly, the setting of the discharge level results in the
respective volume of the droplets as well as the droplet size. The discharge level is set
as a percentage and is controlled by the travel of the screw during the generation of
droplets. In order to investigate the influence of the discharge level on the building
process, it is first doubled. According to the basic settings, the doubling of the dis-
charge level corresponds to an increase from 35 to 70%. Since this adjustment changes
the droplet geometry, a new LT must be defined. To do this, the droplet strand is
discharged into the build chamber without being deposited on the building platform
and then measured by means of microscopy (see Fig. 10).
Fig. 10. Droplet strand with DL of 70%
Rounding results in an average droplet length of 0.25 mm. This value is used for an
initial setting of the LT. Subsequently, the procedure of varying the FF is to be applied
so that a parameter range can be identified to create a porous structure. For the mea-
sured values presented in Fig. 11 and 12, the lowest coefficient of correlation is
r = 0.99. As a result, a linear correlation between the form factor and the pore size is
still established.
An increase in the pore size range can be achieved by doubling the discharge level.
The pore widths are in a range between 63 and 280 µm and the pore lengths between
76 and 282 µm. Compared to the previously produced samples, larger pores can be
produced by increasing the discharge. An extension of the pore size range in negative
direction is not possible.
300
LT 0.15
250
Pore width [μm]
200 LT 0.20
150 LT 0.25
100
Linear (LT 0.15)
50
Linear (LT 0.20)
0
1.0 1.5 2.0 2.5 3.0 3.5
Form factor
Fig. 11. Pore widths for DL of 70%
300
LT 0.15
250
Pore length [μm]
200 LT 0.20
150 LT 0.25
100
Linear (LT 0.15)
50
Linear (LT 0.20)
0
1.0 1.5 2.0 2.5 3.0 3.5
Form factor
Fig. 12. Pore lengths for DL of 70%
To determine the maximum possible pore size that can be produced with APF, the
discharge is increased further. The original value of 35% is tripled to 105%. To
determine the initial layer thickness, the measurement of the droplet strand yields an
average droplet length of 300 µm. Accordingly, the variation of the FF should be
started at LT of 0.30 mm. The results of the investigations are shown in Fig. 13 and 14.
The evaluation of the measurement results again shows that the range for the
production of a porous structure with LT of 0.15 mm has increased considerably. The
pore size range is so large that the achievable sizes are also covered by the layer
thickness of 0.20 mm and 0.25 mm. The pore widths are in the range of 84 to 466 µm
and the pore lengths in the range of 58 to 440 µm. This is a further extension of the
pore size range in the positive direction. It should be noted that the droplets are in a
deformed state. This deformation results from the forced discharge of more material in
a constant time interval. A further increase in the degree of deformation does not allow
reproducible categorization or measurement of the pore geometry. Therefore, no further
increase of the discharge level is performed.
500
LT 0.15
400
LT 0.20
Pore width [μm]
300 LT 0.25
200 Linear (LT 0.15)
Linear (LT 0.20)

100
Linear (LT 0.25)
0
1.5 2.0 2.5 3.0 3.5 4.0 4.5 5
Form factor
Fig. 13. Pore widths for DL of 105%
500
LT 0.15
400
Pore length [μm]
LT 0.20
300 LT 0.25
Linear (LT 0.15)

200
Linear (LT 0.20)
100
Linear (LT 0.25)
0
1.5 2.0 2.5 3.0 3.5 4.0 4.5 5
Form factor
Fig. 14. Pore length for DL of 105%
4 Reproducibility of Open-Pored Structures
In this section, the reproducibility of the porous structures in the APF process is
investigated. The influence of the position on the building platform and the multipli-
cation of the components is considered. With the aim of investigating the influences on
a porous structure, the FF is increased from the value 1.48 to 2.00. This setting shows a
uniform pore structure on the surface.
4.1 Positioning on the Building Platform

Unevenness is indicated by an inconsistent depositing position of the plastic droplets in
the z-direction. Furthermore, it is assumed that a changed distance between the dis-
charge nozzle and the building platform (z-coordinate) leads to a different pressing of
the first layer. To detect the degree of contact pressure, the strand width of the lined-up
plastic droplets is used as a measured variable. If the distance between the discharge
nozzle and the building platform decreases, the strands are pressed onto the building
platform with increased force. It should be possible to determine the result by means of
a widened strand geometry.
To investigate the evenness, 13 different positions are defined for the build-up of
test specimens (see Fig. 15). The coordinates shown indicate the respective center point
of the element to be built. Three building jobs are built, so that each test specimen is
available for examination in three versions. It should be noted that only one test
specimen per building job is produced, as the influence of a multiplication of the
building job has not yet been investigated. As with the measurement of the pore
geometry, the widths of five strands per sample are recorded and the mean value is
calculated (see Fig. 15).
115.0
Color Strand
gradient width [μm]
87.5
60.0
32.5
5.0
210.0 167.5 125.0 82.5 40.0
Fig. 15. Averaged strand widths depending on the positioning on the building platform,
measured on the platform side and evaluated with a color scale
The assumption regarding unevenness on the building platform can thus be con-
firmed. The averaged strand widths at the different positions vary in a range between
176 and 228 µm. In addition to the strand width, the manufactured test specimens are
used to assess the reproducibility of the pore sizes at the nozzle side. For the pore width
an average value of 120 µm with a standard deviation of 4.8 µm is available. The
lengths of the square pore geometry are given including the standard deviations, with a
nominal value of 115 µm and a tolerance of ±10 µm.
4.2 Multiplication of the Test Specimen

Previously, a maximum of one test specimen was produced per building job. In this
section the influence of a multiplication of the test specimen is examined. When
building several components simultaneously within one building job, a layer is first
built for all components before the following layer is deposited. This results in a longer
layer time. In addition, the nozzle is not permanently located above the part to be built
up, as is the case in the production of only one component. This eliminates the
additional heating effect of the heat radiated by the nozzle.
To investigate the above-mentioned influences on the pore size, five test specimens
are built up simultaneously in three runs. The tolerance range of 115 ± 10 µm pre-
viously determined in the investigations serves as a reference. Due to the assumption of
a square pore geometry, the pore length is not shown in the following evaluations. The
determined pore widths are in a range between 106 and 115 µm (see Fig. 16) and the
pore lengths between 119 and 125 µm. Accordingly, the specified tolerance range is
maintained. The multiplication of a component has no influence on the size of the
generated pore geometries.
140
130
Pore width [μm]
120 Tolerance-
110 range
100
90
80
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Test specimen number
Fig. 16. Determined pore widths when multiplying a building job by the factor 5, measured at
the nozzle side
5 Liquid Permeability of Open Porous Structures
Based on the results from macroscopy, components with macropores are to be examined.
Macropores correspond to the largest area within the pore classification. Accordingly,
medium to high permeabilities are expected, so that a stationary measurement with a
liquid is to be carried out. The following conditions must be observed [14]:
1. Cylindrical or square test pieces shall be used
2. No reactions with the pore wall take place
3. The flow is single-phase (homogeneous liquid)
4. The flow is stationary
5. The flow is formed
Condition (1) is ensured by the possibility of a freely selectable geometry by means

of APF during the production of the test specimens. Demineralized water is used to
ensure compliance with conditions (2) and (3). Aspects (4), (5) shall be considered in
the design of the test setup and the performance of the test. A possible setup is shown in
Fig. 17.
Δp Δh
Buffer
tank
V̇
Specimen holder
Fig. 17. Experimental setup for a stationary measurement of liquid permeability [14]
The basic principle of the measurement is to measure the resulting volume flow, which
flows through a sample under the application of a constant pressure. The volume flow
changes depending on the resistance of the sample to the medium flowing through it.
The specimen holder (see Fig. 18) has the function of clamping the specimens to be
tested and ensuring that the fluid flows through them. The liquid flow passes through
the inlet (3) on the upper side, through the porous structure and then into the outlet on
the lower side.
Fig. 18. Upper side (left) and lower side (right) of the specimen holder. 1: Through-bore,
2: Groove for inner O-rings, 3: Inlet and outlet of the test liquid, 4: Specimen contact area,
5: Groove for outer O-ring
The results of the tests are divided for evaluation. First, the evaluation of the
structures with DL of 35% is shown (see Fig. 19). With increasing FF as well as with
increasing LT linear relationships between the mass flow rate and the parameters can be
seen. The correlation coefficients are above r = 0.98. These correlations simultaneously
describe an increase in volume flow with increasing pore openings within a parameter
variation.
500
Mass flow rate [g/min]
400
300
200
100
0
1.5 1.7 1.9 2.1 2.3 2.5
Form factor
Fig. 19. Mass flow rate with different FF (DL = 35%, LT = 0.15 mm)
The lowest flow rate is to be assigned to a sample from the variation of the form
factor (FF = 1.60/pore width = 37 µm) with a value of 0.7 g/min. The lowest value
from the variation of the layer thickness (LT = 0.16 mm/ pore width = 22 µm) is
8.7 g/min. It is noticeable that the test specimens show a higher flow rate with smaller
pore size when the layer thickness is varied compared to the variation of the form
factor. The measurements of the test specimens, produced with a DL of 70%, also show
a linear relationship between the FF and the mass flows (r 0.97).
700
LT 0,15
SD 0.15
Mass flow rate [g/min]
600
500 LT 0,20
SD 0.20
400
LT 0,25
SD 0.25
300
200 Linear(SD
Linear (LT0,20)
0.15)
100
Linear(SD
Linear (LT0,25)
0.20)
0
1.5 2.0
2 2.5 3.0
3 3.5
3.5 4.0
4 4.5
Form factor
Fig. 20. Mass flow rate with different FF and LT (DL = 105%)
While a linear trend can be seen at LT of 0.25 mm and LT of 0.20 mm (r 0.96),

there is a dispersion of the measured values at LT of 0.15 (see Fig. 20). The combi-
nation of the low layer thickness and the too high discharge level does not allow an
even and reproducible droplet deposition. This results in an uneven porous structure for
which there is no linear trend in liquid permeability. At this point it should be noted
that liquid permeability is not only dependent on the size of the pore openings. The free
pore volume in the transverse direction of the samples also influences the volume flow.
The highest mass flow rate determined is 587.7 g/min (DL = 105%, FF = 2.90,
LT = 0.20 mm).
Due to the mainly linear relationships between the process parameters and the
resulting mass flows, the permeability coefficients are calculated only for the smallest
and largest mass flows. The permeability coefficients are kf = 5.09 10−3 m/s
(m_ = 587.7 g/min) and kf = 6.06 10−6 m/s (m_ = 0.7 g/min). According to DIN
18130-1 (see Table 1), components with the following properties can be manufactured:
• Highly permeable
• Permeable
• Low permeability
6 Summary and Outlook
The investigations have shown that it is possible to produce porous structures with the
APF process. The free spaces of the components form a lattice structure of open pore
channels, which is uniformly structured in horizontal and vertical direction. Due to a
connection between the cavities the inner structure is called an “interconnected pore
structure”. The pore width at the surface of the components is adjustable in the size
range 22 to 466 µm and thus belongs to the range of macropores. According to
ISO 4793, this corresponds to laboratory filters of the category P 3 to P 00, which are
used for the analytical filtration of medium-fine precipitates or as a support for solid
samples in flow systems.
The variable size adjustment results in a porosity between 7 and 66%. When ABS
is used, this corresponds to a weight reduction of up to 70%. At the same time, the
material savings lead to a cycle time reduction of up to 75%. As a result of the
reproducibility tests, a tolerance range of the pore widths of ±10 µm is specified. This
value cannot be transferred to the pore openings on the platform side, as the geometry
is influenced by varying degrees of pressure on the droplet strands in the first layer.
This influence is so significant that average deviations of the pore widths of −38 µm
and +14 µm occur. If the geometry or the number of components is varied within a
building job, this has no influence on the pore size.
The level of liquid permeability is not directly related to the size of the openings of
the surface pores. In addition, the volume flow rate that occurs depends on the selected
layer thickness. At constant layer thickness there is a linear relationship between
permeability and the parameter form factor. With the manufactured components, per-
meability coefficients between 5.09 10−3 and 6.06 10−6 m/s were achieved for the
flow of water. In accordance with DIN 18130-1, these values are assigned to the areas
of high permeability, permeability and low permeability.
The application possibilities of open-pored structures are numerous. In addition to
components already in use, such as compressed air ejectors in injection molds or filters
for separating media, research is being conducted into further possible applications.
Examples are porous electrode or battery components within accumulators for the
automotive sector or new types of structures for use in heat and air exchangers [16]. In
addition to the characteristics already investigated, further properties must be tested
according to the field of application. One aspect to be investigated is the mechanical
strength of the structures. For the application as a filter, the absorption capacity of
compressive forces is of particular importance. When a medium flows through the
filter, the filter material must be able to resist the pressure applied by the flow.
References
1. Gaub, H.: Customization of mass-produced parts by combining injection molding and
additive manufacturing with Industry 4.0 technologies. Reinf. Plast. 60(6), 401–404 (2016)
2. Neff, M., Kessling, O.: Geschichtete Funktionsteile im industriellen Maßstab. Kunststoffe,
Nr. 08/2014, pp. 64–67 (2014)
3. Keßling, O.: AKF - Neues industrielles additivesVerfahren. RTejournal - Forum für Rapid
Technologie, vol. 2015. https://www.rtejournal.de/ausgabe11/3963. Accessed 12 Feb 2020
4. Kloke, A: Tröpfchen im Millisekundentakt. Kunststoffe 11/2018. Carl Hanser Verlag,
München (2018)
5. Duffner, E.: Industrielle additive Fertigung funktionsfähiger Kunststoffteile. Lightweight
Des. 9, 22–27 (2016)
6. Ramezani Dana, H., Ben Azzouna, M., Breteau, T., Delbreilh, L., Guillet, A., Barbe, F.:
Développement de matériaux cellulaires avec mise en oeuvre par fabrication additive
bimatière (2017)
7. Hirsch, A., Hecker, F., Moritzer, E.: Process parameter optimization to improve the
mechanical properties of Arburg Plastic Freeformed components. In: Conference Proceed-
ing, 30th Annual International Solid Freeform Fabrication Symposium (SFF), Austin, Texas
(2019)
8. Kaut, F., Cepus, V., Grellmann, W., Lach, R.: Structure-property relationship of additive
manufactured thermoplastic polymers processed with ARBURG Freeformer Technology.
Rapid.Tech – Fachkongress, Forum AM Science, pp. 217–235 (2018)
9. Guessasma, S., Nouri, H., Roger, F.: Microstructural and mechanical implications of
microscaled assembly in droplet-based multi-material additive manufacturing. Polymers 9,
372 (2017)
10. Spiller, Q., Müller, T., Plewa, K., Hanemann, T., Fleischer, J.: Eignung des Kunststoff-
Freiformens für metallische Bauteile (2016)
11. Klahn, C.: Laseradditiv gefertigte luftdurchlässige Mesostrukturen - Herstellung und
Eigenschaften für die Anwendung. Springer, Hamburg (2015)
12. Thienel, K.-Ch.: Werkstoffkunde 1. Lecture script, University of the Bundeswehr, München
(2011)
13. Menges, G., Haberstroh, E., Michaeli, W., Schmachtenberg, E.: Menges Werkstoffkunde
Kunststoffe, 5th edn. Carl Hanser Fachbuchverlag, Wien (2014)
14. Häfner, F., Wagner, S., Freese, C.: Durchlässigkeitsmessung - Bestimmung der absoluten
Durchlässigkeit pröser Stoffe mit Flüssigkeit und Gas. TU Freiburg - Institut für Bohrtechnik
und Fluidbergbau, Freiburg (2015)
15. Günther, K., Sonntag, F., Moritzer, E., Hirsch, A., Klotzbach, U., Lasagni, A.F.: Universal
Micromachining Platform and Basic Technologies for the Manufacture and Marking of
Microphysiological Systems. Micromachines 8, 246 (2017)
16. Hagl, R.: Das 3D-Druck-Kompendium - Leitfaden für Unternehmen, Berater und
Innovationstreiber, 2nd edn. Springer, München (2015)
Novel 4-Axis 3D Printing Process to Print
Overhangs Without Support Material
Michael Wüthrich(B) , Wilfried J. Elspass, Philip Bos, and Simon Holdener
Zurich University of Applied Sciences,

Institute of Mechatronic Systems, Winterthur, Switzerland
michael.wuethrich@zhaw.ch
https://www.zhaw.ch/en/engineering/institutes-centres/ims/
Abstract. This paper describes a novel 4-axis FDM printing process

with a newly designed printhead, for the printing of overhangs without
support structures. With conventional FDM printing, overhangs of more
than 45◦ –60◦ must be supported.
For this novel printing process, the printhead is rotated 45◦ around a
horizontal axis and equipped with a vertical, rotational axis. The print-
head no longer follows layers parallel to the build platform, but moves
on the surface of a 45◦ cone. The printing cone increases in diameter
from layer to layer. With this cone-shaped layers, the printable angles
increases by 45◦ , which leads to printable overhangs of up to approxi-
mately 100◦ .
New slicing strategies for this printing process have been developed to
slice the parts for the novel printing process. The feasibility of the con-
cept has been prototypically demonstrated. The novel design achieves
the advantages of higher speed and quality with lower cost at the same
time.
Keywords: FDM printing · Overhang · Supportless printing · 3D

printing process
1 Introduction
3D Printing (3dp), also known as Additive Manufacturing (AM), is based on
adding material layer by layer to build a geometric model. The most common
3dp technologies are Fused Deposition Modeling (FDM), Selective Laser Sintering
(SLS) and Selective Laser Melting (SLM), Stereolithography (SLA) and Material
Jetting (MJ). These different technologies can process many different materials
ranging from plastics to metals. A good overview on AM is provided in [1].
Compared to known production technologies, 3dp has some major advantages
like the production of individual parts and the integration of functions at no
extra cost or the production of geometries, which are impossible to produce
with conventional technologies. Therefore, 3dp is a strongly growing market,
reaching an expected market size of 21 billion US dollars by 2021 according to
Statista [2].
https://doi.org/10.1007/978-3-030-54334-1_10
Novel 3D Printing Process to Print Overhangs Without Support Material 131
In this paper, the focus is on the improvement of the Fused Deposition Mod-
eling (FDM) technology, which was the most used 3D printing technology in
2017 and 2018, with a share of 46% [3]. Since the FDM process is comparatively
simple and does not require expensive components like a laser, affordable FDM
printers are available, which accounts for their popularity. There are many FDM
desktop models as well as professional printing systems. An overview on the
FDM process and the recent developments in this field is provided in [4]. For the
current research topics, refer to [4,5] or [6].
1.1 Limits of Conventional FDM Printing and Motivation

Conventional FDM printing has some considerable limits, discussed in [7]. The
major disadvantage is the impossibility of printing overhangs with more than
45◦ –60◦ without support structures. Figure 1 shows a printed test object with
overhang angles from 30◦ –85◦ and Fig. 2 shows the resulted quality on the down-
skin of the print, where inaccurate edges can be found near 45◦ and larger errors
after 65◦ . Reference [8] provides a comprehensive overview on the topic of sup-
port structures.
Fig. 1. Printed test object with overhangs Fig. 2. Quality of printed overhangs on
from 30◦ to 85◦ . downskin of test part.
As mentioned in Sect. 1, 3dp builds a model by generating one layer onto the
previous. In FDM, these layers are made by extruding a thin strand of melted
thermoplastic through a nozzle, while moving this nozzle along a defined path.
Therefor each layer consists of an array of strands. Figure 3 shows the nozzle, a
single strand in the current layer and multiple strands, combined to the previous
layers, for the situation where no overhang is printed. So each strand is fully
supported by the strand below.
In order to print an overhang (Fig. 4), the extruded strand is no more fully
supported, which leads in the best case, to a poor printing quality or, in the
132 M. Wüthrich et al.
worst case to unusable results. It turns out that possible printable angles do not
depend on the orientation in space, since gravitational forces have a negligible
influence. This can easily be shown by rotating a printer by 45◦ which does not
change the quality of the printed overhangs. The decisive influence is the flow
of the melted plastic, leaving the nozzle. The viscous material must be deflected
by 90◦ after exiting the nozzle. Therefore, a sufficiently large overlap of about
50% to the previously printed material of the lower layer is necessary to obtain
feasible results. This leads to printable overhangs between 45◦ and 60◦ depending
on different parameters like e.g. layer height. Figure 4 shows the printed layers
of a 45◦ overhang. More details on printable angles and the affecting factors can
be found in [9].
max
Nozzle Nozzle 60°
°-
45
Strand Strand
Layers Layers
Fig. 3. Printing with 0◦ overhang. Fig. 4. Printing with 45◦ (–60◦ max.)
overhang.
There are different methods to print overhangs. The following sections explain
possible solutions for dealing with overhangs.
1.2 Support Structures
The most commonly used method to print overhangs is to use support structure.
This method can be used with every FDM printer and is handled by all major
slicing software packages. When the slicing software detects overhangs over a
defined angle, it generates the support structure below automatically. Support
structure is generated with the respect to easy removal and efficient printing,
both time-wise and material-wise. Figure 5 shows a sliced part (blue) and the
generated support structure (green). There is one layer of vertical gap between
the support structure and the part. This results in a weak joint, where the sup-
port structure can be brocken away after printing. Figure 6 shows the downskin
surface of the printed part after removing the support structure. The smooth-
ness of this surface does not reach the same quality compared to other surfaces
shown in this picture.
Fig. 5. Sliced part with generated sup- Fig. 6. Supported surface after removing
port structure (green). the support structure.
Besides the huge advantage that overhangs can be printed with support
structure, this technique has three major downsides. Firstly, the printing
time increases considerably and secondly it needs additional printing material.
Thirdly, the quality of the supported surfaces degrades notably. Table 1 shows a
comparison of the sliced part shown in Fig. 5 with respect to time and material
used to print the part with and without support structure. The part has been
sliced in PrusaSlicer [10] with a layer height of 0.2 mm and an infill of 10%. With
this settings, it results in an additional 55% printing time and an additional 59%
of used material.
Table 1. Comparison of sliced part with and without support structure.
No support Support Difference

Printing time 34 min 53 min + 55%
Material used 8.3 g 13.2 g + 59%
Although there are optimized supporting techniques like the water-soluble

support structure, the downsides of time, material waste and quality persists.
1.3 BlackBelt
The BlackBelt printer [16] (Fig. 7) has a printing plane, that is rotated 45◦
around the y-axis. With this the nozzle and the layers are rotated by 45◦ . Hence,
a printable 45◦ overhang is rotated into a 90◦ overhang, which is printable with-
out support structure. Figure 8 shows the 45◦ nozzle and layers and the 90◦
overhang. These overhangs can only be printed in one direction. In all other
directions, support structures are still required.
Nozzle
Strand
45°
90° Overhang
Layers
Fig. 7. Blackbelt printer, [16]. Fig. 8. Printing with 45◦ layers.
1.4 Multi Axis 3d Printing

Due to the described downsides using support structure, there are multiple
approaches to print overhangs without support structure. To print without sup-
port structure, more than three moving axes are required. Significant research
is conducted in multi axis FDM printing.
Murtezaoglu et al. show in [11] a geometry-based process planning for multi-
axis support-free AM. Wu et al. have an analogous approach in [12], where the
printing platform is mounted on a robotic arm, so that it can be tilted in two
axis and enable support-free printing. Wang et al. choose in [13] a conventional
5-axis design to implement a non-supporting method. Rieger et al. show in [14]
an approach where the printhead is mounted on a robotic arm.
The Institute of Mechatronic Systems at the Zurich University of Applied
Sciences has developed a novel FDM 6-axis printer (Fig. 9) with a tilting build
platform, called MaxBot [15]. With this technology, overhangs of around 80◦ are
achievable.
As shown in different research projects, multi-axis printing offers a solution
to support-free FDM printing. However, the multi-axis approach requires slicing
algorithms, which are of a much higher complexity due to questions regarding
printing strategies and collision avoidance problems.
Fig. 9. 6-axis printer MaxBot of IMS, ZHAW.
The solution shown in this paper attempts to close this gap due to a more
simple kinematic setup and printing strategy.
2 Novel Solution: RotBot, a Printer with Rotational

Printhead with 45◦ Tilt Nozzle
Printing with a conventional FDM 3-axis printer allows to print in a range of

± 45◦ (Fig. 10). As described in Sect. 1.3, rotating the printing plane by 45◦
enables printing 90◦ overhangs in one direction (Fig. 11). Adding a rotational
axis along the z-axis enables the printer to print overhangs in any direction
(Figs. 12 and 14). Due to the rotational axis, this concept is called “RotBot”.
0° -90° / +90°
± 45° -
90
°
Fig. 10. Conventional Fig. 11. FDM printer with Fig. 12. FDM printer with
FDM printer. a 45◦ printing plane. rotated nozzle and rota-
tional z-axis.
2.1 Conceptual Solution

This idea leads to a novel 3d printing process, where overhangs can be printed
support-free in any direction with only one additional axis. The development of
this printing concept is described in [17].
The moving x-, y- and z-axes of the printer remain the same as in a con-
ventional printer. The nozzle is rotated 45◦ around a horizontal axis and an
additional rotational axis along the z-axis is added, which intersects the nozzle
axis at the exit of the nozzle (Fig. 13). Hence, if the nozzle rotates precisely
around the vertical axis, no compensation motion in x and y is required. This
concept has the big advantage, that it can be an upgrade to existing printer
kinematics, where only the exchange of the printhead is necessary.
Heat block
Rotational Cone
axis shaped
layer
Nozzle
Fig. 13. 45◦ tilted nozzle with the rota- Fig. 14. Printing strategy with the cone
tional axis. shaped layers.
In contrast to the conventional FDM printer, where the layers are in a plane,
with this new concept, the layers become cone shaped. This results in the com-
bination of the 45◦ rotated plane and the vertical rotational axis (Fig. 14). For
detailed information see Sect. 2.3.
2.2 Hardware Setup

As pointed out in Sect. 2.1, an additional rotational axis has to be provided to
print overhangs in any direction. There are two choices to realize a rotational
movement. The first solution is to add this axis to the printhead, which then
rotates around the nozzle. The downside to this solution is, that both, the signal
and the energy as well as the filament must be transferred through a rotation
axis, which makes the design challenging. Alternatively, a rotational axis to
the build platform is added, while the 45◦ nozzle is fixed in its position. The
advantage hereby is, that the filament does not have to cross the rotating axis.
The major downside is the need of a compensating motion of the printhead, if
the slicing cone in the part is not perfectly colinear with the rotational axis of
the printhead.
Section 3.1 describes the mechanical setup, which was implemented in the
prototype.
2.3 Printing and Slicing Strategy
As mentioned in Sect. 2.1, the part is no longer built up in layers parallel to

the build platform, but in a 45◦ cone or frustum of cone shaped layers. As
described below, also other layer shapes are chosen, depending on the shape of
the part being printed. To describe the printing strategy in general, parts with
a rotational symmetry axis are chosen and the overhangs are 90◦ towards the
inside or the outside.
The strands in these layers are printed top down. Every strand is a circle
with radius R and parallel to the build platform at the distance Z. Each layer
has a distance to the next layer of distance δ. After printing the strand n in one
layer, the printhead prints strand n + 1 according to Eqs. (1) and (2):
Rn+1 = Rn + δ (1)
Zn+1 = Zn − δ (2)
Depending on the overhang direction, the cone must be chosen correctly. If

the overhang is towards the outside of the part, a regular cone with the printhead
on the outside of the part must have the radius Rn+1 according to (1). For
overhangs towards the inside of the part, the cone must be upside-down with
the nozzle on the inside. Therefore Rn+1 changes to Eq. (3),
Rn+1 = Rn − δ (3)
as shown in Figs. 15 and 16. With a cone shaped layer finished, the printhead
moves up and starts the new layer again top down.
If a part consists of a combination of inside and outside overhangs, a com-
bination of normal and upside-down cones has to be chosen to build the part.
Figure 17 shows a part, where the lower section of the part consists of inside
cones (for the overhangs towards the inside) and the upper section consists of
outside cones (for the overhangs towards the outside). Another possible combi-
nation is presented in Fig. 18, where the overhangs are at the same distance to
the build platform, so the cones are combined in different manners. The inside
of the part consists of inside cones, and the outside consists of outside cones.
Outside cones
Nozzle
Nozzle
Inside
Layers Layers cones
Fig. 15. Cone and nozzle for outside over- Fig. 16. Cone and nozzle for inside over-
hangs. hangs.
Outside cones Outside cones
Inside cones
Inside cones
Fig. 17. Vertical seperation of inside and Fig. 18. Horizontal seperation of inside
outside cones. and outside cones.
As mentioned, the above described printing strategy with the cones works
for parts with a rotational symmetry axis and represents the principle printing
strategy for the novel printing process the best. Nevertheless, real world parts
have an almost infinite multitude of possible geometries. Therefore, other slicing
strategies are necessary to succeed with the different geometries. Figure 19 shows
an overall workflow architecture of the slicing strategy for the part shown in
Fig. 17. Here the part is split into subparts, where the subparts all have the
overhangs in the same direction. Therefore, it can be sliced with the same inside
or outside strategy. Then the sliced subparts are merged together and the gcode
for the final part is generated.
If the overhang in a section is only in one direction, no rotation of the print-
head is required and the printhead is in a locked rotational position during this
section of the part. This is the same behavior as the BlackBelt printer (Sect. 1.3),
so support-free overhangs can be printed in one direction. In general, if a part
is angular and has no rotational symmetry, the layers are of a prismatic or a
pyramidal shape.
Basically, circular parts with a rotational symmetry require cone and frustum
of a cone shaped layers, angular parts require prismatic, pyramidal or frustum of
a pyramid shaped layers and parts with arbitrary shapes require combinations of
cone, prismatic, pyramidal or frustum of these shaped layers. Section 3.3 provides
details to the printing.
detect overhangs and split parts into subparts with

.stl part their direction same direction of overhang
combine sliced subparts into slice subparts accoring to

print part main part and generate gcode their overhang direction
Fig. 19. Overall workflow architecture of the slicing strategy.
2.4 Challenges and Limits of This Printing Strategy

As described in Sect. 2.3, to slice a body with a more complex geometry, it
has to be split in subbodies with overhangs in only one direction. To perform
this process automatically, sophisticated algorithms are required, which have not
been developed yet. As soon as these algorithms are available, a slicer software
can be implemented to be able to print different parts, that are not generated
manually as the examples shown in Sect. 3.2.
Nozzle
approx. 2mm
Fig. 20. Conflict area when printing a Fig. 21. Possible solution to avoid con-
horizontal inside overhang. flict area.
One known limit is the impossibility of printing a horizontal closed inside

overhang. The part shown in Fig. 16 can not be fully printed, since towards the
center the nozzle would come in conflict with already printed material. This
conflict area is shown in Fig. 20, marked in red color. One approach to solve
this problem is to print this area with a cone shape (Fig. 21) and not horizon-
tally. This cone has a diameter of approximately 2 mm and therefor a height
of approximately 1 mm. It is also possible to adjust the outer geometry of the
nozzle, so that it becomes smaller in the front and only a smaller area is affected.
An other approach to reduce the conflict area is to generally adjust the angle
of the printhead to 30◦ wich will result in smaller overhang angles, that can be
printed.
2.5 Option: Variable Angle of Nozzle

As an option to solve the just described limitation, a variable tilt angle of the
nozzle can be considered. Simple parts with no overhangs can be printed and
sliced with horizontal layers, parts with overhangs can be printed and sliced with
the presented concept. Even combinations of both concepts within one part are
possible. The downside is, that an additional axis to adjust the nozzle-angle
increases the mechanical complicacy of the printhead considerably.
3 Implementation
Two iterations of prototypes are currently being built to validate the concept.
Thus, an upgrade of an existing printer with a modified printhead has been
accomplished. The frame of the printer with the orthogonal kinematics could be
used. The following sections describe the printhead, the slicing of selected parts
as well as the accomplished results.
3.1 Printhead
To print in the described manner, the printhead must be able to fulfill multi-
ple 360◦ rotations. Beside the additional stepper motor for the rotational axis
a slipring is included, which establishes the electrical connection between the
rotating printhead and the controller, enabling unlimited rotational movement.
To reduce the number of connections, being passed through the slipring, the
extruder motor is placed above the rotational axis without rotation. The filament
is transferred to the nozzle, passing through the stepper motor with a hollow
shaft. This design makes sure, that no rotational force is applied to the filament
before it melts in the heater block. Prohibiting the filament from twisting and
hindering the unrestricted rotation.
As mentioned in Sect. 2.1 the nozzle alignment must be as precise as possible
with the rotational axis to avoid compensational motion in the x- and the y-
axis when the printhead is rotating. To fulfill this requirement, the heater block
is rotated by 45◦ and adjusted such that the nozzle exit is on the rotational
axis. Additionally, a system is implemented which allows to accurately align the
nozzle position with screws.
The whole printhead with the rotational axis, the extruder, the slipring, the
adjustment screws the heater block and the nozzle are shown in Fig. 22.
Rotational axis
Extruder
Rotational motor
with hollow shaft
Slipring
Screws to adjust
nozzle position
Heat sink
45° rotated heater block

45° rotated nozzle
Fig. 22. The printhead with the rotational axis.
3.2 Slicer
As described in Sect. 2.3, the printing strategy differs to a conventional strategy

and varies depending on the geometry of the part. Therefore, no available slicing
software can be used to slice the part. A software needs to be developed, which
is able to slice the parts accordingly. Different algorithms must be developed,
which will enable the software to slice the part (see Sect. 2.4).
To test the whole concept and the prototype, the so called gcode, which
defines the motion of the printer has been generated with MATLAB scripts.
The boundaries of the body that should be printed are defined manually. Then
the MATLAB script goes trough the body layer by layer and calculates the gcode
for the single strands. For example the body shown in Fig. 14 would be defined
as a first cylinder with a specified diameter and height and a second cylinder
with a bigger diameter and the same height on top of the first cylinder. The
MATLAB script now starts in the center at the bottom. There it generates the
gcode for a small circle directly on the print bed. It then moves one layer (which
is according to Sect. 2.3 one unit of the length δ) up and generates a second
circle with the same diameter. Then it moves one unit out and one unit down
and generates a new circle with a diameter that is two units bigger than the
first circle. So the 45◦ cone shaped layer is generated. Then it moves one layer
up, which is one unit higher than the pervious one and generates the gcode for
the stings in this layer until it reaches the level of the print bed or the defined
boundary (radius of cylinder) of the body. Then it moves back up and this
process is repeated until the whole body is filled with material. Since this gcode
generation is comparatively complex, only gcodes for simple geometries can be
generated and printed up to now. Nevertheless, the proof of concept with parts
successfully produced was shown, as described in the next section.
3.3 Demo Parts
Figure 23 shows four different parts, that have been printed on the prototype
printer, using different shape dependent strategies.
(a) Test part with rotational symmetry (b) Test part with rotational symmetry
and outside overhang only. and overhangs to the inside and the out-
side.
(c) Testpart with a mostly non-rotating (d) Squared test part with inside and out-
printhead. side overhangs.
Fig. 23. Printed test parts

Figure 23 (a) shows a part with a rotational symmetry and rising overhang
angle. At the outermost perimeter, the overhang has an angle of 100◦ . Through-
out the whole part, there is an outside overhang, thus all layers are of a normal
cone shape as shown in Fig. 15.
Figure 23 (b) also shows a part with a rotational symmetry, but with changing
direction of overhangs. The lower section is an inside overhang, which requires
an upside-down frustum of a cone. The upper section has an outside overhang,
which requires a normal frustum of a cone so the layers are the same, as shown
in Fig. 17
Figure 23 (c) has no rotational symmetry. The printhead was not permanently
rotated, while printing the part. For the lower section of the part, the layers were
planes from top right to bottom left with the printhead on the left side of the
part. For the second section, where there is an overhang towards the right side,
the printhead was rotated by 180◦ and the new layer is now from top left to
bottom right with the printhead on the right side. For the last section, where
the overhang is to the left side, the printhead is again rotated by 180◦ and the
layers are again from right top to left bottom. Since the nozzle was not perfectly
aligned, the position, where the printhead is rotated by 180◦ is clearly visible.
Figure 23 (d) shows an similar part as Fig. 23 (b) but with a square foot print.
So the layers are no longer frustums of cones but of pyramids. The strands are
no longer circles but squares. Moreover, the composition of first an upside-down
and then a normal pyramid is analogous to Fig. 23 (b).
All these parts show overhang surfaces of an equal quality like the non-
overhang surfaces.
4 Conclusions and Discussion
A novel printing process to print overhangs without supporting structures with

a novel printhead and orientation of the nozzle is proposed. The printhead is
equipped with an additional rotational axis and the nozzle is tilted by 45◦ . With
this printing process we have been able to print different test parts, proving the
potential of the concept. Various printing strategies of different geometries were
developed and tested. The gcodes for the tested parts were generated manually
with the help of a MATLAB Script. Several parts with different geometries have
been printed and it turns out, that support-free FDM printing of overhangs up to
90◦ is possible with an uncompromised surface quality.
Nevertheless, there are some challenges and limitations of the presented app-
roach. It is not possible to print horizontal and fully closed inside overhangs. Fur-
ther tests with different approaches to solve this limitation will be conducted.
Since the generation of gcode for this printing strategy requires sophisticated
and new algorithms, the slicing software is not yet implemented. The next goal
is to provide an algorithm to provide an automatically working slicer to comple-
ment the presented printing process. As soon as the slicing software is available,
intense testing of different geometries has to be done to prove the concept of
both, the mechanical as well as the slicing solution.
In addition, the mechanical properties of the parts printed with this novel
approach needs to be investigated. The mechanical properties of FDM printed
parts can vary significantly, depending on different parameters like the printing
direction and the layer height ([4] and [18]). The mechanical properties of parts
printed with this novel approach will have to be compared to parts with the
traditional approach to be able to decide whether the presented approach leads
to comparable properties.
References
1. Wong, K.V., Hernandez, A.: A review of additive manufacturing. In: ISRN Mechan-
ical Engineering 2012, pp. 1–10 (2012). https://doi.org/10.5402/2012/208760
2. Statista: 3D printing market size worldwide from 2013 to 2021. https://www.
statista.com/statistics/796237/worldwide-forecast-growth-3d-printing-market/
3. Statista: Most used 3D printing technologies in 2017 and 2018. https://www.
statista.com/statistics/560304/worldwide-survey-3d-printing-top-technologies/
4. Rahim, T.N.A.T., Abdullah, A.M., Akil, H.M.: Recent developments in fused depo-
sition modeling-based 3D printing of polymers and their composites. Polym. Rev.
59(4), 589–624 (2019). https://doi.org/10.1080/15583724.2019.1597883
5. Dhinakaran, V., Kumar, K.M., Ram, P.B., Ravichandran, M., Vinayagamoorthy,
M.: A review on recent advancements in fused deposition modeling. In: Materials
Today: Proceedings, January 2020. https://doi.org/10.1016/j.matpr.2019.12.036
6. Liu, Z., Wang, Y., Wu, B., Cui, C., Guo, Y., Yan, C.: A critical review of fused
deposition modeling 3D printing technology in manufacturing polylactic acid parts.
Int. J. Adv. Manuf. Technol. 102(9–12), 2877–2889 (2019). https://doi.org/10.
1007/s00170-019-03332-x
7. Wulle, F., Coupek, D., Schäffner, F., Verl, A., Oberhofer, F., Maier, T.: Work-
piece and machine design in additive manufacturing for multi-axis fused deposition
modeling. In: Procedia CIRP 60, 229–234 (2017). https://doi.org/10.1016/j.procir.
2017.01.046
8. Jiang, J., Xu, X., Stringer, J.: Support structures for additive manufacturing: a
review, September 2018. https://doi.org/10.3390/jmmp2040064
9. Zhao, H.M., He, Y., Fu, J.Z., Qiu, J.J.: Inclined layer printing for fused deposition
modeling without assisted supporting structure, vol. 51. Elsevier BV, June 2018.
https://doi.org/10.1016/j.rcim.2017.11.011
10. PrusaSlicer: https://www.prusa3d.com/prusaslicer/
11. Murtezaoglu, Y., Plakhotnik, D., Stautner, M., Vaneker, T., van Houten, F.J.:
Geometry-based process planning for multi-axis support-free additive manufactur-
ing. Procedia CIRP 78, 73–78 (2018). https://doi.org/10.1016/j.procir.2018.08.175
12. Wu, C., Dai, C., Fang, G., Liu, Y.J., Wang, C.C.: RoboFDM: a robotic system for
support-free fabrication using FDM, May 2017. https://doi.org/10.1109/icra.2017.
7989140
13. Wang, M., Zhang, H., Hu, Q., Liu, D., Lammer, H.: Research and implementa-
tion of a non-supporting 3d printing method based on 5-axis dynamic slice algo-
rithm. Robot. Comput. Integr. Manuf. 57, 496–505 (2019). https://doi.org/10.
1016/j.rcim.2019.01.007
14. Rieger, M., Johnen, B., Kuhlenkötter, B.: Analysis and development of the fused
layer manufacturing process using industrial robots. In: Proceedings of ISR 2016:
47st International Symposium on Robotics (2016). https://ieeexplore.ieee.org/
document/7559182/
15. Herrman, D., Tolar, O.: Weiterentwicklung der Gerätetechnik im FDM-Verfahren.

Internal report, ZHAW (2015)
16. Blackbelt. https://blackbelt-3d.com
17. Bos, P., Holdener, S.: Konzepte für FDM Druck von Überhängen ohne
Stützmaterial. Bachelor’s thesis, internal report, ZHAW (2019)
18. Huang, B., Meng, S., He, H., Jia, Y., Xu, Y., Huang, H.: Study of processing param-
eters in fused deposition modeling based on mechanical properties of acrylonitrile-
butadiene-styrene filament. Polym. Eng. Sci. 59(1), 120–128 (2018). https://doi.
org/10.1002/pen.24875
Hybrid Manufacturing: A New Additive
Manufacturing Approach
for Closed Pump Impellers
Robin Rettberg(&) and Thomas Kraenzler
Sulzer Management Ltd., Pumps Equipment, 8401 Winterthur, Switzerland

robin.rettberg@sulzer.com
http://www.sulzer.com
Abstract. Additive Manufacturing enables the production of pump impellers

with a much shorter lead-time than conventionally used casting processes. This
is of particular importance for the maintenance and repair business, as it can lead
to a reduction of down time or the necessity to keep spare impellers on stock.
Sulzer is developing an impeller manufacturing process which combines
additive Laser Metal Deposition (LMD) with subtractive 5 Axis CNC Milling.
The present paper describes the basic process steps, the advantages compared to
commonly used manufacturing processes and the challenges to implement and
to qualify this new manufacturing technology.
Keywords: Directed Energy Deposition (DED) Hybrid manufacturing

Impeller Multi-axis machining
1 Introduction
Like most mechanical engineering companies, Sulzer started to investigate the potential
of Additive Manufacturing (AM) for the production of their products many years ago.
In the area of centrifugal pump manufacturing several parts have been identified that
can benefit from the many advantages that AM promises. With most pump components
being either not very complex in shape or large in overall size, the part with the highest
identified potential is sitting right at the heart of a pump, the impeller. Especially closed
impellers had only one technically and commercially feasible manufacturing route in
the past due to the channel geometries, which was casting. After further investigations
and trials, the various advantages and disadvantages of different AM technologies have
been evaluated [1], with the unfortunate outcome that none of the trialed technologies
alone were able to fulfill the requirements of such components. Major reasons that lead
to the need to re-think the manufacturing process again were overall size limitations as
well as surface quality limitations in inaccessible areas. Luckily, during these times the
first hybrid machine tools that combine DED and multi-axis milling just entered the
market and opened up a new way of additive part-manufacturing with basically no size
limitations and high deposition rates, whilst maintaining precision and surface quality
of milled parts. This paper presents how a closed impeller can be manufactured and
benefits from the hybrid manufacturing approach. Furthermore, technology-specific
development areas as well as qualification specialties are discussed.

https://doi.org/10.1007/978-3-030-54334-1_11
Hybrid Manufacturing 147
2 Technology Overview
2.1 Casting of Pump Impellers
The predominant process for the manufacturing of pump impellers is casting [2]. This
is particularly applicable for closed pump impeller, as those cannot be milled from a
billet due to the high wrapping angle of the impeller channels. Casting of closed
impellers is related to a number of disadvantages, starting with a poor surface quality,
the risk of casting porosity and geometrical deviations due to core shifts, ending with a
commonly long lead time. While techniques like ceramic shell investment casting yield
better surface quality and accuracy than sand casting processes the quality is still
limited when compared to subtractive manufacturing methods like milling and turning.
Thus, machining, grinding and polishing of the cast surface is necessary to fulfill the
requirements, especially for high energy applications like water injection or boiler feed
pumps. However, due to a lack of technically and commercially viable alternative
manufacturing technologies for closed impellers, casting is still the production process
of choice.
2.2 5 Axis CNC Milling

Subtracting material from existing geometries by milling is a common method to
produce components with an accuracy and surface quality far better than casting. The
commonly achievable range is, depending on the nominal size, in the µm-regime for
geometrical tolerances and an Ra of 0.4 µm for surface roughness [3]. The kinematics
of the used milling machine determines the complexity of achievable geometries. More
axis to manipulate the tool relatively to the workpiece allow machining of more
complex shapes.
In contrast to compressor impellers, where milling from solid ingots is meanwhile
state of the art, the milling of closed pump impellers is prevented by the high wrapping
angle of these components. The wrapping angle describes the angle between the inlet
and the outlet of an impeller channel relatively to the impeller rotary axis. Paired with
the limited cross sections at the channel in- and outlet the wrapping angle prevents
access to the full channel by straight milling tools.
A strategy to overcome these limitations is to combine 5-axis CNC milling with
Electrical Discharge Machining (EDM). In this case a wrought billet is milled down as
far as possible. Areas which aren’t accessible by straight tools are subsequently
removed by EDM.
2.3 Electrical Discharge Machining (EDM)

Electrical Discharge Machining (EDM) is a subtractive manufacturing process where
material is removed by means of spark erosion in dielectric fluid. A tool electrode,
usually made out of graphite, copper, or tungsten-copper, is approached towards the
work piece. If the distance falls below a certain value, the capacitor consisting of the
tool electrode and the workpiece is discharged by an arc-over, which removes material
from the workpiece.
148 R. Rettberg and T. Kraenzler
The advantage of EDM is that very complex shapes can be achieved with a very
high accuracy. The machining speed is almost independent of the alloy and there is no
mechanical and very low thermal stress induced in the workpiece. The disadvantage is
that component specific complex shaped electrodes are required, which are consumed
in the machining process. Manufacturing of pump impellers furthermore requires a
sophisticated kinematic to enter the impeller channels.
Sulzer has produced large double suction impellers in a combined process of 5-
Axis-CNC-Milling and EDM (Fig. 1). But due to the high individual efforts for dif-
ferent geometries and still some geometry limitations in impellers with very narrow
channels and at the same time high wrapping angles, the EDM process is only used in
special cases.
Fig. 1. Double suction impeller produced by 5-axis-milling and EDM
2.4 Laser Metal Deposition

Laser metal deposition is known since many years for metallic part repair. Sulzer
Innotec, a former Sulzer division started working with Laser Metal Deposition
(LMD) already in 1988. The main application at that time was the repair of worn shafts
and turbine blades.
Laser Metal Deposition, which is a specific manufacturing method in the group of
Directed Energy Deposition (DED), is a versatile method to add metallic material to a
metal surface. LMD can be considered as a laser welding process. Energy is brought to
a spot on a part surface by means of a laser beam. The same spot is fed via a nozzle by
a constant powder stream transported by an inert gas. Frequently an additional channel
is integrated in the nozzle, directing further inert shielding gas towards the welding
zone. The laser beam melts a part of the surface as well as the injected powder thus
building up a weld bead after the nozzle has moved on and the melt has solidified. By
adding multiple weld beads next to each other a dense coating can be applied. Due to
the lower heat input compared to conventional welding and the resulting low dilution
rate LMD is increasingly used for the application of wear and/or corrosion resistant
coatings. 3D geometries can be achieved by adding several layers upon each other via a
nozzle guided by a robot. The quality of the as deposited surface is however usually not
suited for a usage without prior machining, limiting LMD to the manufacturing of less
complex parts.
3 The Hybrid Process
To combine different manufacturing methods in a production process is natural since

every method has its own benefits, be it commercially or technically. However, for
parts with specific geometries it is technically necessary to switch between different
manufacturing methods very frequently and pump impellers can be considered as such
components.
The manufacturing of a pump impeller by a combination of LMD and milling
would require frequent changeovers from one machine to the other, consuming a lot of
time for realignment and acting as a source of machining errors. Therefore, only an
integrated machine which has the capability to do LMD, milling, and turning in a single
working chamber is able to utilize the full potential of a hybrid impeller manufacturing
process.
The present project was executed on a Lasertec 65 3D hybrid machine tool offered
by the machine manufacturer DMG Mori. The machine uses a standard 5-axis con-
figuration with 3 axis movement of the tool head (X, Y, Z) and two axes manipulating
the table bearing the workpiece (A, B). A schematic representation of the machine axes
and the iterative principle can be seen in Fig. 2.
DMG Mori was the first company to release such a hybrid machine, followed by
several other suppliers offering meanwhile similar solutions. However, the installed
base of Hybrid Machines in commercial enterprises is still rather low. The majority of
installations is up to now in institutes and universities.
The whole machine including additive and subtractive operations is programmed
by means of a CAD/CAM software. In a first step a model of the part has to be prepared
and each single section for later LMD build-up needs to be separated. Subsequently,
the milling and LMD strategies are applied to the separated bodies in order to generate
tool-paths, related tool-vectors, and parameter settings. Those are translated by a
machine specific post-processor to compile the so-called G-code which operates the
CNC machine.
Fig. 2. Schematic representation of the machine and iterative manufacturing principle
4 Impeller Manufacturing
Because of the specific geometrical features of closed pump impellers additive man-
ufacturing (AM) technologies can potentially be a viable alternative to casting.
Notwithstanding that the most common AM technologies like SLM (Selective Laser
Melting), LMD, and WAAM (Wire Arc Additive Manufacturing) are capable to pro-
duce such impellers, the required geometrical and surface quality cannot be met with an
AM process alone. And once a part is generated the restricted accessibility makes post
processing a lot more challenging. Potential processes to improve the surface finish of
AM generated impellers are EDM, ECM (electrochemical machining) and electro
polishing. All these processes require specific machines, tools and process knowledge,
adding further complexity to the manufacturing process.
That is the reason why the capability to apply an iterative process of additive and
subtractive methods in a single machine is such a big benefit. This approach as
described in Sect. 3 allows to produce an impeller based on a milled core part where
the LMD portions are added section by section intermitted by subtractive milling steps
until the final component is complete. All surfaces of the impeller, even those being not
accessible by any milling tool in the final state, can be machined with very high
accuracy and surface quality.
A schematic build process of such an approach is shown in Fig. 3. When defining
the manufacturing strategy, it is important to consider the capabilities but also the
limitations of the processes. This led to the complex shaped milled core (Fig. 3) as
starting point for the hybrid process.
Fig. 3. Schematic iterative impeller manufacturing process (patent pending)
5 Development Areas
In this chapter the two basic development areas for the manufacturing technologies
milling and LMD are described followed by a section that highlights the specific
development areas when using LMD as an AM method.
5.1 Milling Parameters and Strategies

Milling is one of the most common and well-known subtractive metal manufacturing
methods and thus not a specific focus topic in this paper. Nevertheless, it has to be
mentioned that for each material and geometrical characteristic a dedicated milling
process has to be developed for optimal results. Such processes include the use of
optimized tools, cutting parameters, and milling strategies. The most challenging area
with respect to closed impeller manufacturing is the milling of the inner cavities which
form the impeller channels. Dedicated technologies such as fast feed milling can be
used to reduce machining time.
5.2 LMD Parameter Development

For every alloy a specific parameter set has to be developed. The systematic approach
to get optimized parameters is shown in Fig. 4.
Fig. 4. General optimization order for LMD weld parameter development
Assuming the laser optics and powder nozzle are fixed via the selection of the
machine equipment the first parameters to be evaluated are the laser power, feed rate
(velocity of the nozzle and laser), and the powder feed which in combination determine
the energy input per unit length, as well as the inert gas flow of the powder feed system
and if applicable any additional shielding gas flows. The primary measures to be
optimized are the density of the added material and the dimensions of the weld. These
are determined in a parameter study based on single track welds. The single tracks are
evaluated based on cross-sections for their density and general quality. The weld
quality is rated by surface oxidation, the number of defects such as pores or cracks, and
the shape of the cross-Section 100% density as well as a regular outer shape is
required. Further quality criteria are the overall height and width of the weld bead, the
bead angle, and the dilution with the base material. These criteria have to be optimized
while trying to maintain a high deposition rate.
Once a set of basic weld parameters have been determined, the interaction of
subsequent weld tracks is investigated. For this purpose, several weld tracks are applied
parallel to each other with a specific overlap, also called step-over. These tracks are
produced in a meandering way. The step-over is varied to find an optimal value that
provides an even top surface without having any incomplete fusion or pores between
the single tracks. Evaluation is again done via cross-sections.
As a third step for the generic AM material build up the single layers of step two are
put on top of each other each shifted by 90° several times resulting in a cubical sample.
A very important step is the determination of the exact layer height in such a build.
Being a fixed input parameter for the tool path generation an inaccurate layer height can
potentially result in an increasing error in the subsequent layers. Since the welding
process is done in such a case further and further out of the intended laser focus
position and outside the powder nozzle focus the effect can be either self-enhancing or
self-healing. If the layer height is programmed too big the welding will happen further
and further away from the base material and ultimately fail. If the layer height is
programmed too low the theoretical ideal welding area will move further inside the
base material. Although this is not ideal for the process it will result in less deposited
weld material which is in favor of the subsequent layer and thus a kind of steady state
can be reached after a certain number of built layers. For this reason, it is necessary to
determine the real layer height as accurate as possible and to enable the program to
compensate for slight process variations. To ensure a stable process even with geo-
metrical variations of the component, it is recommended to use a slightly lower layer
height for machine programming.
5.3 Build Strategy Development

The development of basic build parameters for an alloy to be applied by a specific
LMD equipment and the subsequent material qualification can be done with simple test
geometries. However, in order to build an actual component, critical areas which need
additional investigation have to be identified and evaluated. Those areas, usually with a
more complex geometry require specific parameters that need to be developed and
verified. Luckily most of these geometries bear generic problems and the developed
parameters can be used for similar geometrical features in other components. In this
paper only geometrical shapes which apply for impeller manufacturing as described in
Sect. 4 are covered.
A geometric constellation that needs to be looked at carefully for most additive
technologies is an overhang (see Fig. 5). An overhang is created if the area in build
direction is larger than the starting cross-section of the build. While other AM tech-
nologies often require support structures to back the overhangs, the LMD process
allows to build overhangs without additional support structures up to a certain degree,
as long as the kinematics of the used equipment enable the necessary tool angles.
Nevertheless, in case of an overhang build the welding conditions change and need to
be compensated. Adding material right on top of an overhanging face increases the
powder losses, while some of the existing material can be molten away, which
decreases the yield of deposited material further. This effect needs to be compensated
by changing the angle of the nozzle, the welding parameters, and/or by depositing
additional layers at the overhanging edge. The required compensation measures are
dependent on the angle of the actual overhang. Figure 6 shows a sample that was used
for overhang build parameter optimization.
Fig. 5. Schematic example of an overhang with an angle a

Fig. 6. Example of a 45° overhang build
Another area which needs special attention is side-walls that will be integral parts
of the created geometry after an LMD-step (see Fig. 7). A side-wall changes the
conditions of the feeding- and shielding-gas as well as the heat convection in the
welding spot. In such regions the build parameters need to be adjusted according to the
shape and angle of the side-wall. Compensation measures can be additional laser
passes, adjustments to the toolpath-overlap, or in general adjustments to the basic
welding parameters in the connection area. Figure 8 shows a sample used for parameter
optimization in side-wall areas.
A more general question that needs to be answered based on the geometry of the
impeller is the build order and sequence. To avoid or to counteract distortion and to
keep internal stresses at a minimum, strategies from conventional welding processes
can be utilized. As an example, it is often beneficial to keep the part on a uniform
temperature level by adding a subsequent AM segment not on a channel next to the
recent one but on the opposing side.
Fig. 7. Schematic example of a side-wall connection

Fig. 8. Example of a build connected to a side-wall
6 Qualification
Currently there are no material standards that are specifically aimed at AM via LMD.
But contrary to other AM methods LMD is similar enough to conventional welding
technologies to be qualified under the same standards framework in many aspects.
Customer requirements and specifications are dictating the qualification process
depending on the application and thus it is important to develop the qualification
process and all requirements in close collaboration with the end users of the
components.
6.1 Powder Qualification

In additive manufacturing the quality of the powder is determined by much more than
only the chemical composition. Particle size and shape play an important role for
achieving defect free components. Flowability is crucial for a steady powder feed.
Trace elements in the powder have a much bigger influence on the component quality
and the resulting material properties than in casting [4]. The goal is therefore to use
only specific powders designed for LMD. Due to the fact that respective standards are
not available yet, it is the responsibility of the AM manufacturer to specify the required
powder properties and to ensure with the powder manufacturer an unchanging quality
of the powder. To avoid any unforeseen problems with variations in the powder quality
the entire qualification process was executed with a 316L powder according to the
specification of one selected supplier. The importance of very tight powder specs was
proven, when the production of test specimen with identical build parameters from
316L powder with equal chemical composition but from a different manufacturer led to
inacceptable results due to hot cracking.
6.2 Material/Welding Qualification

In quality terms welding is identified as a “special process” where the quality of the
product cannot be fully determined in the finished part by non-destructive testing but
has to be ensured by a comprehensive specification and a stringent process control.

This approach should be adopted for the LMD process as the quality of the deposited
material is mainly determined by the process parameters.
The basic material qualification should be carried out in the form of a classic
welding qualification via a welding procedure specification (WPS) and a supporting
procedure qualification record (PQR) according to an acknowledged welding code like
ASME BPVC Sect. IX [5]. The specification covers all production parameters and their
applicable ranges. The parameter ranges are additionally tied to specific component
sizes. Mandatory verification tests include tensile-, Charpy-, and hardness-tests as well
as the examinations of macro sections. These mandatory tests are complemented by
application specific tests such as stress corrosion or pitting corrosion tests, which
become part of the WPS/PQR.
All tests are carried out for bulk AM material as well as for transition areas between
base and AM material including the heat affected zone.
It has to be mentioned that any change in the material, the equipment, or the process
will trigger a new qualification process. The standard AWS D20.1/D20.1 M [6] pub-
lished by the American Welding Society provides a good guideline how to qualify
LMD processes.
6.3 Heat Treatment

A heat treatment can be necessary either to achieve the intended material properties or
to reduce internal stresses in the part. Specific heat treatment cycles, deviating from
known regimes for forged and cast materials may be required due to the different
micro-structure. In all cases the material testing has to be carried out in the final heat-
treated condition.
6.4 In-process Monitoring

Because LMD is a continuous process it is important to monitor the most important
parameters during the manufacturing process. An example of such a monitoring tool is
shown in Fig. 9. The intention is to ensure that the process is running within the pre-
defined boundaries as well as to have means of documentation for reference after the
build. Triggers to interrupt the AM process might be for example abnormal temperature
readings or changes in the powder supply. This way potential damage to the equipment
as well as defective parts can be prevented. All monitored parameters are stored in a
data file that has a unique identifier to the build job. It is important that all measured
parameters are stored together with the axis information of the machine tool in order to
be able to track back the exact position within the manufactured part. In case an error or
irregularity is discovered in a quality check or throughout the lifetime of the part a
correlation between the recorded process parameters and the irregularity can be
established.
Fig. 9. Process control panel
6.5 Part Quality Checks

The qualification of a component produced via the Hybrid process does not differ
significantly from a qualification of conventionally manufactured parts like castings.
The part is checked for its geometrical tolerances with templates or 3D scans. The
soundness is proven by a dye penetration tests (PT). In addition to these standard
procedures, components produced with new AM strategies or new materials should be
completely checked by volumetric methods like ultrasonic testing (UT) or computed
tomography (CT).
7 Results
Sulzer progressed through all qualification steps described in the previous chapters and
produced first demonstrator impellers in fully dense 316L material. All mechanical
properties have exceeded the requirements of conventionally manufactured 316L parts
while conserving the required corrosion resistance.
Major improvements in geometrical accuracy and surface quality have been
achieved up to the level of integrally milled parts manufactured from wrought raw
material. Along with the geometrical accuracy balancing efforts have been reduced
significantly.
Apart from component quality improvements the main goal of significant lead time
reduction has been realized. A production time from raw bar material to final impeller
of 80 h without any milling speed optimizations has been achieved already with the
first ever hybrid manufactured impellers (Fig. 10).
Fig. 10. Example of a pre-machined core part and the finished Hybrid Impeller
8 Conclusions
It has been demonstrated that the combination of LMD with multi-axis milling is a
viable solution for the production of closed impellers and other complex shaped parts.
Both technologies are well-known and have been optimized over many years. Nev-
ertheless, the specific use of LMD for AM and the direct combination of both additive
and subtractive technologies bear some additional challenges that need to be considered
in order to produce fully qualified components with good properties. Examples are the
parameter development for 3-dimensional builds described in Sect. 5.2 and special
geometrical constellations as described in Sect. 5.3.
The major difference of the Hybrid process, compared to the most common metal
AM technology (SLM), is how freedom of design can be utilized to produce large
complex shaped parts. By SLM very fine and detailed structures can be produced and
with the aid of support structures there are almost no limitations to the freedom of
design. However, there are still limitations to the overall size of the parts and due to the
way the process works, they are always built in parallel planes. LMD on the other hand
when controlled with a multi-axis machine tool has the freedom to build in any
direction and the direction can be changed according to the geometry of the
part. Especially for iterative processes like the one highlighted in this paper this
additional degree of freedom is a big benefit. A further advantage of the hybrid process
is the achievable high surface quality due to the integrated milling capabilities. Fur-
thermore, the hybrid process allows a potential multi material processing which may be
used to create specific functional features.
For the presented scope of work the production time even without machining time
optimizations was already a huge step towards a very short lead time. But as briefly
described in Sect. 3 the manufacturing program for each component needs to be tailor-
made. In order to do the programming efficiently the currently available software
solutions need to be developed further. This is done by well-known CAM-software
providers in cooperation with users but also customized software automations are
developed by end-users to allow for a quick and accurate CAM programming of
individual components. This is an important development area which can be the
decisive factor for the overall production lead time.
Commercial aspects of the hybrid approach with LMD are very promising.
Although the component cost when compared to the conventional casting process is a
bit higher the significant lead-time reduction by 4–6 weeks can be the deciding factor
when parts are needed quickly. Especially spare parts are an attractive field for this
technology. A fast on-demand production can save a lot of inventory in spare part
warehousing but also the storage of physical casting models can be superseded. Fur-
thermore, the improved properties like the surface quality can be a deciding factor to go
for this alternative production method.
References
1. Huber, et al.: Process setup for manufacturing of a pump impeller by selective laser melting,
industrializing additive manufacturing. In: Proceedings of Additive Manufacturing in
Products and Applications – AMPA 2017. Springer, Cham
2. Sulzer Pumps Ltd.: Sulzer Centrifugal Pump Handbook. Elsevier Science Publishers Ltd.,
Essex (1989)
3. Fischer, U., et al.: Tabellenbuch Metall, 45th edn. Europa-Lehrmittel, Haan-Gruiten (2011)
4. Oerlikon, O.C.: Effect of minor elements on properties of C300 Maraging steel parts by SLM
process. In: Additive World Conference on Ind. 3D Printing, Eindhoven, NL, 5th edn. (2017)
5. ASME: BPVC Section IX-Welding, Brazing, and Fusing Qualifications. ASME, New York
(2017)
6. AWS D20.1/D20.1 M 2019. Standard for Fabrication of Metal Components using Additive
Manufacturing, AWS, Miami, FL, USA
Adaptive Slicing and Process
Optimization for Direct Metal Deposition
to Fabricate Exhaust Manifolds
Daniel Eisenbarth1(B) , Alessandro Menichelli2 , Fabian Soffel1 ,

and Konrad Wegener2
1
Inspire AG, ETH Zurich, Zurich, Switzerland
eisenbarth@inspire.ethz.ch
2
Institute of Machine Tools and Manufacturing, ETH Zurich, Zurich, Switzerland
https://www.inspire.ethz.ch/
Abstract. Direct metal deposition is an additive technology that has

the potential to fabricate large parts in multiple buildup directions. Espe-
cially curved, thin-walled geometries such as exhaust manifolds are a
promising use case: In theory, direct metal deposition allows nearly arbi-
trary shapes. Internal surfaces that are not accessible with the final part
could be inspected and machined in a stepwise buildup process. However,
the successful production of such parts requires suitable algorithms for
five-axis tool path planning as well as for the optimization of the param-
eters for the specific process. Herein, an adaptive slicing algorithm is
presented that aligns the direction of each layer for minimized overhangs
and creates the tool path under consideration of the process capabili-
ties and limits. By a variation of the scan speed, the deposited powder
per length and therefore the layer height can be modified continuously.
A model-based feedforward control of the laser power accounts for the
varying thermal conduction in thin walls. These approaches are inte-
grated in a fully automated CAM software that generates a suitable tool
path with locally adapted parameters. The fabrication of an exemplary
exhaust manifold shows that the software reduces the manual prepara-
tion effort and enables a flexible additive manufacturing process.
Keywords: Direct metal deposition · Adaptive slicing · Tool path

planning · Process optimization · CAM programming
1 Introduction
Additive manufacturing (AM) of metals is currently dominated by powder-bed
technologies. Although they show various advantages, there are process-inherent
drawbacks such as the predefined, vertical buildup direction and the necessity of
support structures to fabricate overhanging walls above a critical angle. While
powder-bed AM is characterized by the two steps of powder placement and
fusion, there are various deposition welding technologies that feed the mate-
rial directly into the melt pool. With these AM processes, both the energy and
https://doi.org/10.1007/978-3-030-54334-1_12
Adaptive Slicing 161
material are delivered by a single processing head, which is mostly attached to

a robot or CNC machine. Thus, only the axis range limits the design space, and
multi-axis systems allow an arbitrary orientation of the processing head relative
to the workpiece. This flexibility promises the additive production of large parts
without the need for support structures as demonstrated by Greer et al. [1]. As
of today, deposition welding is applied in the industry mainly for coatings and
the repair of geometrically simple elements as shown by Petrat et al. [2]. The
fabrication of complex, multi-layer structures in arbitrary buildup directions is
still in an early research phase, as it poses various challenges that need to be
solved. These challenges can be divided into digital and physical issues: First, a
suitable and collision-safe tool path needs to be calculated from a CAD model,
ideally in an automated approach. Second, the applied deposition welding pro-
cess needs to be able to manufacture the part with the desired microstructure,
surface roughness, and geometrical accuracy. A common depositing welding tech-
nology is direct metal deposition (DMD) as used in here, where metallic powder
is blown by a gas stream into a melt pool created by a laser beam.
Figure 1 presents eight generic slicing approaches to create layers from a 3D
model, depending on the ability of the tool path to change the buildup direction
and on the ability of the process to vary the layer height locally. Three levels
of complexity are distinguished here. The easiest approach of a constant layer
direction and height is shown by illustration (a), with the layers marked in gray
and black on a white substrate. By dividing the part into sub-volumes as shown
in (b), the buildup direction can be optimized for each geometrical element in
order to prevent overhangs or to improve the surface roughness as performed
by Ding et al. [3] and Murtezaoglu et al. [4]. The tool path calculation becomes
most complex when the direction changes within one layer. This approach is
required for AM processes that start on a curved substrate as shown by Zhao et
al. [5]. With a constant layer height as illustrated in (c), each layer is an offset
of the initial substrate curvature.
If it is possible to change the deposition height during the process, the stair-
case effect that occurs at slopes and overhangs can be reduced by a variable
height from layer to layer as proposed by Mao et al. [6] and shown in (d). Illus-
trations (e) and (f) depict combinations of the previously explained approaches.
Multidirectional but plane slicing is illustrated in (g). By a continuous adapta-
tion of the buildup direction, overhangs and the staircase effect can be prevented
for parts that tilt gradually from the bottom to the top. This approach requires
a local adaptation of the layer height, which can be achieved for instance by a
variation of the track overlap or the deposition rate per length. An alternative
is to generate a sloped surface by a staircase, consisting of multiple sub-layers.
Chalvin et al. [7] and Wang et al. [8] implement the slicing approach (g) for
polymer-based filament AM. Ruan et al. [9] perform adaptive slicing with met-
als, but create the required slope by milling of each layer. The difficulty of
adaptive slicing in combination with DMD is the dynamic control of the layer
height while ensuring similar process conditions without over- or underheating
162 D. Eisenbarth et al.
Fig. 1. Generic slicing approaches with a varying tool path and process complexity
of the part. Varying both the buildup direction and the layer height as shown in
(h) is demonstrated with a simple tube by Wang et al. [10].
This publication deals with adaptive slicing according to approach (g) in combi-
nation with a five-axis tool path, demonstrated by the fabrication of a twisted,
thin-walled exhaust manifold. Material scientific issues are not discussed. How-
ever, the part geometry influences the processing conditions as outlined by
Eisenbarth et al. [11]. A nearly steady-state DMD process can only be achieved
by an adaptation of the critical process parameters. Here, a model-based feed-
forward control method is applied to optimize both the laser power and the scan
speed according to the local process requirements. These approaches are inte-
grated in a fully automated CAM software as developed in-house using MAT-
LAB. Figure 2 shows a simplified flowchart of the adaptive slicing algorithm
within the CAM software architecture, highlighted in gray. The subsequent mod-
ules calculate the tool path and the process parameters before the NC-code is
created. The depicted algorithm is explained in the following sections.
Fig. 2. Flowchart of the adaptive slicing loop and its integration in the CAM software
2.1 Concept of Adaptive Slicing
An inclined slicing plane with layer number i is defined by a position and a

normal vector mi , which are calculated from the 3D model to be built. The
STL file format is used as input, as it is widespread, independent from specific
CAD modeling methods, and simple to decompose. Various approaches exist to
determine the normal vector: The centerline of a part can be imported from
a CAD software or calculated explicitly as done by Ruan et al. [12]. Problems
occur as soon as the part splits into or merges from multiple sub-volumes. An
alternative is to focus on the DMD process requirements, namely on minimizing
the overhang angles independent of any centerline. It is possible to minimize the
maximum overhang angle as done by Ding et al. [3], to use a certain cost function,
or to minimize the average overhang angle per path length as performed herein.
In addition, the process limits in terms of the minimum and maximum achiev-
able layer heights δhmin and δhmax need to be considered by the slicing algo-
rithm. The maximum tilt angle βa,max is an inverse tangent function of the
maximum layer height difference divided by the maximum part width wprt in
the tilting direction:

δhmax − δhmin
βa,max = tan−1 (1)
wprt
The most inward wall of the part can show a minimum surface radius ra,min ,
approximated as
δhmin
ra,min = (2)
βa,max
Thus, the achievable surface radius is constrained by the DMD process capa-
bilities and the part width in the tilting direction. The normal vector mi is
determined in an iterative process: In the first iteration, the part is sliced paral-
lel to the previous layer i − 1 in a default distance Δh as shown in illustration
(a) of Fig. 3. The gray volume depicts the cross-section of an arbitrary part in
the xz-plane.
Fig. 3. Adaptive slicing in an iterative process: Calculation of the intended normal

vector mi (a), translating and rotating until a convergence criterion is met (b)
The intersection of the plane with the triangles of the STL geometry results in
one or more closed contour paths, defined by set points pk,i for each intersection
k in layer i. The total number of points in one layer is denoted as N . The
path from one set point to the next can be written as path segment vector
uk,i with a related normal vector nk,i of the respective triangle that defines the
wall orientation. The average wall direction is calculated as the sum of the cross
products of uk,i and nk,i , corresponding to normal vector of the new slicing plane
mi :
N

mi = uk,i × nk,i (3)
k=1
Since n is a unit vector and u is the path segment length, the direction of mi
is weighted inherently according to the respective path distances. The average
wall inclination βp,avg is the angle between the normal vectors of two iterations.
As long as this angle is larger than 0, a mismatch exists between the slicing
orientation and the average wall inclination. The intersection of the new slicing
plane with the STL geometry creates a different contour path that can result
in a varying normal vector mi . Thus, this algorithm is repeated until βp,avg is
0 or complies to a certain convergence criterion as shown in illustration (b) of
Fig. 3. From the previous to the next slicing plane as well as from one iteration
to another, the angle βa between two normal vectors mi−1 and mi is calculated
with the dot product:

−1 mi · mi−1
βa,i = cos (4)
|mi | · |mi−1 |
At the same time, the algorithm needs to ensure that the local layer height
δhk,i at each set point pk,i lies within the layer height limits of the process. If
these limits are exceeded, the plane is translated and rotated accordingly, con-
sidering the maximum tilt angle βa,max . The local layer height can be calculated
with the law of conservation of mass, assuming a constant total powder flow
rate ṁ, a constant powder catchment efficiency η as the fraction of the powder
reaching the melt pool, and a constant density ρ of the solid material. If a thin
wall is made of stacked DMD tracks with an average wall width wm , the local
layer height is inversely proportional to the local scan speed v at a set point pk,i :
ṁ · η
δhk,i = · vk,i −1 (5)
wm · ρ
The volumetric energy density ev describing the heat input for a certain
volume of deposited material is calculated as:
P P ·ρ
ev = = (6)
vk,i · wm · δhk,i ṁ · η
It can be seen that ev does not depend on the local scan speed. Since a local
variation of the layer height does not influence the amount of deposited material
per time, it is reasonable to apply a constant laser power P to maintain a constant
volumetric energy density. However, the specific energy ea as a measure for the
heat input on a certain workpiece area is a function of v:
P
ea,k,i = · vk,i −1 (7)
wm
Thus, the process window for a varying layer height is constrained by the
minimum and maximum scan speeds that allow a proper bonding to the last
layer. Furthermore, the maximum tilt angle βa,max of a plane is limited by
the dynamics of the DMD process: Since the melt pool is liquid, the amount
of deposited powder per length is averaged within the area of the melt pool.
Thus, a rapidly changing scan speed will not lead to the same dynamic response
of the local layer height.
If the radius ra of the CAD model is smaller than the minimum achievable
radius ra,min of the DMD process, the adaptive slicing will result in a growing
mismatch between the ideal and the actual slicing planes. A certain overhang
can be addressed by a five-axis tool path as shown in the following section.
Unfeasible overhangs require either a redesign of the part, or a segmentation
into sub-volumes with different buildup directions according to slicing approach
(b) of Fig. 1.
2.2 Five-Axis Tool Path Creation

A machine tool with a minimum of three linear and two rotational axes is
required to rotate the processing head in any direction. Here, a CNC machine
with three linear axes X, Y, and Z as well as a tilt axis B (around the Y-axis)
and a rotation axis C (around the Z-axis) is used. In order to modify the tool
axis orientation while maintaining the tool center point (TCP) at the position
of the melt pool, the NC controller calculates compensation moves of the linear
axes when rotations are performed. In contrast to most robots, a five-axis CNC
machine allows an unambiguous transformation between the position and orien-
tation of the TCP and the axes and joints of the machine. A singularity exists
at B = 0◦ , as it allows an arbitrary C-angle without an effect on the tool axis
orientation.
Although the adaptive slicing ensures a normal vector mi parallel to the
average wall direction, parts with a varying cross-section show local overhangs in
different directions. For instance, the average wall direction of a cone is constant
and parallel to the axis of revolution, but there is still a remaining overhang
due to the taper angle as illustrated in Fig. 4 (a). For the fabrication of a thin-
walled part by DMD, the direction of the laser beam needs to be parallel to the
local wall direction. Otherwise, small deviations of the part height will lead to
a mismatch between the position of the laser beam and the actual part surface.
Thus, the B and C angles need to be calculated for each tool path segment based
on the local STL triangle.
Figure 4 (b) shows the contour path at an arbitrary position k and layer i,
intersecting a triangle with a normal vector nk,i . The normalized path direction
ûk,i is given as the cross-product of mi at a position k and nk,i :
ûk,i = mi × nk,i (8)
The local wall direction m̃k,i has an overhang angle βp relative to the normal
vector mi of the slicing plane and is calculated as:
m̃k,i = nk,i × ûk,i (9)
With m̃k,i as the tool axis orientation, the B-tilt angle Bk,i of the machine
tool is determined by the z-component of m̃k,i in the machine coordinate system
(x, y, z):
Bk,i = cos−1 (m̃k,i 3 ) (10)
Fig. 4. Exemplary cone with overhangs despite adaptive slicing (a), geometric deriva-
tion of the tool axis m̃k,i parallel to the wall (b)
The C-rotation angle Ck,i is given by the x- and y-components of m̃k,i with
the four-quadrant inverse tangent:

m̃k,i 2 π m̃k,i 2 m̃k,i 1
Ck,i = tan−1 + 1− (11)
m̃k,i 1 2 |m̃k,i 2 | |m̃k,i 1 |
The local angles (B, C)k,i are assigned to the current TCP (x, y, z)k,i , thus
the position and orientation of the processing head are defined for each path
segment. Knowledge of the actual machine kinematics is not required as far as
the NC controller transforms the TCP coordinates into machine coordinates. As
a further step before completion of the NC-code, suitable process parameters
need to be determined.
2.3 Laser Power Adaptation

Especially for the fabrication of thin-walled parts, the processing conditions vary
significantly between the bottom and the top of a part: The first layer deposited
onto a cold, massive substrate is subjected to high thermal conduction through
the workpiece. In contrast, the heat flow and the thermal gradient at the top of
a heated, thin-walled part are significantly lower as revealed experimentally by
Akbari and Kovacevic [13] for laser/wire AM: They show that the microstructure
between the bottom and the top of a thin wall varies highly due to the different
cooling rates in different production stages.
In order to ensure comparable melt pool properties for different part geome-
tries and workpiece temperatures, the heat input in terms of the laser power
needs to be adapted. This can be done by melt pool monitoring and a closed-
loop control, or by a model-based feedforward control as further explained by
Eisenbarth et al. [11]: During generation of the NC-code, a digital twin of the
part is created that shows the deposited material in each stage of production.
Simultaneously, the local part geometry around the current position of the melt
pool is analyzed. Figure 5 illustrates the principle: The algorithm spans a control
volume of a fixed size (green) around the melt pool (red). A geometric factor
κ indicates the amount of material inside the control volume that is available
for thermal conduction. If processing is performed on a massive part, κ equals
1. For a thin wall, the workpiece does not fill the control volume entirely and κ
decreases accordingly.
Fig. 5. Control volume (green) during fabrication of an exemplary part (blue) on a

massive substrate (gray) according to [11]
κ is therefore a measure for the massiveness of a part. The geometric factor is

then correlated to the required laser power by an experimental calibration with
certain test geometries such as a massive block and a thin wall. With suitable
laser power values P1 for κ = 1 and P0 for κ = 0 for a specific material and pro-
cess window, values within these limits are interpolated linearly. Compared to
a thermodynamic simulation, a geometry-based algorithm as shown here is fast,
can handle complex geometries, and does not rely on advanced material prop-
erties. The laser power adaptation is independent of the scan speed adaptation
for the adaptive slicing, but all possible combinations of P and v are considered
for the definition of the process window.
2.4 Experimental Setup
For the fabrication of parts, a prototype machine for combined DMD and milling
from the Swiss company GF Machining Solutions is used. The CNC machine has
a linear axis range of X = 600 mm, Y = 450 mm, and Z = 450 mm. The table
rotates 360◦ around the C-axis and turns from −120 to +45◦ around the B-axis.
A DMD system type AMBIT from company Hybrid Manufacturing Technologies
with a nominal laser spot size of 3 mm and a maximum laser power of 1000 W
at a wavelength of 1070 nm is integrated in the machine. The initial distance
from the workpiece to the nozzle is 8 mm. The nickel-base alloy Inconel
R
(IN)
718 from Carpenter Additive is applied exemplarily, as it is easy to weld and
suited for high-temperature applications in the energy and aerospace industry.
The powder has a particle size distribution from 44 to 105 μm. DMD process
parameters are listed in Table 1.
Table 1. DMD process parameters for material IN718
Parameter Value
Min. laser power P0 [W] 480
Max. laser power P1 [W] 1000
Min. scan speed vmin [mm/min] 150
Max. scan speed vmax [mm/min] 450
Powder flow rate ṁ [g/min] 4.6
3.1 Layer Height Calibration
A suitable process window for thin-walled parts from IN718 was determined
in pretests with the goal of a pore- and crack-free microstructure. For the cal-
ibration of the layer height model according to Eq. (5), straight, thin-walled
tubes with a height of 9 mm are fabricated with different scan speeds and a
constant laser power P1 = 1000 W. Figure 6 plots the resulting layer height.
The hyperbolic fit proves that δh is inversely proportional to v. At a scan speed
of 150 mm/min, the local layer height is 0.8 mm with a wall thickness wm of
3.6 mm. At v = 450mm/min, δh decreases to 0.3 mm and wm decreases slightly
to 2.9 mm. Within this window, the specific energy ea ranges between 46 and
111 J/mm2 . The average volumetric energy density ev is 147 J/mm3 with a
standard deviation of ± 15 J/mm3 . Deviations of ev are due to the fact that the
powder catchment efficiency η increases with the melt pool size. According to
Eq. (6), the volumetric energy density is slightly reduced in areas with a higher
wall thickness and powder catchment efficiency. Considering the laser power
adaptation, ev decreases to 71 J/mm3 if the minimum laser power P0 = 480 W
is applied.
Fig. 6. Local layer height δh as a function of the scan speed v
3.2 Exhaust Manifold Fabrication
The exhaust manifold is modeled as a sweep of a varying cross-section along a

3D guide curve as shown in Fig. 7 (a). The cross-section starts with an oval shape
with a size of 40 times 30 mm, and transitions into an hourglass shape with a
size of 96 times 40 mm. In the lower section, the guide curve follows a circular
arc in the xz-plane up to a tilt angle of 60◦ from the initial z-direction. In the
upper section, the guide curve is reversed and follows a plane that is rotated by
30◦ relative to the previous xz-plane. The top layer has a tilt angle of −80◦ . The
exhaust manifold has a total height of 250 mm and a minimum surface radius
ra of 18 mm in the tilting direction, which is also the achievable radius ra,min of
the chosen process window.
Figure 7 (b) depicts the tool path, consisting of stacked contours. For clarity
reasons, only every 10th layer is displayed. The algorithm calculates adapted
slicing planes that can follow the curvature of the part, since the model geome-
try complies with the process limitations. Due to the varying cross-section, local
overhangs exist with a maximum angle of βp = 15◦ . Picture (c) shows the geo-
metric factor κ in pseudo-colors: In the first layer, processing starts at κ = 1 with
a laser power of P1 = 1000 W to ensure a proper bonding to the substrate. After
reaching 20 mm wall height, the influence of the substrate on thermal conduction
in the thin wall is negligible and κ converges to 0.13 with a corresponding laser
power of 550 W, which remains nearly constant until the end of the process.
The fully automated CAM software does not require any engineering work,
except of the CAD modeling of the part and the initial parameter development
for a specific material. The calculation of the tool path for the exhaust manifold
takes 14 min with a desktop PC, using a single core of a 2.5 GHz processor. The
DMD process allows a buildup rate of 154 g/h with an average powder catchment
Fig. 7. CAM modeling of the exhaust manifold: CAD (a), tool path (b), and geometric
factor κ (c)
Fig. 8. Exhaust manifold during processing (a), as-built part with cut top layer (b)
efficiency of 57%, resulting in a total weight of the exhaust manifold of 1.2 kg. The
fabrication process as depicted in Fig. 8 takes 7 h 50 min. The top layer shows
a certain waviness due to process irregularities. The photograph in (b) shows
the as-built exhaust manifold, deposited on a 8 mm thick substrate from steel
S235JRC. The top layer is cut to show the difference to the dark wall surface,
indicating a certain oxidation of IN718. Compared to a planar part, the shield
gas stream is less effective on a thin wall. Thus, prevention of any oxidation
would require a processing chamber with a controlled shield gas atmosphere.
Deposition welding technologies in combination with a multi-axis CNC machine

allow the additive fabrication in arbitrary directions. This publication addresses
both the digital and physical challenges of multi-directional AM: An adaptive
slicing algorithm is presented that changes the buildup direction gradually. Pro-
cess parameters in terms of the scan speed and laser power are optimized to
account for the characteristics of the DMD process. Thus, the virtual tool path
planning is connected to the actual capabilities and limits of the DMD process,
shown exemplary for the fabrication of an exhaust manifold. With a CAD model
as input, an algorithm calculates the adaptive slicing planes based on the average
wall inclination. The remaining overhang is considered by an alignment of the
tool axis orientation to the local wall direction. With a five-axis machine tool,
the required rotation angles can be calculated unambiguously.
It is shown that the layer height is inversely proportional to the scan speed.
Thus, a variable layer height can be realized by a varying scan speed within a
certain process window. A model-based feedforward control of the laser power
is proposed to adapt the heat input to the changing thermal conduction in thin-
walled parts. Finally, the fabrication of a 250 mm high, twisted exhaust manifold
from IN718 is shown. The calculation of the required NC-code is performed
fully automated based on the CAD model and process parameters for a certain
material, reducing the preparation time and effort for a build job significantly.
In the future, the adaptive slicing algorithm should be enhanced for more
complex geometries such as splitting and merging parts as well as varying wall
thicknesses. The goal is to gain full control of the local layer height for any tool
path strategy while ensuring a dense microstructure. A stepwise buildup with
intermediate machining and inspection of the internal surfaces would improve
the flow properties inside the exhaust manifold.
Acknowledgements. The authors would like to acknowledge the contribution of the

funding agency Innosuisse (grant number 25498) and of the companies GF Machining
Solutions, GF Precicast, and ABB Turbo Systems AG.
References
1. Greer, C., Nycz, A., Noakes, M., Richardson, B., Post, B., Kurfess, T., Love, L.: Intro-
duction to the design rules for metal big area additive manufacturing. Additive Man-
ufacturing 27, 159–166 (2019). https://dx.doi.org/10.1016/j.addma.2019.02.016
2. Petrat, T., Graf, B., Gumenyuk, A., Rethmeier, M.: Laser metal deposition as
repair technology for a gas turbine burner made of inconel 718. Phys. Procedia 83,
761–768 (2016). https://doi.org/10.1016/j.phpro.2016.08.078
3. Ding, D., Pan, Z., Cuiuri, D., Li, H., Larkin, N., van Duin, S.: Automatic multi-
direction slicing algorithms for wire based additive manufacturing. Robot. Comput.
Integr. Manuf. 37, 139–150 (2016). https://doi.org/10.1016/j.rcim.2015.09.002
4. Murtezaoglu, Y., Plakhotnik, D., Stautner, M., Vaneker, T., van Houten, F.J.A.M.:
Geometry-based process planning for multi-axis support-free additive manufactur-
ing. Procedia CIRP 78, 73–78 (2018). https://doi.org/10.1016/j.procir.2018.08.175
5. Zhao, G., Ma, G., Feng, J., Xiao, W.: Nonplanar slicing and path generation meth-
ods for robotic additive manufacturing. Int. J Adv. Manuf. Technol. 96(9-12),
3149–3159 (2018). https://dx.doi.org/10.1007/s00170-018-1772-9
6. Mao, H., Kwok, T., Chen, Y., Wang, C.: Adaptive slicing based on efficient profile
analysis. Comput. Aided Des. 107, 89–101 (2019). https://doi.org/10.1016/j.cad.
2018.09.006
7. Chalvin, M., Campocasso, S., Baizeau, T., Hugel, V.: Automatic multi-axis path
planning for thinwall tubing through robotized wire deposition. Procedia CIRP
79, 89–94 (2019). https://doi.org/10.1016/j.procir.2019.02.017
8. Wang, M., Zhang, H., Hu, Q., Liu, D., Lammer, H.: Research and implementa-
tion of a non-supporting 3D printing method based on 5-axis dynamic slice algo-
rithm. Robot. Comput. Integr. Manuf. 57, 496–505 (2019). https://doi.org/10.
1016/j.rcim.2019.01.007
9. Ruan, J., Eiamsa-ard, K., Liou, F.W.: Automatic process planning and toolpath
generation of a multiaxis hybrid manufacturing system. J. Manuf. Processes 7(1),
57–68 (2005). https://doi.org/10.1016/S1526-6125(05)70082-7
10. Wang, X., Liu, Z., Guo, Z., Hu, Y.: A fundamental investigation on three-
dimensional laser material deposition of AISI316L stainless steel. Opt. Laser Tech-
nol. 126, 106107 (2020). https://dx.doi.org/10.1016/j.optlastec.2020.106107
11. Eisenbarth, D., Soffel, F., Wegener, K.: Geometry-based process adaption to fab-
ricate parts with varying wall thickness by direct metal deposition. In: Progress
in Digital and Physical Manufacturing, Progress in Digital and Physical Manufac-
turing, pp. 125–130. Springer, Leiria (2020). https://doi.org/10.1007/978-3-030-
29041-2 16
12. Ruan, J., Sparks, T.E., Panackal, A., Liou, F.W., Eiamsa-ard, K., Slattery, K.,
Chou, H.N., Kinsella, M.: Automated slicing for a multiaxis metal deposition sys-
tem. J. Manuf. Sci. Eng. 129(2), 303–310 (2006). http://dx.doi.org/10.1115/1.
2673492
13. Akbari, M., Kovacevic, R.: An investigation on mechanical and microstructural
properties of 316lsi parts fabricated by a robotized laser/wire direct metal deposi-
tion system. Addit. Manuf. 23, 487–497 (2018). https://doi.org/10.1016/j.addma.
2018.08.031
Quality
Drift Detection in Selective Laser Melting
(SLM) Using a Machine Learning Approach
Pinku Yadav1,2(&), Olivier Rigo1, Corinne Arvieu2, Emilie Le Guen2,

and Eric Lacoste2
1
SIRRIS, Rue du Bois Saint-Jean 12, 4102 Seraing, Belgium
pinku.yadav@u-bordeaux.fr
2
University of Bordeaux, CNRS, Arts et Metiers Institute of Technology,
Bordeaux INP, INRAE, I2M Bordeaux, 33400 Talence, France
Abstract. Selective laser melting has seen a growing demand in niche industry
applications for its complexity-free manufacturing capabilities. Besides, all the
advantages over traditional manufacturing techniques, reliability, and repeata-
bility is still a challenge. In-situ monitoring systems comprising high precision
sensors such as photodiodes, High-speed IR cameras both in co-axial and off-
axis positions have been installed in commercially available machines to
improve the overall performance of the process. However, understanding the
correlation between the acquired data from sensors and build quality is a
challenge and time-consuming. Thus, the sensitivity analysis of sensors installed
on the machine to various defects and easy detectability of the drift during the
process is the need of the time.
In this work, a sensitivity analysis of the melt pool monitoring system
installed in the commercial SLM 280HL machine is presented. Also, a super-
vised machine learning approach for in-line detection of the drift in the final part
is used. A balanced labeled dataset for training the support vector machine
algorithm by inducing drifts in the parts artificially (overheating and lack of
fusion) was prepared. Then the trained classifier successfully classifies the layers
of the unlabeled case study samples into “drift” and “no-drift” labels. Here, drift
classification is based on overheating and lack of fusion defects. Thus, the
machine learning approach is robust and cost-effective for techniques like
selective laser melting, where to obtain labeled data is time-consuming and
expensive.
Keywords: Selective laser melting In-situ monitoring Melt pool

monitoring Supervised learning
1 Introduction
The Selective Laser Melting (SLM) is one of the additive manufacturing (AM) process
that allows the manufacturing of three-dimensional complex functional structures of
metals or alloys using high power laser [1–3]. The recoater spreads a layer of metal
powder uniformly across the build platform. A high power laser beam scans the
individual cross-sections of the parts to be built using a galvo scanner with an f-theta
lens. In the scanning regions, the metal powder melts, consolidates, and fuses the

https://doi.org/10.1007/978-3-030-54334-1_13
178 P. Yadav et al.
scanned layer to the previous layer already built. The whole process is repeated until
the part is fabricated. The L-PBF process typically carried out in an inert atmosphere to
prevent oxidation of the material and chemical reactions between the molten material
and the environment.
Due to the large numbers of controllable and uncontrollable parameters, it is
challenging to control the process for better repeatability and reliability. For example,
Galy et al. summarized all the possible defects arising in the Al alloy due to a variation
in process parameters [4]. Due to the layer-wise principle of the process, these defects
are complicated to detect after the part is completed. An expensive technique such as
computed tomography is performed to ensure the internal quality of the final
part. Since CT is expensive and time-consuming, it is not viable to use it for every
part. Therefore, the need for an on-line monitoring system is inevitable. Berumen et al.
developed the in-situ monitoring system consists of a CMOS camera and a photodiode,
which captures the radiation in the wavelength range of 700–940 nm [5]. Similarly,
Lott et al. [6] developed a coaxial setup based on the illumination of the laser beam to
study the melt pool dynamics during printing. Doubenskaia et al. [7, 8] developed a
monitoring system consisting of two photodiodes with a detection range of 900–
1700 nm. Nowadays, commercial systems are also equipped with monitoring systems
such as Melt Pool Monitoring (MPM) from SLM Solutions, EOSTATE from EOS
GmbH, InfiniAM from Renishaw.
Due to the enormous size of the in-situ data acquired and the difficulties encoun-
tered in analyzing it, processing the data to qualify the final part is a difficult task. Also,
manually detecting the drift or defect is time-consuming and laborious work. Therefore,
there is a need for applying Machine Learning (ML) techniques for processing the data
to detect drift. Thus, in recent years many researchers have used machine learning
approaches on large datasets for decision making [9, 10]. Grasso et al. [11] demon-
strated the use of the T-mode Principal Component Analysis (PCA) for image data
obtained by OlympusTM I-speed 3 camera mounted outside the build chamber at an
angle of 400 w.r.t build plate to define a spatial statistical descriptor to detect local
over-heating phenomena along the scan path by analyzing the intensity profile of each
pixel. Ikenna et al. [12] used a semi-supervised machine learning approach for auto-
matic fault detection in enormous data obtained using photodiodes. They have used the
ultimate tensile strength as their descriptor to label their tensile parts as acceptable or
faulty to train a semi-supervised learning algorithm. Scime et al. [13] used one-
megapixel Photron FASTCAM Mini AX200 high-speed camera in the visible range to
capture melt pool images. The machine learning technique, called Scale Invariant
Feature Transforms (SIFT), was used to extract features from the captured melt pool
signatures. The bag-of-words (or Keypoints) were used to obtain a scale-agnostic
description of melt pool morphology. The melt pool morphology represents the shape
of the melt pool, spatter, and vapor plume. With the ML approach, the author classified
the individual melt pool morphologies into four categories, such as desirable, balling,
under-melting, and key-hole porosities. Supervised classification ML technique called
Support vector machine (SVM) was used to train and test the model. In another study,
Scime et al. [14] used a computer vision algorithm to predict the percentage of types of
defects present in part based on the powder bed images. Scime et al. [14] extracted the
regions of the images and classifies such as anomaly free, recoater hopping, recoater
Drift Detection in Selective Laser Melting (SLM) 179
streaking, debris, superelevation, part failure, and incomplete spreading. These cate-
gories were used as the fingerprints for the computer vision algorithm.
In this study, a co-axial melt pool monitoring (MPM) system installed on a com-
mercial SLM 280HL machine was used to acquire the in-situ data. It shall be noted that
no additional modifications were made to the hardware available. A balanced labeled
dataset (equal number of “drift” and “no-drift” data points) is prepared for training a
supervised Support Vector Machine (SVM) classifier to detect the drift in the parts. For
testing the trained algorithm, specific geometries that incorporate the overeating and
lack of fusion drift were printed by varying the volumetric energy density.
2 Definitions
• Drift: Non-uniformity in the melt pool signatures results in the “hotspots,” which
are the areas where the intensity of the signal is higher compared to the rest of the
layer. These hotspots are an indication of drift and the areas of highest probability to
generate real defects in part. Recently, Mohr et al. linked hotspots in the MPM layer
to the final porosity in part using computer tomography [15]. Therefore, in our
study, the whole layer is termed as “Drift” if there is a presence of hotspot con-
firmed via the MPM visualization tool.
• No-Drift: The layer with no significant hotspots are termed as no-drift layer.
• Labeled Data: The layer for which the labels such as “drift” and “no-drift” are
known.
• Unlabeled Data: The layer for which the labels, i.e., “drift” and “no-drift,” are not
known.
• Balanced Dataset: The dataset contains an equal number of data points for each
label.
3 Machine Learning Theory
In recent years, data-driven methods have been widely used to monitor and improve the
overall performance of the AM process. The challenges, such as post processability of
the enormous size data and automatic drift detection associated with the in-situ mon-
itoring systems, can be overcome by using the ML approach. Moreover, ML can
monitor the process and provide feedback in real-time. ML approaches can broadly be
classified into two categories, called “Supervised” and “Unsupervised” approaches.
The supervised approaches work for classification and regression problems, whereas
unsupervised approaches work for clustering problems. The main difference between
both approaches mainly depends on the type of data required as input for training. In
supervised, one has to provide a labeled dataset in which each data point belongs to a
specific class as categorized by the label associated with it. In contrast, unsupervised
learning is used when we do not have the labels for the dataset or, in other words, when
we do not have ground-truth knowledge about the dataset. For the AM process, it is
challenging and expensive to obtain ground truth dataset for training. Therefore, a
180 P. Yadav et al.
small dataset is prepared for training the algorithm. In this study, a supervised learning
model called “Support vector machine” (SVM), which works sufficiently good for
small datasets, is used.
Feature Set 2
Feature Set 1
Fig. 1. Schematic of the hyperplane that maximally separates the support vectors corresponding
to each of the two classes [16].
SVM is a classification algorithm that divides the dataset into multiple classes. In
this study, the in-situ data is treated as a binary classification problem, i.e., only two
classes (“drift” and “no-drift”) are identified, so the SVM classifier perfectly fits for this
case. During training the SVM classifier, the decision function amounts to identifying a
suitable reproducible hyperplane that maximizes the distance (also called “margin”)
between the support vectors of both class labels (Fig. 1). SVM classifier can work for
both linear or non-linear classification problems. Further, the linear SVM problems
range in their complexity depending upon their number of features selected. For
example, in two feature dimensions, the hyperplane corresponds to a line, whereas in
the case of three features, the hyperplane is a two-dimensional plane. Regardless of the
SVM’s complexity (i.e., dimensionality), classification problems are often linear,
which means that hyperplane used is a straight line, not a curved line. If the features
selected for the SVM classifier are linearly separable, then we can draw a straight
hyperplane to separate the two labels of the class of interest. Usually, there are two
types of margins: hard margins and soft margins. With a hard margin, we restrict the
classifier to make any error while training. Although the hard margin is computa-
tionally less expensive, but not always the linear separability of the features is so easy.
Therefore, by allowing the classifier to misclassify, the greater generalizability to a new
data can be obtained by a larger margin. This misclassification can be obtained by the
so-called “Soft margin,” which relies on the variable n. The values of n lies in the range
0 n 1. The non-zero values of n, in turn, can allow classification error that can
result when outliers in the training data lead to the hyperplane making mistakes, i.e.,
misclassification (Fig. 2). This hard margin is a particular case of soft margin when the
slack variable (n) is zero. A soft-margin constant denoted by C is introduced to incur a

penalty on n. The parameter C helps to choose the trade-off between the tanning errors
and complexity and reduces the chances of overfitting, which means fine-tuning the
classifier for maximum accuracy in the training dataset. The decision boundary of a
non-linear classifier depends on the data in a non-linear way. In these cases, the kernel
method is used to transform the support vectors into a high dimensional space. A de-
tailed explanation for SVM can be studied in the literature [16, 17].
Fig. 2. (a) Hard margin on linearly separable dataset where no training errors are allowed. (b) the
soft margin where two training errors are permitted to make data non-linearly separable [18].
Typically, there are three stages in the SVM classifier: (a) feature selection and
preparing feature dataset, (b) training the classifier and testing, and (c) checking the
accuracy of the classifier. All three stages for our case are discussed in the next
sections.

4.1 Material and Machine
The gas atomized AlSi7Mg0.6 spherical powder supplied by SLM Solutions was used
to print specimens with varying process parameters. The particle size distribution was
20–63 µm as specified by the supplier. The apparent density of the powder was
1.53 g/cm3, and the chemical composition of the as-received powder is tabulated in
Table 1.
182 P. Yadav et al.
Table 1. Elemental composition of as received AlSi7Mg0.6 powder (All the values are given in
wt%).
Element Al Cu Fe Mg Mn Si Ti Zn Others
Minimum [wt%] Balance – – 0.45 – 6.50 – – –
Actual [wt%] Balance <0.01 0.08 0.55 <0.01 6.90 0.07 0.01 <0.03
Maximum [wt%] Balance 0.05 0.19 0.70 0.10 7.50 0.25 0.07 0.03
The commercial SLM 280HL (SLM Solutions Group AG, Lübeck, Germany)
equipped with 700 W twin continuous wave (CW) ytterbium fiber lasers with an
emitting wavelength of 1070 nm and a spot diameter of 80–115 lm was used for
printing. The build envelope volume is 280 280 365 mm3, and the build chamber
was maintained in the Ar gas environment with an oxygen level below 0.1%. The Al
base plate was preheated to a temperature of 150 °C prior to printing to reduce the
effect of residual stresses in part.
The SLM 280HL machine is equipped with an in-situ monitoring device called
‘Melt Pool Monitoring,’ which consists of two on-axis photodiodes. The specifications
and working principles are discussed in Sect. 4.3.
4.2 Part Geometry and Process Parameters

To prepare a training dataset for the SVM classifier model, a balanced dataset comprising
an equal number of drift and no-drift layers is necessary. To obtain the balanced dataset,
an artificially drift is introduced in the samples. Therefore, the unique geometrical
specimens were printed, as shown in Fig. 3. The process parameters tabulated in Table 2
were varied to obtain varied volumetric energy density in the range of 40–73 J/mm3 for
each shown geometry. The overhang samples (Fig. 3a, and 3b) with an overhang of 8 mm
was printed without any support structure and stripes rotation of 67° for the overhang
layer. No down-skin parameters were used for the overhang layer. For lack of fusion
samples (Fig. 3d), an internal cuboid type groove with dimensions 10 8 0.09 mm3
was printed. The thickness of the groove set to 0.09 mm, i.e., three times the layer
thickness of 30 l. The printed parts were cut along the build direction (z-direction), and
3 l polished for microscopic analysis. The optical microscope supplied by Leica systems
was used for analyzing the porosity.
10 mm
2 mm 2 mm 15 mm
10 mm
10 mm
(10 x 8 x 0.09) mm3
10 mm 10 mm
10 mm
10 mm
(a) (b) (c)
Fig. 3. Sketch of the specimens (a) cubic overhang (size 10 10 10 mm), (b) cylindrical
overhang (diameter: 10 mm and height 10 mm), and (d) specimen with inner groove.
Table 2. Varied process parameters for printing.

Varied Parameter Values
Power (W) 300, 350, and 400
Scanning speed (mm/s) 1200, 1400, 1650, and 1900
Hatch distance (mm) 0.13, 0.26, and 0.52
Layer thickness (mm) 0.03
Scanning strategy Stripes
Rotation in layers (o) 67
4.3 Melt Pool Monitoring System

The co-axial melt pool monitoring (MPM) system installed on the commercial machine
SLM 280HL was used to collect thermal emissions from the melt pool formed due to
laser-powder interaction. The melt pool systems are co-axial systems, which means it is
in the alignment of the laser path and collects the real-time emissions from the laser
path at acquisition frequency of 100 kHz. The MPM module consists of two photo-
diodes with different sensitive areas. The spectral range of the photodiodes cannot be
revealed due to confidentiality issues. However, both the photodiodes capture the
thermal emission in the near-infrared region. The schematic diagram of the MPM
system is shown in Fig. 4. Only the emissions traveling perpendicular to the build
platform are taken into account. The thermal radiations follow the same path as of laser
and directed into the MPM module with the help of a semi-transparent mirror, which
does not allow laser wavelength to pass. The signal is split into two different spectral
ranges and captured by the installed photodiodes, respectively. The received signal is
Fig. 4. Schematic of melt pool monitoring system installed on SLM 280HL [19].
184 P. Yadav et al.
forwarded to associated ADCs (Analog to digital convertor) and provided in an FPGA

(field-programmable gate array) by the individual photodiodes. The captured thermal
emissions from photodiodes 1 and 2 are stored along with the x/y-coordinates (16-bit).
The values are stored in parallel with the laser on/off signal from FPGA to PC in every
10 µs. All the data is stored for every layer in a data file, which can be accessed as 2D
representation in MPM software provided by the SLM Solutions [19]. The new file is
created automatically for each layer after the complete exposure. For this work, no
additional modifications are made to the installed hardware. Also, only the data from
ADC1 (Photodiode 1) are further post-processed. The reason to select the ADC1 is due
to its specific detection range.
Volume Energy Density Sensitivity Analysis
The sensitivity of the MPM system for the volume energy density was studied. The
cylindrical specimens with the sandwich structure were printed, where the bottom and
upper part of the cylinder was printed with the optimized process parameters. In con-
trast, the middle part was subjected to varied volumetric energy density. The geometry
of the samples is shown in the right corner of the graph depicted in Fig. 5. It shows the
mean thermal counts for every layer, and it can be observed the photodiode is sensitive
to change in volume energy density in the samples, which ranges from 40–73 J/mm3.
The first 50 layers that account for a higher signal represents the support structure of the
part. Therefore, it is necessary to remove support structure data from the data preparation
before processing. The layers from 300 to 580 indicate the variation in the input vol-
umetric energy region. Sample 3 has the lowest energy density of 40.48 J/mm3, which
contributes lower mean thermal emissions, whereas sample 7 has the highest volume
energy density of 73.26 J/mm3, which shows higher mean thermal emissions. So it can
be concluded that the MPM system signal is proportional to input volumetric energy
density.
Non optimized parameters

μ (Thermal counts)
Sample geometry
Optimized parameters
Optimized parameters
Layer number
Fig. 5. The mean thermal emissions recorded by MPM systems for the induced drift detection in
the specimens shown in right corner of graphs (dark red color: drift area, red area: optimized
processed parameters).
Down-Skin Sensitivity Analysis

To study the sensitivity of the MPM system for down-skin, overhang samples was
printed, as shown in Fig. 3a and Fig. 3b. It was observed that the photodiode signal
shows a gradual increase in the thermal emissions as the laser exposure moves from the
printed part to the overhang part (Fig. 6). This is in conjunction with the theory that the
powder has lower thermal conductivity compared to the consolidated part. The pres-
ence of voids in loose powder makes it an inferior heat conductor compared to the
printed part. Thus, the consolidated part acts as a major heat sink in the SLM process.
Therefore when the region above the powder exposed, the melt pool thermal emissions
are higher compared to the melt pool region on the printed part. The same phenomenon
could also occur during a lack of fusion defects. As sometimes, due to bad powder
spreading, there is non-uniformity in the powder bed, and some regions are not covered
with the powder uniformly for a few layers.
On powder
On powder
Thermal Emissions
Thermal Emissions
On consolidated part
On consolidated part
Scan vector
Scan vector
(a) (b)
Fig. 6. Scan vector wise MPM signal for downskin layer for (a) cylindrical sample, (b) cuboid
sample.
Nevertheless, in the next passes of recoating, the regions with lack of powder are
uniformly covered, but now the laser prints thicker layer compared to other regions
with uniform powder spreading. This phenomenon can lead to a lack of fusion defects
as laser energy is not enough to fully melt the powder layer. So it can be concluded that
the MPM systems are also sensitive to lack of fusion defect.
4.4 Training Data Preparation

Recently, the Machine learning approach has gained much attention in AM due to its
capabilities to resolve the issue of big data and easy post processability of the in-situ
monitoring data. To perform supervised machine learning, the need for labeled data is
vital. However, it is challenging to obtain ground truth certified data for the AM parts
as the drift in the process can be linked to many parameters [20]. Usually, to obtain a
labeled dataset commonly referred to as certified dataset, the expensive techniques such
as computer tomography of the part is done. For this study, an artificially drift
(overheating and lack of fusion) in the parts with special geometrical features and by
186 P. Yadav et al.
varying process parameters was created. The careful selection of the layers from the
build parts (81 parts) and labeling it as ‘Drift’ and ‘No-drift’ was performed by ana-
lyzing the layers in the MPM software provided by the SLM solutions and also by
statistical analysis. It was noticed that the layers which having the hotspots show a right
shift in the histogram compared to the layers without any hotspots. This relation led us
to decide the mean and median as our features for every layer. A balanced labeled
dataset of 600 data points, which comprises an equal number of drift and no drift layers
so that the biasing of the SVM model can be avoided was prepared. Before preparation,
the SVM classifier has to be optimized for best-fit parameters, as discussed in the next
section.
4.5 Bayesian Optimization and Training of SVM Classifier

The best-fitting parameters are to be selected for the SVM classifier to increase the
success rate of the classifier. To find the best-suited hyperparameters for the model, a
bayesian optimization algorithm was used. A hyperparameter is an internal parameter
of an algorithm that needs to be optimized. For example, in our case (SVM model), the
box constraint, kernel-function, and kernel-scale are the hyperparameters. These
parameters can significantly influence the performance of the algorithm. Thus, opti-
mization of the hyperparameters is advisable. However, optimization is difficult and
time-consuming. Therefore, Bayesian optimization is well suited for classification and
regression algorithms in machine learning. The Bayesian optimization algorithm
minimizes the objective function f(x) for x in a bounded domain. The f(x) can be
scholastic or deterministic, which means it can return different results for the same
point x. The overall working principle of Bayesian optimization can be found else-
where [21, 22].
The cross-validation of the optimized SVM classifier was performed on the training
dataset. The whole dataset was partitioned into 70% and 30% sub-datasets to cross-
validate and check the performance of different hyperparameters of the SVM model.
The performance and accuracy % of the different hyperparameters are tabulated in
Table 3. It can be seen that the fine gaussian SVM has the highest accuracy with a
cross-validation success rate of 95.5%. So, for our dataset, fine gaussian hyperpa-
rameter to train our SVM classifier was used.
Table 3. Bayesian optimization results of SVM classifier for different hyperparameters.

Linear SVM 94.2%
Quadratic SVM 93.8%
Cubic SVM 78.7%
Fine Gaussian SVM 95.5%
Medium Gaussian SVM 93.5%
Coarse Gaussian SVM 91.0%
Finally, the SVM classifier was trained on the whole dataset, as shown in Fig. 7a
and predicted labels are shown in Fig. 7b. Further, testing of the SVM classifier is
performed using the different datasets (which is not used during training) as discussed
in the results section.
Feature 2 (Median)
Feature 2 (Median)
Feature 1 (Mean) Feature 1 (Mean)
(a) (b)
Fig. 7. (a) The un-clustered training dataset, and (b) labels predicted by the trained fine gaussian
SVM classifier model.
To check the performance of the classification at different thresholds, the Receiver

Operating Characteristic- Area under the curve (ROC-AUC) was analyzed. ROC
represents the threshold probability of the classifier, and AUC indicates the separability
of the classifier, i.e., how well the classifier can distinguish between different classes.
Higher the AUC, better is the model separability. ROC curve is plotted between TPR
(True positive rate) vs FPR (False positive rate). For the model, the ROC curve was
plotted, as shown in Fig. 8. The threshold probability is 95.5% (marked by an asterisk
(*)), and the AUC for the model is 98.53%. In other words, the proposed model is
98.53% capable of allocating each data point with the correct label.
TPR
AUC: 98.53 %
Fig. 8. ROC-AUC curve for the trained SVM classifier on 600 training data.
188 P. Yadav et al.
5.1 Overheating Drift

The trained SVM classifier was tested for the overheating samples. The feature dataset
(mean and median) was prepared for two different overhang samples (Fig. 3a and 3b),
which failed due to the overheating of the layers. The specific samples were used for
the case study as the location of the overheating drift is known and can visually be
verified. The prepared feature dataset was passed to a trained SVM classifier, and labels
were predicted, as shown in Fig. 9. For the cubic overhang sample, the layer numbered
from 360 to 378 was labeled as “drift,” which can also be visually verified (printed part
is shown in the top left corner of Fig. 9a). The failure of the particular layers was due to
poor heat transfer in the absence of proper support structure for the overhang
part. Similarly, the labels were predicted for the cylindrical overhang part (Fig. 9b) and
are in conjunction with the visual inspection of the part.
Feature 2 (Median)
Feature 2 (Median)
Feature 1 (Mean) Feature 1 (Mean)

(a) (b)
Fig. 9. SVM model predicted labels for (a) cubic overhang specimen, and (b) cylindrical
specimen.
5.2 Lack of Fusion Drift

Lack of fusion is the most common defect in the SLMed parts, which results in internal
porosity in the final part. The detection of internal porosity is a challenging task and often
requires expensive techniques such as computer tomography. We prepared samples that
will lead to internal porosity in the final part, and the location of the porosity is known to
check the trained SVM classifier for lack of fusion defect. The feature matric (mean and
median) for the lack of fusion part, as shown in Fig. 3d is prepared and passed to the
trained SVM classifier. The SVM classifier predicts the labels for each layer, as shown in
Fig. 10a. The three layers numbered from 190 to 192 were predicted as “drift,” which
correspond to the location of porosity in the sample. The same was also verified by the
optical micrograph, as shown in Fig. 10b. Therefore, it can be concluded that the trained
classifier works well for the presented case studies. However, it shall be noted that the
effectiveness of the model for other types of drifts shall be studied in the future.
Feature 2 (Median)
Lack of fusion defect
Feature 1 (Mean)
(a) (b)
Fig. 10. (a) Predicted labels for the lack of fusion defect, (b) optical micrograph of the defect.
6 Conclusion
An attempt to classify the in-situ monitoring data into drift and no-drift was made using
a supervised machine learning model (SVM classifier). For this study, the sensitivity of
the installed monitoring system was studied for artificial defects in the built-up spec-
imen. The varied process parameters were purposely chosen to obtain a variety of drift
and no drift data points for training the SVM classifier. The in-situ monitoring data is
treated as a two-class problem (“drift” and “no-drift”), and the SVM classifier used to
predict the drift layers in the parts studied. The feasibility of the trained algorithm to
detect the drift in the process was verified by comparing it with the Melt pool moni-
toring system. The post processability of the enormous sized in-situ data can be sig-
nificantly improved with the help of the ML models. Also, the use of ML models can
be a revolutionary step in improving and monitoring the SLM process in real-time.
The future aim of this study is to test the classifier on real case complex parts and
verify it with the micro-computer tomography dataset. Also, to study the reliability of
the ML model for other materials.
Acknowledgments. This work was conducted as part of the “ENABLE” project funded by the
European Union’s Marie Sklodowska-Curie Actions (MSCA) Innovative Training Networks
(ITN) H2020-MSCA-ITN-2017 under the grant agreement Number 764979.
References
1. Clijsters, S., Craeghs, T., Buls, S., Kempen, K., Kruth, J.-P.: In situ quality control of the
selective laser melting process using a high-speed, real-time melt pool monitoring system.
Int. J. Adv. Manuf. Technol. 75(5–8), 1089–1101 (2014)
2. Kruth, J., Wang, X., Laoui, T.: Lasers and materials in selective laser sintering (2003). http://
emeraldinsight.com
190 P. Yadav et al.
3. Loh, L.E., Liu, Z.H., Zhang, D.Q., Mapar, M., Sing, S.L., Chua, C.K., Yeong, W.Y.:
Selective Laser Melting of aluminium alloy using a uniform beam profile: the paper analyzes
the results of laser scanning in Selective Laser Melting using a uniform laser beam. Virt.
Phys. Prototyping 9(1), 11–16 (2014)
4. Galy, C., Le Guen, E., Lacoste, E., Arvieu, C.: Main defects observed in aluminum alloy
parts produced by SLM: from causes to consequences. Addit. Manuf. 22, 165–175 (2018)
5. Berumen, S., Bechmann, F., Lindner, S., Kruth, J.P., Craeghs, T.: Quality control of laser-
and 820 powder bed-based Additive Manufacturing (AM) technologies. Phys. Procedia 5,
617–622 (2010)
6. Lott, P., Schleifenbaum, H., Meiners, W., Wissenbach, K., Hinke, C., Bültmann, J.: Design
of an optical system for the in situ process monitoring of selective laser melting (SLM).
Phys. Procedia 12, 683–690 (2011)
7. Doubenskaia, M., Pavlov, M., Grigoriev, S., Tikhonova, E., Smurov, I.: Comprehensive
optical monitoring of selective laser melting. J. Laser Micro/Nanoeng. 7(3), 236–243 (2012)
8. Doubenskaia, M., Grigoriev, S., Zhirnov, I., Smurov, I.: Parametric analysis of SLM using
comprehensive optical monitoring. Rapid Prototyping J. 22, 40–50 (2016)
9. Purtonen, T., Kalliosaari, A., Salminen, A.: Monitoring and adaptive control of laser
processes. Phys. Procedia 56, 1218–1231 (2014)
10. Grasso, M., Laguzza, V., Semeraro, Q., Colosimo, B.M.: In-process monitoring of selective
laser melting: spatial detection of defects via image data analysis. J. Manuf. Sci. Eng. Trans.
ASME 139(5), 051001 (2017)
11. Grasso, M., Demir, A.G., Previtali, B., Colosimo, B.M.: In situ monitoring of selective laser
melting of zinc powder via infrared imaging of the process plume. Robot. Comput. Integr.
Manuf. 49, 229–239 (2018)
12. Okaro, I.A., Jayasinghe, S., Sutcliffe, C., Black, K., Paoletti, P., Green, P.L.: Automatic fault
detection for laser powder-bed fusion using semi-supervised machine learning. Addit.
Manuf. 27, 42–53 (2019)
13. Scime, L., Beuth, J.: Using machine learning to identify in-situ melt pool signatures
indicative of flaw formation in a laser powder bed fusion additive manufacturing process.
Addit. Manuf. 25, 151–165 (2016)
14. Scime, L., Beuth, J.: Anomaly detection and classification in a laser powder bed additive
manufacturing process using a trained computer vision algorithm. Addit. Manuf. 19, 114–
126 (2018)
15. Mohr, G., Altenburg, S.J., Ulbricht, A., Heinrich, P., Baum, D., Maierhofer, C., Hilgenberg,
K.: In-situ defect detection in laser powder bed fusion by using thermography and optical
tomography—comparison to computed tomography. Metals 10(1), 103 (2020)
16. Pisner, D.A., Schnyer, D.M.: Support vector machine. In: Machine Learning, pp. 101–121.
Academic Press (2020)
17. Ding, X. (ed.): Advances in Character Recognition. BoD–Books on Demand (2012)
18. Mahmoudi, A., Takerkart, S., Regragui, F., Boussaoud, D., Brovelli, A.: Multivoxel pattern
analysis for FMRI data: a review. In: Computational and Mathematical Methods in Medicine
(2012)
19. Alberts, D., Schwarze, D., Witt, G.: High speed melt pool & laser power monitoring for
selective laser melting (SLM®). In: 9th International Conference on Photonic Technolo-
gies LANE, vol. 2016 (2016)
20. Gibson, I., Rosen, D.W., Stucker, B.: Additive Manufacturing Technologies, vol. 17.
Springer, New York (2014)
21. Hastie, T., Tibshirani, R., Friedman, J.: The Elements of Statistical Learning: Data Mining,
Inference, and Prediction. Springer, New York (2009)
22. Pelikan, M., Goldberg, D.E., Cantú-Paz, E.: BOA: the Bayesian optimization algorithm. In:
Proceedings of the genetic and evolutionary computation conference GECCO 1999, vol. 1,
pp. 525–532, July 1999
Influence of the Inert Gas Flow on the Laser
Powder Bed Fusion (LPBF) Process
Florian Wirth(&), Alex Frauchiger, Kai Gutknecht,

and Michael Cloots
Irpd AG, Lerchenfeldstr. 3, 9014 St. Gallen, Switzerland

florian.wirth@irpd.ch
Abstract. The inert gas flow is known to have a significant impact on the laser
powder bed fusion (LPBF) process in terms of process stability and consistent
process results across the whole build platform. Thus, the optimization of the
inert gas flow leads to both direct and indirect improvements of the part quality
as follows. If the gas flow can steadily and efficiently remove soot particles
emerging from the melting process, scattering and attenuation of the laser beam
or debris on the laser windows can be avoided, which would indirectly impair
the quality of the built parts. Spatter particles should be removed as well because
they can directly lead to bonding defects inside the produced parts. Therefore,
the gas flow in a self-constructed LPBF machine has been optimized system-
atically based on computational fluid dynamics (CFD), particle tracking and
experimental studies. Herein the effect of the process gas flow and gas type on
spatter and soot particles is presented in detail. According to the simulation
results, the soot removal is improved by a smaller shielding gas inlet height at
the cost of a potential deterioration of the soot removal at extreme process
parameters. The simulation results have been validated by measurements of the
gas flow velocity and of the density of the built parts. The advantages and
disadvantages of different process gas types are shown and recommendations for
the gas flow design are derived.
Keywords: Laser powder bed fusion Shielding gas flow Process gas type
Spatter Soot
1 Introduction and State of Research
The LPBF process is based on an iterative procedure to manufacture parts in layers

inside a powder bed. Once a layer of powder has been spread by the coater on a plane
surface, one or several laser beams are guided by laser scanners over the areas which
are to become molten and form a single layer of the additively manufactured parts after
solidification. Within the immediate environment of its interaction with powder, molten
metal and solidified metal, each laser beam creates a melt pool, which follows the scan
path. Upon completion of each layer, the build plate is lowered, a new powder layer is
spread and the next layer of the parts to be manufactured is molten. The final parts
consist of a multitude of overlapping weld seams. In the build chamber, a shielding gas
flow prevents the metal from oxidation. Moreover, the shielding gas flow shall carry

https://doi.org/10.1007/978-3-030-54334-1_14
Influence of the Inert Gas Flow on the LPBF Process 193
away process by-products such as spatter or soot and it shall protect the laser windows
from contamination with soot. If the shielding gas flow is not suitably designed to
perform these tasks, the laser windows are contaminated with debris leading to an
attenuation of the laser beam, defocusing of the laser beam or even damage to the laser
windows. Furthermore, airborne particles inside the build chamber likewise attenuate
the laser beam and provoke scattering or defocusing of the laser beam so that the LPBF
process becomes instable. LPBF is used for applications in aerospace, energy, tool and
mold making such as lightweight structures with topology optimization, turbine blades
or forging tools. Nevertheless, according to the National Institute of Standards and
Technology (NIST) [1], the Additive Manufacturing Special Interest Group [2] and
Caltanisetta et al. [3], a broader application is limited by the relative instability and
poor repeatability of the process and by the low productivity in current LPBF
machines. Therefore, multi-laser systems have been developed and high power lasers
are increasingly employed for enhanced productivity. However, more lasers and higher
laser power lead to an increase in process by-products such as spatter and soot, which
impaired the process stability already before. Consequently, it has become even more
important to efficiently remove soot and spatter from the build chamber by an opti-
mized shielding gas flow. Because of this reason, herein the interaction between the
shielding gas flow and spatter or soot particles is investigated in order to optimize the
shielding gas flow in terms of the gas type and the shielding gas flow design. Prior to
these investigations, it was observed with the shielding gas flow settings of Table 3,
that the density of the parts produced with both parameter sets of Table 1 was 0.5%
points lower in the case of build setup 1 according to Fig. 2 in comparison to a build
job where only seven cubes were built at once. This effect can be explained by the
accumulation of soot particles inside the build chamber, when the ratio of scan time to
recoating time is high. Therefore, the successful approaches shown in the following
sections had to be found to improve the soot removal from the build chamber.
Several authors have already shown an influence of the shielding gas flow on
spatter and soot particles by experimental studies and simulations. Anwar and Pham [4]
observed an increased amount of spatter at the gas outlet when the shielding gas flow
velocity was increased because then the gas flow was able to carry the spatter particles
further and less particles fell into the powder bed. Furthermore, when the scanning
direction was against the gas flow, more spatter particles could be seen at the gas outlet
as the spatter particles are presumably ejected from the rear of the melt pool in the
opposite direction of the scanning direction, i.e. towards the shielding gas outlet. In a
further study, Anwar and Pham [5] could measure a higher ultimate tensile strength in
the aforementioned cases with more spatter particles deposited close to the gas outlet.
Moreover, they could detect a higher number of sparks and other particles interacting
with the laser beam when scanning in the gas flow direction, leading to an attenuation
of the laser beam and thus to bonding defects. Renishaw [6] showed that the airborne
by-products of the LPBF process have a negative effect on a laser beam that is scanning
in the gas plume of another laser, where it is attenuated so that the mechanical prop-
erties of the built parts are impaired.
There is also an influence of different gas types on the LPBF process. Pauzon et al.
[7] could not detect a significant difference in the mechanical properties of parts built
from 316L stainless steel when comparing argon with nitrogen as the shielding gas. But
194 F. Wirth et al.
Pauzon et al. [8] could achieve an increase by 44% in the build rate of Ti-6Al-4 V
when using a mixture of 50% argon and 50% helium. This mixture combines the
beneficial properties of both gases. The high thermal conductivity and heat capacity of
helium reduces the accumulation of process by-products, while the latter are removed
more efficiently thanks to the high density of argon.
Anwar [9] developed a simulation model to investigate the influence of the
shielding gas flow on the trajectories of the spatter particles, which showed a limited
influence of the shielding gas flow on the spatter particles in accordance with the
experimental data. However, the influence of different gas flow designs and gas types
was not investigated. The behavior of smaller particles such as soot particles inside the
build chamber was not considered either.
When the movement of process by-products inside the build chamber has to be
investigated, the gas flow velocity in the process region, i.e. the velocity of the
evaporated metal has to be known especially for the simulation of the trajectories of
small particles. Bidare et al. [10] developed a 2D axisymmetric simulation model to
investigate the gas flow velocity in the vicinity of the LPBF process. Accordingly, a
laser beam with a spot size of 50 µm can induce a velocity up to 1200 m/s (300 m/s) at
a laser power of 200 W (50 W). Similar 2D simulations of Mayi et al. [11] show results
in the same order of magnitude. However, the indicated velocities seem rather high,
which might be caused by the neglect of the feed motion of the laser beam as Jakumeit
et al. [12] found a velocity of 100 m/s when processing IN718 in their advanced 3D
simulation. Masmoudi et al. [13] even state a velocity as low as 40 m/s in an argon
atmosphere with a pressure of 1.0 bar for 316L stainless steel.
A self-developed, modularly constructed, ready for various measurement equipment

and vacuum suitable LPBF laboratory machine with a build plate size of
400 400 mm2 was used for the experiments. A small build plate for reduced
quantities can also be inserted. The machine can be equipped with up to two lasers. For
the herein mentioned experiments, only one 1 kW continuous wave fiber laser type
IPG YLR-1000-WC with a wavelength of 1070 nm was used. Nitrogen was employed
as the shielding gas, which was kept at room temperature by a cooling system.
The produced samples were evaluated by Archimedean density measurements to
see the influence of process by-products and thus the effect of the shielding gas flow,
which is visualized in Fig. 1. The shielding gas flow consists of a purge gas flow from
the ceiling and a horizontal flow from an inlet nozzle with a cross sectional area of
78 578 mm2, which can be reduced. Due to geometrical constraints, the flow
direction of the inlet nozzle is 6.8° inclined downwards. In the beginning, 40% of the
total shielding gas flow was assigned to the purge gas flow based on the findings of
Chen et al. [14]. A Testo Ø 16 mm vane probe anemometer was used for gas flow
velocity measurements.
outlet
inlet nozzle
Fig. 1. Shielding gas flow inside the LPBF machine.
Stainless steel 1.4404 powder from LPW with a particle size of 10-45 µm was
processed using the process parameters listed in Table 1. For the experimentally
investigated shielding gas flow settings, build jobs with the build setups shown in
Fig. 2 were run, which cover the whole build plate. These build setups consisted of
several cuboids with a size of 10 10 10 mm3 or 20 20 10 mm3 with the
200 W or 500 W parameters of Table 1, where the parameter set with high laser power
produced a significant amount of fume. As it was already possible to achieve a high
density of the produced parts in build setup 1 after the first modifications of the
shielding gas setup, further experiments were carried out with the build setup 2, where
a significantly higher amount of fume is generated, so that the shielding gas flow
system is put under more strain.
Table 1. Process parameters.

Parameter set 1 Parameter set 2
Laser power [W] 200 500
Scan speed [m/s] 1.2 1.8
Laser spot size [µm] 86 136
Hatch distance [µm] 70 100
Layer thickness [µm] 30 60
196 F. Wirth et al.
200 W
500 W
Fig. 2. Build setup 1 (left) and build setup 2 (right) for experimental investigations.
3 Modeling
The simulation model was set up in ANSYS Fluent using the k-x shear stress transport
(SST) turbulence model with the default values for the turbulence model parameters
and assuming an incompressible fluid. Additionally, the temperature field and the
concentration of gas contaminated with fume is calculated. The velocity field, the
temperature field and the concentration field are calculated in the steady state. The
discrete phase model (DPM) is applied to investigate the influence of the shielding gas
flow on spatter and soot particles. The drag force is calculated according to the
spherical drag law by Morsi and Alexander [15]. Only a 150 mm thick center slice of
the volume inside the build chamber is considered in the developed simulation model
to investigate the effect of the shielding gas flow on spatter and soot particles as shown
in Fig. 3. Symmetry boundary conditions are applied at the front and backside of this
slice. It is assumed that the plume is generated during the process on the build plate in a
circle with a radius of 5 mm similar to an island and with its center at the indicated
position of the plume velocity towards the front of the model. Due to its symmetry
boundary condition, the model comprises only a half circle. The plume is emulated by
hot gas streaming into the build chamber at this position. The plume velocity indicates
the speed of the gas rising from the process zone. The aforementioned literature data for
the velocity of an undisturbed gas plume vary in a broad range and it is not known, how
the unsteady movement of the gas plume e.g. on an island influences its velocity in the
upper spheres. Therefore, the plume velocity was set to upl ¼ 10 m=s so that the
behavior of the soot particles in the simulation corresponds to the observations at the
initial machine configuration, where some smoke clouds bounced against the rear wall
of the build chamber and were deflected upward. The gas is assumed to obey the ideal
gas law
pV ¼nRT ð1Þ

with the pressure p, volume V, number of moles n, gas constant R ¼ 8:31 J=ðmol KÞ
and temperature T. While the shielding gas enters the process chamber at ambient
temperature T ¼ T0 ¼ 293 K, the plume gas is assumed to emerge with the evapora-
tion temperature Tv ¼ 3130 K of the powder material.
rear wall
shielding gas flow velocity
=0
outlet,
plume velocity
Fig. 3. Simulation model geometry.
Two kinds of particles were considered in the investigations. The soot particles
inside the fume are very small so that their behavior is dominated by surface forces such
as the drag force due the high surface to volume ratio. On the contrary, spatter particles
show a comparably low surface to volume ratio. Therefore, they mainly follow the
inertial forces. In the simulation model, the particles are injected at the plume inlet in the
upward direction with the ejection velocity ue . The initial velocity of the soot particles is
equal to the plume velocity, whereas the spatter particles with a size of 50–200 lm are
injected with a velocity of 3 ± 1.5 m/s based on the experimental measurement data of
Gunenthiram et al. [16]. Barrett et al. [17] indicate higher velocities up to 50 m/s, but
these high velocities can be attributed to small particles entrained by the plume. The
physical properties of the considered gases according to the ANSYS Fluent database are
summarized in Table 2. Nitrogen is commonly used due to its comparably low price, the
high thermal conductivity of helium can be beneficial to remove fumes from the process
chamber and the high density of argon leads to a comparably high drag force exerted
upon the spatter particles to blow them away.
198 F. Wirth et al.
Table 2. Physical properties of considered gases.

Nitrogen (N2) Helium (He) Argon (Ar)
Molar mass [g/mol] 28.0134 4.0026 39.948
Viscosity [µPas] 17.2 18.6 2.125e−05
Thermal conductivity [W/(mK)] 0.0454 0.1513 0.0158
Heat capacity [J/(kgK)] 1040.67 5193 520.64
The default and initial settings of the simulation model are listed in Table 3. The
shielding gas velocity us ¼ 0:93 m=s is equivalent to a shielding gas flow rate of
150 m3/h, while the purge gas velocity up ¼ 0:11 m=s results from a purge gas flow
rate of 100 m3/h.
Table 3. Default and initial settings of the simulation model.

Gas type Nitrogen (N2)
Inlet height 78 mm
Shielding gas flow velocity us 0.93 m/s
Purge gas flow velocity up 0.11 m/s
Initially, the influence of the purge gas flow rate was investigated by increasing it by
50% as shown in Fig. 4, where the gas flow velocity field and the soot particle tracks are
depicted in the symmetry plane that goes through the plume inlet. The vortex at the rear
wall is obviously suppressed when the purge gas flow rate is increased. The plume gas
percentage is reduced from 0.40% to 0.34%, which describes the volume percentage of
gas in the symmetry plane that has entered the chamber through the plume inlet.
Gas flow velocity[m/s]
Fig. 4. Gas flow velocity and soot particle tracks at a purge gas velocity of 0.11 m/s (left) and
0.165 m/s (right).
The reduced accumulation of soot inside the process chamber is confirmed by the
results of the Archimedean density measurements in Fig. 5, which shows the achieved
density in the three experiments for the two process parameter sets, which can primarily
be distinguished by the laser power. For both laser powers, the density is increased
when the purge gas flow rate is raised by 50%, even though the corresponding build
setup generates more soot. The increase in density can be traced back to less fume
inside the build chamber, which scatters and attenuates the laser beam.
100.00%
99.50%
99.00% reference (build

density
setup 1)
98.50%
150% purge gas
flow (build setup 2)
98.00%
half inlet height
(build setup 1)
97.50%
200 W 500 W
laser power
Fig. 5. Influence of shielding gas flow setting on density of produced parts.
The aforementioned finding raises the question, if the purge gas flow rate can also
be increased at the expense of a reduced horizontal shielding gas flow rate. Therefore,
the purge gas flow rate was increased from 100 m3/h to 150 m3/h, while the shielding
gas flow rate was reduced from 150 m3/h to 100 m3/h. Both simulations were carried
out with the optimized inlet height of 39 mm. The results are depicted in Fig. 6 and
show that the inverted ratio of shielding gas flow rate to purge gas flow rate leads to a
higher amount of soot particles inside the process chamber. Thus, the plume gas
percentage increases from 0.394% to 0.428%.
Fig. 6. Gas flow velocity and soot particle tracks at a shielding and purge gas flow rate of
150 m3/h and 100 m3/h (left) or vice versa (right).
200 F. Wirth et al.
If the shielding gas flow rate is limited or fixed, a trade-off has to be found between
the inlet height and the shielding gas flow velocity. An increased inlet height means a
more homogeneous shielding gas flow velocity field across the build plate and a wider
gas jet has a higher capability to carry away the soot particles. Furthermore, a high
shielding gas flow velocity is also beneficial as it can transmit a higher momentum to
the plume in the direction towards the outlet. According to the results presented in
Fig. 7, more soot is removed from the process chamber, if the inlet height is decreased
while the volume flow rate is kept constant, leading to an increased shielding gas flow
Inlet Plume gas

height percentage
78 0.401%
mm
Gas flow velocity [m/s]
39 0.394%
mm
25 0.369%
mm
Fig. 7. Gas flow velocity and soot tracks at different inlet heights with same volume flow rate.
velocity. However, at the inlet height of 25 mm the gas flow becomes more turbulent,
which can be seen from some soot tracks that extend to the upper region of the process
chamber. This means that there is an increased risk in temporary vortices occurring,
which carry soot particles into the upper region of the process chamber or that the
plume even passes partially through the thin gas jet. This can occur especially when
processing special materials or when using extreme process parameters such as a small
spot size combined with a high laser power and a low scan speed, which lead to
significantly more evaporation and thus higher plume velocities as compared to the
situation that was assumed in the simulations. Therefore, an inlet height of 39 mm was
chosen as a conservative compromise, which was tested during the validation exper-
iments. The shielding gas inlet height was reduced from 78 mm to 39 mm by an insert
in the shielding gas nozzle. This also allows for the upkeep of a sufficiently high gas
flow rate when other powder materials such as aluminum or comparably small powder
particles are processed, which are carried away by the shielding gas at lower velocities
compared to the herein considered stainless steel powder. According to Fig. 5, the
halving of the inlet height leads to an increase of the part density by 0.4–0.5% points.
This means that the same density is achieved as if only seven cubes are produced at
once and there is no significant accumulation of soot particles inside the process
chamber any more.
The comparison between nitrogen and helium as shielding gas does not indicate a
significant difference in Fig. 8. The plume gas percentage is also similar with 0.4011%
in the case of nitrogen and 0.4009% for helium. The advantage of helium thanks to its
high thermal conductivity, which leads to a faster cooling and therefore reduced
buoyancy of the plume gas, is used up by the disadvantage of its low density and thus
the low momentum that can be transferred to the plume gas. That is also the reason,
why Pauzon et al. [8] employed gas mixtures that represent a compromise regarding the
physical properties of the different gases.
Fig. 8. Gas flow velocity and soot tracks with nitrogen (left) and helium (right) as shielding gas.
In Fig. 9 the spatter particle tracks are depicted for three different ejection velocities
ue . According to the simulation results, the spatter particles can hardly be removed and
kept away from the powder bed, even if argon is the process gas, if the inlet height is
78 mm and the shielding gas flow velocity is 3.5 m/s. 3.5 m/s is the experimentally
determined maximum velocity, above which the herein used powder is carried away
202 F. Wirth et al.
from the powder bed. Only the smaller spatter particles with the lower ejection velocity
ue are carried away by the shielding gas flow. It has to be admitted, that the ejection
angle was not taken into account. If the spatter particles are already ejected from the
melt pool towards the outlet, there is a higher chance that they do not land in the
powder bed. Nevertheless, if the spatter particles are ejected towards the shielding gas
inlet, it becomes almost impossible to prevent them from landing in the powder bed.
Therefore, the vertical ejection at least does not make the problem worse than it really
is and it is sufficient to investigate the principal mechanisms of spatter removal from the
powder bed area. However, the simulation model represents a worst case scenario as
the spatter particles are ejected close to the shielding gas inlet and have to fly over the
whole powder bed. It is so that in practice the particles ending in the powder bed are
reduced, but still considerable in volume.
Fig. 9. Tracks of spatter particles with an ejection velocity ue of 1.5, 3.0 and 4.5 m/s in an argon
atmosphere.
In order to validate the simulation results further, the gas flow velocity of nitrogen
10 mm above the build plate was measured at 9 points evenly distributed across the
whole build plate at a shielding gas flow rate of 150 m3/h and with an inlet height of
78 mm. Accordingly, the difference between measurement and simulation data is in the
range of ±10%.
5 Conclusion
A simulation model has been developed in order to investigate the influence of the
shielding gas flow in a LPBF machine on soot and spatter particles. It can be concluded
from the simulation results that a trade-off has to be found for the shielding gas inlet
height at a given maximum shielding gas flow rate. In general, a larger inlet height leads
to a more homogeneous velocity field across the powder bed. Soot is however more
efficiently removed with a smaller inlet height, which leads to higher shielding gas flow
velocities. But at the same time, the shielding gas flow becomes more turbulent and there
is an increasing risk, that the plume passes through the gas jet. Furthermore, it has to be
taken into account, that powder is carried away from the powder bed if the shielding gas
velocity exceeds the limit, which is determined by the size and weight of the powder
particles as well as by the density and viscosity of the gas.
The simulations did not reveal any advantages in the use of pure helium shielding
gas with regard to its ability in removing soot particles from the process chamber.
According to the simulation results, the shielding gas flow has only a limited
capability to carry away spatter particles; especially the critically large ones usually end
up in the powder bed.
References
1. National Institute of Standards and Technology. NIST: Measurement Science Roadmap for
Metal-Based Additive Manufacturing (2013)
2. Additive Manufacturing Special Interest Group: Shaping our National Competency in
Additive Manufacturing (2012)
3. Caltanissetta, F., Grasso, M., Colosimo, B.M.: In-situ defect detection and correction in laser
powder bed fusion. In: XIV AITeM Conference, pp. 1–11 (2019)
4. Anwar, A.B., Pham, Q.C.: Effect of inert gas flow velocity and unidirectional scanning on
the formation and accumulation of spattered powder during selective laser melting. In:
Proceedings of the 2nd International Conference on Progress in Additive Manufacturing
(Pro-AM), pp. 531–536 (2016)
5. Anwar, A.B., Pham, Q.C.: Selective laser melting of AlSi10Mg: effects of scan direction,
part placement and inert gas flow velocity on tensile strength. J. Mater. Process. Technol.
240, 388–396 (2017)
6. Renishaw. https://resources.renishaw.com/en/details/White%20paper%3a%20%20Multi-
laser%20processing%20strategies%20for%20high-integrity%20component%20manufacture
(112228). Accessed 24 Apr 2019
204 F. Wirth et al.
7. Pauzon, C., Hryha, E., Forêt, P., Nyborg, L.: Effect of argon and nitrogen atmospheres on the
properties of stainless steel 316 L parts produced by laser-powder bed fusion. Mater. Des.
179, 107873 (2019)
8. Pauzon, C., Forêt, P., Hryha, E., Arunprasad, T., Nyborg, L.: Argon-helium mixtures as
laser-powder bed fusion atmospheres: towards increased build rate of Ti-6Al-4 V. J. Mater.
Process. Technol. 116555 (2019)
9. Anwar, A.B.: Large scale selective laser melting: study of the effects and removal of spatter
by the inert gas flow. Dissertation Nanyang Technological University (2019)
10. Bidare, P., Bitharas, I., Ward, R.M., Attallah, M.M., Moore, A.J.: Fluid and particle
dynamics in laser powder bed fusion. Acta Mater. 142, 107–120 (2018)
11. Mayi, Y.A., Dal, M., Peyre, P., Bellet, M., Metton, C., Moriconi, C., Fabbro, R.: Laser-
induced plume investigated by finite element modelling and scaling of particle entrainment
in laser powder bed fusion. J. Phys. D Appl. Phys. 53(7), 075306 (2019)
12. Jakumeit, J., Laqua, R., Zielinski, J.: Effect of gas flow on the melting and solidification of
IN718 powder in a LPBF process. In: II International Conference on Simulation for Additive
Manufacturing - Sim-AM 2019 (2019)
13. Masmoudi, A., Bolot, R., Coddet, C.: Investigation of the laser–powder–atmosphere
interaction zone during the selective laser melting process. J. Mater. Process. Technol. 225,
122–132 (2015)
14. Chen, Y., Vastola, G., Zhang, Y.W.: Optimization of inert gas flow inside laser powder bed
fusion chamber with computational fluid dynamics. In: Proceedings of the 29th Annual
International Solid Freeform Fabrication Symposium (2018)
15. Morsi, S.A.J., Alexander, A.J.: An investigation of particle trajectories in two-phase flow
systems. J. Fluid Mech. 55(2), 193–208 (1972)
16. Gunenthiram, V., Peyre, P., Schneider, M., Dal, M., Coste, F., Koutiri, I., Fabbro, R.:
Experimental analysis of spatter generation and melt-pool behavior during the powder bed
laser beam melting process. J. Mater. Process. Technol. 251, 376–386 (2018)
17. Barrett, C., Carradero, C., Harris, E., Rogers, K., MacDonald, E., Conner, B.: Statistical
analysis of spatter velocity with high-speed stereovision in laser powder bed fusion. Progress
Addit. Manuf. 4(4), 423–430 (2019). https://doi.org/10.1007/s40964-019-00094-6
Artificial Intelligence for Monitoring and
Control of Metal Additive Manufacturing
Giulio Masinelli, Sergey A. Shevchik, Vigneashwara Pandiyan, Tri Quang-Le,

and Kilian Wasmer(B)
Laboratory of Advanced Materials Processing, Swiss Federal Laboratories

for Materials Science and Technology (Empa), 3602 Thun, Switzerland
kilian.wasmer@empa.ch
https://www.empa.ch/web/s204/dynamical-processes
Abstract. Quality monitoring in Additive Manufacturing (AM) is cur-

rently mostly based on temperature measurements of the process zone or
on layer/piecewise high-resolution surface imaging. To this aim, various
sensors, such as pyrometers, photodiodes, and matrix CCD detectors,
have been tested. These standard temperature measurements, however,
do not provide a comprehensive description of the process dynamics,
as they are just limited to surface observations. Furthermore, they are
often used for post-factum inspection, i.e., after the piece is partially or
even completely produced. No robust and low-cost methods are so far
known to monitor the quality of laser processes in real-time. To close this
gap, we propose an innovative approach for online quality monitoring of
additive manufacturing employing acoustic emissions (AE). In fact, AE
signals can provide in-depth information about the process, e.g., melting,
resolidification, delamination, and cracking of the workpiece. Moreover,
the sintering or melting of the metal powder has several unique acous-
tic signatures that can be detected and interpreted in terms of quality.
In our approach, the correlation of the acoustic signals to the quality
of the produced pieces is made by Artificial Intelligence (AI) methods.
Specifically, AI in the form of Machine Learning is used to perform a
data-driven extraction and recognition of the unique acoustic signatures
from different sintering or melting events. In this contribution, we present
a summary of our results in the fields of selective laser melting and laser
welding, which have similar underlying mechanisms. At first, we discuss
how, by using AE, we can classify different types of defects and porosity
content in both processes. Afterward, with the aid of high-speed X-ray
imaging, we demonstrate the real-time performance of our approach in
the classification of transient events/regimes that are critical for the final
quality—in particular conduction, stable keyhole, unstable keyhole, pore
formation, and blowout. Finally, we present the future possibilities in
terms of control of AM processes based on AI.
Keywords: Additive Manufacturing · Process monitoring · Process

control · Acoustic emission · Artificial Intelligence

https://doi.org/10.1007/978-3-030-54334-1_15
206 G. Masinelli et al.
1 Introduction
Additive Manufacturing (AM) is a technology developed to manufacture three-
dimensional components based on computer models with great flexibility on
the choice of materials, including metals, ceramics, and polymers in different
forms, such as liquids, powders, or wires. AM offers several advantages compared
to traditional manufacturing methods [1], including very few geometric shape
limitations [2], innovative multi-material manufacturing, and low running costs.
These characteristics meet the needs of modern production, to the point that
AM is considered to be the new industrial revolution [1], drawing the attention
of a growing number of industries, such as automotive [3], medical [4], aerospace
[5], jewelry [6], and food [7].
In this article, we focus on Selective Laser Melting (SLM) [8]—an AM sub-
branch that allows building metallic components from powders. Similarly to
other AM technologies, the underlying physical phenomena that enable the func-
tioning of SLM, are highly complex. Indeed, they involve the interaction between
the intense laser beam with multiple material phases (solid, liquid, and vapor),
as well as very fast heating and cooling cycles of the base material (from 103 to
107 K/s) [9].
Therefore, despite many developments of this technology, SLM still suffers
from imperfect process repeatability [10–12], limiting its applications in mass
production, in particular in the ones with high-quality standards. Indeed, many
processing parameters—e.g., laser energy, scanning speed, hatch distance, and
powder layer thickness—must be carefully chosen according to the specific pow-
der used [2,13,14]. An improper choice of the parameters can lead to an accu-
mulation of residual stress inside the workpiece or to the formation of defects,
such as significant porosity, cracks, or lack of fusion [11], resulting in weakened
mechanical properties [15].
Currently, the industrial standard for the quality monitoring of the parts
produced by AM in terms of porosity or cracking is X-ray tomography [16].
Being a costly and time-consuming process, this type of analysis is generally
performed post-mortem—after the part has already been produced. In the liter-
ature, there are also some attempts to design near real-time quality monitoring
systems [11,17]. The two main adopted approaches are: i) image analysis of the
surface of each layer of the workpiece [18], and ii) temperature measurements of
the melting zone [11]. The main drawbacks of both techniques derive from their
limitation to surface measurements, with limited access to in-depth information.
Furthermore, their temporal resolution is not ideal for the analysis of the tran-
sient events during the process, which have been reported to take place at time
scales down to hundreds of microseconds [19]. For these reasons, the accuracy of
those commercial approaches is debatable [11].
In this context, there is a clear need for a reliable, cost-effective, and effi-
cient detection system that can provide subsurface information about the pro-
cess. One solution can be found in Acoustic Emission (AE) sensing technology.
Indeed, the use of AE signals to trace the onset or propagation of cracks/defects
in non-transparent materials is well-known for its efficiency, high sensitivity, and
AM Quality Monitoring and Control with AI 207
non-destructive nature [20,21]. Moreover, it can be easily implemented in indus-

trial environments for online monitoring due to low-cost hardware. The main obsta-
cles that prevent the wide-spread use of AE technology for AM monitoring derive
from weak useful signals and loud background noise.
To this extent, we proposed a new methodology where AE is combined with
Machine Learning (ML). Indeed, highly sensitive AE sensors combined with
ML enable the detection of meaningful information in AM signals even in a
very loud environment. This approach has already been successfully applied to
several applications with a high level of background noise, including tribology
[22], and fracture mechanics [23]. It has to be noted that—even though the
selected examples used a highly sensitive fiber-based detector—the proposed
approach is very flexible in terms of sensing devices, from piezo-based contact
sensors to airborne microphones.
This contribution summarizes the work done so far by our research group
in the field of in situ AM and Laser Welding (LW) monitoring, combining AE
and ML. We start with the retrospective of the preliminary feasibility study
and then highlight the recent developments in correlating AE with the hidden
physical mechanisms responsible for the defects formation.
Firstly, we begin with an overview of the work of Shevchik et al. [24], which is
a preliminary study to investigate the feasibility of discerning, from AE signals,
between three different final build qualities in terms of porosity. The different
quality grades were intentionally provoked by using three sets of process param-
eters. Secondly, the application of our approach for the monitoring of different
LW regimes is presented [25]. Even though this work refers to a different appli-
cation, the same physical phenomena are involved in melting and subsequent
solidification of the metallic material. Thirdly, we move to a more recent work,
[26], where the AE based monitoring system was pushed a step further, aiming
at identifying the momentary events leading to the formation of defects, such as
porosity and blowout. In this case, the system was guided, during its training, by
high-speed X-ray radiography. In particular, it was shown that our methodology
could distinguish between stable and unstable keyhole regimes with high confi-
dence levels. Finally, we conclude with a discussion on how the precedent results
can be used for an AI-based closed-loop system for defect-free laser processes.
2 Detection Tool for Additive Manufacturing
The work presented hereafter is focused on SLM, but the same concepts can also
be applied to many other AM techniques. We start the survey from the work of
Shevchik et al. [24] that investigated the feasibility of in situ monitoring of SLM
processes combining AE and ML.
2.1 Experimental Setup
To validate the robustness of the method, several experiments were carried out
using a commercially available machine Concept M2 (Concept Laser GmbH,
Germany), equipped with a fiber laser that operates continuously at a wave-
length of 1071 nm. The diameter of the focused laser spot was 90 µm, and the
beam quality factor was M 2 = 1.02. The powder material was CL20ES stainless
steel (1.4404/316L) with a particle size distribution ranging from 10 to 45 µm.
The produced experimental sample had a cuboid shape with dimensions of
10 × 20 mm3 .
In this work, the quality aspect of interest was the concentration of pores in
the workpiece. Their presence is one of the most critical defects as they may not
be noticed during a visual inspection, and they harm the mechanical integrity of
the sample. Different porosity levels were induced by using diverse laser scanning
velocities during the manufacturing of the various layers of the workpiece, all the
while fixing other process parameters (laser power P = 125 W, hatch distance
h = 0.105 mm, powder layer thickness t = 0.03 mm). Figure 1 shows a typical
optical image taken at the cross-section of the workpiece.
Fig. 1. (a) Test workpiece produced with three porosity concentrations. The regions
with different qualities can be distinguished by their unique colors. Typical light micro-
scope cross-section images of the regions produced with a laser scanning speed of (b)
300 mm/s (medium quality), (c) 500 mm/s (high quality), and (d) 800 mm/s (poor
quality). Reprinted by permission from Elsevier License: Elsevier [24].
Poor quality refers to the process parameters leading to the highest pore con-
centration (1.42%) and is obtained with the highest scanning speed (800 mm/s).
In this case, the cause of the porosity is often referred to as lack of fusion, and
occured due to insufficient energy to sinter all the particles within the laser
beam. As far as the medium quality is concerned, the porosity level was 0.3%,
and the pores were caused by the overheating of the material due to the lower
laser scanning speed and, consequently, higher dose of deposited energy, resulting
in the formation of deep keyhole channels. Finally, good samples were obtained
using an average scanning speed between the two aforementioned ones, and the
achieved porosity level was 0.07%. In this case, the speed was selected to pro-
vide, on the one hand, sufficient energy to sinter all the particles within the laser
beam, and on the other hand, to prevent the material from overheating. More
details on the process parameters and their effect on the working piece can be
found in Shevchik et al. [24] and in Fig. 1.
The AE signals during the entire SLM process were recorded with a Fiber
Bragg Grating (FBG). The latter was placed inside the printing chamber, at a
distance of 20 cm from the process zone. Detailed information about the sensing
principle of the FBG sensor can be found in Shevchik et al. [24].
2.2 Correlation of AE with AM Quality

The collected AE signals were divided into three categories according to the
quality of the obtained pieces. The obtained dataset was stochastically split
into two sets, one for training and one for testing. Each category of the dataset
consisted of 300 signals of 160 ms, and there was no shared data between the
two datasets. This approach simulated a real-life condition in which the trained
system has to operate with unseen input data.
2.2.1 Feature Extraction
The mathematical transform known as Wavelet Package Transform (WPT) was

used to extract meaningful information from the signals. In particular, the WPT
allows calculating the spectrograms of each signal, consisting of the relative
energies of different narrow frequency bands [27,28]. An example of a raw AE
signal and its spectrogram computed by means of WPT can be found in Fig. 2.
Fig. 2. Example of AE signal recorded when using the optimal process parameters (a)
and the corresponding spectrogram made up of the signal’s relative energies in several
narrow frequency bands localized in time-frequency domain (b). The red marker in (a)
shows the window of 160 ms used to partition the signals. Reprinted by permission
from Elsevier License: Elsevier [24].
The signal representation provided by the WPT has several significant advan-
tages. First, it reduces the amount of data and ambient noise by allowing the
selection of specific frequency bands. In addition, it enables the interpretation of

the AE signals as an evolution over time of the chosen frequency bands. Finally,
it adapts the signals so that they can be used as inputs to existing Convolu-
tional Neural Network (CNN) implementations. In fact, most existing CNNs are
developed for image processing, where the 2D spatial domain of a picture is ana-
lyzed. Since the wavelet spectrogram domain is two-dimensional as well (time
and frequency), the existing CNNs can be directly applied to them. More details
on this technique can be found in Shevchik et al. [24].
2.3 Spectral Convolutional Neural Networks (SCNN)
The main disadvantage of using raw AE signals is their inherent complexity that
originates from the complicated nature of the underlying mechanisms. Practi-
cally, this means that the differences between signals from different categories
are not evident, making the classification task harder.
A conventional dimensionality reduction technique (Principal Component
Analysis, PCA, [29]) makes it evident that data from different categories is very
entangled, and no discrete clusters (data groupings) can be found (see Fig. 3).
For these reasons, simple, but effective ML techniques such as SVM cannot be
used.
80
60
40
20
Good quality
Medium quality
0 Poor quality -200
-150
20 30 40 -100
50 60 70 -50
80 X1
X2
Fig. 3. Projection of the acoustic features into a 3D feature space using Principal
Component Analysis (PCA). The visualized dataset includes thirty data points from
each quality category. Reprinted by permission from Elsevier License: Elsevier [24].
Deep CNNs are well-known for being able to tackle this problem by successive
projections of the data into higher dimensionality spaces [30]. The main disad-
vantage is that their training is very demanding in terms of both computational
resources and amount of required data. To overcome these limits, a Spectral
Convolutional Neural Network (SCNN) has been adopted. SCNN inherits all
the advantages of traditional convolutional neural networks, with the additional
benefit of being able to process data with more complex structures (or geome-
tries) in a lightweight fashion in terms of computational effort. These benefits
are obtained by using irregular convolutional operations as a feature extraction
tool. In particular, the irregularity of the input data is captured using graphs.
This external tool guides the network during the training procedure to optimize
its structure. As a result, it gives the possibility to process highly irregular data
in cases where the application of linear classifiers is not effective. More details
on the network architecture can be found in Shevchik et al. [24].
2.3.1 Results and Limitations

As described in Shevchik et al. [24], the overall accuracies of the SCNN applica-
tion towards porosity detection ranged from 83% to 89%. These results showed
the potential of the proposed approach, especially considering that this was only
the first feasibility study. In other words, we can conclude that the AE signals
recorded by an FBG and their processing with the SCNN have the potential to
be a solution for in situ quality monitoring of AM processes.
It has to be noted that the classification accuracy was high, but it can be
furtherly improved. Indeed, medium quality samples are occasionally misclassi-
fied with the other categories. One of the aspects that limits the potential of
the developed technique is the length of the acoustic signal acquisition window.
In the presented study, it was observed that the optimal time span was 160 ms.
However, the window length is a trade-off between stability and spatial resolu-
tion. Indeed, reducing the window span increases the temporal resolution of the
detection of the momentary events, but, unfortunately, so does the sensitivity
to noises. To overcome those limitations, in Shevchik et al. [31], the structure of
the developed SCNN is further developed to be able to process data from the
combination of two scanning windows with different time spans.
2.4 Double Scanning Strategy
As previously mentioned, the choice of the time span of the scanning window
is crucial. In fact, the AM process incorporates multiple events with different
time scales, and a window of a fixed size can hardly capture all the transients
involved. For this reason, in Shevchik et al. [31], the collected AE signals were
scanned with two running windows that were characterized by different time
lengths, indicated as short and long running window (SRW and LRW, respec-
tively). By using two running windows at the same time, an analysis—which is
stable to the presence of noise—is carried out at different time scales. This fea-
ture is particularly useful in the application under examination. Therefore, the
CNN structure has been adapted to process the spectrograms calculated from
both SRW and LRW simultaneously. Specifically, the wavelet spectrograms for
SRW and LRW were constructed separately and were used as inputs for the
SCNN classifier. Finally, an exhaustive search has been performed to identify
the optimal windows sizes and resulted in timespans of 80 ms and 160 ms for
SRW and LRW, respectively. An example of the two optimal running windows
and a typical AE signal can be found in Fig. 4.
Fig. 4. Visualization of a typical raw AE signal and the optimal windows sizes of 80 ms
and 160 ms for SRW and LRW. Reprinted by permission from IEEE License: IEEE [31].

The classification accuracies obtained by this method range from 78% to 91%.
These results demonstrate that it is possible to combine highly sensitive AE
sensors with an SCNN to classify the sample qualities in terms of porosity con-
centration. It has to be noted that the classification was carried out without
using imaging techniques or altering the setup and with minimum manual inter-
ventions. Besides, the results have shown that the continuous SLM process can
be divided into a finite number of momentary events with unique acoustic sig-
natures. These signatures can be analyzed by the developed Machine Learning
framework and interpreted in terms of quality. The developed technique is also
able to detect porosity within the material, which cannot be achieved with exist-
ing image-based methods [11].
Given the similarities of SLM to laser welding of metal workpieces, a similar
approach has been used for the detection of the different welding regimes during
laser welding.
3 Laser Welding Quality Monitoring

As described in Sect. 2.2, the developed ML technique allowed discerning between
different operating regimes responsible for different qualities during the AM pro-
cess. In the successive work [25], the authors aimed at exploiting the previously
reported approach for a more challenging task, namely, to apply the monitoring
system to LW to identify the relationships between the different welding regimes
and the AE signals. The usage of the same methodology to this different applica-
tion comes as no surprise, as LW is an industrial process with many similarities
to SLM, involving the same physical mechanism for metal melting and succes-
sive solidification. In particular, in Shevchik et al. [25], the investigations were
carried out using laser welding of titanium workpieces.
In this specific contribution, the signals collected during laser welding were
assigned to four categories: (i) no illumination, (ii) conduction welding, (iii)
keyhole without porosity, and (iv) keyhole with porosity. The first category rep-
resents the absence of laser radiation. The second one is characterized by a low
laser intensity leading to the so-called conduction welding regime (shallow). The
third and fourth categories refer to the keyhole (deep) welding regimes with and
without the presence of pores in the workpiece.
The distinction between the different welding regimes is crucial because, in
real applications, the transition between them can be provoked by local changes
in the material properties or in process parameters. To be specific, the capability
to distinguish between keyhole with and without porosity is of particular inter-
est for industrial applications. Indeed, the keyhole welding regime, despite its
advantages [32,33], is prone to defects. In particular, the high instability of the
keyhole channel in deep welding can lead to porosity in the workpiece [34,35].
In this work, the ground truth for the said categories was estimated a posteriori
by the visual inspection of the cross-sections of the welding joints.
On par with the previous works, the WPT was used to compute the relative
energies of the selected narrow frequency bands to form the spectrograms of
each signal. In contrast, the ML classifier that processes the spectrograms was
modified to boost the computational speed to enable the online monitoring of
the process.
3.1 Graph SVM with Adaptive Multi-kernel

As already discussed in paragraph Sect. 2.3, the described classification task
is challenging because of the similarities between data of different categories.
Even though the problem has been solved by using an SCNN, its relatively high
number of parameters and the computational resources needed for building a
graph for every new input sample to be classified, make it not suitable for real-
time applications.
For this reason, an original multi-kernel approach has been proposed in
Shevchik et al. [25] to relax the computational constraints while keeping a high
classification accuracy.
Specifically, in Shevchik et al. [25], the classification of the collected AE sig-
nals was carried out by a modification of the Laplacian Support Vector Machine
(LapSVM) [36]. The attractiveness of LapSVM relies on resolving sophisticated
features configurations by searching for the optimal decision cut in the feature
space to differentiate between the various categories [37]. Nevertheless, on the
other side, its main limitations derive from the general usage of a single Gaus-
sian kernel. Indeed, even though it can adapt to complex geometries, it fails to
recognize separated categories if the corresponding features are not well clus-
tered. For these reasons, LapSVM was improved by involving several adaptive
Gaussian kernels.
As shown in Shevchik et al. [25], this modification enhanced the algorithm
abilities in separating features from different categories. A depictive example on
how the multi-kernel LapSVM acts on the data can be found in Fig. 5.
It is noteworthy to notice that the developed ML technique inherits from the
conventional LapSVM a very high computational efficiency. Indeed, the classifi-
Fig. 5. Algorithm performance with synthetic data made up of two categories whose
features do not form distinct groups: (a) original feature space, where the two colors
represent the two categories; (b) decision boundaries defined by LapSVM with a single
kernel; as can be seen, there are plenty of misclassifications, as several blue points
fall inside the green region; (c) decision boundaries defined by LapSVM with multi
Gaussian kernels; (d) 3D visualization of the learned Gaussian kernels after multi-
kernel LapSVM training. Reprinted by permission from IEEE License: IEEE [25].
cation of a new input data is reduced to just a matrix multiplication, allowing

the monitoring system to work in real-time. To be specific, the computational
time needed per measurement was as fast as 70 ms.

As reported in Shevchik et al. [25], the classification accuracies of the real-time
monitoring systems are very high, ranging from 87% to 99%. However, the cat-
egories that were misclassified the most are keyhole with porosity and keyhole
without porosity. These results indicate that there are overlapping of the AE fea-
tures for those categories, which can be improved by optimizing the algorithms.
On the other hand, the false classifications can also be due to errors in the iden-
tification of the ground truth. Indeed, it is almost impossible to detect all the
porosities by optical inspection of the cross-sections, due to the small size of the
pores. Moreover, the post mortem analysis cannot identify the pores that have
been generated during the process and which have disappeared afterward when
reaching the liquid/air interface [19,38].
In support of these assertions, it can be noted in Table 1 that almost 11%
of the input data labeled by visual inspection as keyhole without porosity is
classified as keyhole with porosity, while just 8% of the input data labeled as
keyhole with porosity is mistaken for keyhole without porosity.
Table 1. Percentage classification accuracies per category versus ground truth.

Adapted from Shevchik et al. [25].
Ground truth
Conduction welding
Keyhole without
No illumination
Keyhole with
pores
pores
Classification
accuracy
[%]
No illumination 98.8 1.2 0.0 0.0
Conduction welding 0.0 92.0 3.0 5.0

Keyhole without
0.0 2.0 87.1 10.9
pores
Keyhole with
0.0 1.8 8.0 90.2
pores
3.2 Quality Monitoring with High-Speed X-Ray Radiographic

Guidance
It is noteworthy that all the research work done so far has been focused on
a statistical approach to the analysis of the AE signals with the primary goal
of monitoring. Therefore, the nature and the physical explanation of the AE
contents were not investigated. In particular, the developed method allowed to
stop a process when a non-acceptable defect is detected. However, if this happens,
there are no other solutions than to discard the workpiece, despite the waste of
materials and machining time. To overcome this issue, a fully automatic control
loop is required. Nevertheless, to achieve this goal, we need to understand the
physics of defect creation and to determine whether it is possible to classify the
transients leading to artifacts such as the transition between stable and unstable
keyhole.
Given the importance and relevance of the matter, the successive contri-
butions of our research group were focused in this direction. In particular, in
Shevchik et al. [26], the AE signals were analyzed employing ML techniques to
identify the relationships between the momentary events that have an impact
on the quality of the weld, and the AE signals. To provide the ML algorithms
with precise labeling of the data, the real-time monitoring system was combined
with high-speed hard X-ray radiography. In particular, the process dynamics
have been visualized by X-ray images of the process area. From the X-ray data,
the following events are extracted and used for the training of the ML classifier:
conduction welding, stable keyhole, unstable keyhole, blowout, and pores.
Compared to the previously described works, the AE signals were sensed by
a piezoelectric sensor (PICO HF-1.2, Physical Acoustic, Germany) attached to
the sample holder. Moreover, the laser was additionally equipped with an optical
system for the collection of the back-reflected light (LBR) from the process
zone. Figure 6 shows a sequence of X-ray images of the process zone during the
experiment, together with the recorded AE and LBR signals.
Fig. 6. X-ray images of the process zone (a) and the corresponding AE and optical
signals (b) during the experiment at 250 W. The scale bar equals 300 µm. Reprinted
by permission from Elsevier License: Elsevier [39].
The spectrograms computed utilizing the WPT (see Sect. 2.2.1) of these two
signals were then combined, and the corresponding classification accuracy is
compared to the results obtained with the individual sensors. As can be seen
in Table 2, the unification of LBR and AE improves the accuracy of the classi-
fication, although the increment is only marginal compared to what it can be
obtained with the AE only.
Table 2. Percentage classification accuracies per category versus ground truth for (a)
LBR sensor, (b) AE sensor, and (c) combination of the two sensors. Adapted from
Shevchik et al. [26].
Ground truth
Conduction welding
Conduction welding
Conduction welding
Unstable keyhole
Unstable keyhole
Unstable keyhole
Stable keyhole
Stable keyhole
Stable keyhole
Blowout
Blowout
Blowout
Pores
Pores
Pores
Classification
accuracy
[%]
Conduction welding 88 7 5 0 0 99 1 0 0 0 99 1 1 0 0
Stable keyhole 5 82 13 0 0 1 91 8 0 0 0 93 6 1 0
Unstable keyhole 4 5 87 4 0 0 2 92 6 0 0 6 94 0 0
Blowout 0 0 5 95 0 0 0 1 99 0 0 0 1 99 0
Pores 0 10 10 7 73 0 8 7 0 85 0 12 0 0 88
(a) (b) (c)
3.2.1 Results and Future Work

There are two remarkable results in this work. First, the X-ray guidance allowed
us to provide the ML framework with high quality labeled data, increasing the
certainty of the system in distinguishing between stable keyhole and unstable
keyhole. Since the latter is subject to defects, the ability to detect the stable -
unstable transition is of great interest. Indeed, it is a prerequisite for the imple-
mentation of a closed-loop process to avoid the formation of defects. Second, it
was noticed that it is not only possible to detect the formation of pores, but also
their removal through a successive scan of the laser. This outcome is particu-
larly satisfactory since it enables the realization of a device that can first detect
defects and successively repair them.
4 Beyond Classification: Process Control
As previously mentioned, the developed monitoring system allows the detection

of several momentary events responsible for the production of workpieces with
different qualities. From the initial offline analysis, we have seen how the system
has evolved both in terms of classification (from the generic regimes to the
detection of the physical events) and computational speed, making it ready for
online applications.
The natural continuation of this work is the development of a system that is
not only capable of monitoring the quality of the workpiece but also to predict the
upcoming events and to take corrective actions to prevent the onset of defects.
Our research group is already working in this direction. In particular, the goal is
to apply Machine Learning in this situation as well, in the form of Reinforcement
Learning (RL). RL, indeed, is able to learn, via trial and error, an optimal
feedback controller for the monitored process.
5 Conclusions
The present contribution provides an overview of our research efforts on in situ

and real-time monitoring of SLM and LW processes. Our approach primarily
relies on the combination of acoustic emission sensing technology and state-
of-the-art Machine Learning algorithms. Indeed, even though AE signals can
provide information that is not accessible to standard optical methods—such as
the volumetric behavior of the process zone—their complicated nature makes it
hard for conventional approaches to excerpt meaningful information from them.
ML, in contrast, is able to perform a convenient data-driven feature extraction
and recognition of the unique acoustic signatures from different quality critic
operational regimes.
In particular, our previous works show that the use of ML algorithms for the
analysis of the AE signals enables not only the detection of the final qualities of
the manufactured pieces but also the momentary events leading to the formation
of defects. Furthermore, the system was also able to identify if a subsequent laser
pulse has removed a pore, preparing the ground for a defect removal apparatus.
As of now, the computational time per classification on a PC equipped with
a Graphics Processing Unit (GPU) can be as low as 2 ms. This fact, combined
with the cost-effectiveness of the AE technology, makes the system ready for the
integration in commercially available machines with minimum effort for in situ

and real-time monitoring of AM processes.
Quality monitor is only one of the many outcomes of the presented method-
ology. The described works also open the doors for an AI-based control loop of
laser processes. Indeed, based on the gained experience, it is possible to design
an adaptive ML control unit the can detect changes in the process dynamics
and automatically adjust the process parameters to manufacture high-quality
workpieces reproducibly.
In order to achieve this goal, additional investigations are planned for future
improvements to the system performance. In fact, despite the promising results
based on AE, it is of our belief that the monitoring system can benefit from a
combination of different types of sensors, e.g., pyrometer or photodiodes. Specifi-
cally, the sensor aggregation can increase the robustness of the system in a noisier
environment and enable it to work with a more complicated process. Secondly,
an expansion of the quality categories is needed to investigate the applicability of
the algorithm to other defect types such as lack of fusion, balling, delamination,
and cracking. Finally, the generalization of the monitoring system’s capabili-
ties has to be scrutinized by investigating how the algorithm can adapt to new
materials or to different geometries.
References
1. Zhai, Y., Lados, D.A., Lagoy, J.L.: Additive manufacturing: making imagination
the major limitation. JOM 66(5), 808–816 (2014)
2. King, W.E., Anderson, A.T., Ferencz, R.M., Hodge, N.E., Kamath, C., Khairallah,
S.A., Rubenchik, A.M.: Laser powder bed fusion additive manufacturing of metals;
physics, computational, and materials challenges. Appl. Phys. Rev. 2(4), 041304
(2015). http://aip.scitation.org/doi/10.1063/1.4937809
3. Reddy, S.N., Maranan, V., Simpson, T.W., Palmer, T., Dickman, C.J.: Application
of topology optimization and design for additive manufacturing guidelines on an
automotive component. In: Proceedings of the ASME Design Engineering Technical
Conference, vol. 2A-2016. American Society of Mechanical Engineers (ASME),
December 12 2016
4. Salmi, M., Tuomi, J., Paloheimo, K.S., Björkstrand, R., Paloheimo, M., Salo, J.,
Kontio, R., Mesimäki, K., Mäkitie, A.A.: Patient-specific reconstruction with 3D
modeling and DMLS additive manufacturing. Rapid Prototyp. J. 18(3), 209–214
(2012)
5. Türk, D., Kussmaul, R., Zogg, M., Klahn, C., Spierings, A.B., Könen, H., Ermanni,
P., Meboldt, M.: Additive manufacturing with composites for integrated aircraft
structures. In: International SAMPE Technical Conference, pp. 1404–1418 (2016)
6. Klotz, U.E., Tiberto, D., Held, F.: Optimization of 18-karat yellow gold alloys for
the additive manufacturing of jewelry and watch parts. Gold Bull. 50(2), 111–121
(2017)
7. Blurhapsody launches 3D printed pasta service (for real this time)/video -
3D printing media network, [ONLINE] https://www.3dprintingmedia.network/
blurhapsody-launches-3d-printed-pasta-service-real-time-video/. Accessed 24 Apr
2020
8. Frazier, W.: Metal additive manufacturing: A review. J. Mater. Eng. Perform. 23,
1917–1928 (2014)
9. Das, M., Balla, V.K., Basu, D., Bose, S., Bandyopadhyay, A.: Laser processing of
SiC-particle-reinforced coating on titanium. Scripta Mater. 63(4), 438–441 (2010)
10. Gu, D.D., Meiners, W., Wissenbach, K., Poprawe, R.: Laser additive man-
ufacturing of metallic components: materials, processes and mechanisms. Int.
Mater. Rev. 57(3), 133–164 (2012). http://www.tandfonline.com/doi/full/10.
1179/1743280411Y.0000000014
11. Everton, S.K., Hirsch, M., Stravroulakis, P., Leach, R.K., Clare, A.T.: Review
of in-situ process monitoring and in-situ metrology for metal additive manufac-
turing. Mater. Des. 95, 431–445 (2016). http://www.sciencedirect.com/science/
article/pii/S0264127516300995
12. Tapia, G., Elwany, A.: A review on process monitoring and control in metal-based
additive manufacturing. J. Manuf. Sci. Eng. 136, 060801 (2014)
13. Tammas-Williams, S., Zhao, H., Léonard, F., Derguti, F., Todd, I., Prangnell, P.B.:
XCT analysis of the influence of melt strategies on defect population in Ti-6Al-4V
components manufactured by Selective Electron Beam Melting. Mater. Charact.
102, 47–61 (2015)
14. Shifeng, W., Shuai, L., Qingsong, W., Yan, C., Sheng, Z., Yusheng, S.: Effect of
molten pool boundaries on the mechanical properties of selective laser melting
parts. J. Mater. Process. Technol. 214(11), 2660–2667 (2014)
15. Lewandowski, J.J., Seifi, M.: Metal additive manufacturing: a review of mechan-
ical properties. Ann. Rev. Mater. Res. 46(1), 151–186 (2016). http://www.
annualreviews.org/doi/10.1146/annurev-matsci-070115-032024
16. Thompson, A., Maskery, I., Leach, R.: X-ray computed tomography for additive
manufacturing: a review. Meas. Sci. Technol. 27, 072001 (2016)
17. Sharratt, B.M.: Non-destructive techniques and technologies for qualification of
additive manufactured parts and processes: a literature review. Technical report
(2015)
18. Scime, L., Beuth, J.: Anomaly detection and classification in a laser powder bed
additive manufacturing process using a trained computer vision algorithm. Addit.
Manuf. 19, 114–126 (2018)
19. Martin, A.A., Calta, N.P., Khairallah, S.A., Wang, J., Depond, P.J., Fong, A.Y.,
Thampy, V., Guss, G.M., Kiss, A.M., Stone, K.H., Tassone, C.J., Nelson Weker, J.,
Toney, M.F., van Buuren, T., Matthews, M.J.: Dynamics of pore formation during
laser powder bed fusion additive manufacturing. Nat. Commun. 10(1), 1–10 (2019)
20. Krampikowska, A., Pala, R., Dzioba, I., Šwit, G.: The use of the acoustic emission
method to identify crack growth in 40CrMo steel. Materials 12(13), 2140–2147
(2019)
21. Mazal, P., Vlasic, F., Koula, V.: Use of acoustic emission method for identification
of fatigue micro-cracks creation. In: Procedia Engineering, vol. 133, pp. 379–388.
Elsevier Ltd. (2015)
22. Saeidi, F., Shevchik, S.A., Wasmer, K.: Automatic detection of scuffing using acous-
tic emission. Tribol. Int. 94, 112–117 (2016)
23. Shevchik, S.A., Meylan, B., Mosaddeghi, A., Wasmer, K.: Acoustic emission for in
situ monitoring of solid materials pre-weakening by electric discharge: a machine
learning approach. IEEE Access 6, 40313–40324 (2018)
24. Shevchik, S.A., Kenel, C., Leinenbach, C., Wasmer, K.: Acoustic emission for in situ
quality monitoring in additive manufacturing using spectral convolutional neural
networks. Addit. Manuf. 21, 598–604 (2018)
25. Shevchik, S.A., Le-Quang, T., Farahani, F.V., Faivre, N., Meylan, B., Zanoli, S.,
Wasmer, K.: Laser welding quality monitoring via graph support vector machine
with data adaptive kernel. IEEE Access 7, 93108–93122 (2019)
26. Shevchik, S., Le-Quang, T., Meylan, B., Farahani, F.V., Olbinado, M.P., Rack, A.,
Masinelli, G., Leinenbach, C., Wasmer, K.: Supervised deep learning for real-time
quality monitoring of laser welding with X-ray radiographic guidance. Sci. Rep.
10(1), 1–12 (2020)
27. Mallat, S., Hwang, W.L.: Singularity detection and processing with wavelets. IEEE
Trans. Inf. Theory 38(2), 617–643 (1992)
28. Daubechies, I.: Ten Lectures on Wavelets. Society for Industrial and Applied Math-
ematics (1992)
29. Jolliffe, I., Cadima, J.: Principal component analysis: a review and recent develop-
ments. Philos. Trans. Roy. Soc. A Math. Phys. Eng. Sci. 374, 20150202 (2016)
30. Goodfellow, I., Bengio, Y., Courville, A.: Deep Learning. MIT Press (2016). http://
www.deeplearningbook.org
31. Shevchik, S.A., Masinelli, G., Kenel, C., Leinenbach, C., Wasmer, K.: Deep learn-
ing for in situ and real-time quality monitoring in additive manufacturing using
acoustic emission. IEEE Trans. Industr. Inf. 15(9), 5194–5203 (2019)
32. Piveta, A.C.G., Montandon, A.A.B., Ricci, W.A., Nagle, M.M.: Mechanical
strength and analysis of fracture of titanium joining submitted to laser and tig
welding. Mater. Res. 15(6), 937–943 (2012)
33. Svenungsson, J., Choquet, I., Kaplan, A.F.: Laser welding process - a review of
keyhole welding modelling. In: Physics Procedia, vol. 78, pp. 182–191. Elsevier
B.V., January 2015
34. Zhou, J., Tsai, H.L., Wang, P.C.: Transfort phenomena and keyhole dynamics
during pulsed laser welding. J. Heat Transfer 128(7), 680–690 (2006)
35. Berger, P., Hügel, H.: Fluid dynamic effects in keyhole welding an attempt to
characterize different regimes. In: Physics Procedia. vol. 41, pp. 216–224. Elsevier
B.V. , January 2013
36. Belkin, M., Niyogi, P., Sindhwani, V.: Manifold regularization: a geometric frame-
work for learning from labeled and unlabeled examples. J. Mach. Learn. Res. 7,
2399–2434 (2006)
37. Melacci, S., Belkin, M.: Laplacian support vector machines trained in the primal.
J. Mach. Learn. Res. 12, 1149–1184 (2011)
38. Courtois, M., Carin, M., Le Masson, P., Gaied, S., Balabane, M.: A complete model
of keyhole and melt pool dynamics to analyze instabilities and collapse during laser
welding. J. Laser Appl. 26(4), 042001 (2014). http://lia.scitation.org/doi/10.2351/
1.4886835
39. Le Quang, T., Shevchik, S., Meylan, B., Vakili-Farahani, F., Olbinado, M., Rack,
A., Wasmer, K.: Why is in situ quality control of laser keyhole welding a real
challenge?, vol. 74, pp. 649–653, October 2018
Simulation
Development of a Process Model for Bead
Deposition Rates and Cooling Behavior
of Large Scale Additive Manufacturing Parts
Michel Layher(&), Lukas Eckhardt, Andreas Hopf, and Jens Bliedtner
Ernst-Abbe-University of Applied Sciences Jena, 07745 Jena, Germany

michel.layher@eah-jena.de
Abstract. Large Scale Additive Manufacturing (LSAM) has been under constant
development during the last years. While granule extrusion technologies made a
variety of materials available for additive manufacturing, they also introduced new
challenges in processing different polymers. Therefore, systems at medium
extrusion rates (1–6.5 kg/h) have been in the focus of investigations. Even though
current setups can be utilized for individual manufacturing processes, they are
lacking a generalized description for defining repeatable bead parameters. Espe-
cially layer height is a crucial parameter since, in opposite to Fused Filament
Fabrication (FFF), a significant deviation between height setting and the resulting
height occurs. Based on that observation empirical investigations were conducted
in order to derive repeatable process parameters for bead deposition. As a result, a
model was found to describe different bead formations. Now, desired input
parameters such as bead height and width directly lead to manufacturing settings
like extrusion rate, feed rate and the necessary z-increment. Especially the devi-
ation of the latter parameter can be compensated by the found function. Besides
this important peculiarity of LSAM, cooling mechanisms differ significantly to
FFF. Due to the larger bead diameter, the ratio between volume and cooling surface
area increases and therefore, material solidification requires longer cooling times.
Excessive cooling intervals for large geometries on the other hand do not appear to
affect part accuracy or bonding quality in a negative way. This behavior has been
investigated on shell-models (hollow cuboids and hollow cylinders) and the
minimum temperature difference necessary for achieving accurate parts was
gained. In addition, bonding quality for different bonding temperatures was
determined in a flexural test for the materials ABS and PMMA.
Keywords: Large scale additive manufacturing Screw extrusion Bead

deposition model Cooling behavior Thermal analysis
1 Introduction
Fused Layer Modelling (FLM) is known as one of the most utilized additive manu-
facturing (AM) technologies. The majority of systems realize part generation by the
melting of a filament through a heated nozzle and its deposition on a platform,
according to the part’s geometry. Besides this concept, developed by Crump [1],
granule based processes have been introduced during the last years. First developments

https://doi.org/10.1007/978-3-030-54334-1_16
224 M. Layher et al.
were applied in scientific research for processing ceramics [2], bone tissue [3] as well
as fiber-composites [4]. Especially in the last years granule based experimental setups
have evolved to huge manufacturing devices for industrial applications such as Big
Area Additive Manufacturing (BAAM) [5] or combinations of printing and milling
machines [6].
Despite different analyzation- and processing approaches like fractographic
investigations of FLM-parts at tensile loads [7], numerical simulations of residual stress
[8] or the reduction of voids by the addition of thermally expandable microspheres into
the polymer [9], an overall understanding of the process parameters’ influence on the
part quality is still missing. Further reasons can be found in the plethora of heat transfer
models based on de Gennes [10] (neck growth, diffusion, randomization, reptation–
model, etc.) [11–17], which have only been validated partly by recent practical
investigations [18]. Studies show that especially residual stress needs to be avoided in
order to compensate effects like warping or delamination. At the moment this can only
be achieved by means of parameter studies as exemplarily shown in [19–21]. Even
though correlations between parameters and quality properties are depicted, no uni-
versal and comprehensive process model has been presented, yet [22].
In the focus of the investigations is the development of a model which describes the
relation between controllable process parameters such as nozzle velocity in the x-y-
plane (feed rate) and z-increment as well as the resulting bead geometry (height and
width) in large scale additive manufacturing. Furthermore, cooling mechanisms of shell
components including single layer cooling, part stability of hollow cylinders and the
influence of cooling time on bond strength were examined. The materials chosen for
the experiments are acrylonitrile-butadiene-styrene-copolymer (ABS – Polylac PA–
758) as a standard polymer in FLM, as well as poly(methyl methacrylate) (PMMA -
Altuglas VSUVT), which provides perfect properties regarding post-processing steps,
such as laser beam polishing [23].
2 Background Large Scale Additive Manufacturing (LSAM)
Over the last years Fused Layer Modelling has been under constant improvement.
Nevertheless, current extrusion processes are not completely understood. Especially,
since large scale additive manufacturing (LSAM) is not only a younger technology but
also different in processing behavior. This significantly affects the deposition of defined
bead geometries and hence, the creation of components in general.
Several approaches were developed in order to describe the bead geometry
depending on defined parameter settings. Spinnie and Smith [24] derived a linear model
connecting flow rate to rotational speed of the extruder, nozzle height and table speed.
The results showed a high disparity between extrusion rate and platform velocity as
well as changes in shape at higher and lower velocities. It was noticed that bead height
showed values approaching unity as the nozzle height was successively increased.
Kumar et al. [25] investigated how parameters like screw speed, barrel (nozzle)
temperature, bed temperature, deposition speed and stand-off distance between the
nozzle tip and bed surface effect the melt flow rate, layer thickness and road width.
Thereby, ‘one factor at a time’ approach was used. Results showed an increase of melt
Development of a Process Model for Bead Deposition Rates 225
flow rate by increasing screw speed and barrel temperature. Also, road width and layer
thickness were affected by changes in barrel temperature, bed temperature, deposition
speed and stand-off distance between the nozzle and the platform. Deposition speed
and stand-off distance were considered being the most influential parameters on the
layer thickness.
The most accurate approach was found by Schmidt et al. [26] since a clear relation
between rotational speed and mass output was found. The strand geometry was eval-
uated with a target-performance-comparison which lead to a process window of suit-
able parameters to define strand width and height. Both showed maximum deviations
of <10%. Nevertheless, an overall model for deriving bead geometry, especially in
height, is not provided. The occurring gap between the desired height setting and
gained height after the process is not described by the depicted relations.
A different way to define a certain bead geometry is the utilization of a roller for
consolidation. This allows to set beads to a certain height and a corresponding width.
A drawback is found in the temperature gradient between bead and roller, which might
influence bonding behavior or residual stresses [27, 28].
It is seen that LSAM is still lacking an overall model for bead description in relation
to the user defined parameter settings. This affects the processing conditions during a
print as well as achievable part quality.
3 Experimental Setup and Characteristics
The experimental setup utilized for extrusion process analysis and the fabrication of
polymer based AM parts consists of a three-axis milling machine which has been
converted into a 3D-printer by replacing the milling spindle by an extrusion plasticizing
system (Fig. 1).
Additionally, a heating platform was implemented to provide sufficient adherence
between extruded polymer and the build platform. Part dimensions are limited to a
maximum of 210 mm 210 mm 250 mm (x, y, z) and an exhaust system ensures
capturing emerging gases during the printing process. The maximum provided feed rate
is vf = 1200 mm/min. In order to exploit the merits of granule-based extrusion a
conventional single-screw extruder (ExOn8 by Dohle) was chosen. Dohle provides
several types of extrusion systems and gained first experiences in AM direct extrusion
already. The system provides two heater bands and thus, two temperatures zones for
melting the polymer. It is very flexible regarding material selection and unaffected by
its vertical mounting position. Therefore, a continuous process including a homoge-
nous melt flow as well as little thermal degradation is enabled. The nozzle has an outer
diameter of 10 mm, an inner diameter of 3 mm and provides a mass output of up to
6 kg/h. The overall mechanical design of the 3D-printer allows further extensions such
as adding infrared sensors or cameras.
In order to run repeatable extrusion processes it is necessary to have a well aligned
and characterized system. The material dependent temperature settings were predefined
by the extruder manufacturer (ABS: Tscrew = 195 °C; Tnozzle = 240 °C/PMMA:
Tscrew = 145 °C; Tnozzle = 205 °C) and are thus, optimally aligned to the used mate-
rials. Nevertheless, investigations regarding the occurrence of temperature deviations
were conducted. Results show that the deposition rate is not distinctively affected by a
shift of nozzle and screw temperatures of 10 K above and below the materials’ optimal
processing conditions. Hence, there is no need for a further adaption or optimization of
the extruder’s heating zones.
Material supply
Extruder
y Exhaust system
Build platform
z
x 3-axis system (x, y, z)
Fig. 1. Experimental setup
While deposition rates do not appear to be very temperature-sensitive, they are

dependent on material characteristics and properties of the extrusion system. The
extruder provides rotational speeds of n = 15–80 1/min. Falling below or exceeding
those values will lead to malfunctions and might cause damages to the system. Con-
sequently, the extruder was operated at different rotational speeds, starting at the lowest
(n = 15, 17, 19 – 30 1/min). The deposited material was weighed afterwards. This
procedure was repeated ten times for each setting. The addressed large scale printing
process should provide an approximate output of 2 kg/h. This amount was reached at
n = 30 1/min for both materials. Thus, the upper limit of investigation settings was
defined. According to the gained information a linear relation between rotational speed
(n in 1/min) and volumetric material flow rate (yABS/yPMMA in mm3/s) was gained
(Fig. 2). Through regression two valid functions for deposition rates of ABS and
PMMA were found for n:
yABS ¼ 14:1 n þ 116 ð1Þ
yPMMA ¼ 15:7 n ð2Þ
and validated by a correlation coefficient of R2ABS = 99.88% and R2PMMA = 99.69%.

550
500 y = 14.1x + 116

Volumetric flow rate y [mm³/s]
R² = 0.9988
450
400
350
300
250 y = 15.7x PMMA 145°C / 205°C

R² = 0.9969 ABS 195°C / 240°C
200
12 17 22 27 32
Rota onal speed extruder n [1/min]
Fig. 2. Material deposition of ABS and PMMA according to rotational speed of the extruder
The present results lay the foundation for upcoming investigations regarding the
realization of user defined bead geometries.
4 Development of a Bead Geometry Model for ABS

and PMMA
Schmidt et al. [26] developed an approach for granulate-based strand deposition. He

assumes a stadium-shaped bead exhibiting a certain height (h) and width (w) (Fig. 3
left) and is therefore able to calculate the cross-sectional area for given bead dimen-
sions. The optimal feed rate is then deduced by equating extruded volume (cross-
sectional area times feed rate) and strand volume (deposition rate). A reduced height-
setting leads to a reduced bead height and increased bead width, when extruded volume
and feed rate are kept constant. Thresholds such as depositing too little or too much
material were investigated. However, there is no discrepancy between the desired
height-setting (z) and the effective bead height (h) specified [26]. In order to ensure
consistent part quality and a homogenous bead deposition along any direction, this
consideration becomes necessary.
The findings of Schmidt et al. [25] were applied to calculate the feed rates for a test
series with a constant rotational extruder speed of 15 1/min, a layer height of h = 4 mm
and a bead width w between 4 mm and 10 mm.
Contrary to the results of Schmidt et al. [26] the resulting bead height was found not
to be equal to the z height-setting. Therefore, the height-setting was varied within an
interval of 1.6 mm to 5 mm by Dz = 0.2 mm. The dimensions of the generated
geometries were measured and geometrical limits derived. For instance, a low z height-
setting can cause a bead width larger than 10 mm.
Fig. 3. Cross-sectional view of bead geometries
This exceeds the outer diameter of the nozzle and therefore, leads to material
accumulation at the nozzle surface and hence, defects at deposited strands (Fig. 3 right).
On the other hand, the height of the deposited bead cannot be larger than the free bead
diameter in order to avoid undesired shapes as illustrated in Fig. 3, center. For the used
3 mm nozzle the free bead diameter is 4 ± 0.1 mm due to the die swell phenomenon
[29] which limits the scope of further investigations to z 4 mm. The minimal cross-
sectional area of a bead is limited by the machine’s maximal feed rate of 1200 mm/min.
Any cross section smaller than 10 mm2 requires feed rates >1200 mm/min or rotational
extruder speeds <15 1/min which cannot be realized with the available system.
The gained values provide evidence that the variables height-setting (z), bead
height (h) and feed rate (vf) are linked to each other. The correlation between bead
height and height-setting is approximately linear with a varying offset. However, it is
rather described by a logarithmic function, especially when approaching the mentioned
process limits. The relation between height-setting and feed rate on the other hand is
more intricate, since the feed rate directly relates to the cross-sectional area of the bead.
Geometric properties are therefore largely influenced by the melt flow between nozzle
and build platform. While these rheological characteristics of the polymer are not easily
assessed in a single equation, a fractional rational function proved to be a good way to
describe the dependencies. Ultimately the z-height setting for a desired bead geometry
can be calculated with the following numerical value equation:
b d lnðhÞ
z ¼ aþ þ c ln h þ 2 þ f ln h2 þ g ð3Þ
vf vf vf
Depending on the materials’ properties the coefficients of the associated model differ,
which is depicted in Table 1. Providing a correlation coefficient of R2 = 99.26% for
ABS and R2 = 98.94% for PMMA, the model’s validity within the defined process
window is proven.
Table 1. Determined coefficients of ABS and PMMA

a b c d f g R2
ABS 1.97 −766 1.38 −570000 −0.293 1930 99.26%
PMMA 2.21 −933 1.60 −510000 −1.53 2590 98.94%
The material dependent fit functions are displayed in Fig. 4. Therein the relations
between height-setting (z), bead height (h) and feed rate (vf) are visualized.
Fig. 4. Fit function of ABS (left) and PMMA (right) in spatial dimension
The model has been successfully tested for various bead geometries. In order to
narrow the focus of upcoming investigations only beads with w = 7 mm; h = 3 mm
(ABS) and w = 6 mm; h = 3 mm (PMMA) are going to be examined further.
Due to these findings an overall definition of related printing parameters in large
scale additive manufacturing is provided for the first time. A software application was
developed and the derived function implemented. Now it is possible to gain the nec-
essary process parameters (vf, z, n) by inserting the desired bead geometry (h, w).
Nonetheless, the derived function is always depending on specific material properties
and tailored to the present setup. If other than the described materials are being used,
particular coefficients need to be gained again. This requires a repeated investigation of
the geometrical shape of the single beads.
5 Cooling Mechanisms of Large Scale Shell Components
In large scale additive manufacturing as well as in fused layer modelling processes

generally, bead temperature is the governing parameter for bonding quality and
maintaining shape accuracy of manufactured parts. Since every kind of material is
optimally processed at a certain extrusion temperature, bead formations are rather
depending on feed rate, related cooling mechanisms and the amount of deposited
material. Particularly beads of large scale parts retain, due to the bigger volume, more
heat than classic FLM parts. Consequently, an enhanced understanding of the pro-
cessing conditions, especially the cooling behavior becomes necessary.
5.1 Layer Cooling of Hollow Cuboids

The cooling mechanisms in LSAM were investigated by means of a temperature
analysis for both acrylonitrile-butadiene-styrene-copolymer and poly(methyl
methacrylate). An infrared camera (Optris PI640) was implemented into the experi-
mental setup in order to measure the radiance emitted by individual layers of hollow
cuboids in the long wave infrared range (Fig. 5). Based on this data layer temperatures
were derived. Printing parameters were determined using the established relations
between deposited volume and desired strand volume as well as the numerical value
Eq. (3). Specimens’ parameters are delineated in Table 2.
Table 2. Specimens’ parameters for thermal analysis

Part circumference Part height w h vf z
[mm] [mm] [mm] [mm] [mm/min] [mm]
ABS 860 80 7 3 1032 3.91
PMMA 650 80 6 3 848 3.66
Fig. 5. Thermal analysis of a hollow cuboid (PMMA)
For the measurements an emission coefficient of e = 0.95 was assumed [30].

According to the camera’s data sheet an inaccuracy of ±2% (or at least ±2 K) needs to
be taken into account. Applying the bead geometry model, a hollow cuboid consisting
of one wall was manufactured as reference model for further investigations. Generally,
heat dissipation occurs through free convection as well as conduction to previously
deposited strands. Measuring points of distinct layers at different layer numbers provide
comprehensive insights into cooling mechanisms of single strands as part of the ref-
erence geometry. Since reheating occurs once the subsequent bead is deposited above
any given measurement position, the temperature is only recorded for the duration of
one circulation. Circulation times are DtABS = 50 s and DtPMMA = 46 s. The temper-
ature measurement ends a few seconds earlier. Once the nozzle approaches the mea-
surement point its heat radiation is reflected at the lower bead and impedes the
measurements’ accuracy. Merely measurement time at the nth bead exceeded this time
duration since no subsequent bead is being deposited afterwards. The collected data is
displayed in Fig. 6.
Fig. 6. Cooling of deposited PMMA-beads for a hollow cuboid
It is noticed that cooling functions T(t) of different layer numbers show only little
deviations to each other. The temperature curve generally agrees with existing
numerical models that incorporate radiation, convection and conduction, yet there is no
convenient analytic function to generalize the gathered data. Newton´s law of cooling
[31] cannot be applied as temperatures throughout the bead´s cross section are not
constant. Analytic models that account for the geometric boundary conditions as well
as transient thermal conduction have been proposed [16] but cannot easily be adapted
to gathered data. Instead, a straightforward approach to generalize and assess mea-
surements is required. A suitable function needs to consider the ambient temperature
(T∞) and the extrusion temperature (T0) in order to accurately describe the cooling
process. As it appears that the cooling rate is not only proportional to the temperature
difference between bead and environment but also to some extend time-dependent, a
further, time-dependent exponent has been added to Newton’s equation:
pffi
T ðtÞ ¼ T1 þ ðT0 T1 Þ eat þ b t ; ða; b\0Þ ð4Þ
While this approach is so far not corroborated by an underlying model it does provide
an easy and accurate way to fit data and can be substantiated with numerical FE–
calculations. A transient thermal simulation with the bead’s geometry, physical prop-
erties and added convection cooling along the lateral surface e.g. provides a surface
temperature curve that is perfectly described by the presented function (R2 = 99.97%).
Despite containing a certain error in measurement, the data, gathered by thermal
imaging, can also be fitted providing a correlation coefficient of R2 = 98.98% for ABS
and R2 = 99.53% for PMMA with the displayed, material dependent coefficients
(Table 3). These results were reproduced for a number of parts made from vertically
stacked single strands. Also the cooling of thermally thin, conventional FLM-beads
with layer heights of 0.3 mm was investigated and fitted substantially more successful
(R2 = 99.90%) than by utilizing the unmodified Newton´s law (R2 = 82.9%).
Table 3. Coefficients describing the cooling function of the ABS and PMMA specimens
w h T0 T∞ a b Dt R2
pffiffi
[mm] [mm] °C °C [1/s] ½1= s] [s] [%]
ABS 7 3 222 20 −0.00163 −0.0825 50 98.98
PMMA 6 3 195 20 −0.000583 −0.063 46 99.53
Despite those promising results it is important to mention that this approach cannot
be applied to other kinds of materials, bead dimensions or part geometries without
adapting the coefficients. The function is solely intended to describe the cooling of
specific parts mathematically in order to make the evaluation of thermographic mea-
surements workable. In practice it is often beneficial to combine this approach with
existing cooling models that are not solely based on Newton´s law of cooling as
presented by Compton et al. [32] or Costa [18].
5.2 Part Stability of Hollow Cylinders

In order to gain comprehensive knowledge about the relation of cooling mechanisms
and part stability in LSAM, further examinations at cylindrical shell parts were con-
ducted. Applying the bead geometry model again, hollow cylinder geometries of
H = 45 mm in height and made of ABS (bead geometry: w = 7 mm; h = 3 mm) and
PMMA (bead geometry: w = 6 mm; h = 3 mm) were manufactured at a constant feed
rate. The cylinder diameter was varied according to predefined values so the circulation
time changed as well. While printing, deposited layers were analyzed by an infrared
camera (Optris PI640).
Fig. 7. Thermal investigations regarding cooling behavior of PMMA hollow cylinders of

various diameters
Layer temperatures as well as temperature differences between preceding layers and

subsequent layers were gained by evaluating the fitted cooling functions of individual
cylinders. Depending on the circulation time, parts showed deviations in shape and
height (Fig. 7). Especially hollow cylinders of small diameters exhibited strong
deformations since insufficient bead cooling leads to heat accumulation and hence,
mechanical instabilities. Dimensional accuracy is only achieved with sufficient cooling
time for deposited strands. A decrease of the polymer’s temperature leads to an
increased shear modulus. Thus, the weight of subsequently generated layers can be
borne by lower layers without any alteration of their height or width. Generally, this
state of rigidity is reached below the polymer’s glass transition temperature (Tg).
Compared to standard FLM-processes extruded strands in LSAM contain a higher heat
capacity due to their increased mass. As a consequence, the temperature reduction due
to convection proceeds slower and is therefore the limiting factor for achievable
printing speeds, especially at smaller parts. Also bonding quality might be affected
negatively when temperature of a preceding bead drops under Tg and hence, neck
growth and randomization processes cannot take place to the full extent.
The printed components’ height was measured (Hmeas) and compared to its nominal
value of Hnom = 45 mm. Since the materials’ thermal shrinkage needs to be considered,
expected heights of Hshrink = 44.2 mm (ABS) and Hshrink = 44.4 mm (PMMA) con-
firm a deviation free printing process. All related and gained values for parts made of
ABS and PMMA are depicted in Table 4.
Table 4. Determined cylinder parameters for ABS and PMMA

h w a Hnom Hshrink Hmeas Dt DT
[mm] [mm] [1/K] [mm] [mm] [mm] [s] [K]
ABS 3 7 8.5E−05 45 44.2 44.33 ± 0.05 36.5 81 ± 1.7
PMMA 3 6 8.0E−05 44.4 44.25 ± 0.05 27 56 ± 1.4
h – bead height w – bead width a – expansion coefficient
Hnom – nominal height Hshrink – shrink-height Hmeas – measured height
Dt – circulation time DT – temperature difference (layer to layer)
In Fig. 8 the connection between the circulation time/ temperature difference and
the associated achievable component height is delineated. Each circulation time leads
to a certain temperature difference of two consecutive strands. Is it increased, the
deposited bead gains more time to cool down and to solidify. Thus, the temperature
difference between two consecutive beads raises and part stability and thereby, height is
increased. The experiment revealed that parts are considered mechanically stable at a
circulation time of Dt = 36.5 s for ABS and Dt = 27 s for PMMA. This time span is
related to a bead temperature difference (preceding to subsequent layer) of DT = 81
± 2 K at ABS and DT = 56 ± 2 K at PMMA, respectively. Accordingly, a stable
part is reached at Hshrink, depicted by the dashed line (for easier visualization only one
line is shown since shrink-height of ABS and PMMA is similar).
90 40.0
PMMA (ΔT)
T of consecu ve beads [°C]
80 ABS (ΔT) 35.0
Circula on me t [s]
70 PMMA (Δt) 30.0
ABS (Δt) 25.0
60
20.0
50
15.0
40
10.0
30 5.0
20 0.0
10 20 30 40 50
Component height Hmeas [mm]
Fig. 8. Temperature difference and circulation time as well as the associated component height
The displayed results describe a defined cooling time of LSAM layers of hollow
components. It is important to mention that those findings are only valid for the
specified terms. Alterations in bead geometry, material selection as well as environ-
mental conditions, such as an enclosed build chamber or additional heating systems,
will strongly influence the gained printing results. Hence, any kind of modification will
lead to different results and consequently further investigations.
5.3 Influence of Cooling Time on Flexural Strength

In LSAM as well as generally in FLM-processes, bonding quality is one of the most
important criteria. Anisotropy and orthotropy, respectively, occur due to the layered
manufacturing in accordance with bead orientation. Nevertheless, components printed
in standard FLM processes are often compared to injection molded parts as shown in
[33–36]. Also, cooling mechanisms are supposed to have the biggest impact on bond
quality [22, 37].
Currently, testing methods for additively manufactured parts are provided in

DIN EN ISO 17296-3 [38] which recommends to determine flexural properties of
polymers according to DIN EN ISO 178 [39]. However, in LSAM it is very difficult to
match the required specifications. That makes an overall comparison to injection
molded parts challenging. Nevertheless, parts built at certain process parameters and
under repeatable manufacturing conditions are comparable to each other. Conse-
quently, merely specimen dimensions and test-settings were derived from DIN EN ISO
178 [39] and utilized for the non–standardized investigation. The interlayer bonding
quality was examined for the already displayed parts. Hollow cuboids were generated
with specifically programmed interruption times between each deposited layer. It was
expected that longer cooling leads to decreased interlayer bond, since wetting, diffusion
and randomization mechanisms cannot take place to their full extent [40]. Specimens
are cut out having a thickness of 7 mm and width of 15 mm at ABS as well as a
thickness of 6 mm and width of 15 mm at PMMA. Due to the grooved surface
structure the effective specimen thickness is reduced causing systematically underes-
timated flexural strengths. A Length of 80 mm was achieved by printing 27 layers of
3 mm in height, considering the materials’ shrinkage.
Fig. 9. Flexural strength measurement with Ametek LS5
The difference in layer width between ABS and PMMA as well as the number of
investigated conditions is again due to the limited feed rate of the experimental
setup. A 4-point-bending test was conducted using an Ametek LS5 universal testing
machine with a YLC 5 kN load cell (Fig. 9).
Results debunked the stated expectations. Cooling time independent, the tested
ABS and PMMA specimens show no significant difference in bonding strength
(Fig. 10).
35.0
PMMA
30.0 [ZELLBEZ]s [ZELLBEZ]s
ABS [ZELLBEZ]s
Flexural strength σ [MPa]
25.0
20.0
15.0
[ZELLBEZ]s [ZELLBEZ]s [ZELLBEZ]s [ZELLBEZ]s
10.0
5.0
0.0
40 60 80 100 120 140 160 180
ΔT of consecu ve beads [K]
Fig. 10. Temperature (and time-) difference of consecutive beads and resulting flexural strength
The bonding quality seems to be unaffected by cooling time variations. A possible

cause is the increased amount of heat inside the deposited bead due to its bigger
volume. This energy reheats adjacent beads and therefore, still enables fusing to colder
strands. The gained values of DT = 157 ± 5 K (ABS) and DT = 142 ± 5 K (PMMA)
provide evidence that even beads below glass transition temperature are still optimally
fused with the subsequently deposited material. This result seems very promising for
the manufacturing of very large parts but is also in contrast to Ajinjeru et al. [41] who
claimed the opposite when investigating the rheological properties of high-performance
polymers in LSAM. He assumed that, due to the longer cooling times, in LSAM
interlayer bonding is reduced since the substrate’s temperature decreases below Tg.
However, this particular bonding behavior was not examined and proven in this study.
Therefore, the statement needs to be reconsidered regarding the latest results.
On average a flexural strength of r = 26.3 ± 1.4 MPa for ABS and
r = 9.0 ± 1.0 MPa for PMMA is achieved, which makes ABS the favorable material
regarding external loads. Since the chosen testing method is not in accordance with
DIN EN ISO 178 [39], results of printed parts are not comparable to solid specimens.
Generally, bulk PMMA [42] possesses better mechanical properties than bulk ABS
[43]. Based on the results it is assumed that due to the layered manufacturing, certain
effects are superimposed. PMMA is a very brittle material and known for a high notch
sensitivity [44]. Its notch impact strength (Charpy) is approximately ten times smaller
compared to ABS [45]. In addition, part failure is significantly influenced by the
notch’s geometry. Smaller radii of the present notches lead to a higher sensitivity
regarding bending loads [46]. When considering the emerging interlayer connection of
LSAM geometries and the related notches between the beads, the gained results for
PMMA parts in contrast to ABS components become comprehensible.
6 Results and Prospects
Based on a developed experimental setup for large scale additive manufacturing

(LSAM), process characteristics such as bead geometry generation and cooling
behavior of LSAM parts were investigated.
Initially, the utilized setup was characterized and deposition rates of ABS and
PMMA gained. A model to generate user defined bead geometries was developed and
associated functions for ABS and PMMA derived. Providing a correlation coefficient of
R2 = 99.26% for ABS and R2 = 98.94% for PMMA, results confirm the developed
model. A software program provides printing parameters based on user defined bead
geometry data.
Layer cooling was investigated at hollow cuboids. Thermal analyses showed that
the beads surface temperature does not decline purely exponentially as assumed by Li
et al. [47]. Reasons can be found in the immense differences regarding the bead size
gained through the considered printing processes. In LSAM strands have an increased
volume, which provides a bigger heat capacity. Consequently, cooling occurs slower
and temperature differences across the bead’s diameter cannot be neglected. Instead a
distinct cooling function was necessary to describe the cooling of individual beads
analytically. This fitting function proved to be applicable to any of the tested specimens
and does agree with existing numerical cooling models as proposed by Compton et al.
[32] or Costa [18].
Hollow cylinders were printed to examine the part stability. Data of circulation time
and resulting temperature difference between consecutive beads was gathered and part
height examined. Stable components are archived with a circulation time of Dt = 36.5 s
for ABS and Dt = 27 s for PMMA which corresponds to a bead temperature difference
between preceding and subsequent layer of DT = 81 ± 1.7 K for ABS and DT = 56
± 1.4 K for PMMA, respectively. Since ABS is processed at higher temperatures a
longer cooling is required.
Interlayer bonding quality was investigated for hollow cuboids under consideration
of varying cooling times between the deposited beads. A 4-point-bending test revealed
that interlayer bonding is cooling time independent. The tested ABS and PMMA
specimens do not show considerable differences in bonding strength. Even deposited
beads which cooled below glass transition temperature are still optimally fused with the
subsequently deposited material. This suggests that the heat of extruded beads is
sufficient to initiate wetting, diffusion and randomization processes [10]. Therefore,
bonding quality is considerably increased and polymer surfaces strongly fuse together.
On average a flexural strength of r = 26.3 ± 1.4 MPa for ABS and
r = 9.0 ± 1.0 MPa for PMMA is reached. PMMA provides worse results than ABS.
Effects of manufacturing in layers as well as high brittleness, little notch impact
strength (Charpy) and particularly the notched geometry, due to bead deposition at
LSAM, seem to be superimposed and lead to premature specimen failure. Nonetheless,
a conclusive comparison between the differently manufactured specimens, relatively to
each other, was possible.
All of the presented results in this study are considered valid only under certain
requirements. Alterations in e.g. bead geometry, material selection as well as changes
of environmental conditions or the experimental setup, will strongly influence printing

results. Consequently, depicted findings might be adopted but cannot directly trans-
ferred to any kind of LSAM machine.
Further investigations are going to consider massive LSAM structures. For this
purpose, new deposition strategies will become necessary. Part accuracy and density is
expected to be dominated by even more complex cooling mechanisms.
Acknowledgment. This research and development project (funding number 03WKDE02D) is

funded by the German Federal Ministry of Education and Research (BMBF). The author is
responsible for the contents of this publication.
We would also like to thank our partner Granula Deutschland GmbH for supporting us
regarding material sourcing.
References
1. Crump, S.S.: Apparatus and method for creating three-dimensional objects Patent US
5121329 A (1989)
2. Bellini, A.: Fused deposition of ceramics. A comprehensive experimental, analytical and
computational study of material behavior, fabrication process and equipment design. Ph.D.
thesis (2002)
3. Shor, L., Gordon, J., An, Y., Guceri, S., Sun, W.: Precision extruding deposition of
polycaprolactone and composite polycaprolactone/hydroxyapatite scaffolds for tissue
engineering. In: IEEE 32nd Annual Northeast Bioengineering Conference, pp. 73–74 (2006)
4. Yuan, L., Gibson, I.: A framework for development of a fiber-composite, curved FDM
system. In: International Conference on Manufacturing Automation, pp. 93–102 (2007)
5. Cincinnati Incorporated: BAAM (Big Area Additive Manufacturing) (2015). https://
3dprintingindustry.com/news/research-institutions-collaborate-local-motors-big-area-
additive-manufacturing-composite-materials-111446/. Accessed 10 Feb 2020
6. Thermwood: LSAM (Large Scale Additive Manufacturing) (2017). http://thermwood.com/
lsam/brochures/lsam_2017_brochure/index.html#?page=8. Accessed 10 Feb 2020
7. Riddick, J.C., Haile, M.A., Wahlde, R.V., Cole, D.P., Bamiduro, O., Johnson, T.E.:
Fractographic analysis of tensile failure of acrylonitrile-butadiene-styrene fabricated by
fused deposition modeling. Additive Manuf. 11, 49–59 (2016)
8. Talagani, M.R., Dor Mohammadi, S., Dutton, R.: Numerical simulation of big area additive
manufacturing (3D printing) of a full size car. SAMPE J. 51, 27–36 (2015)
9. Wang, J., Xie, H., Weng, Z., Senthil, T., Wu, L.: A novel approach to improve mechanical
properties of parts fabricated by fused deposition modeling. Mater. Des. 105, 152–159
(2016)
10. de Gennes, P.G.: Reptation of a polymer chain in the presence of fixed obstacles. J. Chem.
Phys. 55(2), 572–579 (1971)
11. Yardimci, M.A., Güçeri, S.: Conceptual framework for the thermal process modelling of
fused deposition. Rapid Prototyping J. 2(2), 26–31 (1996)
12. Pokluda, O., Bellehumeur, C.T., Vlachopoulos, J.: Modification of Frenkel’s model for
sintering. AIChE J. 43(12), 3253–3256 (1997)
13. Cogswell, F.N.: Polymer Melt Rheology. A Guide for Industrial Practice. Woodhead Publ.
Ltd., Cambridge (1997)
14. Bellehumeur, C.T., Kontopoulou, M., Vlachopoulos, J.: The role of viscoelasticity in
polymer sintering. Rheol. Acta 37(3), 270–278 (1998)
15. Rodriguez, J.F., Thomas, J.P., Renaud, J.E.: Maximizing the strength of fused-deposition
ABS plastic parts. In: Tenth Solid Freeform Fabrication Symposium (SFF), Austin, TX,
pp. 9–11 (1999)
16. Thomas, J.P., Rodriguez, J.F.: Modeling the fracture strength between fused-deposition
extruded roads. In: Solid Freeform Fabrication Symposium (SFF), Austin, TX (2000)
17. Yan, Y., Zhang, R., Hong, G., Yuan, X.: Research on the bonding of material paths in
melted extrusion modeling. Mater. Des. 21(2), 93–99 (2000)
18. Costa, S.F.: Free form extrusion - extrusion of 3D components using complex polymeric
systems. Ph.D. thesis (2012)
19. Li, H., Wang, T., Sun, J., Yu, Z.: The effect of process parameters in fused deposition
modelling on bonding degree and mechanical properties. Rapid Prototyping J. 24(1), 80–92
(2018)
20. Onwubolu, G.C., Rayegani, F.: Characterization and optimization of mechanical properties
of ABS parts manufactured by the fused deposition modelling process. Int. J. Manuf. Eng.
2014, 13 (2014)
21. Ang, K.C., Leong, K.F., Chua, C.K., Chandrasekaran, M.: Investigation of the mechanical
properties and porosity relationships in fused deposition modelling-fabricated porous
structures. Rapid Prototyping J. 12(2), 100–105 (2006)
22. Bähr, F., Westkämper, E.: Correlations between influencing parameters and quality
properties of components produced by fused deposition modeling. Procedia CIRP 72,
1214–1219 (2018)
23. Layher, M., Hopf, A., Eckhardt, L., Bliedtner, J.: Laser beam polishing of polymers.
PhotonicsViews 16(3), 83–87 (2019)
24. Spinnie, N.K., Smith, D.E.: Large scale fused deposition modeling: the effect of processing
parameters on bead geometry. In: Solid Freeform Fabrication Symposium – An Additive
Manufacturing Conference (2016)
25. Kumar, N., Jain, P.K., Tandon, P., Pandey, P.M.: Investigation on the effects of process
parameters in CNC assisted pellet based fused layer modeling process. J. Manuf. Processes
35, 428–436 (2018)
26. Schmidt, L., Schricker, K., Bergmann, J.P., Hussenöder, F., Eiber, M.: Characterization of a
granulate-based strand deposition process in the FLM-method for definition of material-
dependent process strategies. Rapid Prototyping J. 25(1), 104–116 (2018)
27. Hertle, S., Wörz, A., Drummer, D.: Analyse der Temperaturführung in der roboterbasierten
Materialextrusion mit Polypropylen. Rapid.Tech – Fachkongress, pp. 244–257 (2019)
28. Eyercioglu, Ö., Aladag, M., Sever, S.: Temperature evaluation and bonding quality of large
scale additive manufacturing thin wall parts. Sigma J. Eng. Nat. Sci. 36(3), 645–654 (2018)
29. Richardson, S.: The die swell phenomenon. Rheol. Acta 9(2), 193–199 (1970)
30. Baird, D.G., Collias, D.I.: Polymer Processing. Principles and Design. Wiley, Hoboken
(2014)
31. Merker, G.P.: Konvektive Wärmeübertragung. Springer, Berlin and Heidelberg (1987)
32. Compton, B.G., Post, B.K., Duty, C.E., Love, L., Kunc, V.: Thermal analysis of additive
manufacturing of large-scale thermoplastic polymer composites. Addit. Manuf. 17, 77–86
(2017)
33. Ahn, S., Montero, M., Odell, D., Roundy, S., Wright, P.K.: Anisotropic material properties
of fused deposition modeling ABS. Rapid Prototyping J. 8(4), 248–257 (2002)
34. Casavola, C., Cazzato, A., Moramarco, V., Pappalettere, C.: Orthotropic mechanical
properties of fused deposition modelling parts described by classical laminate theory. Mater.
Des. 90, 453–458 (2016)
35. John, C., Kausch, M., Blase, J., Kroll, L., Holzinger, M., Reinhardt, A.: Additive Herstellung
von Kunststoffbauteilen im High-Speed-Verfahren unter Einsatz von Standardgranulat.
Konstruktion - Zeitschrift für Produktentwicklung und Ingenieur-Werkstoffe 05-2018 (2018)
36. Tseng, J.-W., Liu, C.-Y., Yen, Y.-K., Belkner, J., Bremicker, T., Liu, B.H., Sun, T.-J.,
Wang, A.-B.: Screw extrusion-based additive manufacturing of PEEK. Mater. Des. 140,
209–221 (2018)
37. Sun, Q., Rizvi, G.M., Bellehumeur, C.T., Gu, P.: Effect of processing conditions on the
bonding quality of FDM polymer filaments. Rapid Prototyping J. 14(2), 72–80 (2008)
38. [DIN EN ISO 17296-3] Additive manufacturing - Main characteristics and corresponding
test methods; December 2016. DIN Deutsches Institut für Normung e.V., Berlin, Ref. Nr.
17296-3:2016-12, Beuth Verlag GmbH, 10772 Berlin
39. [DIN EN ISO 178] Plastics - Determination of flexural properties; August 2019. DIN
Deutsches Institut für Normung e.V., Berlin, Ref. Nr. 178:2019-08, Beuth Verlag GmbH,
10772 Berlin
40. Wool, R.P., O’Connor, K.M.: A theory crack healing in polymers. J. Appl. Phys. 52(10),
5953–5963 (1981)
41. Ajinjeru, C., Kishore, V., Liu, P., Lindahl, J., Hassen, A.A., Kunc, V., Post, B., Love, L.,
Duty, C.: Determination of melt processing conditions for high performance amorphous
thermoplastics for large format additive manufacturing. Addit. Manuf. 21, 125–132 (2018)
42. PMMA: Datenblatt Altuglas VS UVT (PMMA). Altuglas International (2013)
43. ABS: Datenblatt Polylac PA-758 ABS. Chi Mei Polylac® (2019)
44. Zheng, X., Wang, H., Yan, J.: Notch strength and notch sensitivity of polymethyl
methacrylate glasses. Mater. Sci. Eng., A 349(1–2), 80–88 (2003)
45. Grellmann, W., Seidler, S.: Kunststoffprüfung. München: Carl Hanser Verlag (2015). http://
dx.doi.org/10.3139/9783446443907
46. Frick, A., Stern, C.: Praktische Kunststoffprüfung. Hanser, München (2011)
47. Li, L., Gu, P., Sun, Q., Bellehumeur, C.: Modeling of bond formation in FDM process. In:
31st, North American Manufacturing Research Conference, Hamilton, Canada, pp. 612–620
(2003)
Estimations of Interlayer Contacts in Extrusion
Additive Manufacturing Using a CFD Model
Raphaël Comminal(&) , Sina Jafarzadeh , Marcin Serdeczny ,

and Jon Spangenberg
Department of Mechanical Engineering, Technical University of Denmark,

2800 Kgs Lyngby, Denmark
rcom@mek.dtu.dk
Abstract. Numerical modeling is one of the key components in the develop-

ment of digital twins of Additive Manufacturing (AM) processes, encompassed
in the concept of Industry 4.0. Numerical simulations also have a role to play in
the improvement of contemporary AM techniques, among which material
extrusion is very popular. Computational Fluid Dynamics (CFD) models have
recently proven successful for simulating the deposition flow in extrusion AM.
Previous modeling works using CFD have been able to predict the influence of
processing parameters on the cross-sectional shape of the printed strand, as well
as the mesostructure formation resulting from their deposition and fusing. The
present study focuses on using CFD simulations to quantify the waviness of a
strand extruded on top of a previous layer with an orthogonal print direction,
which often occurs when printing a part with a rectilinear infill pattern and
alternate 0°/90° raster angles from layer to layer. The variations in the strand
width and thickness of the second layer were found to depend on the strand-to-
strand gap in the first layer. The CFD model was also used to determine the
interlayer penetration depth, which enhances mechanical performances.
Keywords: Numerical simulations Material extrusion additive

manufacturing Computational fluid dynamics Deposition flow Interlayer
contact
1 Introduction
Material extrusion is one of the prominent techniques used for Additive Manufacturing
(AM). The concept of material extrusion AM consists in depositing a continuous strand of
material along a computer-generated toolpath, to build a part layer by layer. It has
successfully been applied with a broad range of materials, including several types of
plastics (thermoplastics, thermosets and rubber), hydrogels, ceramic pastes, molten
metals, concrete, etc. Depending on the material, different extrusion technologies have
been used: filament-fed extrusion, screw extruder, syringe extruder, and hydraulic pump.
Some of the current challenges with material extrusion AM include the improvement
of the geometrical conformity (dimensional tolerance and surface quality) and mechan-
ical properties of the manufactured components. Generally, the parts manufactured by
material extrusion AM have porosities, and that reduces their mechanical performances

https://doi.org/10.1007/978-3-030-54334-1_17
242 R. Comminal et al.
when compared to the equivalent parts manufactured by conventional methods (e.g.

injection molding) [1–3]. The porosity directly depends on the mesostructure formed by
the successive deposition of the strands, which is influenced by the printing parameters
[4]. Moreover, the mechanical properties of the component are intrinsically linked to the
inter- and intra-layer bonds in the mesostructure [5–8].
A voxelized model of material deposition that is solely based on volume conser-
vation was developed by Gleadall et al. [9] to predict the micro-architectures of 3D
printed scaffolds. A detailed flow analysis of material spreading during extrusion AM
was proposed in [10]. Computational Fluid Dynamics (CFD) simulations have been
used in [11–22] to model the deposition of strands. Du et al. [11] used CFD to analyze
the thermal field and morphology of a 3D printed thin wall of ABS. Comminal et al.
[12] employed an isothermal Newtonian creeping flow model to investigate the
influence of the printing parameters on the cross-sectional shape of the strand deposited
on a planar surface. The numerical predictions of the model were later validated
through experimental measurements by Serdeczny et al. [13]. In a complementary
study, the same team [14] used a power-law shear-shinning model and showed that the
numerical results were insensitive to the rheological model, confirming that the
Newtonian creeping flow assumption was a valid modeling hypothesis. The corner
rounding and swelling at turns was also modeled by Comminal et al. [15, 16]. Xia et al.
[17–20] simulated the deposition of multiple strands and layers, with a non-isothermal
and non-Newtonian fluid flow model. In their model, the deposited strands have the
possibility to deform under the pressure applied by the layers deposited on top of them.
Several cases were simulated, included the multi-layer deposition of parallel and
orthogonal strands. In contrast, Serdeczny et al. [21, 22] modeled the multi-layer
deposition of parallel strands with an isothermal Newtonian creeping flow model.
Moreover, the mesostructure formation was simulated by successive simulations where
the previously deposited strands were treated as solid bodies excluded from the flow
domain. The cases of aligned and skewed layers of parallel strands were considered.
These simulations estimated the porosity, surface roughness, and inter-/intra- layer
contact areas, and agreed qualitatively well with the experimental results.
The current work presents novel numerical analysis of the strand deposition on top
of a previous layer with an orthogonal print direction. This situation frequently occurs
when printing a part with a rectilinear infill pattern and alternate 0°/90° raster angles
from layer to layer. The numerical model is applied to investigate the influence of the
distance between the strands of the first layer on the waviness of the second layer, as
well as the interlayer penetration depth.
2 Numerical Methods
This work uses the same modeling assumptions as in the previous works of Comminal
et al. [12, 16] and Serdeczny et al. [13, 22], but the numerical solutions are obtained
with a different numerical scheme, using the CFD software FLOW-3D® [23]. The flow
is modeled as an isothermal Newtonian fluid flow, which is governed by the continuity
equation and the momentum conservation equation:
Estimations of Interlayer Contacts in Extrusion Additive Manufacturing 243
ru¼0 ð1Þ

@u
q þ u ru ¼ rp þ gr2 u þ qg ð2Þ
@t
where u is the velocity vector field, p is the pressure, q is the material density, g is the
dynamic viscosity, g is the gravity acceleration vector, and t is the time. The following
material parameters, q = 1000 kg/m3 and g = 1000 Pas were used. As discussed in
[12, 13], the extrusion flow typically has a very low Reynolds number Re * 10−3,
corresponding to the creeping flow regime. This means that the flow is virtually
insensitive to the material density and viscosity – as long as Re ¼ qUD=g 1, where
U and D are the characteristic speed and characteristic length of the flow, respectively
(e.g. the extrusion speed and the nozzle diameter). For that reason, the temperature-
variation of the viscosity is neglected, and the flow is solved as isothermal.
The material deposition was simulated inside a build volume of 5.6 1.1 2.0
mm, along the X-Y-Z directions. The build volume contains the extrusion nozzle and the
substrate, which are both included as solid objects into the computational domain, as
shown in Fig. 1. The geometry of the extrusion nozzle is a cylindrical tube with an inner
diameter D = 0.4 mm and a wall thickness of 0.25 mm. In addition, the top boundary of
the computational domain was covered by an upper plate with a hole coinciding with the
nozzle orifice. The substrate comprises a planar build surface and one layer of parallel
strands, previously printed on that surface. Six different strand configurations of the first
layer were used in the simulations. The different geometries of the first layer were
generated by duplicating the geometry of a single strand, obtained from a previous
numerical simulation of strand deposition [13] (case g/D = 0.8 and V/U = 1.0). All the
substrate geometries have the same layer thickness T1 = 0.30 mm (obtained for a nozzle
height H1 = 0.32 mm), but different strand repetition distances S, varying from 0.4 to
1.6 mm. The 3D geometries of the first layer were extruded through the computational
domain along the Y direction. For the deposition simulations of the second layer, the
nozzle was positioned at the distance H2 = 2H1 = 0.64 mm above the build plane.
During the simulations, the extrusion nozzle moved along the X direction inside the
computational domain (thus a 90° raster angle with the first layer), with a constant
travel speed V = 20 mm/s, while extruding material (in the Z direction) with a steady
volumetric flux U = 20 mm/s. The same values of the nozzle travel speed and extru-
sion rate were used in [13] to simulate the deposition of the strand constituting the first
layer. Note that the upper plate (that moves together with the nozzle) is used for
restricting the material inlet to the nozzle, as the inflow boundary condition was applied
to the entire top surface of the computational domain. The computational cost was
reduced by placing a symmetry boundary condition on the transversal X-Z plane
passing through the axis of the nozzle. Thus, the actual computational domain included
only half of the nozzle geometry, and was 5.6 0.55 2.0 mm. A continuative outlet
boundary condition was applied at the other boundaries of the computational domain.
Moreover, the no-slip boundary condition was applied to the surface of all solid
objects.
Fig. 1. Geometry of the CFD simulations. Computational domain delimited by the cyan box.
The front plane is a symmetry boundary.
The computational domain was meshed with a Cartesian grid that has a uniform
grid size of 20 lm. The governing equations of the flow were discretized with the finite
volume method. The partial blockage of the grid cells by the solid objects (i.e. the
moving nozzle and substrate) was taken into account with the immersed boundary
method. In addition, the free surface of the extruded material was captured with the
volume-of-fluid method. The simulations used a single-phase solver that only resolved
the flow of extruded material, and neglected the surrounding air (in contrast with the
previous works [12–16, 21, 22]). An adaptive time step size was used and the total
simulation time was 0.27 s.
3 Numerical Results
Snapshots of the extruded strand at different time points of the simulations are shown in
Fig. 2, where S = 1.0 mm. The final shapes of the extruded strands in the other sim-
ulations with different strand repetition distances S are also represented in Fig. 3. The
strand of the second layer presents a wavy shape, which contrasts with the strands of
the first layer (deposited on a planar surface) that have a uniform width W1 and
thickness T1; see Fig. 4. The width W2 and thickness T2 of the second layer strand are
functions of the position along the strand, as shown in Fig. 4. The strand thickness T2,
corresponds the distance between the top of the first layer (represented by the dashed
red line in Fig. 4b) and the highest point of the free surface of the second layer strand.
The variations in the strand size has the same periodicity as the strand repetition
distance. The minimum strand width W2,min is located at the middle of the gaps
between the strands of the first layer, while the maximum strand width W2,max occurs
on top of the first layer strands, where the nozzle-to-substrate distance is the lowest.
When moving along the second layer strand in the same direction as the printing head,
the lowest strand thickness T2,min occurs at the end of every strands in the first layer
(i.e. the beginning of the strand-to-strand gap). Moreover, the largest strand thickness
T2,max is located at the beginning of each strand in the first layer (i.e. the end of the
strand-to-strand gap).
t = 0.00 s t = 0.02 s t = 0.04 s
t = 0.06 s t = 0.08 s t = 0.10 s
t = 0.12 s t = 0.14 s t = 0.16 s
t = 0.18 s t = 0.20 s t = 0.22 s
t = 0.24 s t = 0.26 s t = 0.27 s
Fig. 2. Free surface of second layer strand, printed on top of a first layer with a strand repetition
distance S = 1.0 mm. Snapshot every 0.02 s and at the final simulation time (t = 0.27 s).
The maximum and minimum values of the strand width and thickness of the second
layer are plotted as a function of the strand repetition distance of the first layer in Fig. 5.
To avoid the end effects of extrusion starting and finishing, the dimensions of the strand
were only measured within the central region of the simulation domain, represented by
the black dashed lines in Fig. 4. For ease of comparison, the dimensions of the first
layer strands are also represented by the blue dashed lines in Fig. 5. It should be noted
S = 0.4 mm (S/W1 = 0.72)
S = 0.48 mm (S/W1 = 0.85)
S = 0.6 mm (S/W1 = 1.07)
S = 1.0 mm (S/W1 = 1.78)
S = 1.3 mm (S/W1 = 2.31)
S = 1.6 mm (S/W1 = 2.84)
Fig. 3. Side and top views of the strand at the final simulation time, for different strand
repetition distance S of the first layer. Axis units are in mm.
Fig. 4. Definitions of the strand dimensions. Top view (a) and side view (b). The red dashed line
represents the top of the first layer. The black dashed lines show the measuring region.
that the strand dimensions reported in Fig. 5 are normalized by the nozzle diameter
(D = 0.4 mm), while the strand repetition distances are normalized by the first layer
strand width (W1 = 0.56 mm). The results show that the waviness (i.e. the difference
between the maximum and minimum dimensions) of the second layer strand remains
limited when S/W1 < 1. This is explained by the fact that the overlap in the first layer
strands makes the first layer closer to a flat surface. However, when S/W1 > 1, the
waviness tends to increase with the strand-to-strand gap (i.e. the strand repetition
distance) in the first layer.
Fig. 5. Maximum and minimum normalized strand width W2/D and thickness T2/D as a function
of the normalized strand repetition distance S/W1 of the first layer.
The interlayer penetration depth d (defined as the distance between the top surface
of the first layer and the lowest point of the second layer, see Fig. 4b) is reported in
Fig. 6. The interlayer penetration depth is normalized by the thickness of the first layer
(T1 = 0.30 mm), and according to the numerical results, the interlayer penetration
depth increases almost linearly with the strand repetition distance. Moreover, the
maximum value d/T1 = 1 is attained when the second layer completely collapses into
the first layer and touches the build plate. This is visible in the case where S/W1 = 2.84.
Fig. 6. Normalized interlayer penetration depth d/T1 as a function of the normalized strand
repetition distance S/W1 of the first layer.
This work presents a novel numerical simulation of the strand deposition on top of a
previously printed layer of parallel strands with an orthogonal printing direction. The
numerical model is based on the isothermal Newtonian creeping flow assumptions,
which have proven sufficient in our previous works [12, 13, 16, 22], to investigate the
influence of the printing parameters on the strand morphology and the mesostructure of
3D printed parts.
The mechanical strength of 3D printed parts depends on the contact area between
the extruded strands, as well as the bonding strength of those contacts. The current
CFD model does not predict the bonding strength of the interlayer contacts, which, in
the case of thermoplastics, is a temperature-driven molecular diffusion process. Nev-
ertheless, the isothermal CFD simulations provide information about interlayer con-
tacts. The numerical results show that the maximum strand width of the second layer
and the interlayer penetration depth – which both promote the interlayer contact
between individual strands – increase with the strand repetition distance of the first
layer. At the same time, however, the number of interlayer contacts per unit length of
printed strand decreases with the strand repetition distance, which needs to be taken
into account to evaluate the total interlayer contact area.
The main message of these numerical results is that printing layers with alternate
printing directions from layer to layer can result in non-uniform strands. Future works
should validate the simulated strand deformations with experiments. Finally, this type
of numerical simulations could be useful for improving slicer software and exploring
novel deposition strategies, toward more reliable and efficient material extrusion AM
techniques.
Acknowledgments. The authors would like to acknowledge the support of the Danish Council
for Independent Research (DFF), Technology and Production Sciences (FTP) (Contract
No. 7017-00128).
References
1. Ahn, S.H., Montero, M., Odell, D., Roundy, S., Wright, P.K.: Anisotropic material
properties of fused deposition modeling ABS. Rapid Prototyping J. 8(4), 248–257 (2002).
https://doi.org/10.1108/13552540210441166
2. Koch, C., Van Hulle, L., Rudolph, N.: Investigation of mechanical anisotropy of the fused
filament fabrication process via customized tool path generation. Addit. Manuf. 16, 138–145
(2017). https://doi.org/10.1016/j.addma.2017.06.003
3. Dawoud, M., Taha, I., Ebeid, S.J.: Mechanical behaviour of ABS: an experimental study
using FDM and injection moulding techniques. J. Manuf. Processes 21, 39–45 (2016).
https://doi.org/10.1016/j.jmapro.2015.11.002
4. Rodriguez, J.F., Thomas, J.P., Renaud, J.E.: Characterization of the mesostructure of fused-
deposition acrylonitrile-butadiene-styrene materials. Rapid Prototyping J. 6(3), 175–186
(2000). https://doi.org/10.1108/13552540010337056
5. Rodríguez, J.F., Thomas, J.P., Renaud, J.E.: Mechanical behavior of acrylonitrile butadiene
styrene fused deposition materials modeling. Rapid Prototyping J. 9(4), 219–230 (2003).
https://doi.org/10.1108/13552540310489604
6. Bellini, A., Güçeri, S.: Mechanical characterization of parts fabricated using fused deposition
modeling. Rapid Prototyping J. 9(4), 252–264 (2003). https://doi.org/10.1108/135525403
10489631
7. Ang, K.C., Leong, K.F., Chua, C.K., Chandrasekaran, M.: Investigation of the mechanical
properties and porosity relationships in fused deposition modelling-fabricated porous
structures. Rapid Prototyping J. 12(2), 100–105 (2006). https://doi.org/10.1108/135525406
10652447
8. Abbott, A.C., Tandon, G.P., Bradford, R.L., Koerner, H., Baur, J.W.: Process-structure-
property effects on ABS bond strength in fused filament fabrication. Addit. Manuf. 19, 29–
38 (2018). https://doi.org/10.1016/j.addma.2017.11.002
9. Gleadall, A., Ashcroft, I., Segal, J.: VOLCO: a predictive model for 3D printed microarchi-
tecture. Addit. Manuf. 21, 605–618 (2018). https://doi.org/10.1016/j.addma.2018.04.004
10. Agassant, J.F., Pigeonneau, F., Sardo, L., Vincent, M.: Flow analysis of the polymer
spreading during extrusion additive manufacturing. Addit. Manuf. 29, 100794 (2019). https://
doi.org/10.1016/j.addma.2019.100794
11. Du, J., Wei, Z., Wang, X., Wang, J., Chen, Z.: An improved fused deposition modeling
process for forming large-size thin-walled parts. J. Mater. Process. Technol. 234, 332–341
(2016). https://doi.org/10.1016/j.jmatprotec.2016.04.005
12. Comminal, R., Serdeczny, M.P., Pedersen, D.B., Spangenberg, J.: Numerical modeling of
the strand deposition flow in extrusion-based additive manufacturing. Addit. Manuf. 20,
68–76 (2018). https://doi.org/10.1016/j.addma.2017.12.013
13. Serdeczny, M.P., Comminal, R., Pedersen, D.B., Spangenberg, J.: Experimental validation
of a numerical model for the strand shape in material extrusion additive manufacturing.
Addit. Manuf. 24, 145–153 (2018). https://doi.org/10.1016/j.addma.2018.09.022
14. Serdeczny, M.P., Comminal, R., Pedersen, D.B., Spangenberg, J.: Numerical study of the
impact of shear thinning behaviour on the strand deposition flow in the extrusion-based
additive manufacturing. In: Billington, D., Leach, R.K., Phillips, D., Riemer, O., Savio, E.
(eds.) Proceedings of the 18th International Conference of the European Society for
Precision Engineering and Nanotechnology, pp. 283–284. The European Society for
Precision Engineering and Nanotechnology (2018)
the material deposition and contouring precision in fused deposition modeling. In:
Proceedings of the 29th Annual International Solid Freeform Fabrication Symposium,
pp. 1855–1864. Laboratory for Freeform Fabrication, Austin, TX (2018)
16. Comminal, R., Serdeczny, M.P., Pedersen, D.B., Spangenberg, J.: Motion planning and
numerical simulation of material deposition at corners in extrusion additive manufacturing.
Addit. Manuf. 29, 100753 (2019). https://doi.org/10.1016/j.addma.2019.06.005
17. Xia, H., Lu, J., Dabiri, S., Tryggvason, G.: Fully resolved numerical simulations of fused
deposition modeling. Part I: fluid flow. Rapid Prototyping J. 24(2), 463–476 (2018). https://
doi.org/10.1108/RPJ-12-2016-0217
18. Xia, H., Lu, J., Tryggvason, G.: Fully resolved numerical simulations of fused deposition
modeling Part II–solidification, residual stresses and modeling of the nozzle. Rapid
Prototyping J. 24(6), 973–987 (2018). https://doi.org/10.1108/RPJ-11-2017-0233
19. Xia, H., Lu, J., Tryggvason, G.: Simulations of fused filament fabrication using a front
tracking method. Int. J. Heat Mass Transf. 138, 1310–1319 (2019). https://doi.org/10.1016/j.
ijheatmasstransfer.2019.04.132
20. Xia, H., Lu, J., Tryggvason, G.: A numerical study of the effect of viscoelastic stresses in
fused filament fabrication. Comput. Methods Appl. Mech. Eng. 346, 242–259 (2019).
https://doi.org/10.1016/j.cma.2018.11.031
21. Serdeczny, M.P., Comminal, R., Pedersen, D.B., Spangenberg, J.: Numerical prediction of
the porosity of parts fabricated with fused deposition modeling. In: Proceedings of the 29th
Annual International Solid Freeform Fabrication Symposium, pp. 1849–1854. Laboratory
for Freeform Fabrication, Austin (2018)
22. Serdeczny, M.P., Comminal, R., Pedersen, D.B., Spangenberg, J.: Numerical simulations of
the mesostructure formation in material extrusion additive manufacturing. Addit. Manuf. 28,
23. Flow Science, Inc., Santa Fe, NM, FLOW-3D® Version 12.0 (2019)
Influence of Fibers on the Flow Through
the Hot-End in Material Extrusion Additive
Manufacturing
Marcin Serdeczny, Raphaël Comminal, David Bue Pedersen,

and Jon Spangenberg(&)
Department of Mechanical Engineering, Technical University of Denmark,

2800 Kgs Lyngby, Denmark
josp@mek.dtu.dk
Abstract. Advancements in material extrusion additive manufacturing are

driven by the need of fast production of high-quality parts. A key innovation was
the introduction of short fibers into the built filament, which substantially
improve the strength of manufactured components. However, the presence of
fibers also affects the viscosity and thermal conductivity of the filament, which
are important parameters for the extrusion flow and thus the maximum printing
speed. In this paper, we numerically study the effect of fibers on the polymer
flow and pressure drop inside the nozzle, which determines the maximum
extrusion rate. The thermoplastic polymer flow is simulated with a non-
isothermal computational fluid dynamics model and the inclusion of fibers is
treated with a continuum approach. The simulations are performed for ABS
polymer with short carbon fibers, however, other thermoplastic systems with
short fibers (e.g. glass, wood, or nylon) can be integrated into the model. The
model provides a virtual window into the process and illustrates that while fibers
increase the viscosity of the filament and the pressure drop, they also improve
the thermal conductivity and lead to faster melting of the polymer, which has an
opposite effect on the pressure drop. Finally, the model quantifies the relation-
ship between the fiber volume fraction and the maximum extrusion rate.
Keywords: Fused deposition modeling Fiber reinforcement Computational

fluid dynamics
1 Introduction
Material extrusion Additive Manufacturing (AM), or Fused Deposition Modeling

(FDM), is currently undergoing a transition from being a prototyping technique to
producing functional parts with industrial-grade mechanical strength. One of the key
drivers behind this progress is the inclusion of fiber reinforcement inside the filament
material, which improves the mechanical performance of the produced components [1].
Short carbon fibers are a popular type of reinforcement used in material extrusion AM
[2, 3]. In addition to increasing the mechanical strength, the carbon fibers also improve
the thermal conductivity and reduce the coefficient of thermal expansion of the com-
pound material, which leads to less thermal warpage and lower residual stresses [4, 5].

https://doi.org/10.1007/978-3-030-54334-1_18
252 M. Serdeczny et al.
Functional parts made by carbon reinforced material extrusion AM can potentially be

applied in medical equipment [6], aerospace [7–9], automotive [10] and consumer
products [11].
Experimental studies have been conducted to quantify the improvement in tensile
strength and elastic modulus of 3D printed components, due to the presence of short
carbon fibers [1, 3]. Computational Fluid Dynamics (CFD) simulations have been
widely used to model the polymer flow during material extrusion AM [12–18], among
which the works by Heller et al. [19, 20] focused on the prediction of the fiber
orientation in the extrusion and deposition flow. Thermo-mechanical simulations were
conducted to predict the deformations and residual stresses in the finished
carbon/polymer composites [5, 21]. However, there is a limited research on the
influence of fiber reinforcement on the melting and flow of polymer composite inside
the print head. The addition of fibers alters the rheology and the thermal properties of
the fluid, which are both highly important for the flow behavior.
Heat transfer effects and rheology are kesy aspects to increasing the build rates in
material extrusion AM [22]. Mackay et al. [23] determined the failure velocity that was
defined as the filament feeding rate at which the heater was not able to melt efficiently
the polymer to perform successful extrusion. Nienhaus et al. [24] used a load cell to
measure the dependence of the filament feeding force on the feeding rate. They found
that the feeding force becomes unstable before reaching the failure velocity. This
observation was confirmed by Serdeczny et al. [25], where a model to estimate the
maximum feeding rate before the extrusion becomes unstable was presented. It was
hypothesized that the maximum feeding rate is limited by the heat transfer and the
ability to melt the filament completely before it reaches the nozzle contraction. A more
detailed explanation of the extrusion instability was given in a follow-up work [26]
presenting numerical simulations of the flow inside the print head. The feeding force
instability was hypothesized to be due to a fluctuation of the melt zone position inside
the flow channel.
In this work, we modify the CFD model presented in [26] to investigate the
influence of fiber additives on the flow inside the printing head and the melting of
filament. The inclusion of fibers is simulated with a continuum approach by modifying
the viscosity and thermal conductivity of the fluid. A parametric study on the influence
of the fiber content and feeding rate is performed.
2 Methodology
2.1 Physics of the Numerical Model

The numerical model simulates the polymer flow inside the hot-end channel of a 3D
printer. A sketch representing the hot-end is shown in Fig. 1a. The essential elements
of the hot-end are: the nozzle that extrudes the material, the liquefier that melts the
polymer, the heat break that isolates thermally the liquefier from the upper parts of the
print head, and the heat sink, which releases the excess heat to the environment. In the
numerical model, the internal channel of the hot-end was implemented, as shown in
Fig. 1b. The filament entered the channel with a constant velocity V, termed the
Influence of Fibers on the Flow Through the Hot-End 253
feeding rate and the inlet temperature TIN , equal to the room temperature. The material
remained solid until its temperature reached the glass temperature of the polymer Tg .
The channel was subdivided into five sections with dimensions as indicated in the
Fig. 1b. It is noted that the filament had a slightly smaller diameter DF than the barrel
section DB . It was assumed that the wall of the channel had a steady temperature, which
varies along the channel, as shown in Fig. 1c. The geometry of the channel was based
on the E3D v6 hot-end and its dimensions are given in Appendix A.
Fig. 1. a) Sketch of the hot-end. b) Internal channel of the hot-end. c) Wall temperature of the
channel.
The flow and temperature of material is governed by the following equations [27]
ru¼0 ð1Þ
Du
q ¼ rp þ r g/ S þ qga ð2Þ
Dt
DT
qCv ¼ r k/ rT ð3Þ
Dt
S ¼ ru þ ruT ð4Þ
where u is the velocity vector, q is the density, t is the time, p is the pressure, S is the
strain rate tensor, ga is the gravity acceleration vector, Cv is the specific heat capacity,
T is the temperature, k/ is the effective thermal conductivity, and g/ is the apparent
dynamic viscosity. Thus, it was assumed that the flow is incompressible and the fluid
had a constant density. The fiber suspension in the polymer was approximated with a
continuum approach and the flow of only one phase was simulated. However, the free
surface of the fluid was resolved and the remaining part of the domain was filled with
void that had a constant pressure and temperature. The presence of fibers was
accounted in the formula of the apparent fluid viscosity and thermal conductivity,
which are discussed in detail in Sects. 2.3 and 2.4. The influence of fibers on the fluid
density and specific heat was assumed negligible for the flow inside the print head, and
their values are given in Table 1 together with the remaining simulation parameters.
Table 1. Simulation parameters.

Parameter Symbol Unit Value
Fluid density q kg=m 3 1060
Specific heat capacity Cv J=ðkg KÞ 1750

Inlet temperature TIN C 25

Glass transition temperature Tg C 105

Liquefier temperature TL C 200
Feeding rates V mm=min 40, 60, 80, …, 280
2.2 Implementation Details

The numerical simulations were performed using the commercial CFD software Flow-
3D [28]. The implicit coupled pressure-velocity solver was used to solve the governing
equations for the material flow. The energy equation was solved with an implicit time-
discretization scheme that is second order accurate in space. The problem was assumed
to be axisymmetric and a 2D mesh was constructed as represented in Fig. 2. The
average cell size was 20 lm and the mesh was refined near the channel walls. At the
top of the domain, the velocity inlet boundary condition was prescribed. The contin-
uative boundary, which specifies all spatial derivatives of the fluid properties equal to
zero, was used at the outlet located outside of the nozzle orifice. The wall of the
channel had a prescribed temperature, as shown in Fig. 1c. The heat flux between the
channel wall and the fluid was calculated as
q_ w ¼ hðTw T Þ ð5Þ
where h is the heat transfer coefficient at the channel wall while Tw is the channel wall
temperature. The value of h is unknown and was found through comparison with
experimental measurements of the filament feeding force during extrusion of neat ABS.
The measurements and their results are discussed in detail in [25]. It wasfound that the
simulation represents the measurements most closely, when h 600 W ðm2 KÞ in the

fully developed flow region, and h 8000 W ðm2 KÞ close to the inlet of the channel
where a recirculation of the flow occurs. The simulations were run until a steady state
was reached, or for 60 s, when the flow was inherently unsteady. A more in-depth
discussion about the implementation of the numerical model can be found in [26].
Fig. 2. 2D axisymmetric cylindrical mesh used in the study. Left: outline of the entire mesh.
Right: close-up on the contraction section.
2.3 Rheological Properties of the Polymer with Suspended Fibers

The polymer used in this study is ABS, which is one of the most commonly used
thermoplastics in material extrusion AM. The apparent dynamic viscosity of a neat
ABS can be described with a temperature dependent power law
T n
gðc_ ; T Þ ¼ K c_ n1 a ð6Þ
Tr T
where K is the consistency index, n is the power law index, c_ is the shear rate
magnitude, T is the polymer temperature, Tr is the reference temperature, and aT is the
time temperature shift factor. The temperature dependency of aT is modeled with the
Williams–Landel–Ferry (WLF) equation

C1 ðT Tr Þ
aT ðT Þ ¼ exp ð7Þ
C 2 þ ð T Tr Þ
where C1 and C2 are material constants. The inclusion of fibers changes the apparent
viscosity of the fluid. One of the commonly used simple model to capture this
dependence is the equation by Maron and Pierce [29, 30]

/ 2
gr ¼ 1 ð8Þ
A
where gr is the relative viscosity, which is defined as the ratio of filled system viscosity
g/ to the neat polymer viscosity g at the same shear stress (not the same shear rate); /
is the fiber volume fraction; A is the parameter related to the aspect ratio of the fibers
[29]. Taking into account the definition of gr and combining it with Eq. (6), the
viscosity of the filled system can be written as
T n n
g/ ¼ K c_ n1
/ a g ð9Þ
Tr T r
The derivation of Eq. (9) is given in Appendix B. The strain rate magnitude relates
to the velocity vector in the numerical model as
rffiffiffiffiffiffiffiffiffiffiffiffiffi
1
c_ / ¼ S:S ð10Þ
2
The rheological material parameters used in this study are given in Table 2. The oscil-
latory measurements of neat ABS were done for temperatures ranging from 125 °C to
250 °C and a master curve was created at Tr = 200 °C. The measurement points and the
power law fit are shown in Fig. 3. Moreover, we have estimated the dynamic viscosity for
ABS filled with fibers at different volume fractions (the weight fractions are also shown
for reference). We have assumed that the fibers have an average aspect ratio b ¼ lf =df
equal to 23, where lf is the fiber length and df is the fiber diameter. For this aspect ratio, the
parameter A in Eq. (8) equals 0.25, as given in [29]. It is assumed that the material behaves
as a Generalized Newtonian Fluid (GNF), meaning that the Trouton ratio (the ratio of
elongational viscosity to shear viscosity) is equal to 3. A large fiber content may invalidate
this assumption, necessitating the use of a viscoelastic constitutive model [30]. In such
case, the complexity and computational cost of the numerical simulations would be
greatly increased.
Table 2. Rheological material parameters.

Parameter Symbol Unit Value Source

Reference temperature Tr C 200 –
Consistency index K Pasn 30104 Measured
Power law index n – 0.24 Measured
WLF constant 1 C1 – 8.97 Measured

WLF constant 2 C2 C 155.2 Measured
Average fiber aspect ratio b – 23 [29, 31]
Parameter in Eq. (8) A – 0.25 [29]
Fiber density qf kg/m3 1800 [32]
Fig. 3. The master curve for neat ABS polymer at Tr = 200 °C and estimated dynamic apparent
viscosity for fiber suspension with varying fiber volume fraction.
2.4 Thermal Conductivity of the Reinforced Filaments

The reinforced filaments used in material extrusion AM are themselves produced by
screw extrusion of a matrix (thermoplastic) and fibers compound mixed together. Due
to the nature of extrusion compounding, most of the fibers align with the axis of the
filament [33–35]. However, there will always exist some deviation of the fiber orien-
tations from the extrusion direction, which can be quantified by a Fiber Orientation
Distribution (FOD). Many models to predict the thermal conductivity of fiber rein-
forced composites were developed, among which we chose to use the approach by Fu
and Mai [36], which accounts for the influence of FOD on the thermal conductivity.
A sketch representing the method is shown in Fig. 4. The effective thermal con-
ductivity of a composite consisting of misaligned fibers with known FOD can be
calculated as [36]
Z 180
k/ ¼ kðhÞ gðhÞdh ð11Þ
0
where kðhÞ is the thermal conductivity of a unidirectional fiber composite with its
principle axis (‘1’ longitudinal and ‘2’ transverse, cf. Fig. 4b) oriented at an angle h to
the heat transfer direction x, and gðhÞ is the FOD function, that is approximated with a
normal probability density function, as suggested by experimental observations [34]
!
1 ðh hmean Þ2
gðhÞ ¼ pffiffiffiffiffiffi exp ð12Þ
rh 2p 2r2h
where hmean is the mean fiber orientation angle (that is aligned with the filament axis,
i.e. extrusion direction), and rh is the standard deviation of the orientation distribution.
The thermal conductivity of a unidirectional fiber composite at an angle h can be

calculated as a superposition of its longitudinal thermal conductivity k1 (parallel to the
fibers) and transverse thermal conductivity k2 (perpendicular to the fibers) (cf. Fig. 4b)
kðhÞ ¼ k1 cos2 h þ k2 sin2 h ð13Þ
where the longitudinal and transverse thermal conductivities are calculated using
Halpin-Tsai [36] equation, respectively
1 þ 2bl1 /
k1 ¼ km ð14Þ
1 l1 /
1 þ 2l2 /
k2 ¼ km ð15Þ
1 l2 /
kf 1 =km 1
l1 ¼ ð16Þ
kf 1 =km þ 2b
kf 2 =km 1
l2 ¼ ð17Þ
kf 2 =km þ 2
where km is the thermal conductivity of the polymer matrix, kf 1 and kf 2 are thermal
conductivities of a single fiber in parallel and perpendicular direction to the fiber axis.
Fig. 4. Sketch representing the thermal conductivity model. a) Filament with misaligned fibers
with predominant direction z. b) Decomposition into unidirectional fiber composites with fibers
aligned at different directions. c) Definition of fiber angle h.
During the flow through the hot-end, the polymer is heated due to radial conduction
from the channel walls while the heat flux due to the axial conduction is usually
negligible, as compared to the axial convection. Moreover, as the problem is
axisymmetric, k/ can be assumed to represent the thermal conductivity of the filament
in all directions. The parameters used in the estimation of the effective thermal con-
ductivity are given in Table 3. The calculated dependence of the effective thermal
conductivity on the fiber volume fraction is shown in Fig. 5. For reference, the thermal
conductivity of the filament with perfectly aligned fibers (rh ¼ 0 Þ is also shown. It is
observed that the FOD increases the thermal conductivity and its influence becomes
larger with higher volume fractions. In the end, we note that compared to the approach
of Fu and Mai [36], we have not accounted for the fiber length distribution, however its
influence on the thermal conductivity was found to be negligible for the present case.
Table 3. Parameters used in the estimation of the effective thermal conductivity of the fiber
reinforced filament.
Parameter Symbol Unit Value Source
ABS thermal conductivity km W=ðm KÞ 0.33 [37]
Fiber thermal conductivity (parallel) kf 1 W=ðm KÞ 10 [38]
Fiber thermal conductivity (transverse) kf 2 W=ðm KÞ 10 [38]

Mean fiber orientation angle hmean 90 Based on [34]

Standard deviation of FOD (Eq. (12)) rh 15 Based on [34]
Fig. 5. Influence of the fiber volume fraction and the standard deviation of the fiber orientation
on the thermal conductivity in the radial direction of the filament.
3 Results
3.1 Comparison with Experimental Results
First, the numerical simulation is validated with experimental measurements of the
filament feeding force when a neat ABS filament (/ ¼ 0Þ is extruded. The filament
feeding force is the force required to push the filament through the hot-end system and
it was measured in [25, 39] using a load cell, where a detailed description can be found.
The comparison of the simulated and measured feeding force as a function of the
feeding rate can be seen in Fig. 6. The error bars show the standard deviation of the
feeding force oscillations during the measurement/simulation of individual data points.
It can be seen that for low feeding rates, no oscillations in the feeding force are
observed and the extrusion is in the stable regime. The temperature profile within the
hot-end for the stable regime can be seen in Fig. 7. At some critical feeding rate, called
the maximum feeding rate, the force curve changes slope and oscillations appear,
which indicates unstable extrusion. It is noted that these instabilities are distinct from
the shark skin and melt fracture extrusion instability, as they occur at a lower frequency
and are not related to change in the appearance of the extrudate surface [25]. A similar
behavior of the feeding force was measured in [24]. It can be seen that the numerical
results are in a good agreement with the measurements and the maximum feeding rate
threshold between the stable and unstable regime is captured by the simulations. The
unstable regime is related to the insufficient heat transfer rate at large feeding rates.
Looking at the evolution of the temperature profile for the unstable regime (cf. Fig. 7),
it can be seen that as the heat transfer rate is too small, the solid filament penetrates the
channel causing fluctuations in the position of the melt line. The magnitude of the
feeding force fluctuations are larger in the simulations than in the experiments. Note
that in the simulations, the solid filament below the glass temperature is modelled as a
rigid solid, neglecting viscoelastic effects that may damp the pressure variations.
Finally, we note that our definition of the maximum feeding rate is based on the
extrusion instability coming from the insufficient heat transfer. There are other factors
limiting the maximum feeding rate in material extrusion AM, such as the buckling of
the filament and the maximum feeding force that can be applied on the filament by the
extruder; however their analysis was outside of the scope of this work.
Fig. 6. Comparison of the measured and simulated filament feeding force for the extrusion of
neat ABS filament at TL = 200 °C.
V 40 mm/min 280 mm/min

t 15 s 3s 5s 6s 8s 24 s
Comment Stable Unstable
Fig. 7. Evolution of the temperature profiles within the hot-end in stable and unstable extrusion
regimes.
3.2 Influence of the Carbon Fibers on the Flow Through the Hot-End
Figure 8 compares the simulated feeding force as a function of the filament feeding rate
for different carbon fiber volume fractions. It can be seen that the increase in the fiber
content leads to a higher feeding force, which was expected as the fibers increase the
material viscosity. However, at large feeding rates, the feeding force for / ¼ 5% and
/ ¼ 10% is slightly lower than for a neat polymer ð/ ¼ 0%Þ. This is explained by the
improved thermal conductivity of the filament, which enhances melting of the polymer.
The trend is better visible in Fig. 9, where the simulated feeding force was plotted as a
function of the fiber volume fraction at different feeding rates. For low and moderate
feeding rates, the force increases monotonically with the fiber volume fraction. How-
ever, at higher feeding rate, a minimum in the feeding force is observed for / from
around 5% to 10%, which indicates that the increased thermal conductivity of the
filament can be beneficial for reducing the pressure inside the hot-end.
Except the change in the magnitude of the feeding force, we observe that for high
fiber volume fractions, the extrusion remains stable at higher feeding rates. This can be
seen in Fig. 8 and 9 by the error bars that indicate the force fluctuations, and in
Fig. 10a, where the simulated feeding force is shown as a function of time for different
fiber volume fractions. The oscillations of the feeding force and the melt zone position
determine the maximum feeding rate before the extrusion becomes unstable, as dis-
cussed in Sect. 3.1. The dependence of the maximum feeding rate on the fiber volume
fraction, as suggested by the simulations, is shown in Fig. 10b. It is observed that the
maximum feeding rate increases linearly as a function of the carbon fiber volume
fraction, due to improved effective thermal conductivity of the filament.
Fig. 8. Feeding force as a function of the filament feeding rate for varying fiber volume fraction.
Simulated for DC = 0.4 mm and TL = 200 °C. The error bars indicate the force oscillations in
time.
Fig. 9. Feeding force as a function of the fiber volume fraction for varying feeding rates.
Simulated for DC = 0.4 mm and TL = 200 °C. The error bars indicate the force oscillations in
time.
Fig. 10. a) Feeding force as a function of time for different fiber volume fractions, at a feeding
rate V ¼ 200 mm=min; b) Simulated maximum feeding force (as defined in Sect. 3.1) as a
function of fiber volume fraction.
4 Conclusions
This work is one of the first attempts to model the influence of fiber reinforcement on
the flow inside the hot-end during material extrusion AM. A CFD model was used to
simulate the non-isothermal flow of a power law fluid. The presence of fibers was
accounted for by modifying the apparent dynamic viscosity and effective thermal
conductivity of the filament. In this work, one of the most commonly used systems was
investigated, that is ABS matrix with carbon fibers. However, the model is suitable to
simulate fibers made of different materials such as glass, carbon, wood, or nylon. The
presented method can be used for short fibers where the aspect ratio (fiber length/fiber
diameter) is moderately low (below 25). For long fibers, the elongational viscosity may
become more important than shear viscosity. Hence, the Generalized Newtonian Fluid
approach may no longer be valid, and more accurate representation of the fluid rhe-
ology is required. The results show that the presence of carbon fibers leads to a higher
apparent viscosity of the fluid, which, in general, requires a larger feeding force.
However, at higher feeding rates, the presence of carbon fibers can be beneficial and
decrease the feeding force, because of the increased thermal conductivity that promotes
melting of the filament–which ultimately decreases the apparent viscosity. The pre-
sented approach could be used to optimize the design of filaments for high printing
speed by utilizing fiber reinforcements or particle fillers with high thermal conductivity.
The future work shall include experimental measurements in order to confirm the
numerical findings. Specifically, the feeding force could be measured for filaments
containing varying fiber fractions.
Acknowledgements. The authors would like to acknowledge the support of the Danish Council
for Independent Research (DFF) | Technology and Production Sciences (FTP) (Contract
No. 7017-00128).
Appendix A
See Table 4.
Table 4. Dimensions of the E3D v6 hot-end channel [40].

Dimension Symbol Unit
Value
Nozzle diameter DC mm 0.40
Barrel diameter DB mm 2.00
Filament diameter DF mm 1.75
Capillary length LC mm 0.60

Contraction angle a 60
Liquefier section length LL mm 15.5
Heat break section length LHB mm 2.00
Heat sink section length LHS mm 1.00
Appendix B
The relative viscosity gr is the ratio of g/ and g, defined at the same shear stress
gU ðsÞ s=_c/ c_
gr ¼ ¼ ¼
gð s Þ s=_c c_ /
The dynamic viscosity of the filled system can be written as
T n
g/ ¼ ggr ¼ K c_ n1 a g
Tr T r
where c_ ¼ gr c_ / . Thus, the composite dynamic viscosity is
n1 T n T n n
g/ ¼ K gr c_ / a g ¼ K c_ n1
/ a g
Tr T r Tr T r
References
1. Brenken, B., Barocio, E., Favaloro, A., Kunc, V., Pipes, R.B.: Fused filament fabrication of
fiber-reinforced polymers: a review. Addit. Manuf. 21, 1–16 (2018). https://doi.org/10.1016/
j.addma.2018.01.002
2. Hofstätter, T., Pedersen, D.B., Tosello, G., Hansen, H.N.: State-of-the-art of fiber-reinforced
polymers in additive manufacturing technologies. J. Reinf. Plast. Compos. 36, 1061–1073
(2017). https://doi.org/10.1177/0731684417695648
3. Tekinalp, H.L., Kunc, V., Velez-Garcia, G.M., Duty, C.E., Love, L.J., Naskar, A.K., Blue,
C.A., Ozcan, S.: Highly oriented carbon fiber-polymer composites via additive manufac-
turing. Compos. Sci. Technol. 105, 144–150 (2014). https://doi.org/10.1016/j.compscitech.
2014.10.009
4. Love, L.J., Kunc, V., Rios, O., Duty, C.E., Elliott, A.M., Post, B.K., Smith, R.J., Blue, C.A.:
The importance of carbon fiber to polymer additive manufacturing. J. Mater. Res. 29, 1893–
1898 (2014). https://doi.org/10.1557/jmr.2014.212
5. Brenken, B., Barocio, E., Favaloro, A., Kunc, V., Pipes, R.B.: Development and validation
of extrusion deposition additive manufacturing process simulations. Addit. Manuf. 25, 218–
6. Liaw, C.Y., Guvendiren, M.: Current and emerging applications of 3D printing in medicine.
Biofabrication 9 (2017). https://doi.org/10.1088/1758-5090/aa7279
7. Wohlers Associates. Wohlers report 2017 (2017)
8. Goh, G.D., Agarwala, S., Goh, G.L., Dikshit, V., Sing, S.L., Yeong, W.Y.: Additive
manufacturing in unmanned aerial vehicles (UAVs): challenges and potential. Aerosp. Sci.
Technol. 63, 140–151 (2017). https://doi.org/10.1016/j.ast.2016.12.019
9. Widden, M., Gunn, K.: Design build test of model aerofoils for engineering education using
FDM. Virtual Phys. Prototyping 5, 189–194 (2010). https://doi.org/10.1080/17452759.2010.
528841
10. Turner, B.N., Gold, S.A.: A review of melt extrusion additive manufacturing processes: II.
Materials, dimensional accuracy, and surface roughness. Rapid Prototyping J 21, 250–261
(2015). https://doi.org/10.1108/RPJ-02-2013-0017
11. Šljivić, M., Fragassa, C., Pavlović, A., Kraišnik, M., Ilić, J., Stanojević, M.: Additive
manufacturing of functional parts based on material extrusion technology. Contemp. Mater.
7, 178–184 (2016)
the strand deposition flow in extrusion-based additive manufacturing. Addit. Manuf. 20, 68–
13. Serdeczny, M.P., Comminal, R., Pedersen, D.B., Spangenberg, J.: Experimental validation
of a numerical model for the strand shape in material extrusion additive manufacturing.
Addit. Manuf. 24, 145–153 (2018). https://doi.org/10.1016/j.addma.2018.09.022
14. Serdeczny, M.P., Comminal, R., Pedersen, D.B., Spangenberg, J.: Numerical simulations of
the mesostructure formation in material extrusion additive manufacturing. Addit. Manuf. 28,
15. Comminal, R., Serdeczny, M.P., Pedersen, D.B., Spangenberg, J.: Motion planning and
numerical simulation of material deposition at corners in extrusion additive manufacturing.
Addit. Manuf. 29, 100753 (2019). https://doi.org/10.1016/j.addma.2019.06.005
16. Xia, H., Lu, J., Tryggvason, G.: Fully resolved numerical simulations of fused deposition
modeling. Part II-solidification, residual stresses, and modeling of the nozzle. Rapid
Prototyping J. 24, 973–987 (2018). https://doi.org/10.1108/RPJ-11-2017-0233
17. Xia, H., Lu, J., Dabiri, S., Tryggvason, G.: Fully resolved numerical simulations of fused
deposition modeling. Part I: fluid flow. Rapid Prototyping J. 24, 463–476 (2018). https://doi.
org/10.1108/RPJ-12-2016-0217
18. Pigeonneau, F.D.F., Vincent, X., Agassant, J.-F.: Heating and flow computations of an
amorphous polymer in the liquefier of a material extrusion 3D printer. Addit. Manuf. 29,
100820 (2019). https://doi.org/10.1016/j.addma.2019.100820
19. Heller, B.P., Smith, D.E., Jack, D.A.: Effects of extrudate swell and nozzle geometry on fiber
orientation in fused filament fabrication nozzle flow. Addit. Manuf. 12, 252–264 (2016).
https://doi.org/10.1016/j.addma.2016.06.005
20. Heller, B.P., Smith, D.E., Jack, D.A.: Planar deposition flow modeling of fiber filled
composites in large area additive manufacturing. Addit. Manuf. 25, 227–238 (2019). https://
doi.org/10.1016/j.addma.2018.10.031
21. Favaloro, A.J., Brenken, B., Barocio, E., Pipes, R.B.: Simulation of polymeric composites
additive manufacturing using Abaqus. In: Proceedings 2017 Science in the Age of
Experience, pp. 103–114 (2017)
22. Go, J., Schiffres, S.N., Stevens, A.G., Hart, A.J.: Rate limits of additive manufacturing by
fused filament fabrication and guidelines for high-throughput system design. Addit. Manuf.
16, 1–11 (2017). https://doi.org/10.1016/j.addma.2017.03.007
23. Mackay, M.E., Swain, Z.R., Banbury, C.R., Phan, D.D., Edwards, D.A.: The performance of
the hot end in a plasticating 3D printer. J. Rheol. (N Y N Y) 61, 229–236 (2017). https://doi.
org/10.1122/1.4973852
24. Nienhaus, V., Smith, K., Spiehl, D., Dörsam, E.: Investigations on nozzle geometry in fused
filament fabrication. Addit. Manuf. 28, 711–718 (2019). https://doi.org/10.1016/j.addma.
2019.06.019
25. Serdeczny, M.P., Comminal, R., Pedersen, D.B., Spangenberg, J.: Experimental and
analytical study of the polymer melt flow through the hot-end in material extrusion additive
manufacturing. Addit. Manuf. 32, 100997 (2020). https://doi.org/10.1016/j.addma.2019.
100997
26. Serdeczny, M.P., Comminal, R., Mollah, M.T., Pedersen, D.B., Spangenberg, J.: Numerical
modeling of the polymer flow in the hot-end of material extrusion additive manufacturing.
Addit. Manuf. (2020, to appear in an issue). https://doi.org/10.1016/j.addma.2020.101454
27. Bird, R.B., Armstrong, R.C., Hassager, O.: Dynamics of Polymeric Liquids. Wiley,
Hoboken (1987)
28. FLOW-3D | We solve the world’s toughest CFD problems. https://www.flow3d.com/.
Accessed 4 Sept 2019
29. Kitano, T., Kataoka, T., Shirota, T.: An empirical equation of the relative viscosity of
polymer melts filled with various inorganic fillers. Rheol. Acta 20, 207–209 (1981). https://
doi.org/10.1007/BF01513064
30. Metzner, A.B.: Rheology of Suspensions in Polymeric Liquids. J. Rheol. (N Y N Y) 29,
739–775 (1985). https://doi.org/10.1122/1.549808
31. Ning, F., Cong, W., Qiu, J., Wei, J., Wang, S.: Additive manufacturing of carbon fiber
reinforced thermoplastic composites using fused deposition modeling. Compos. Part B Eng.
80, 369–378 (2015). https://doi.org/10.1016/j.compositesb.2015.06.013
32. Jeon, Y.P., Alway-Cooper, R., Morales, M., Ogale, A.A.: Carbon Fibers, 2nd edn. Elsevier,
Amsterdam (2013)
33. Fu, S.Y., Lauke, B., Mäder, E., Yue, C.Y., Hu, X., Mai, Y.W.: Hybrid effects on tensile
properties of hybrid short-glass-fiber-and short-carbon-fiber-reinforced polypropylene com-
posites. J. Mater. Sci. 36, 1243–1250 (2001). https://doi.org/10.1023/A:1004802530253
34. Vaxman, A., Narkis, M., Siegmann, A., Kenig, S.: Short-fiber-reinforced thermoplastics.
Part III: effect of fiber length on rheological properties and fiber orientation. Polym. Compos.
10, 454–462 (1989). https://doi.org/10.1002/pc.750100610
35. Guo, R., Azaiez, J., Bellehumeur, C.: Rheology of fiber filled polymer melts: role of fiber-
fiber interactions and polymer-fiber coupling. Polym. Eng. Sci. 45, 385–399 (2005). https://
doi.org/10.1002/pen.20285
36. Fu, S., Mai, Y.: Thermal conductivity of misaligned short-fiber- reinforced polymer
composites. J. Appl. Polym. Sci. 88, 1497–1505 (2003). https://doi.org/10.5772/66336
37. Osswald, T.A., Puentes, J., Kattinger, J.: Fused filament fabrication melting model. Addit.
Manuf. 22, 51–59 (2018). https://doi.org/10.1016/j.addma.2018.04.030
38. Carbon UD HS 194 gsm. www.tohotenax-eu.com. Accessed 31 Jan 2020
39. Serdeczny, M.P., Comminal, R., Pedersen, D.B., Spangenberg, J.: Experimental analysis of
the filament feeding force in material extrusion additive manufacturing. In: Proceedings of
the Joint Special Interest Group meeting between euspen and ASPE Advancing Precision in
Additive Manufacturing, pp. 141–145 (2019)
40. E3D-Online Documentation - E3D-Online. https://e3d-online.dozuki.com/. Accessed 3 Jul
2019
Deploying Artificial Intelligence
for Component-Scale Multi-physical Field
Simulation of Metal Additive
Manufacturing
Ehsan Hosseini1(B) , P. Gh. Ghanbari1,2 , F. Keller1,3 , S. Marelli4 ,

and Edoardo Mazza1,2
1
Empa, Swiss Federal Laboratories for Materials Science and Technology,
Dübendorf, Switzerland
ehsan.hosseini@empa.ch
2
Department of Mechanical and Process Engineering,
Institute for Mechanical Systems, ETH Zürich, Zürich, Switzerland
3
Department of Mathematics, ETH Zürich, Zürich, Switzerland
4
Department of Civil, Environmental and Geomatic Engineering,
Institute of Structural Engineering, ETH Zürich, Zürich, Switzerland
Abstract. Additive manufacturing is evolving as a prominent aspect of

the fourth industrial revolution and offers unprecedented design freedom
for creating solid objects from 3D digital models. Effective contribution
of metal additive manufacturing (MAM) to realization of Industry 4.0
requires improved mechanical integrity of the AM parts at a lower cost.
This study aims to exploit the potential of multiscale modelling and arti-
ficial intelligence (machine learning and metamodelling) to realise reliable
component-scale multi-physical field simulations for MAM. The thorough
quantitative understanding of the various physical phenomena involved in
MAM processes would enable a systematic optimisation of process condi-
tions for achieving ‘first-time-right’ high-quality production.
Keywords: Metal additive manufacturing · Multi-physics simulation ·

Surrogate modelling
1 Introduction
Commerical additive manufacturing (AM) first emerged with the stereolithogra-
phy (SL) process from 3D Systems in 1987 [1]. Arguably, the first reported metal
AM parts were made from copper, tin and Pb–Sn solder powders through a selec-
tive laser sintering (SLS) process in 1990 [2]. German company EOS (Electro-
Optical Systems), Sandia National Laboratories, Swedish company Arcam-AB
and Fraunhofer institute in germany later developed metal additive manufactur-
ing (MAM) techniques such as direct metal laser sintering (DMLS), laser engi-
neered net shaping (LENS), electron beam melting (EBM) and selective laser
https://doi.org/10.1007/978-3-030-54334-1_19
Component-Scale Multi-physical Field Simulation of MAM 269
melting (SLM) [1,3]. MAM is currently a rapidly growing industry with revenue
estimates projected to surpass $7’150M by 2026 in comparison to $1’030M in
2016 [4].
In the early phase of its 30-year history, AM had been primarily employed
for the fabrication of conceptual and functional prototypes then known as Rapid
Prototyping (RP) [5]. Recent innovations in processes and materials, and extend-
ing the applicability of the technology to metals, have transformed it to a manu-
facturing technology for the production of end-use parts [5–7]. Important factors
hindering even faster growth and widespread industrial application of MAM are
the high cost and uncertain mechanical performance of the AM builds [8]. Due to
such problems, the business cases for the employment of this technology in many
industries are rather marginal, particularly for the fabrication of components
operating under severe loading conditions. Resolving the described challenges
would thus lead to further progress of digital manufacturing and accelerate the
realisation of the foreseen 4th industrial revolution. Due to the lack of a better
alternative, trial-and-error strategies, that are costly in terms of both time and
money, are often adopted to optimise the MAM process conditions [8–10]. More-
over, the quality of products so designed does not ultimately meet the expected
requirement(s) in most safety-sensitive load-bearing applications. This ongoing
research aims to contribute to this endeavour using advanced multi-physical field
simulations.
2 Multi-physical Field Simulation for MAM
A thorough quantitative understanding of the MAM process requires insight

from different modelling areas, such as thermal, mechanical, metallurgical, fluid,
and thermodynamics simulations (Table 1). The extremely high computational
cost of such multi-field simulations has been addressed in previous research by
either decreasing the simulation domain size and/or by assuming gross simplifi-
cations. Instead, this study proposes to exploit the power of multiscale modelling
and artificial intelligence (machine learning and metamodelling) to realise reli-
able component-scale simulations for MAM. The underlying idea originates from
the incremental build characteristic of MAM and how the involved phenomena
are repetitive and deterministic in nature. The variation in the characteristics of
the deposited material increments does not originate from a significant change
Table 1. Relevant simulation fields for metal additive manufacturing
Simulation type Predictions Simulation techniques

Thermal simulation Thermal profiles, hot spots Continuum finite element
Thermomechanical simulation Residual stress, distortion Continuum finite element
Thermal-fluid simulation Defect, surface roughness Computational fluid dynamics
Microstructural simulation Grain structure, texture Cellular automata, phase field
Mechanical prediction Deformation, failure Crystal plasticity finite element
270 E. Hosseini et al.
in the involved physics but only from a (modest) change in the imposed bound-
ary conditions. Therefore, component-scale simulations of MAM process can
be considered as solving a large number of nominally similar small-scale mod-
els, each with slightly different boundary conditions. The computational cost of
such repetitive simulations can be significantly reduced by employing a dedicated
machine learning toolbox. Systematic design of experiments would first involve
a limited number of small-scale simulations to identify the sensitivity of the out-
put parameters to the boundary conditions. Results from these simulations can
then be exploited to train and cross-validate a computationally cheap surrogate
(meta)model without compromising the process efficacy. This leads to a sig-
nificant cost reduction for component-scale multi-physical field simulations and
allows their direct implementation in construction of numerical optimisations for
the fabrication of AM components with desired properties and performance.
It should be noted that this study uses metamodels as a very efficient machine
learning regression tool. Devised from the field of uncertainty quantification
(UQ), surrogate (meta) models are replacements of expensive computational
models which can be used in analyses that require a large number of model
evaluations, such as uncertainty propagation and design optimization, in a rea-
sonable time [11]. In contrast to the typical machine learning applications which
deal with a large amount of (noisy) data, a metamodel approximates a com-
putational model by analysing the outcomes of a small number of simulation
runs.
This ongoing research intends to develop a flexible simulation platform which,
for a given set of MAM process conditions, trains and validates a metamodel on
the basis of few small-scale high-fidelity simulations that incorporate thermal,
solid mechanics, microstructural, thermodynamics, and fluid dynamics compu-
tations (Fig. 1). For a given geometry and process parameter set, the plat-
form will use a specific metamodel for predicting a range of relevant param-
eters/characteristics within the MAM part – such as temperature and resid-
ual stress profiles, part distortion, spatial distribution of defects and porosity,
and microstructural parameters. The prediction results will be subsequently
exploited by a crystal plasticity finite element framework for assessing the
mechanical integrity of MAM components. Notably, various high-end experimen-
tal facilities will be employed for designing dedicated experiments to 1) under-
stand the involved physical phenomena, 2) derive relevant numerical models
based on observations, and 3) ultimately evaluate the reliability and effective-
ness of the proposed simulation strategy.
Employing a metamodel as an alternative for the conventional simulation
techniques is associated with some levels of uncertainty for the obtained predic-
tions. This uncertainty mainly depends on the robustness and representativeness
of the training data set. Primarily, the uncertainty of metamodel predictions are
quantifiable, and thus an uncertainty index would be included in the applica-
tion of the metamodel for simulation of a component. For conditions where the
involved uncertainty is unacceptable (i.e. high levels of uncertainty at critical
Rough component-scale continuum thermal simulation
Multi-physical fields metamodel

Thermal metamodel
Temporal temperature profiles
Mech metamodel CA metamodel PF metamodel CFD metamodel
Inherent-strain and Grain size, shape and Precipitate and Porosity and surface
dislocation density texture parameters segregation state roughness parameters
Component-scale crystal plasticity finite element simulation
Residual stress and distortion Risk of crack formation Stress and strain distribution Macroscopic deformation response
Fig. 1. Metamodelling for component-scale multi-physical field simulations of MAM

(CA: cellular automata, PF: phase field, CFD: computational fluid dynamics)
locations of the assessed component), additional small-scale simulations for rep-

resenting the situation at those locations, can be designed and added to the
training pool to reduce the uncertainty level.
This paper presents the preliminary results for application of the described
methodology for thermal simulation of selective laser melting (SLM) process.
3 Continuum Thermal Modelling

Temperature profile evolutions are perhaps the most critical information required
for any type of MAM simulations. The temperature profile experienced by
the molten material during MAM is very different from other manufacturing
techniques, and to a great extent controls the state of defects, residual stress,
microstructure, and properties of the product. Finite element (FE) analysis
solves the below equation for calculation of the time-dependent temperature
fields over the discretised domain:
ρcp Ṫ − ∇.(k∇T ) = qvol (1)
where ρ, cp , and k are respectively density, specific heat, and thermal conduc-
tivity of the material; and qvol is the volumetric heat generation, e.g. due to
the laser beam exposure. Contrary to the conventional continuum thermal anal-
yses, for simulation of SLM, 1) employment of element activation or the quiet
element strategy is required to simulate incremental material addition to the
model, and 2) each material point can take one of the states of powder, solid
or liquid. Implementation of these SLM specific considerations into commercial
FE packages are often executed through user-defined subroutines. Ref [12] com-
prehensively describes the details for continuum thermal simulation of SLM in
ABAQUS FE package. In Fig. 2 an overview of the model setup for 2D thermal
simulations is shown.
Fig. 2. Finite element model setup for 2D continuum thermal modelling of the SLM
process [12].
High-fidelity thermal analysis of the SLM process requires a high level of dis-
cretisation in time and space (e.g. element size of 10–20 µm and time-increments
of a few μs) which hinders the full-scale analysis of a real-size component due
to extremely high computational costs. However, this level of discretization is
only required near the melt-pool where temperature gradients are steep and the
thermal evolution rates are high. Therefore, the proposed approach breaks down
the problem into two scales of local and global, where the global calculations
employ a coarse mesh and larger time increments to determine the temperature
profiles in regions far away from the melt-pool, while fine-mesh local simulations
adaptively follow the beam location and estimate the thermal profiles at the
vicinity of the melt-pool. This particular type of modelling technique combines
the results of the local simulations and the global solution, thus providing reliable
temperature predictions at a significantly reduced computational cost. Figure 3
shows the various steps that are taken in this multiscale modelling approach.
For concept verification, a computationally expensive finite element model
that solves the whole simulation region using small time increments and a fine
mesh is defined as the ‘reference’ solution. From comparing the multiscale results
with the reference model, an assessment of maintained accuracy and the reduc-
tion in computational costs can be made. For instance, in the case of 2D simu-
lation of a 2 × 2 mm2 SLM deposition, the nodal temperatures in the multiscale
approach remain nearly identical to those from the reference solution as observed
Rough component-scale continuum thermal simulation

Geometry of Vtotal
Continuum thermal model Temporal temperature
Material model parameters
Solver: ABAQUS AM Plug-in profile for Vtotal ‒ iVlocal
Laser parameters
Local small-scale continuum thermal simulations

Laser parameters Geometry of Vlocal
Initial temperature for iVlocal ∩ i-1Vlocal
from previous local solution
Initial temperature for iVlocal ‒ i-1Vlocal Local continuum thermal model Temporal temperature
from rough global solution Solver: ABAQUS, FORTRAN & Python profile for iVlocal
Boundary temperatures for iSlocal
from rough global solution Material model parameters
Integration of local and global solutions

Temporal temperature profile for Vtotal
Temporal temperature Temporal temperature
profile for Vtotal ‒ iVlocal profile for iVlocal
Mech Sim CA Sim PF Sim CFD Sim
Fig. 3. Overview of the proposed strategy for continuum thermal modelling of MAM
process
in Fig. 4a. Meanwhile, the computational costs from adopting the multiscale
approach are reduced by roughly an order of magnitude as shown in Fig. 4b.
Extension of the adaptive multiscale idea to strongly reduce the computational
costs of larger simulation domains is ongoing.
Although such a strategy significantly decreases the computational costs, it
will still neither be affordable for the current industry nor for consideration in
numerical process optimisation exercises. The effectiveness of employing machine
learning-based approaches for further moderating the cost of MAM numerical
analyses have been discussed in the following.
Employment of the described multiscale approach for thermal analysis of
an SLM part involves thousands of local simulations. Due to the repetitive
and deterministic nature of the heat transfer problem, it is possible to replace
the numerous computationally expensive local finite element simulations with a
much cheaper surrogate (meta)model.
A number of local thermal analyses were performed to generate the temper-
ature profiles in the vicinity of the melt-pool for a number of different build con-
ditions. The generated data were then exploited to develop a cheap metamodel-
based on sparse polynomial chaos expansion (PCE) [13] combined with principal
component analysis (PCA) [14] for dimensional reduction. The effectiveness of
surrogate modelling based on the PCE method has been demonstrated in a
number of studies e.g. [15–17]. PCE surrogate modelling is typically used to
replace computationally expensive models with uncertain input parameters and
to efficiently compute statistics of the uncertain model outputs. The polynomial
chaos expansion theory is established on the basis of polynomials orthogonal
3,000
2,000 Reference Reference Approach
2,000
Multiscale Multiscale Approach
Temperature (◦ C)
5×5
1,500
Temperature (◦ C)
1,500 2,000
1,000
Time (hours)
4×4
1,000 500
40.101 40.1015 40.102 40.1025 1,000

500 Time (s)
3×3
2.4×2.4
1.6×1.6
2×2
1×1
0
35 40 45 50 55 60 65 0
0 50 100 150 200 250 300
Time (s)
Number of Nodes (×1000)
(a) (b)
Fig. 4. (a) Comparison of temperature evolution for identical coordinates between

the multiscale and reference simulation. (b) Comparison of calculation time for 2D
simulation of the SLM process of various squares. The labels show the size of the
model in mm2 . The horizontal axis is based on the number of nodes in the reference
simulation [12].
with respect to the distribution of the probabilistic model input. The mapping
between input and output data that should be substituted by the metamodel
is seen as a stochastic model on these data spaces and it is then expanded in
terms of the orthogonal polynomial basis functions. Depending on the accuracy
needed, the infinite-dimensional expansion is truncated after a certain number
of elements. During the training phase, the coefficients of the PCE for the given
polynomial basis will be determined using a set of training data points with
corresponding model responses. Because of the high dimensionality of the data
spaces (temperature data on two dimensional grid), principal component analy-
sis is used to reduce the dimensions of both the input and the result space. The
PCE metamodel is then used to model the dependence of the reduced result
data on the reduced input data.
The metamodel is built to replace the local small-scale finite element simu-
lations of the multiscale approach presented before. Thus, the results obtained
through application of the metamodel can directly be integrated in the existing
workflow of thermal modelling of the MAM process.
The validation of the metamodel involves evaluation of its ability to predict
the temperature evolution for input settings which were not revealed during
the training phase. As an example, the SLM process of a 1 × 1 mm2 block is
considered. The multiscale simulation of this setup consists of 165 local models
out of which, only ten are used to train the surrogate model. Afterwards, the
temperature evolution for the remaining local models is predicted using the
surrogate model and a value for the prediction error (with the multiscale finite
element simulation as the reference) can be computed. In Fig. 5, comparison
of the nodal temperature between the surrogate model and multiscale finite
FE
2,000 Emulated
Temperature (◦ C)
1,500
1,000
500
0 0.2 0.4 0.6 0.8 1

Time (ms)
(a) (b)
Fig. 5. Comparison of temperature predictions from surrogate modelling and local

FE thermal analysis, a) peak temperature temperature distributions b) temperature
evolutions.
element simulation at a point in the center of the deposition domain is shown.

Only minor differences between the two temperature evolution curves are visible.
In this exemplary setting, the surrogate model results show good agreement with
the finite element simulation results while being produced at a much lower cost.
This study aims to use the potential of multiscale finite element simulations in
conjunction with machine learning algorithms to create a framework for reliable
component-scale simulations of MAM. First, since the highly involved multi-
physics phenomena is present only in the vicinity of the melt pool, numerical
simulations can be separated into different scales to reduce computational costs
and maintain the small degree of discretization required for capturing the phys-
ical developments during the SLM process. Second, the deterministic nature of
the phenomena involved in MAM allows the repetitive small-scale local finite ele-
ment simulations to be replaced by a surrogate model after generating enough
data through FE simulation, for training the algorithm.
The proposed strategy has been employed for thermal simulation of SLM
process which demonstrated its efficiency for reliable prediction of temperature
profiles at significantly reduced computational costs. Extension of the proposed
modelling strategy for radically improving the efficiency of other types of MAM
simulations are under investigation.
References
1. Wohlers, T., Gornet, T.: History of additive manufacturing. Wohlers Rep.
24(2014), 118 (2014)
2. Manriquez-Frayre, J.A., Bourell, D.L.: Selective laser sintering of binary metallic

powder (1990)
3. Sames, W.J., List, F.A., Pannala, S., Dehoff, R.R., Babu, S.S.: The metallurgy
and processing science of metal additive manufacturing. Int. Mater. Rev. 61(5),
315–360 (2016)
4. Barroqueiro, B., Andrade-Campos, A., Valente, R., Neto, V.: Metal additive man-
ufacturing cycle in aerospace industry: a comprehensive review. J. Manuf. Mater.
Process. 3(3), 52 (2019)
5. Mellor, S., Hao, L., Zhang, D.: Additive manufacturing: a framework for imple-
mentation. Int. J. Prod. Econ. 149, 194–201 (2014)
6. Hague, R., Mansour, S., Saleh, N.: Material and design considerations for rapid
manufacturing. Int. J. Prod. Res. 42(22), 4691–4708 (2004)
7. Huang, Y., Leu, M.C., Mazumder, J., Donmez, A.: Additive manufacturing: current
state, future potential, gaps and needs, and recommendations. J. Manuf. Sci. Eng.
137(1), 014001 (2015)
8. Mani, M., Lane, B., Donmez, A., Feng, S., Moylan, S., Fesperman, R.: Measurement
science needs for real-time control of additive manufacturing powder bed fusion
processes. Technical report NIST IR 8036, National Institute of Standards and
Technology, February 2015
9. Arrizubieta, J.I., Lamikiz, A., Cortina, M., Ukar, E., Alberdi, A.: Hardness, grain-
size and porosity formation prediction on the Laser Metal Deposition of AISI 304
stainless steel. Int. J. Mach. Tools Manuf 135, 53–64 (2018)
10. Markl, M., Körner, C.: Multiscale modeling of powder bed-based additive manu-
facturing. Annu. Rev. Mater. Res. 46(1), 93–123 (2016)
11. Schöbi, R.: Surrogate models for uncertainty quantification in the context of impre-
cise probability modelling. IBK Bericht 505 (2019)
12. Gh Ghanbari, P., Mazza, E., Hosseini, E.: Adaptive local-global multiscale app-
roach for thermal simulation of the selective laser melting process (Under Review)
13. Blatman, G., Sudret, B.: Adaptive sparse polynomial chaos expansion based on
least angle regression. J. Comput. Phys. 230(6), 2345–2367 (2011)
14. Jolliffe, I.T.: Principal Component Analysis. Springer, Heidelberg (2002)
15. Collaboration, E., Knabenhans, M., Stadel, J., Marelli, S., Potter, D., Teyssier,
R., Legrand, L., Schneider, A., Sudret, B., Blot, L., et al.: Euclid preparation: II.
The euclidemulator–a tool to compute the cosmology dependence of the nonlinear
matter power spectrum. Mon. Not. R. Astron. Soc. 484(4), 5509–5529 (2019)
16. Torre, E., Marelli, S., Embrechts, P., Sudret, B.: Data-driven polynomial chaos
expansion for machine learning regression. J. Comput. Phys. 388, 601–623 (2019)
17. Nagel, J.B., Rieckermann, J., Sudret, B.: Principal component analysis and sparse
polynomial chaos expansions for global sensitivity analysis and model calibration:
application to urban drainage simulation. Reliab. Eng. Syst. Saf. 195, 106737
(2020)
Prototyping and Testing
Experimental Investigation of Filament
Behaviour in Material Extrusion Additive
Manufacturing
Mark Golab(&) , Sam Massey , and James Moultrie
Department of Engineering, University of Cambridge, Cambridge, UK

mrg62@cam.ac.uk
Abstract. Fused Deposition Modelling (FDM) is a relatively low-cost additive

manufacturing process that is commonly characterised by poor dimensional
accuracy and precision. Prior work indicates that dimensional errors may be the
result of fundamental issues relating to how the filament behaves during
deposition and not due to limitations in FDM machine design. Some studies
have examined the cross-sectional area of single strands, highlighting the
dimensional variation, but there have been no studies examining these changes
in other orthogonal planes and along the whole strand length. This study seeks
to characterise the behavior of deposited material, to explore the impact of
different process parameters on single and multiple strand geometries. Single
layer strands of Poly-Lactic Acid (PLA) are deposited through a 0.4 mm
diameter nozzle at varying nozzle gap heights, filament volumetric velocities,
and print head velocities. Image analysis is used to quantify the width, height,
cross sectional shape, and qualitatively analyse strand geometries. Basic print
repeatability is shown to be in the order of 75 lm at a strand length of 20 mm.
By altering processing parameters, strand width variation of 2.6 mm has been
demonstrated. Variations at the start and end of each strand are more significant
than in the middle. Variations in shape becomes much more pronounced when
printing multiple strands on top of each other. These results demonstrate the
importance of careful process parameter selection on part quality, which to date
has not been sufficiently well acknowledged.
Keywords: Fused Deposition Modelling Fused filament fabrication

Additive Manufacturing Filament behaviour Dimensional accuracy
1 Introduction
Fused deposition modelling (FDM) also known as fused filament fabrication (FFF) is a
Material Extrusion (ME) Additive Manufacturing (AM) technique, whereby a feed-
stock thermoplastic filament is extruded through a liquefier via two pinch rollers and is
deposited in a layer-by-layer fashion on a build plate [1]. FDM has the ability to create
complex geometries that would otherwise be impossible to fabricate via conventional
means, thus making it an attractive alternative. However, for this ME technique to be
widely adopted as a preferred manufacturing process, printed parts must satisfy the
dimensional/geometric tolerances and quality requirements of accepted standards [2].

https://doi.org/10.1007/978-3-030-54334-1_20
280 M. Golab et al.
It is widely acknowledged that ME AM can yield errors such as dimensional accuracy

and precision, poor surface finish, and poor inter-layer bond formation of adjoining
strands, thus affecting overall part quality [1, 3, 4]. In order to elucidate these phe-
nomena and mitigate these errors, multiple studies have investigated as to how the
processing parameters generate these defects. Anitha et al. analysed the effects of
processing variables on the surface roughness of FDM parts, it was concluded that
layer thickness was the most significant contributor [5]. This staircase effect as a result
of layer thickness was also observed by Pérez [6] and Pandey et al. [7]. Nancharaiah
et al. [8] conducted an experimental approach using Taguchi’s design of experiments
and ANOVA analysis, which determined that layer thickness and “road-width” affected
both part accuracy and surface quality significantly. Empirically, it is recognised that a
smaller layer thickness can improve surface quality of printed parts, however, this adds
to build time and is not the only parameter that drives surface quality.
Build orientation is also known to significantly affect part quality along with
additional characteristics [9–11]. Ollison and Berisso [12], studied the effects of build
orientations 0 , 45 , and 90 on the cylindricity of printed parts, the results indicated
that the 90 orientation had a statistically significant reduction on cylindricity error.
Vasudevarao et al. [13] also investigated the effects of part orientation on part quality
and concluded that a better surface finish can be achieved at angles orthogonal to the
build envelope. An array of studies have implemented theoretical approaches to pre-
dicting part quality from resultant processing parameters, however, these are typically
semi-empirical and are only comparable with specific printers nor holistically trans-
ferrable [14]. Sheoran and Kumar [15] reviewed the most recent research on process
parameter optimisation for improving part quality in FDM. It was established that the
aforementioned parameter, layer thickness, has a significant influence on dimensional
accuracy. Additionally, a reduced print speed, low extrusion speed, and a reduced
number of outer shells has been proven to increase dimensional accuracy.
Specifically, layer thickness has been the most explored parameter for influencing
part quality, however, there is significant scope to experimentally investigate other
processing parameters such as, road-width, filament volumetric velocities, print head
velocities, and outer shells along with the characterisation of strand formation accuracy.
Two recent studies [16, 17] have demonstrated how the processing conditions have
affected cross sectional morphologies of deposited filament, however, single strands
were only analysed in one orthogonal plane. Serdeczny et al. [16] validate their
numerical model via experimental characterisation of single strand cross sectional
measurements. As nozzle gap height is reduced the cross-sectional strand shape per-
pendicular to the print direction transitions from an oval shape to a cuboid with
rounded corners. A nozzle gap height of 0.4 mm and print head velocity of 1.0 mm/s
produced a rectangular strand at a width of 2.4 mm and height 0.4 mm. Whereas, a
nozzle gap height of 1.2 mm and print head velocity of 1.0 mm/s produced a rectan-
gular strand at a width of 1 mm and height 0.8 mm. As the print head velocity was
reduced, a larger amount of filament was deposited for a set filament volumetric
velocity. These results are statistically significant, as they demonstrate the local
dimensional variance that can be achieved via parameter modification, thus affecting
overall global part quality and geometry. Hebda et al. [17] also measured the cross-
sectional perimeters of single strands but varied the filament volumetric velocity along
Experimental Investigation of Filament Behaviour in Material Extrusion 281
with print head velocity and nozzle gap height, similar findings were recorded. It is to
be noted that both of these studies performed measurements only in the plane
orthogonal to the print direction, on specific printers, and only single strands were
deposited.
Extensive research has been undertaken to explore how the processing conditions
affect part quality and dimensional accuracy and precision; however, further research is
still required, thus we aim to build on this existing knowledge. In this study, we
investigate X/Y orthogonal plane cross sectional measurements of deposited strands
and present findings on repeatability, single strand and multiple strand behaviour. We
use a single filament feedstock material, PLA. In this study we are setting the foun-
dations for proposing a multiple-input variable optimisation for strand geometry. By
understanding the variations in printing parameter selection, this will enable control
over filament behaviour for enhancing part quality and dimensional accuracy.
2 Methodology
In order to investigate the behaviour of molten polymer as a function of processing

parameters in ME AM, an experimental approach was undertaken. Transparent rect-
angular extruded acrylic sheets were mounted to the build envelope of a Prusa 3D
printer. Single layer strands of filament were extruded and deposited onto the sheets at
varying processing conditions, nozzle gap height, print head velocity, and filament
volumetric velocity. The acrylic sheets represent the primary layer for deposited fila-
ment and ensure that there is enough contrast during image analysis. The deposited
sheets were encapsulated within resin and then cut and polished in the X and Y
orthogonal planes to expose the strand’s cross-sectional areas. Optical microscopy was
used to realise and perform measurement analysis of the perimeters, surface areas, and
geometries of the cross sections.
2.1 Experimental Set-Up

The experiments were conducted using a Prusa i3 MK3 Multi Material desktop 3D
printer with a nozzle diameter of 0.4 mm. Fluorescent red 1.75 mm diameter PLA
filament manufactured by MG Chemicals was used at an extrusion temperature of
215 °C. This temperature was held constant during experimentation as it has previously
been proven to have no effect on the filament shape after deposition [16]. The filament
strands were deposited directly onto the acrylic sheets (width 35 mm length
25 mm thickness 1.5 mm) at an ambient temperature of 19 °C. The sheets were
mounted to the build plate during deposition. The build plate temperature was set at
60 °C. Individual toolpath commands were executed in G-code to initiate the extrusion
process. Prior to strand deposition, the build envelope was probed nine times at various
X and Y coordinates using the Prusa’s automatic bed levelling and height calibration
function after each sheet was mounted. Thus, minimising the potential in accumulated
errors in Z positioning. The printing parameter values used are provided in Table 1.
Values were selected to provide significant strand morphological variation for each
process parameter, whilst also operating within the Prusa 3D printer limits.
282 M. Golab et al.
Table 1. Printing parameter values and their respective symbols

Variable parameter Unit Value
Nozzle gap height mm 0.15, 0.3
Nozzle diameter mm 0.4

Filament volumetric velocity mm3 s 2.4, 48.1
Print head velocity mm/s 1, 20
Feedstock filament diameter mm 1.75 ± 0.05

Extrusion temperature C 215

Bed temperature C 60
The printing parameters used were comparable to those used in previous studies
[16, 17], but we note that alternative parameters would be selected if we were using the
prusa slicer’s system presets. The filament deposition was executed systematically line-
by-line using custom G-code commands as opposed to automatically generated code
from the PrusaSlicer software commonly used with Prusa printers. The printed samples
were placed in sacrificial containers, as shown in Fig. 1, whereby MetPrep EPO-SET
resin was poured to encapsulate the acrylic sheets and PLA strands. The resin capsule
was cut and polished to reveal the strands cross sections at set depths in the X planes
and along the Y longitudinal plane. Image analysis of strand cross-sections was per-
formed using a Zeiss Olympus BX51 Optical Microscope at 5 magnification. In this
study, we extend upon prior work by examining the printed filaments across two
orthogonal planes and at the start and end points of the strand. Printed samples were
produced to enable examination of the repeatability of a single strand and the properties
of single and multiple strands.
Fig. 1. Resin holders for filament encapsulation. a) Holder to cut in the X orthogonal plane, b)
holder to cut in the Y and Z orthogonal planes.
2.2 Repeatability Strands

To validate the repeatability of this printing process, 32 strands were printed across 4
acrylic slides (8 strands per slide) using constant processing conditions. Some
dimensional variation in filament was expected, however the magnitude of this vari-
ation was not known. By characterising the innate repeatability of the printing process,
we are able to determine whether any variation seen in subsequent experiments is a
direct result of changing processing conditions and not the underlying process
repeatability. The parameters used were; nozzle gap height 0.3 mm, print head velocity
5 mm/s, and filament volumetric velocity 24.05 mm3 s as shown in Fig. 2. This
filament volumetric velocity was chosen as it is a relatively typical value used in the
PrusaSlicer’s printing system presets. Each strand was cut three times across the X
orthogonal plane along the longitudinal length of the strand. The cross sections were
characterised at the start, middle and end of the strands as demonstrated in the depo-
sition strategy in Fig. 2.
Fig. 2. Repeatability deposition and cutting schematic strategy presenting the chosen printing
parameters and the three X cutting planes along the length of the strand.
2.3 Single Strands

Individual strands were deposited at varying processing conditions in order to under-
stand the effects the parameters have on the dimensional geometry of printed single
filament. Figure 3 presents the specific parameters that were used for each individual
strand in addition to the deposition and cutting strategy.
Fig. 3. Singular strands deposition and cutting schematic strategy presenting the chosen printing
parameters and two orthogonal (X and Y planes) cutting planes
284 M. Golab et al.
The aforementioned X-cutting plane locations used in the repeatability experiment

were reproduced for this experiment. Furthermore, the centre of each of the eight
strands were cut and polished in the Y orthogonal plane along the longitudinal length.
2.4 Multiple Strands

Three strands were printed on top of each other to determine if the morphology of the
strands was altered as a result of subsequent strands being deposited. The processing
conditions and X cutting planes used in the aforementioned single strands experiment
were also used for the multiple stands experiment as shown in Fig. 4. The G-code for
this experiment was executed systematically, whereby the bottom first eight strands
were deposited, then the middle eight strands, and then the final eight strands as
opposed to depositing each sample in turn layer-by-layer.
Fig. 4. Multiple strand deposition schematic strategy presenting the chosen printing parameters,
quantity of layered strands, and one orthogonal (X plane) cutting plane
The cross sections of the strands produced in all three experiments were positioned
under an optical microscope in order to obtain the X and Y coordinates, cross sectional
area, perimeter, and maximum height and width. Figure 5 presents a cross sectional
profile of a strand cut in the middle of the strand about the X plane.
Fig. 5. a) Optical microscopy image of a cross sectional profile and b) a contour graph of the
same cross-sectional profile.
Raw data in the form of X and Y coordinates was analysed and plotted in Microsoft
Excel 2019, thus presenting a contour graph also shown in Fig. 5. In order to centre the
strand cross section X and Y coordinates, a lateral translation calculation was used.
A rotation calculation was also used to ensure that the strand’s base lines were pre-
sented horizontally. The experiments conducted foster an understanding as to how
deposited PLA filament behaves in ME AM.
3 Results
In this section, we have demonstrated how repeatable the material extrusion additive
manufacturing process is and how varying the processing conditions can significantly
affect the overall dimensional geometry of the resultant PLA deposited strands using a
Prusa 3D Printer.
3.1 Repeatability Strands

Figure 6 shows the maxima, average, and minima perimeter cross sections of PLA
deposited strands at the start, middle, and end of the strand. The cross-sectional profiles
of 32 strands were averaged, whilst the largest and smallest profiles were plotted
against the average to demonstrate the variation. Constant processing conditions were
used as shown in Fig. 2.
a) Start b) Middle c) End

300 300 300.
250 250 250.
200 200 200.
150 Maxima 150 Maxima 150.
Z axis (μm)
Maxima
Z axis (μm)
Z axis (μm)
100 Average 100 Average 100. Average

Minima Minima Minima
50 50 50.
0 0 0.
-400 -200 0 200 400 -400 -200 0 200 400 -400. -200. 0. 200. 400.
X axis (μm) X axis (μm) X axis (μm)
Fig. 6. Maxima, average and minima contour plots of cross-sectional profiles at constant
processing conditions. a) Start of strand, b) middle of strand, c) end of strand.
The repeatability experiment demonstrates that more significant variation in

dimensional accuracy has occurred at the beginning of the strand, whereas, the middle
of the stand exhibits greater precision. Additionally, the end of the strand shows lower
precision than the centre but greater than the beginning. The height and width of the
strand start shows a tolerance of ±30 lm and ±75 lm respectively. The height and
width of the strand middle exhibits a tolerance of ±15 lm and ±5 lm respectively.
The height and width of the strand end displays a tolerance of ±25 lm and ±30 lm
respectively. Surprisingly, the strands’ heights did not reach the set nozzle gap height,
however, this is due to the combined print head and filament volumetric velocities
depositing an insufficient amount of filament relative to the set nozzle gap height. At a
reduced print head velocity and an increased filament volumetric velocity, a 0.3 mm
strand height would be achieved. The tolerances were determined from the obtained
raw data points. The visual observations show that greater variation takes effect at the
start and end of the strand which is hypothesised to be a result of variability in the flow
of extrudate when the filament feeding mechanism initiates and finishes relative to the
286 M. Golab et al.
acceleration and deceleration of the print head. Once the feeding mechanism is in a
constant flow and the specified print head velocity has been achieved, greater accuracy
is observed.
An investigation by Bähr and Westkämper stated that an uncontrolled flow can
affect shape and position thus decreasing the dimensional accuracy, however, an array
of mechanisms can affect the repeatability of printed parts, most significantly nozzle
gap height with regards to printing parameters [18]. Dantan et al. [19] presented a
mathematical model that predicts repeatable geometrical behaviour, however, there are
currently no studies that have experimentally proven the repeatability and of an indi-
vidual strand in three cutting planes. Serdeczny et al. concluded that the FDM process
is repeatable using PLA. However, they only characterised three independent extru-
sions on one cutting plane [16].
It is believed that repeatable ME AM can be achieved during a constant extrudate
flow and the set print head velocity has been reached. Thus, this study investigates the
effects of the processing conditions on filament behaviour during deposition. It is to be
noted that this is the most rigorous study on strand repeatability currently known to the
authors.
3.2 Single Strands

Figures 7, 8, 9 and 10 show the experimental findings from eight single strands that
were deposited at a range of printing parameters. It is to be noted that the produced
results were comparable with a previous study [17]. Typically, it can be observed that
the higher nozzle gap height of 0.30 mm results in a higher Z dimension, however, this
geometry is also significantly influenced by the ratio between filament volumetric
velocity and print head velocity. Figures 8 and 9 demonstrate how a low nozzle gap
height to high filament volumetric velocity ratio generates an elliptical strand shape as
opposed to a more desired cylindrical shape, these strand morphologies are also
exhibited in previous studies [16, 17]. Under certain processing conditions, the nozzle
can obstruct and distort the deposited filament. This will be discussed in greater detail
in the subsequent section.
It is widely acknowledged that lower print head velocities yield better quality
depositions [20], as can be observed in the Y plane if one compares Figs. 7 and 10. In
Figs. 7, 8, 9 and 10 it can be seen that the A-A cross sections morphologies were
significantly smaller and varied in shape compared to the rest of the cross sections. This
was also somewhat observed in the repeatability study. The cross-sectional geometries
along the Y plane changed from the start to the end of the strand for all processing
conditions. It is postulated that this phenomenon is a result of an inconsistent filament
flow rate relative to the print head velocity, thus ‘blobbing’ of the filament can occur.
Fig. 7. X and Y cross sections of deposited strands at a filament volumetric velocity of

2:4 mm3 s and a print head velocity of 1 mm/s
Decreasing the print head velocity to filament volumetric velocity ratio can lead to
an increase in the cross-sectional area of the strands as shown in Fig. 9, as more
material is deposited per unit length in comparison to Fig. 8 where the print head
velocity is 1 mm/s. In Figs. 8 and 9, the results demonstrate that despite a nozzle gap
height of 0.3 mm, the morphology of the strand was able to exceed a height of 0.5 mm.
This is not an effect of the build envelope’s flatness as it was calibrated prior to each
print, however, the filament volumetric velocity value was significantly high, thus the
filament not only expanded laterally but vertically. The strands’ heights and widths are
highly susceptible to changes in these parameters.

Figure 10 demonstrates that a low filament volumetric velocity to high print head
velocity ratio exhibits significantly poor strand quality as shown in the Y plane. This is
288 M. Golab et al.
a result of a mismatch in ratios; thus, the filament is not deposited quickly enough
relative to the print head velocity. This has also induced a lack of adhesion contact
between the build envelope and deposited strand.

To demonstrate how the processing conditions can significantly affect the strand
size, Figs. 7 and 9 were compared. Both figures have the same print head velocity of
1 mm/s, however, the filament volumetric velocities were 2:4 mm3 s and 48:1 mm3 s
respectively. The middle B-B cross section in Fig. 7 has a height of 450 lm and width
of 600 lm, whereas the middle B-B cross section in Fig. 9 has a height 700 lm and
width of 2800 lm for the 0.15 mm nozzle gap height. By tailoring the processing
conditions, one can control not only the geometry and shape but the overall quality of
part including surface finish and inter-layer bonding.
3.3 Multiple Strands

Printed components are usually produced in multiple layers, so it is important to
characterise the behaviour of multiple stands as this better reflects reality in terms of
real component creation. Figure 11 presents the resultant analysed cross-sections,
whereby the processing conditions used in the single strand study were replicated for
multiple strands. Empirically, it was acknowledged that the nozzle gap height in this
instance had a significant influence on the morphology of the subsequent strands. The
greatest deformation
was realised in Fig. 11d), at a filament volumetric velocity of
48:1 mm3 s and a print head velocity of 1 mm/s, where the nozzle had compressed the
third layer during deposition. This appears to have resulted in a side flow of filament to
form around the nozzle, distorting the overall geometry. For this experiment, it was
clear that the nozzle gap height was not suitable for this specified filament volumetric
velocity. Additional noticeable effects can be seen in Fig. 11c), where the measured
strand height for a nozzle gap height of 0.3 mm was smaller than that of the 0.15 mm
nozzle gap height, although the 0.3 mm strand did exhibit increased lateral elongation.
An unexpected finding presented in Fig. 11b) showed the initial strand cross section of
the 0.3 mm nozzle gap height to be smaller in width than that of the second and third
strands. As the nozzle gap height was significantly disproportionate to the combined
filament volumetric velocity and print head velocity, the first strand exhibited little to
no downwards pressure from the nozzle that would typically result in lateral elonga-
tion. Therefore, the initial deposited molten polymer strand was allowed to form to a
greater extent in the Z axis, thus resulting in less vertical space for the subsequent
strands, compressing them laterally.
As previously demonstrated in Fig. 10, a low filament volumetric velocity to high
print head velocity ratio generated limited bonding between strands, also observed in
Fig. 11a). Additionally, it has been observed across all experimentation, that the
morphology of the strand changes from start to end despite the processing conditions
remaining constant. This fluctuation is believed to be a result of an inconsistent flow of
filament over the extrusion length. In some instances, the bonding characteristics
between layers have generated undesired geometries. However, the specified param-
eters were deliberately varied such that significant geometrical variation would be
observed. It would be likely that more uniform geometries would be observed with
alternative processing parameters.
290 M. Golab et al.
Fig. 11. Start, end, and middle X plane cross sections of three-layer deposited strands. a)

Filament volumetric velocity 2:4 mm3 s and print head velocity 20 mm/s. b) Filament

volumetric velocity 2:4 mm3 s and print head velocity 1 mm/s. c) Filament volumetric velocity

48:1 mm3 s and print head velocity 20 mm/s. d) Filament volumetric velocity 48:1 mm3 s and
print head velocity 1 mm/s.
4 Discussion and Conclusion
The results demonstrate that the printing parameters significantly influence filament
behaviour, and that it is therefore of paramount importance to select appropriate
parameters when printing parts.
The underlying machine repeatability is shown to be in the order of ±30 lm
(strand height) and ±75 lm (strand width). In comparison, when varying process
parameters for single strands, the variation is in the order of 500 lm (height) and
2600 lm (width). When multiple strands are deposited on top of each other, there is a
similar variation in width, but more significant defects in the shape of the deposited
strand can be seen in extreme conditions.
Thus, dimensional variation observed when changing process parameters far
exceed the underlying dimensional errors observed during the repeatability test. This
demonstrates a theoretical level of precision achievable, which may be improved with
different processing parameters. The characterisation of strand cross section dimensions
in this study may also form the basis of improved dimensional accuracy and precision
for varying processing conditions.
Parameter selection is not typically undertaken by casual users of 3D printing

machines. Indeed, most ME AM software enables the user to select from a wide variety
of parameters but include a default combination that would most typically be used.
Additionally, there is little guidance as to the impact of changing parameters. The
PrusaSlicer software provides the user with the option to print with a range of system
presets, however, these presets do not take into account the variation of filament
behaviour. The printed parts tend to exhibit dimensional inaccuracies and precision
defects, thus it is important to reduce these errors. The findings from this study may be
incorporated into the software to improve the output for any given selection, regardless
of the user’s understanding of their strand-level effects.
Variations observed at the central section of a strand are far less significant than
those at the start and end of a strand. Variations at the start also exceed those at the end
of a strand. Thus, when a machine is printing at a ‘steady state’, less variation is seen.
To date, little academic attention has been given to ensuring precision is achieved along
the whole strand. This is important, as non-conformities at the strand ends can result in
significant part defects, poor mechanical properties, and print failures.
The multiple strand sections show that examining single strands alone is not suf-
ficient to understand how processing parameters truly influence geometry. When
multiple strands have been considered, large variations between each individual
deposited strand have been observed. This would suggest that producing components
that reliably fit together is a complex problem which entails detailed understanding of
inter-strand interaction. These variations would be further complicated by any joins of
adjacent strands, which would be seen in ‘real’ parts.
Further research is required to better understand the optimal processing conditions
for multiple strands. Further work might also examine variation as a result of strands
joining in orthogonal planes and varied deposition directions. An additional repeata-
bility experiment will be undertaken using the PrusaSlicer’s system presets to establish
the dimensional variation of printed strands when using default settings.
Controlling the start and end conditions or limiting their presence in strand depo-
sition appears to be important. Further research might examine how modifying the G-
code commands (e.g. filament volumetric velocity, nozzle position) locally to com-
pensate for variations could be applied. The retraction speeds at the start of the strands
were not measured, this requires further research to understand how retraction influ-
ences the morphology of the strand. Further work might also determine whether dif-
ferent materials and printer nozzles might also have an impact.
Acknowledgements. The authors would like to thank the financial support of the UK Engi-
neering and Physical Sciences Research Council (EPSRC) through the EPSRC Centre in Ultra
Precision Engineering (EP/K503241/1).
References
1. Peng, F., Vogt, B., Cakmak, M.: Complex flow and temperature history during melt
extrusion in material extrusion additive manufacturing. Addit. Manuf. 22, 197–206 (2018)
292 M. Golab et al.
2. Paul, R., Anand, S.: Optimal part orientation in rapid manufacturing process for achieving
geometric tolerances. J. Manuf. Syst. 30(4), 214–222 (2011)
3. Park, S., Rosen, D.: Quantifying effects of material extrusion additive manufacturing process
on mechanical properties of lattice structures using as-fabricated voxel modeling. Addit.
Manuf. 12, 265–273 (2016)
4. Vishwas, M., Basavaraj, C., Vinyas, M.: Experimental investigation using taguchi method to
optimize process parameters of fused deposition modeling for ABS and nylon materials.
Mater. Today: Proc. 5(2), 7106–7114 (2018)
5. Anitha, R., Arunachalam, S., Radhakrishnan, P.: Critical parameters influencing the quality
of prototypes in fused deposition modelling. J. Mater. Process. Technol. 118(1–3), 385–388
(2001)
6. Pérez, C.: Analysis of the surface roughness and dimensional accuracy capability of fused
deposition modelling processes. Int. J. Prod. Res. 40(12), 2865–2881 (2002)
7. Pandey, P., Venkata Reddy, N., Dhande, S.: Improvement of surface finish by staircase
machining in fused deposition modeling. J. Mater. Process. Technol. 132(1-3), 323–331
(2003)
8. Nancharaiah, T., Ranga Raju, D., Ramach, V., Raju, R.: An experimental investigation on
surface quality and dimensional accuracy of FDM components. Int. J. Emerg. Technol. 1(2),
106–111 (2010)
9. Mohamed, O., Masood, S., Bhowmik, J.: Optimization of fused deposition modeling process
parameters: a review of current research and future prospects. Adv. Manuf. 3(1), 42–53
(2015)
10. Alexander, P., Allen, S., Dutta, D.: Part orientation and build cost determination in layered
manufacturing. Comput.-Aided Des. 30(5), 343–356 (1998)
11. Thrimurthulu, K., Pandey, P., Venkata Reddy, N.: Optimum part deposition orientation in
fused deposition modeling. Int. J. Mach. Tools Manuf 44(6), 585–594 (2004)
12. Ollison, T., Berisso, K.: Three-dimensional printing build variables that impact cylindricity.
J. Ind. Technol. 26(1), 2–10 (2010)
13. Vasudevarao, B., Natarajan, D., Henderson, M., Razdan, A.: Sensitivity of Rp surface finish
to process parameter variation. In: 11th Solid Freeform Fabrication Symposium, Austin,
pp. 251–258 (2000)
14. Bikas, H., Stavropoulos, P., Chryssolouris, G.: Additive manufacturing methods and
modelling approaches: a critical review. Int. J. Adv. Manuf. Technol. 83(1–4), 389–405
(2015)
15. Jaisingh Sheoran, A., Kumar, H.: Fused deposition modeling process parameters optimiza-
tion and effect on mechanical properties and part quality: review and reflection on present
research. Mater. Today: Proc. 21, 1659–1672 (2019)
16. Serdeczny, M., Comminal, R., Pedersen, D., Spangenberg, J.: Experimental validation of a
numerical model for the strand shape in material extrusion additive manufacturing. Addit.
Manuf. 24, 145–153 (2018)
17. Hebda, M., McIlroy, C., Whiteside, B., Caton-Rose, F., Coates, P.: A method for predicting
geometric characteristics of polymer deposition during fused-filament-fabrication. Addit.
Manuf. 27, 99–108 (2019)
18. Bähr, F., Westkämper, E.: Correlations between influencing parameters and quality
properties of components produced by fused deposition modeling. Procedia CIRP 72,
1214–1219 (2018)
19. Dantan, J., et al.: Geometrical variations management for additive manufactured product.
CIRP Ann. 66(1), 161–164 (2017)
20. Zhou, Y., Nyberg, T., Xiong, G., Liu, D.: Temperature analysis in the fused deposition
modeling process. In: 3rd International Conference on Information Science and Control
Engineering (ICISCE), pp. 678–682. IEEE, Beijing (2016)
Debinding and Sintering of Dense Ceramic
Structures Made with Fused Deposition
Modeling
Frank Clemens1(&), Josef Schulz1,2, Lovro Gorjan1, Antje Liersch2,

Tutu Sebastian1, and Fateme Sarraf1
1
Empa - Swiss Federal Laboratories for Materials Science and Technology,
Überlandstrasse 129, 8600 Dübendorf, Switzerland
frank.clemens@empa.ch
2
Hochschule Koblenz Wester Wald Campus,
Rheinstraße 56, 56203 Höhr-Grenzhausen, Germany
Abstract. Shaping and thermal processing of ceramic structures with fused

deposition modeling (FDM) is promising, but still a challenging technique.
Achieving an optimal compromise between the thermoplastic feedstock prop-
erties in terms of 3D printing and debinding behavior is not trivial. In this paper
aluminum oxide (Al2O3), zirconium oxide toughened aluminum oxide
(ZTA) and ZrO2 (zirconium oxide) thermoplastic filaments were developed for
multi-material printing of 2-2 and 3-3 composites using a Bowden extruder with
2.8 mm filament thickness. For all filaments, a thermoplastic binder based on
Ethylene Vinyl Acetate (EVA) and stearic acid (SA) with a ceramic filler
content of 45 vol.% were used. Increasing the SA content, it was possible to use
the solvent debinding process and therefore the thickness of the printed struc-
tures could be increased up to 10 mm. After 2 h in acetone solvent, more than
40 wt% could be removed. With additional thermal partial debinding process at
240 °C for 8 h, up to 82 wt% of the thermoplastic binder could be removed
successfully. The monolithic printed discs achieved a relative density of 98.2
and 99.1% for the ZTA and the ZrO2 samples, respectively. Finally, the
mechanical properties were investigated by ring-on-ring bending test method
based on ASTM standard. For ZrO2 discs a mechanical strength of 429 MPa
could be achieved. By mixing 5% yttrium stabilized ZrO2 with Al2O3, the
mechanical strength could be increased from 220 (pure Alumina) to 297 MPa
(ZTA). For the 2-2 composite, a mechanical bending strength of 214 MPa could
be achieved.
Keywords: Fused deposition modeling Fused filament fabrication

Ceramics ZrO2 Al2o3 ZTA Ceramic Ring-on-ring test
1 Introduction
The additive manufacturing (AM) process can be used to achieve three-dimensional

structures placing layer by layer on a support structure. Different technologies to achieve
layered-like structures have been developed in the past to accomplish monolithic as well

https://doi.org/10.1007/978-3-030-54334-1_21
294 F. Clemens et al.
as composite structures. One of the AM processes which can be used to fabricate a 3D

composite structure is the fused deposition modeling (FDM) also known as fused fil-
ament fabrication (FFF) process. It is a well-established additive manufacturing tech-
nique. In the FDM/FFF process, a thermoplastic-based filament is fed into an extruder
by gears (feeding rollers). The thermoplastic is melted inside the printing head (hotend)
and extruded through a small nozzle [1, 2]. The extruded melted filament is deposited on
a substrate and will solidify by convection of cool air. The main advantages of the
FDM/FFF process include inexpensive equipment, a large variety of materials, easy to
use and the possibility of making multi-material and large parts [3]. The filament
mechanical properties have to meet some special requirements in order to be suitable for
the FDM/FFF printer. It must be strong and hard enough to avoid shearing due to
pinching from the drive wheel [3]. A certain stiffness is required to avoid buckling
between drive wheels and the hot zone of the extrusion die. It should also be noticed that
the melting behavior of the material, as well as processing parameters, have a significant
influence on whether the printing will be successful or not.
In contrast to FDM/FFF of polymers with filler materials, a high filling level of
ceramic particles is essential to ensure proper sintering of the ceramic in the final
processing step. For the densification (reduction of porosity between the ceramic
particles) of the ceramic, the sintering process step is indispensable. In addition, the
thermoplastic materials and other polymeric additives, which are essential for the
shaping process, have to be completely removed before the sintering step. Therefore, a
so-called debindng step is necessary after the printing process.
Ceramic filaments with high solid loading content (>45 vol.%) are more chal-
lenging for processing in FDM/FFF process, because they are more brittle, and they are
more difficult to feed through the nozzle [2, 4]. In addition, the 3D printing of ceramic
filaments depends on the rheological behavior of the feedstocks, the mechanical
properties of the filament, the geometry of the filament, the design of the machine, the
processing parameters and the printing head [2, 3].
Several formulations for the ceramic filaments were reported by researchers for
different materials like tricalcium phosphate TCP [6–8] mullite [9], alumina [5, 6, 10,
11], zirconia [10, 12, 13] silicon nitride [5, 14] and PZT [15]. Nowadays, the strength
of sintered ceramic parts made by the FFF does not achieve the strength values
obtained by other thermoplastic shaping methods, which could be considered as a
limiting factor of this technique.
The mechanical properties of FDM printed and sintered ceramic materials have not
been reported frequently. Conzelmann et al. have investigated the effect of the printing
direction on FDM/FFF fabricated bars [11]. However, the mechanical properties are
affected by the roughness of the surface and the defects from printing, debinding, and
sintering. Based on this knowledge, zirconia toughened alumina and zirconia filaments
were developed for Bowden extruder technology, and dense monolithic, 2-2 and, 3-3
composite ceramic discs have been printed, debinded and sintered to measure the
mechanical properties by using the ring-on-ring bending test. Evaluation of mechanical
properties for FDM/FFF fabricated ceramics will become more important in the future
to evaluate this ceramic processing route for future industrial applications.
Debinding and Sintering of Dense Ceramic Structures Made with FDM 295
2 Experimental
For this study, alumina CT3000 (Almatis GmbH, Germany) and zirconia 3Y-PTZ-SE
(Tosoh Europe B.V., Germany) ceramic powder were used. For the zirconia toughened
alumina a mixture of alumina (CT3000) with 5 vol.% of yttria-partially stabilized
zirconia (3Y-PTZ-SE), based on Barros et al. was used [16]. The thermoplastic binder
consisted of ethylene-vinyl acetate (DuPont, USA) and stearic acid (Sigma Aldrich,
Switzerland) with a ratio of 60:40. For the filaments used for the FDM/FFF printers, the
ceramic filler content was fixed to 45 vol.%.
For the powder analysis, LS 13 320 laser diffraction particle size analyzer and SA
3100 surface analyzer (Beckman Coulter, Germany) were used. For the particle size
analysis, the powders was dispersed in water using an ultrasonic finger. The specific
surface area of the powders was determined from five-point N2 adsorption isotherm
measurements (Beckman Coulter SA3100, Beckman-Coulter Inc.). The morphology of
the powders was investigated by scanning electron microscopy (SEM, VEGA3, Tes-
can, Czech Republic).
Thermoplastic feedstocks were prepared by mixing all components in a high shear
mixer (Rheomix 600, HAAKE Polylab OS, Thermo Electron Corporation, Germany).
Before mixing, ceramic powders were dried overnight at 120 °C. The compounding
was done in three steps to avoid agglomerates. In the first and second steps, the material
was mixed at 90 °C with 10 rpm for 20 min. each. In the third step, the material was
mixed with 40 rpm for another 20 min.
Filaments were made by extrusion of the three different feedstocks at 80 °C through
a die orifice with a diameter of 2.8 mm, using a piston extruder (RH7 Flowmaster,
Malvern Instruments Ltd, UK).
A commercial FDM/FFF printer Sigma BCN3D R17 (3D-Printerstore, CH) was
used for the fabrication of ceramic discs. Round discs were punched out from square
printed discs. For the printing, a nozzle diameter of 0.6 mm, a print head temperature of
170 °C and a print bed temperature of 60 °C was used. Each printed layer was printed
with a height of 0.125 mm. 11 layers with 0°/90° orientation were printed to achieve a
thickness of the green samples of 1.4 mm.
For the debinding step, pure partial thermal debinding method at different temper-
atures, with and without power bed, was investigated [17]. The thermal debinding was
carried out in a static air using a furnace (LT 40/12, Nabertherm GmbH, Germany). For
each different dwell temperature (200, 210, 220 and 230 °C) a dwell time of 8 h was
used. The powder bed was used to draw out the liquid polymeric binder because of
capillary forces [17]. In addition solvent and thermal debinding process methods were
combined later, to remove the organic binder for the round discs. For the solvent
debinding process, acetone was heated up to its boiling point (56 °C) using a distillation
column. After debinding the samples at 500 °C, the final sintering was done in an
electrically heated furnace (LHT 04/17, Nabertherm, Germany) at 1600 °C, with a
heating rate of 5 K/min and dwelling time of 2 h. To investigate the shrinkage of the
filaments, TMA 402 F3 (Netzsch GmbH, Germany) was used. To compare the
shrinkage behavior of the filaments, samples were debinded and the sintering behavior
was evaluated with a heating rate of 5 K/min up to 1550 °C.
The strength of sintered 3D printed discs was evaluated using a standard ring-on-
ring bending test on cylindrical samples [18]. Based on the thickness of the sample, a
support ring with a diameter of 18 mm and a load ring with a diameter of 7 mm were
selected. For the mechanical analysis, a universal tensile testing machine Zwick Z005
(ZwickRoell GmbH & Co. KG, Germany) was used. The microstructure of the fracture
surface was investigated by scanning electron microscopy (SEM, VEGA3, Tescan,
Czech Republic).
3 Results and Discussions
3.1 Power Characterization

The powder was investigated by routine analyses, namely laser diffraction, to analyze
the particle size distribution of the ceramic powders, BET, to determine the specific
area of the ceramic powder. Finally, the habitus of the ceramic powders was investi-
gated by scanning electrical microscope. The results are shown in Fig. 1 for the alu-
mina powder and in Fig. 2 for the zirconia powder, respectively.
Fig. 1. Results of the characterization for the CT3000 alumina powder. a) CT 3000 granule as
received, b) platelet structure which could be observed in some parts.
In some parts of the alumina granulates, primary particles with platelet structure
could be observed. After wet dispersion in water, granulates decomposed and a powder
with the mean particle of 1.2 lm could be observed. The surface area of the powder
was 7.8 m2/g which correlated with the datasheet of Almatis GmbH.
For the ZrO2 powder, a mean particle size of 0.6 lm and a specific surface area of
7 m2/g was observed.
Fig. 2. Results of the characterization for the 3Y-PTZ-SE zirconia powder. a) 3Y-PTZ-SE
granulate as received, b) the structure of the granulate.
3.2 Feedstock and Filament Fabrication

Figure 3 summarizes the results of the filament fabrication. Filaments made with a
mixture of CT3000 and 5% of 3Y-PTZ-SE powder could be extruded with lower
pressure. Because of the higher particle size distribution of the CT3000, this was
expected. Because of the limitation of the Bowden extruder motor power, the higher
content of 3Y-PTZ-SE (ZrO2) was not printable with the selected thermoplastic binder
system.
Fig. 3. a) Pressure needed for the fabrication of the filaments with a diameter of 2.85 mm at
85 °C, b) experiment to evaluate the minimum bending radius of the filaments.
Based on the slightly higher bending radius of the ZTA filaments, a higher powder
content resulted in brittle filaments that were not able to print anymore.
3.3 FDM/FFF Printing of the Ceramic Samples

Both filaments were printed with the Sigma BCN3D R17 desktop printer at 170 °C.
However, an infill of 120% had to be used to be able to achieve dense ceramic
structures (see Fig. 4a). A lower infill value resulted in interfilament porosity (Fig. 4b.).
Fig. 4. a) Optimization of the infill from 100 to 150% in 10% steps, b) Microscopic picture of a
printed sample with 100% infill, c) Picture of a sample with 150% infill.
At a higher infill level, the accumulation of the material resulted in areas where the
printing direction needed to be changed (see Fig. 4c).
3.4 Debinding of the Ceramic Filaments and Printed Samples

Based on the results of Conzelmann et al., a pure thermal debinding process was
planned to used [17]. Figure 5 shows the partial thermal debinding process of the ZTA
and ZrO2 filaments.
Fig. 5. Investigation of the partial thermal debinding process on ZrO2 and ZTA filaments,
a) both filaments were partially debinded with and without a powder bed for 8 h. After debinding
process b) ZrO2 filament keeps the shape, c) while the ZTA filament deforms due to the softening
of the thermoplastic binder before decomposition of the binder occurs.
Due to the larger particle size of the alumina particles, for the ZTA filaments, a
higher binder weight loss was expected, in comparison to the zirconia filaments. Larger
particles result in larger interparticle pore sizes and therefore the decomposition and
diffusion process should be higher for the ZTA filaments. However, the results show
contrariwise behavior. The contrariwise behavior can be explained by the morphology
of the alumina powder (CT3000). A certain amount of particles show a platelet
structure (see Fig. 1). This will limit the diffusion process and lower the mass loss of
the binder during the thermal debinding process.
After debinding at 230 °C a weight loss of 60% could be achieved for the two
different filaments. However, large deformation of the ZTA filaments and printed discs
could be observed. This deformation could not be avoided by reducing the partial
thermal debinding process down to 200 °C. In comparison to this study, Conzelmann
et al. used an alumina filler content of 50 vol.% for the filaments [17]. Unfortunately,
with the Sigma BCN3D R17 ZrO2 filaments with a higher powder content than 45 vol.
% could not be printed anymore. To avoid the deformation of the discs, an additional
solvent debinding step was used for all FDM/FFF printed monolithic, 2-2 and 3-3
composites (Fig. 6).
Fig. 6. a) binder loss and shrinkage during the solvent debinding of 3D printed discs. Solvent
debinded b) ZrO2 filaments show no defect after solvent debinding, while c) ZTA filaments show
a large crack.
Interestingly the ZTA printed discs show the highest mass loss of all the samples.
For the two composites, a lower mass loss and a higher shrinkage were observed
(Fig. 6a). After solvent debinding, the ZTA filament does not show a shape defor-
mation any more. However, a large crack inside the filament can be observed. This
might be the reason for the higher binder loss during the solvent debinding step.
3.5 Sintering of the Ceramic Printed and Debinded Samples

The sintering behavior of the ZTA and the ZrO2 feedstocks were investigated using a
thermal-mechanical analyzer (TMA). The results are shown in Fig. 7. The sintering of
both filaments starts at 1030 °C. As expected, the ZrO2 filament shrinks much faster in
comparison to the ZTA filaments. At 1550 °C the ZrO2 filaments reach the max.
sintering shrinkage of 20%. For the ZTA filaments at 1030 °C, the sintering starts
slowly. Up to 1300 °C, the shrinkage of the materials reaches only 2%. After 1300 °C
sintering of ZTA increase significantly and the sintering (14%) has not been finished at
the 1550 °C (max. temperature of the TMA analytic machine).
The table in Fig. 7 shows the shrinkage of the printed discs after sintering at 1600 °C
for 2 h. For the ZTA a relative density of 98.2% could be achieved, whereas, for the
sintered ZrO2 discs, a relative density of 99.1% revealed. The relative density of 96.7%
and 96% for the 2-2 and 3-3 composites could be achieved, respectively.
Fig. 7. Shinkage behavior of ZTA and ZrO2 filaments with a filler content of 45 vol.%. The
Table shows the shrinkage of 3D printed monolithic, 2-2, and 3-3 composite discs after sintering
at 1600 °C for 2 h.
3.6 Mechanical Properties of the Ceramic Sintered Samples

To investigate the mechanical properties of the different sintered discs, a ring-on-ring
test was selected (see Fig. 8). In the table, the mechanical properties of monolithic, 2-2,
and 3-3 composite discs after sintering at 1600 °C for 2 h are presented. Because of the
low number of samples (<5), only the mean value of the 2-2 and 3-3 composites have
been reported.
Fig. 8. Mechanical properties of monolithic, 2-2, and 3-3 composite discs sintered at 1600 °C
for 2 h. For the composites, only the mean value has been reported, because of the low number of
samples (<5).
The ZrO2 monolithic discs show two different slopes. For the 7 samples with higher
mechanical properties, a mean strength of 483 MPa and a Weibull modulus of 5.1
could be calculated using a confidence interval of 90%. For the 6 samples with lower
mechanical strength, a mean strength of 336 MPa and a Weibull modulus of 17.7 could
be observed.
To better understand the results of the mechanical properties of the different
materials, the microscopic analysis was performed on selected samples (Fig. 9).
(Figure 9(a and b) shows the fracture surface of the monolithic ZrO2 discs. Large
defects based on interfilament porosity is shown. These failures are related to printing
defects and often seen in structures printed with commercial FDM/FFF printers.
Fig. 9. Microscopy images of a) low strength ZrO2 FDM printed sample, b) higher
magnification of printing failure, c) high strength ZrO2 FDM printed samples, d) ZTA FDM
printed sample, e) typical failure inside the ZTA FDM printed samples, f) higher magnification of
the defect in ZTA FDM printed samples, g) 2-2 composite sample with low strength, h) 2-2
composite with high strength, i) 3-3 composite with cracks.
For monolithic ZrO2 discs with high mechanical strength, almost no defects could
be detected in SEM microscopy analysis. For the monolithic ZTA samples defects in
the printed structures looks different (see Fig. 9e, and 9f). The cracks have a similar
shape as shown in Fig. 6c. It can be assumed that the solvent debinding process has to
be optimized or alternative alumina powder must be selected for future studies. In
Fig. 9g and Fig. 9h a 2-2 composite with low strength 150 MPa and high strength
253 MPa are shown. In both samples, interfilament porosity is present, which explains
the lower mechanical properties in comparison to the monolithic ZTA samples. The
mechanical properties of the 3-3 composite were unexpectedly low. Microscopy
analysis revealed, that all samples were full of cracks. This can be explained by the
large difference between the shrinkage behavior of the ZTA and the ZrO2 materials.
4 Summary
In this work, the fabrication of 2-2 and 3-3 composites with ZTA and ZrO2 multi-
filament printing has been investigated for the first time. A two-step debinding process
based on the solvent and thermal debinding processing step are proposed. Further
investigations on the processing route have to be done, to achieve defect-free samples
and to improve the mechanical properties.
The accuracy of the commercial FDM/FFF printer (Sigma BCN3D R17) is the
main limitation to achieve sintered ceramic parts with reliable mechanical properties.
Comparing the results with the work of Barros et al. [16], we can assume, that
FDM/FFF printing is a good alternative technology for the fabrication of monolithic
and 2-2 composites. Typically, infiltration und impregnation techniques are nowadays
needed to achieve 3-3 composite materials. The FDM/FFF technology has the
advantage, to manufacture 3-3 composites which will be interesting to build up

structural as well as functional ceramic composites for different applications. To be able
to bring the ceramics fabricated by FDM/FFF technology into industrial applications, it
is necessary to evaluate mechanical properties for the sintered structures on a more
regular base. Based on our results we assume, that ring-on-ring bending test has great
potential to evaluate the mechanical properties of printed ceramics on a regular base.
Acknowledgment. This project has received funding from the Swiss National Science Foun-
dation (SNF) under grant agreement No 200021 184691 and funding for the instrument from
SNF project No. 206021 164024.
References
1. Clemens, F.: Fused deposition modeling. https://www.euroceram.org/. Accessed 22, 25 2020
2. Clemens, F.: FDM Keramikdruck: Herstellung von keramischen Strukturen mit Hilfe von
kommerziellen Filamentdruckern. Keram. Z. 71(5), 40–45 (2019)
3. Gonzalez-Gutierrez, C.J., Cano, S., Schuschnigg, S., Kukla, C., Sapkota, J., Holzer, C.:
Additive manufacturing of metallic and ceramic components by the material extrusion of
highly-filled polymers: a review and future perspectives. Materials 11(5), 840 (2018)
4. Gibson, I., Rosen, D., Stucker, B.: Additive Manufacturing Technologies. Springer, New
York (2015)
5. Rangarajan, S., Qi, G.: Powder processing, rheology, and mechanical properties of feedstock
for fused deposition of Si3N4 ceramics. J. Am. Ceram. Soc. 69, 1663–1669 (2000)
6. Nötzel, D., Eickhoff, R., Hanemann, T.: Fused filament fabrication of small ceramic
components. Materials 11, 1–10 (2018)
7. Bose, S., Darsell, J., Kintner, M., Hosick, H., Bandyopadhyay, A.: Pore size and pore
volume effects on alumina and TCP ceramic scaffolds. Mater. Sci. Eng., C 23, 479–486
(2003)
8. Gorjan, L., Reiff, L., Liersch, A., Clemens, F.: Ethylene vinyl acetate as a binder for additive
manufacturing of tricalcium phosphate bio-ceramics. Ceram. Int. 44, 15817–15823 (2018)
9. Kalita, S.J., Bose, S., Hosick, H.L., Bandyopadhyay, A.: Development of controlled porosity
polymer-ceramic composite scaffolds via fused deposition modeling. Mater. Sci. Eng., C 23,
611–620 (2003)
10. Gorjan, L., Tonello, R., Sebastian, T., Colombo, P., Clemens, F.: Fused deposition modeling
of mullite structures from a preceramic polymer and c-alumina. J. Eur. Ceram. Soc. 39,
2463–2471 (2019)
11. Conzelmann, N., Gorjan, L., Sarraf, F., Poulikakos, L., Partl, M., Müller, C., Clemens, F.:
Manufacturing complex Al2O3 ceramic structures using consumer-grade fused deposition
modelling printers. Rapid Prototyping J. Accepted 04 Feb 2020
12. Neale, G.H., Nader, W.K., Cano, S., Gonzalez-Gutierrez, J., Sapkota, J., Spoerk, M.,
Arbeiter, F., Schuschnigg, S., Holzer, C., Kukla, C.: Additive manufacturing of zirconia
parts by fused filament fabrication and solvent debinding: selection of binder formulation.
Addit. Manuf. 26, 117–128 (1973)
13. Cano, S., Gonzalez-Gutierrez, J., Sapkota, J., Spoerk, M., Arbeiter, F., Schuschnigg, S.,
Holzer, C., Kukla, C.: Additive manufacturing of zirconia parts by fused filament fabrication
and solvent debinding: selection of binder formulation. Addit. Manuf. 26, 117–128 (2019)
14. McNulty, A., Mohammadi, T.F., Bandyopadhyay, F., Shanefield, A., Danforth, D.J., Safari,
A.: Development of a binder formulation for fused deposition of ceramics. Rapid
Prototyping J. 4, 144–150 (1998)
15. Bach, M., Sebastian, T., Melnykowycz, M., Tony, L., Scharf, D., Clemens, F.: Additive
manufacturing of piezoelectric 3-3 composite structures. In: Meboldt, M., Klahn, C., (eds.)
Proceedings Additive Manufacturing Products and Applications - AMPA2017, pp. 93–103.
Springer, Cham (2018)
16. Barros, M.D., Rachadel, P.L., Fredel, M.C., Janssen, R., Hotza, D.: Mechanical behaviour of
Zirconia-Toughened Alumina Laminates with or without Y-PSZ intermediate layers.
J. Ceram. Sci. Technol. 9(1), 69–78 (2018)
17. Conzelmann, N.A., Gorjan, L., Sarraf, F., Poulikakos, L.P., Partl, M.N., Müller, C.R.,
Clemens, F.J.: Manufacturing complex Al2O3 ceramic structures using consumer-grade
fused deposition modeling printers. Rapid Prototyping J. (2020, accepted)
18. ASTM: Standard test method for monotonic equibiaxial flexural strength of advanced
ceramics at ambient temperature 1. ASTM Stand. 03, 1–16 (2009)
Feasibility Investigation of Gears
Manufactured by Fused Filament Fabrication
Hans-Jörg Dennig(&) , Livia Zumofen , and Andreas Kirchheim
ZHAW Zurich University of Applied Sciences,

ZPP Centre for Product and Process Development, Winterthur, Switzerland
hans-joerg.dennig@zhaw.ch
Abstract. This paper describes the production, quality and tests of plastic gears
made by fused filament fabrication (FFF). Different 3D printers of low and high
quality and four different materials were tested: PLA, ABS, PA and PA carbon
fiber composite. After optimizing the print parameters, various gears were
manufactured and tested. The geometry and the test procedure were executed
based on the VDI 2736 standard. This paper is a feasibility study of the fatigue
strength of different gears. There is no material data set for the specific FFF-
Materials and certain printers.
The results show, that the quality of the printing process and the accuracy of
the geometry are not responsible for the failure mode. The most important
parameter is the friction coefficient of the different materials. Due to friction the
tooth root will heat up and effect the material properties. PLA gears failed due to
the low melting temperature, whereas ABS and PA gears failed on tooth root
fracture.
Keywords: Gears Fused filament fabrication Testing Additive

manufacturing VDI 2736
1 Introduction
The use of additive manufacturing in the field of gears, for example for the production
of metal gears with internal cooling systems is becoming increasingly important [1, 2].
More and more gears can also be produced in the field of additive manufacturing of
plastic gears [3]. Precise and durable materials are produced here using selective laser
sintering. These components can now be regarded as reliable and can exceed the
fatigue strength of injection moulded Polyoxymethylene (POM) wheels.
FFF1 manufactured gears (Fig. 1) made of Polylactide (PLA) or Acrylnitril-
Butadien-Styrol-Copolymere (ABS) are not associated with high strength or precision
and are only used for low-stressed components and mock-ups for gears. Datasheets for
1
Fused filament fabrication (FFF) is an extrusion process that presses a heated thermoplastic material,
preferably continuously, through a nozzle and deposits the resulting strand in a defined manner on a
substrate. To create contours, the extrusion nozzles are moved accordingly in the construction plane
(x, y) [4].

https://doi.org/10.1007/978-3-030-54334-1_22
Feasibility Investigation of Gears Manufactured by Fused Filament Fabrication 305
gear transmission detailed FFF-Material characteristics, the influence of print param-

eters and on fatigue is unknown.
In this study, gears made of different available materials were manufactured on
different printers with different printing parameters and tested on a gear test bench. The
geometry and the test are based on the current European gear standard VDI 2736.
Fig. 1. Fused Filament Fabrication (FFF)-Process [4]
2 Failure Mode of Plastic Gears
Due to the cost-effective production of gears in the hobbing process, the involute shape
of the teeth has become historically accepted. However, with the possibilities of
additive manufacturing, numerous other shapes are conceivable, for example, ones
optimized for ventilation loss or flank contact efficiency. But, the design and modelling
of all current calculation literature, standards and software is based on the involute
shape. That’s why, the involute shape is used in the following investigation. The tooth
in plastic gears can fail in different ways [5, 6]. Figure 2 shows the important failure
modes patterns. The following applies:
– Melting: The melting temperature of the material at the teeth is exceeded; more heat
is added than dissipated. The teeth temperature rises steadily. Due to the temper-
ature increase in the plastic teeth, the mechanical strength decreases continuously
until it is too low for the transmission of force. The teeth soften or pushed away
within a very short time.
– Tooth root fracture: The bending stress is too high at the tooth root: If the bending
strength of the plastic tooth is too low, bending fracture occurs outside the wear
channel, where the flank does not come into contact with the opposite flank.
306 H.-J. Dennig et al.
– Deformation: This occurs in both static and dynamic cases: Static: Deformation is
caused by the flow of the material. The long-term modulus of elasticity is exceeded
when a static load is applied. Dynamic: The short-term modulus of elasticity is
exceeded. In both cases, the tooth head play is reduced and can lead to clamping of
the gear stage.
– Delamination: The laminates created in the production process separate due to the
excessive tangential force on the tooth flank. Additionally, the temperature at the
tooth flank can exceed the melting temperature of the matrix, which can also cause
delamination.
– Pitting: Hertzian pressure is too high with superimposed shear stress; pitting is the
result of material fatigue in the near-surface area. The pitting is also caused by the
shear stresses and the thermal stresses due to the sliding of the flanks [7].
– High wear: Low wear resistance or a highly abrasive surface of the friction partner;
caused in part by chipping and in part by abrasion. This is especially the case on the
tooth root and head, where the highest sliding speed occurs. Wear when mating the
same plastics with each other is particularly high, especially on gears reinforced
with short fibres.
– Tooth fracture on the flank: Crack formation due to Hertzian pressure and shear
stresses at the wear zone, the crack grows into a fracture due to the bending stress.
The design of plastic gears, especially in Europe, is based on VDI guideline 2736
[6]. This compares the material-specific and temperature-dependent permissible tooth
root stress rFlim based on a Woehler-curve with the stress rF that occurs. In addition, the
wear coefficient kw is calculated based on tooth flank distance before and after the tests.
Fig. 2. Potential modes of failure for plastic gears (according to [5, 6, 7]): a-Melting, b-tooth
root fracture, c-wear, d-pitting, e-delamination, f-deformation, g- tooth fracture on the flank
To determine fatigue strength under pulsating stress rFlim Eq. (1)2 the number of
load cycles before tooth fracture are counted [8]: b = face width, Ft = nominal tan-
gential force, mn = normal module. The damage patterns of pitting and tooth fracture
on the flank are more likely to occur with high-strength plastics such as
Polyetheretherketon (PEEK). Therefore, the rolling contact-fatigue-strength rHlim and
thus the Hertzian stress will not be discussed further here.
YSa Ft
rFlimN ¼ YFa Y ð1Þ
YSt b mn
To determine wear coefficient kw, wear is measured after a fixed number of roll-
overs NL. In this investigation, the abrasion coefficient kw is calculated via the local
linear wear Wlokal measured at the flank Eq. (2): Fn,lokal = normal force on the tooth at
the contact, f = local specific gliding.
Wlokal b 1
kw ¼ ð2Þ
NL F n;lokal f
The limit for Wlokal is defined as 20% of the modul mn.

The design of plastic gears, compared to steel gears, is generally challenging due to
the temperature-dependent young’s modulus. VDI 2736 describes the procedure for
designing plastic gears and also contains specific material data. However, since plastics
from different manufacturers have different material properties, despite having the same
name, these specific properties or values rarely exist. In addition, there are different
plastic compositions and so-called fiber additives that modify properties. In the field of
additive manufacturing of plastic gears there are no reference values for a design.
Skawiński (et al.) [9] have investigated the effects of layer heights, layer structure
and numerous other print parameters on resulting mechanical properties of gears.
However, this was not on the basis of an involute form or with regard to a dynamic gear
test. Thus, there are no published research results on FFF-printed and tested gears.
3 Test Samples
3.1 Gearwheel Geometry

The gear geometry of the FFF plastic gears is based on size 2 described in VDI 2736
Part 4 [8], which is slightly modified due to better tooth contact (Table 1). A feather
key connection is used as wheel-hub connection Fig. 3.
2
The values for the Y-factors are defined in the [8].
Table 1. Gear dimensions

a
Symbol Value (FFF-Gear; Steel-Gear)
Centre Distance a 60 mm
Normal module mn 2 mm
Tooth number z1; z2 31; 29
Face width b1; b2 8 mm; 13 mm
Tip diameter da1max; da1min 66 mm; 65.97 mm
Tip diameter da2max; da2min 62 mm; 61.97 mm
Root diameter df1max; df1min 56.85 mm; 56.75 mm
Root diameter df2max; df2min 52.9 mm; 52.83 mm
Tip rounding radius 0 mm; 0.6 mm
Pressure angle an 20°
Helix angle b 0°
Addendum modification x1, x2 0 mm
Tooth Quality > 10e (DIN 58405); 7 (DIN 3961)
h*aP 1
h*fP 1.25
q*fP 0.38
a
Index 1 = FFF-Gear; Index 2 = Steel gear
Fig. 3. Gear wheel geometry with feather key connection
3.2 Material and Production

All gears were printed horizontally without support structures (Fig. 4). This ensures
that the contour lines are as precise as possible. In addition, there is no tensile stress
between the layers in the area of the tooth root.
Fig. 4. Example of a Craftbot section in the slicing software
Three materials commonly used on FFF printers were examined: PLA, ABS and
PA66 (Nylon) (Table 2, Table 3). With PA66, it is possible to integrate endless carbon
fibers in the printing process: The Markforged dual printer Mark two, which can
process various additional materials to reinforce components via a fiber feed. For this
reason, the material PA66 was additionally processed with endless reinforced carbon
fibres (PA66 + CF). The Craftbot+printer is a low-cost PLA printer that has very good
quality results in its price segment. This compares to the professional ABS printer from
Stratasys F270.
Table 2. Overview of materials (the values are based on the information provided by material
suppliers)
Material Tensile/Flexural strength at Youg Heat deflection
yield/MPa modulus/MPa temp./°C
PLA [10] 26/45 2539 63
ABS (plus-P430) [11] 31/35 2200 96
PA [12] -/51 1700 41
PA+CF [12] -/540 60000 105
Table 3. Printers and print parameters used

Material Printer Extr. Layer Contourlines Top/Bo. Infill
width/mm h./mm layers
PLA Craftbot + 0.4 0.2 2 4 75%
ABS Stratasys F270 0.3556 0.25 2 2 22%
PA Markforged 0.4 0.125 2 4 37%
PA+CF Mark Two
The gear geometry from Table 1 is printed with the settings shown in Table 3.
A comparison of the print results is shown in Fig. 5 and Fig. 6.
Fig. 5. Printed gear wheel examples PA66 + CF (left) and PLA (right)
Fig. 6. Section view of a ABS-Gear with two contour lines.
4 Quality
Figure 7 shows the flank of different printed gears with various layer resolutions.
A striking feature of ABS and PA66 is the so-called wrapping. All the gears were
printed without raft. Also pictured is the so-called oozing on the flank surface of
PA66 + CF, which is due to the different materials PA and CF. However, the quality of
the surfaces can be rated as very good for all materials after several optimization loops
of the printing parameters (Fig. 8).
Fig. 7. Flank of different gears: PLA, ABS, PA, PA + CF (from left to right), z = vertical
printing axis
Fig. 8. Top-Layer: Printing results of the gears: PLA, ABS, PA, PA + CF (from left to right)
All the wheels were analysed on a double-flank rolling test device (Fig. 9). The
results indicate the base tangent length, the tooth-to-tooth error and, in particular, the
concentricity.
Fig. 9. Double flank gear tester (left), disc micrometer (right)

According to DIN 3960, the tooth width Wk was measured with a disc-micrometer
at three different positions (Fig. 9) and can be compared to the values as in Table 1.
The quality of plastic gears is defined in the DIN standard 58405. This standard can
be set in the software of the double flank gear tester. Injection-moulded plastic gears
made of POM, for example, have a standard quality of 8 to 10 (out of 12, whereas 1 is
the best). However, this quality can only be achieved by complex tool optimisation.
Concentricity is a particularly important factor here. Poor concentricity leads to local
high pressure on the tooth flank, which in turn leads to increased temperatures in
dynamic operation. Table 4 shows the measurement results of the gears. At least three
wheels for each material were measured.
As expected, the values of all the printed gears were outside the standard. The high-
quality Stratasys printer showed the best geometrical results, followed by the low-cost
Craftbot printer.
Based on the measurement in Table 4, the tolerances of the steel counter gear can
be defined and the steel gear can be produced. The tolerance adjustment on the steel
wheel prevents the gear pair from jamming. The steel counter gear was manufactured
on a DMG Mori NTX 2000sz turning/milling centre with Euklid software.
Table 4. Measurement results from double flank gear tested and disc micrometre: Wk1 = base
tangent length, Fr’’ = the roundness, fi’’ = the tooth to tooth error; Quality 10e SET
All values Fr’’ Fr’’ St. fi’’ fi’’ Wk1 Wk1 SET (max -
in mm ACTUAL SET Deviation ACTUAL SET ACTUAL min)
Fr’’
PLA 0.264 0.052 0.081 0.263 0.037 27.161 27.337 − 27.269
ABS 0.128 0.047 0.128 27.356
PA 0.405 0.113 0.405 27.120
PA+CF 0.674 0.125 0.674 27.477
5 Test
5.1 Test Rig
The test setup is a non-mechanical closed loop test rig (Fig. 10, Fig. 11). The inves-
tigations were executed in the combination between gear 1 (plastic) and gear 2 (steel)
and with a centre distance of 60 mm (also Table 1). Gear 1 is the driving gear. All tests
were driven in dry condition.
Fig. 10. Plastic gear test rig
IR-Thermo camera
TU-Sensor
plastic gear Steel gear
Fig. 11. Gearbox
To calculate the material properties, the temperature to be evaluated was taken at

almost the end of the investigation cycle. The tooth flank/surface temperature TS and
the body temperature TB during operation were measured with a thermal infrared
camera (Fig. 12). TB can be regarded as equivalent to the temperature of the tooth root.
The room temperature TU is controlled by a sensor closed to the gears, so that the TB
stays constant depending on the induced power.
TS
direction
of rotation
TB
Fig. 12. Principle of the temperature measurement

All test results were evaluated according to the specifications of VDI 2736 Part 4.
At least three gears per load were tested. At 1 or 2 ∙ 106 load cycles respectively the
test was aborted. At pre-determined intervals inspection for possible damage (pitting,
tooth fracture) was carried out. Once the type of damage was defined, the torque value
of the further tests was selected depending on the number of load cycles to be run. The
drive speed was always 1500 min−1. The plastic gears were not conditioned, but were
tested within three days of manufacture.
5.2 Results
Figure 13 shows the results of the fatigue tests. The incoming torque Tan is converted
to a tooth root stress rF via the pitch circle and the gear width, as in Eq. (1). This
allows comparisons to be made with other research results.
30
Test results 3, PA66+CF, 55 °C
σF without Lubrication 4, PA66+CF, 55 °C
/MPa z = 31
m = 2 mm
25 b = 7 mm
test interrupted, load cycles achieved
2, PA66+CF, 50 °C
20 67, ABS-V2, 35 °C
54, ABS, 35 °C 66, ABS-V2, 35 °C 1, PA66+CF, 50 °C
55, ABS, 35 °C
68, ABS-V2, 35 °C
12, PA66, 41 °C
15
53, ABS, 35 °C 63, ABS, 35 °C
13, PA66, 41 °C
24, PLA, 65 °C 59, ABS, 35 °C

10 14, PA66, 41 °C
62, ABS, 35 °C 51, ABS, 35 °C
50, ABS, 35 °C
3, PLA, 65 °C 25, PLA, 65 °C 52, ABS, 35 °C
5
1.E+04 1.E+05 number of load cycles 1.E+06
Fig. 13. Result diagram of the fatigue test
For certain gears, the tooth width was measured after a pre-determined number of
cycles and based on this, the wear coefficient was determined according to formula (2)
(Table 5).
Table 5. Wear results

ABS-1 ABS-2 ABS-3 PA66 – 1 PA66 - 2 PA66-CF-1 PA66-CF-2
Fn,Lokal/N 51.4 51.4 51.4 51.4 51.4 85.7 102.9
f 0.1578
Wlokal/mm 0.11 0.10 0.11 0.01 0.02 0.06 0.10
NL 157500 175000 130000 2008500 2008500 2008500 2041500
Ø-kw/mm3/ 1051.7 11.0 36.3
Nm/106
After fatigue test and wear measurements, the following key statements can be
made with respect to the load cycles:
PLA-Gears: The tooth root temperature TS exceeds the heat deflection temperature of 63
°C during the measurements. This causes the teeth to melt away. PLA wheels thus fail on
melting (Fig. 14). The filling structure therefore has no influence on the lifetime of the
gear. The tooth flank temperature is responsible for this type of failure due to melting, and
can only be changed by the accuracy of the teeth and the material. It is possible that the
printed layer height effects the surface and thus the temperature. However, since the
number of cycles was very small, no variations were carried out here. Measurements
regarding wear could not be made because all the gears failed within a very short time.
Fig. 14. PLA Gears with melted teeth

ABS-Gears: The tests with ABS gears were carried out with the steel counter wheel,
which was specifically designed for the PLA wheels. The ABS gears failed at all
measured torques owing to tooth-root fracture. The test rig cannot detect single tooth
breakage, which is why several teeth were chipped off on some gears (Fig. 15). The
tooth root temperature was around 35 °C in all tests.
In further tests the number of outline layers was increased from two to four, which
led to slightly higher cycle numbers under equal loads at approx. rF = 16 MPa (ABS-
V2 in Fig. 13).
One noticeable result was the very high wear. It can be postulated that the print
layers on the flanks are smoothed at the beginning of the tooth engagement, which
causes the tooth form to be reduced more than through normal wear. The local wear
Wlokal is determined by the tooth width before and after the test. kw is accordingly very
high.
Fig. 15. ABS gears with broken teeth
PA66-Gears: The tests of PA66 gears were stopped after a maximum of 2 million
cycles. Apart from signs of wear on the tooth flank, no damage was visible (Fig. 16).
The tooth root temperature was around 41 °C in all tests.
Here too, the wear is rather high in comparison to the usual wear values of kw = 1–
7.8 mm3/Nm/106 [8].
Fig. 16. PA66 gear after two Mio. cycles at a root stress rF = 12 MPa
PA66-CF-Gears: The tests of PA66-CF gears were stopped after a maximum of 2

million cycles. Apart from signs of wear on the tooth flank, no damage was visible.
Since the load and thus the tooth root stress was increased again, the temperatures rose
up to 50 °C. Above the tooth root stress of rF = 23 MPa the gear exceeded a tem-
perature of approx. 60 °C and began to deform plastically. The teeth were bent
(Fig. 17).
In further tests, the tooth root load was increased to approx. rF = 31 MPa, which
caused the gears to fail (Fig. 18). However, it cannot be clearly determined whether this
was due to melting or tooth root fracture.
Fig. 17. PA66-CF gears with deformed teeth and wear on the flanks
Fig. 18. PA66-CF gear with broken or melted teeth
Here too, the wear is rather high in comparison to the usual wear values of kw = 1–
7.8 mm3/Nm/106 [8].
6 Conclusion
The tests carried out provide a preliminary overview of the load capacity and use of
additive manufactured gears made of PLA, ABS, PA66 and PA66+CF by the FFF
process. It shows that gears made of PLA are not suitable due to the high heat gen-
erated by contact of the tooth flanks. While ABS gears can be used for low loads, gears
made of PA66 achieve reasonable cycle times even with high tooth root stress.
However, in order to be able to make reliable statements about the material properties
in tooth contact, Woehler curves must be created at constant tooth root temperatures.
Two Woehler curves at tooth root temperatures, set far apart, are sufficient.
Due to the heat deflection temperatures of the various thermoplastics, a sufficiently
robust cooling system must be used. With current resources this can be done, for
example, with nitrogen cooling, which is available on the above-mentioned test stand.
The cooling must be controlled based on the heat generated at the tooth flank contact
and measured by infrared camera.
All manufactured gears tend to be produced with larger outer diameters and smaller
inner diameters at the wheel-hub connection. This means that the contour lines can be
optimized to enable printers to produce standard-compliant gears. Results demonstrate,
however, that although the geometric print quality is poor in comparison to injection-
moulded gears, the number of load cycles achieved is high, especially with PA66. It
can therefore be assumed that the flank properties with a rough surface have less of an
influence than initially expected. Since ABS wheels fail on tooth breakage, an opti-
mization of the number of outline layers lead to an increase in strength.
The numbers of cycles and the loads that were run with PA66-CF and PA66 were
compared with values of injection moulded PA66 gears from the literature. Compar-
ative values exist in the area of low loads. These show that the PA66-CF wheels
certainly have similar cycle numbers compared to injection moulded PA66 gears. For
higher ranges, in particular rF > 10 MPa, it is not possible to compare them directly,
as values for the tooth root temperature do not exist for the same loads. Therefore, tests
with injection moulded PA66 gears need to be carried out and compared.
The wear values measured in the test cannot be compared with injection-moulded
gears. It can be postulated that the print layers on the flanks are smoothed at the
beginning of the tooth engagement. In order to obtain meaningful results, tooth width
measurements must be taken during the tests: for example, after 100,000 and after
1,000,000 cycles.
In further tests, the aim is to determine the characteristic curves of PA66(+CF)
gears at different temperatures and to conduct wear measurements which provide valid
results. These material characteristics can then be used in calculation software such as
KISSsoft.
References
1. Brecher, C., Löpenhaus, C., Greschert, R., Poprawe, R., Schleifenbaum, J.H., Ziebura, D.:
‘Potenzial generativ gefertigter Bauteile in der Getriebetechnik’, in Antriebstechnisches
Kolloquium ATK. Aachen 2019, 251–264 (2019)
2. Dennig, H.-J., Winterberg, S.: Test stand for internally cooled metal gears. In: International
Conference on Gears 2019, Garching/Munich, Germany, vol. VDI-Berichte 2355, pp. 1263–
1265 (2019)
3. ‘Selektives Lasersintern (SLS) Material|igus®’. https://www.igus.ch/3d-print-material/
selektives-lasersintern-material. Accessed 17 Feb 2020
4. Gebhardt, A.: Additive fertigungsverfahren: additive manufacturing und 3D-Drucken für
prototyping – tooling – produktion, 5th edn. Carl Hanser Verlag GmbH & Co. KG,
München (2016)
5. Roesler, J.: ‘Zur Tragfähigkeitssteigerung thermoplastischer Zahnräder mit Füllstoffen’,
Doctoral Thesis, Technische Universität Berlin, Fakultät V - Verkehrs- und Maschinensys-
teme, Berlin (2005)
6. Verein Deutscher Ingenieure VDI: 2736-1: Thermoplastische Zahnräder - Werkstoffe,
Werkstoffauswahl, Herstellverfahren, Herstellgenauigkeit, Gestalten. Beuth Verlag, Berlin
(2014)
7. Dennig, H.-J., Monn, S., Vodermayer, A.: Thermoplastic high performance composite gears.
In: 3rd International Conference on High Performance Plastic Gears 2019,
Garching/Munich, Germany, vol. VDI-Berichte 2355, pp. 1279–1290 (2019)
8. Verein Deutscher Ingenieure VDI: 2736-2: Thermoplastische Zahnräder - Ermittlung von
Tragfähigkeitskennwerten an Zahnrädern. Beuth Verlag, Berlin (2013)
9. Skawiński, P., Siemiński, P., Błazucki, P.: Applications of additive manufacturing (FDM
method) in the manufacturing of gear. Mechanik, no. 12, pp. 976/173-976/179, December
2015. https://doi.org/10.17814/mechanik.2015.12.582
10. Stratasys: ‘Datasheet PLA Filament’ (2020). https://www.stratasys.com/de/materials/search/
pla
11. Stratasys: ‘Data sheet ABS Filament’ (2020). https://www.stratasys.com/de/materials/search/
abs
12. Markforged: ‘Material Datasheet Composites - REV 3.2 - 9/9/2019’. http://static.
markforged.com/downloads/composites-data-sheet.pdf
Qualification of Additively Manufactured
Blood Vessel Models for the Evaluation
of Braided Stent Designs
Juliane Kuhl1(&) , Ngoc Tuan Ngo2 , Jan-Hendrik Buhk2 ,

Andreas Ding3 , Andrés Braschkat3 , Jens Fiehler2 ,
and Dieter Krause1
1
Hamburg University of Technology, Hamburg, Germany
juliane.kuhl@tuhh.de
2
University Medical Center Hamburg-Eppendorf, Hamburg, Germany
3
Acandis GmbH, Pforzheim, Germany
Abstract. For testing new designs of braided stents, an animal free anatomical
model is needed, in which different Flow Diverter (FD) designs can be
implanted and evaluated. The geometric replication of vessels by additively
manufactured (AM) models has already been proven for aneurysm coiling
within the neurointerventional training model HANNES. Nevertheless, since
FDs closely interact with the vessel wall, AM vessel model materials need to be
qualificated for the use case. Therefore, this paper presents an appropriate test
bench for AM material tests and characterizes AM impact factors. As an out-
look, different AM materials are tested and compared. Additionally, the impact
of the postprocessing is investigated for one material.
The paper shows the qualification of AM models for animal free design tests
of a medical devices, where an objective test environment gives the opportunity
to qualify AM materials for the medical use case.
Keywords: Additive manufacturing (AM) Medical device validation

Product development Biomedical design AM material test setup
1 Introduction
Due to the great freedom of geometry in additive manufacturing (AM), it is possible to

reproduce complex anatomical structures in a physical AM model. These anatomical
AM models can then be used to validate medical devices with regard to their intended
use and usability. In this way, by using AM, a reliable assessment about the geometric
suitability of a medical device design can be carried out repeatedly under constant
conditions without using animal experiments.
However, in order to be able to make a targeted and reliable statement about the
quality of the design of a medical device, AM models have to behave as exactly as
possible to reality. Depending on the type of product, the medical device interacts
differently with the AM model. This interaction must be investigated and brought into
line with reality before the AM model can be used for design validation tests.

https://doi.org/10.1007/978-3-030-54334-1_23
322 J. Kuhl et al.
This paper presents the investigation on AM models, more precisely the AM

materials, which shall be used for testing different Flow Diverter (FD) designs for the
treatment of aneurysms. A FD is a stent like tubular braided structure which is used for
intracranial aneurysm treatment by positioning the implant inside the diseased vessel.
After a short introduction to medical device validation basics and the medical device of
the FD, the validation environment HANNES with its AM vessel models is presented
in Sect. 2. Afterwards, the research method for this paper is proposed (Sect. 3). Fur-
thermore, in Sect. 4.1, the developed test setup for comparing different AM materials
and post processing procedures of AM is introduced. Impact factors which can be
emphasized and first test results are explained in Sect. 4.2 and 4.3. After the summary
of the results and a discussion of limitations (Sect. 5), a conclusion and outlook are
given (Sect. 6).
2 Background
As a background for this work a short introduction to the design validation of medical
devices is given. In addition, the medical device of the Flow Diverter (FD), is intro-
duced. Afterwards, a validation environment for FD designs is presented, in which AM
blood vessel models are used.
2.1 Design Validation for Medical Devices

In Summary and according to the Regulation (EU) 2017/745 of the European Parlia-
ment [1], a medical device is defined as any instrument that can be used for the
detection, prevention, monitoring, treatment or alleviation of disease and that does not
produce its effects primarily by pharmacologically or immunologically active means. In
order to be approved for the European market, a medical device has to bear a CE mark.
To obtain this, a declaration of conformity has to be issued by the medical device
manufacturer, where he proves that the basic requirements for the safety, performance
and effectiveness of the medical device have been met [2].
This paper deals with the technical evaluation of medical devices, where especially
safety issues concerning product behavior and usability are examined. This includes
product verification and validation. While design verification confirms the fulfilment of
the previously defined product requirements, the design validation includes both the
examination whether the intended use can be achieved with the product (product
validation) and the testing whether a specified user can apply the product in such a way
that the intended use is fulfilled (usability validation) [3] (see Fig. 1). Besides the
introduction of a quality management system, technical product tests and risk assess-
ments are essential parts of the technical product validation [2].
Qualification of Additively Manufactured Blood Vessel Models 323
Fig. 1. Medical Device Development Process with special Emphasis on Design Verification and
Validation (according to [3])
By means of a target-oriented design validation, product risks can be identified and

eliminated at an early stage. For executing a design validation, a high degree of user
involvement is necessary [4]. Practical validation tests are often carried out for iden-
tifying product as well as human risks. Thereby, prototypes can be tested and rated by
the user in adequate test environments. Hence the test environment has a huge impact
on the evaluation results, the development of a realistic test environment is of great
importance. Thus, a meaningful result can be obtained and possible risks can be
uncovered as completely as possible.
2.2 Flow Diverter – Medical Device for Aneurysm Treatment

Intracranial aneurysms (IA) are pathological bulges in blood vessels that can lead to
life-threatening bleeding in the event of a rupture, most commonly subarachnoid
hemorrhage (SAH). For the treatment of IA, surgery or multiple different therapeutic
modalities of endovascular treatment are available such as Coiling, Ballon-Assisted
Coiling, Stent-Assisted Coiling, Flow Disruption and Flow Diversion with a FD (see
Fig. 2) [5].
Fig. 2. Schematic Representation of a Blood Vessel with Aneurysm and Implanted FD
For implanting a FD, a guide catheter is advanced through an access in the femoral
artery through the vascular vessel tree up to the cervical artery often followed by an
intermediate catheter. Afterwards a microcatheter is advanced over an microwire at the
324 J. Kuhl et al.
distal landing zone. The FD mounted on a transport wire is then inserted through the
hollow structure of the microcatheter. The FD can be released in the vessel by pushing
it out of the microcatheter with the help of a transport wire. It is possible to retrieve the
FD back into the micro catheter up to a certain point.
By placing the FD in the diseased vessel, the blood flow can be diverted from the
aneurysm into the parent vessel while inducing vessel reconstruction through neoen-
dotheliziation. This subsequently leads to thrombosis of the aneurysm, over time
shrinkage of the aneurysm and finally exclusion of the aneurysm from the circulation
[6]. For this to happen, however, the FD must fit tightly against the vessel wall. Proper
vessel wall apposition is mandatory to promote sufficient endothelialization along the
implant and thrombosis of the aneurysm. On the other hand, poor vessel wall appo-
sition can lead to persistent blood flow into the aneurysm due to endoleaks and
thrombembolic complications due to compromised neoendotheliziation [7]. Moreover,
poor vessel wall apposition is associated with device migration and shortening resulting
in insufficient stent coverage of the diseased vessel and potential delayed aneurysm
rupture [7, 8].
For this reason, newly developed FD designs must be examined carefully, partic-
ularly with regard to their apposition in the vessel and the final wall adaptation
achieved. To validate device design and prove usability, an animal-free validation
environment is desirable that allows realistic testing with reliable evaluation of the
designs.
2.3 Design Validation Within a Neurointerventional Training Model

The neurointerventional training model HANNES, as shown in Fig. 3, was developed
to train neuroradiologists in the catheter-based, minimally invasive treatment of vas-
cular diseases like aneurysms [9, 10]. Therefore, HANNES simulates the whole rele-
vant vessel tree including, among others, the aorta, carotid and cerebral vessel models.
Due to the modular design of the entire model, certain components of the vascular tree
can be exchanged in order to depict different treatment scenarios. [9, 10] Further
features, like the simulation of a human circulation by a fluid system, a tempered blood
substitute, the integration of HANNES in a angiography facility or the application of
original instruments, like catheters or implants, [10] let HANNES become a suitable
Fig. 3. Hamburg Anatomical Neurointerventional Simulator HANNES with Patient-Specific

Cerebral Vessel Models with Aneurysm
environment for the design validation of neurovascular medical devices. In the present
case, different FD designs are to be examined for their suitability, handling and
usability for aneurysm treatment. For this purpose, patient-original blood vessel models
of anatomies treated with a FD have to be developed and integrated into HANNES.
2.4 Additively Manufactured Blood Vessel Models

Aneurysm models used in HANNES (Fig. 3) are additively manufactured. Up to now,
it has proven to be the best method for the production of the geometry of blood vessel
models [11, 12]. AM models offer a great freedom of geometry and a flexible pro-
duction also for small quantities, so that patient-specific anatomies can be simulated
appropriate by a hollow AM model [10]. To do so, patient data is recorded and
segmented. Specifications such as adapters are then added to the anatomy and a layer
model is generated, before additive production takes place. After the post processing,
the AM aneurysm model can be used within HANNES. Spallek et al. [10] introduce the
standardized individualization process (Fig. 4) to manufacture patient specific aneur-
ysm models within a standard procedure.
Fig. 4. Standardized Individualization Process (according to Spallek et al. [10])
For the AM of a blood vessel model, stereolithography is used. Here the AM model
is built up layer by layer from a liquid synthetic resin that hardens under UV light [13].
In our use case, the Form 2 and Form 3 printers of Formlabs, MA, USA are used. For
the aneurysm models, the materials Flexible (black color, flexible behavior) and Clear
(transparent, rigid behavior) have been used so far. However, Formlabs has launched a
new material called Elastic, which is elastic and transparent at the same time and
therefore is very promising for using it for blood vessel models. Nevertheless, further
investigations are necessary. The possible layer thicknesses during printing depend on
the material: Clear can be printed in a layer thickness of 25 lm, 50 lm or 100 lm;
Elastic only in a layer thickness of 100 lm and Flexible can be planned in thicknesses
50 lm and 100 lm. Before printing, a layer model with support structures has to be
prepared. After the printing process, the materials are washed in isopropanol using the
Form Wash device and cured afterwards in the Form Cure device (see Fig. 4). After the
removal of the support structures, the AM model can be used.
326 J. Kuhl et al.
For evaluating the usability of new FD designs, AM blood vessel models shall be
used. However, since the interaction of the FD with the vessel wall is of high
importance, AM materials must be qualified for FD design tests to provide a realistic
and reliable validation environment for FD design tests.
3 Research Method
The aim is to capture the interaction between different AM materials and a FD. Thus, a
suitable material for validation tests for FD designs executed in HANNES should be
determined. In a first step, to qualify the AM materials for the replication of blood
vessel anatomies, an objective test setup has to be developed. Thus, the interaction
between the FD and the AM materials should become measurable.
To do so, test setups are researched in which the interaction between an
endovascular medical device and a certain material is investigated. On this basis,
requirements for the test setup for FD tests can be defined. According to the engi-
neering development process [14], an own test setup concept is developed and
implemented. Derived from the process steps of AM, possible impact factors are
emphasized and a test plan was created. In addition to the test setup, the test procedure
was defined. To increase the reliability of the test results, two identical tubes of each
material configuration shall be tested. Furthermore, each tube is tested up to three
times. The evaluation of the recorded test results is done with the software Diadem, as
presented in detail in Sect. 4.3, where the force-path respectively force-time diagrams
are evaluated qualitatively.
4 Development of a Test Setup for AM Material Qualification
In this section, the developed test setup as well as the test procedure is explained. The
examination on impact factors concerning the AM process is described. Finally, an
outlook on first test results are presented.
4.1 Development of the Test Setup and Test Procedure

In order to reflect the interaction between a neurovascular medical instrument and a
certain material, different experimental setups, such as those by Desai [15], Uyama [16]
or ZwickRoell [17], measure the frictional force that arises when pulling a medical
device, in most cases a catheter, through a hollow structure of a defined material
(Fig. 5).
Fig. 5. Friction Model and Selected Concept for Measuring the Interaction Between a FD and
an AM Blood Vessel Model
Based on the research, setup requirements are defined and an experimental setup for
measuring the interaction between a FD and an AM blood vessel model is developed
(Fig. 5). Although the FD is not pulled through the vessel during a treatment, the vessel
wall interaction is of high importance. Since a test setup, in which different catheter
designs are pulled through a fixed material, has been established for the evaluation of
the haptic of catheters, the setup is to be adapted for AM material qualification for FD
design tests: The FD is implanted with a gliding movement in which the FD is
gradually pushed out of the introducer from the distal to the proximal end and thus
unfolded in the vessel. It is important that the positioned distal end does not move
during or after the implantation procedure (see Sect. 2.2). As soon as a breakaway
force characteristic for each material pairing is overcome, the FD moves in the vessel.
This breakaway force is to be investigated in a suitable setup to choose a material that
has approximately the same small breakaway force as a human blood vessel. In order to
make this statement, physicians must evaluate different materials based on their
experience, so that a desirable, realistic range for the breakaway force can be defined
and AM test results can be compared with this range of values. Physicians can already
perceive very small force values with their medical treatment instruments. The testing
of a human blood vessel is not carried out for ethical reasons. With these thoughts, the
following test setup is developed.
To quantify the interaction between a FD and AM materials, a FD is released
20 mm from the introducer and pulled with a constant speed of 200 mm/min through a
straight, AM test tube of 100 mm length. The force is measured by the tension-
compression force sensor KD24s ±10 N. It has a nominal force of 10 N and an
accuracy of 0.1%, as very small friction forces are expected.
The test tubes are made of the materials by Formlabs and have an inner diameter of
3.5 mm and a wall thickness of 1 mm. To fix and position the tube, an adapter is
attached to the tube digitally and printed directly during production. The counterpart of
the adapter, made of Clear, is glued in the center of a transparent container, filled with
up to 4.4 L of water as blood substitute. The superelasticity of the FD material is
adjusted in such a way that constant conditions can be assumed in a temperature range
of more than 30 °C and therefore small deviations from body temperature (37 °C) are
acceptable. Therefore, the heating element Dennerle Thermo Heater Compact (25 W)
is set to its maximum value of 36 °C A thermometer is used to measure the water
temperature.
328 J. Kuhl et al.
In order to realize the movement, the experimental setup is setup in a Galdabini

tractor and appropriate connectors are selected. To fix the introducer to the moving
crosshead, a mounting bracket is constructed in which the introducer can be clamped
vertically. Figure 6 shows the described test setup.
Fig. 6. Test Setup for Investigating the Interaction Between a FD and an AM Model
4.2 Considerations on Possible AM Impact Factors

When applying additive manufacturing, various impact factors have to be taken into
account. As described in Sect. 3, the AM process includes the preparation of a layer
model with support structures, the printing process itself as well as the washing and
curing of the model.
In the preparation of the layer models, the orientation of the test tube and the
execution of the support structures need to be the same for all test tubes, independent of
the AM material. The models are arranged in the printing platform at an angle of
approx. 30°, with the adapter pointing upwards. This is to exclude any influence of the
manufacturing on the results of the test.
The printing process takes place under the same conditions each time. However,
different materials are chosen, so that tubes are made of Clear, Flexible and Elastic
material. To avoid any influence of the layer thickness and make materials comparable,
all test tubes have the same layer thickness of 100 lm.
The washing process is always supported by the Form Wash device, which uses
isopropanol (90%). For all materials, the washing process is executed according to
manufacturer’s specifications. The curing in Form Cure always takes place according
to manufacturer’s specifications.
Additionally, to understand the impact of the washing process, the washing dura-
tion (2 + 3 min, 5 min) and frequency (one or two washing processes) is varied while
the material, here Elastic, is kept constant (Fig. 7).
The duration of the curing process was also varied for the Elastic material (4 h)
(Fig. 7). In addition, it is tested whether wetting the inner surface of the tube with
mineral oil during the curing process has an influence on the interaction. The mineral
oil is intended to remove the isopropanol absorbed during the washing process from the
material and enable better and more complete curing.
Fig. 7. Investigation on Impact Factors of the AM Postprocessing (Washing and Printing)
In addition to AM material comparison and investigations on post processing

procedures, the test setup can also be used to evaluate coatings that are applied to the
surface of the manufactured AM model.
4.3 Remarks on the Analysis and Interpretation of Measured Data

The analysis of the measured data is done with the software Diadem by National
Instruments, Texas, USA. To better evaluate qualitatively the test results recorded at a
sampling rate of 100 Hz, the smoothing function of the program is used and the 16
neighboring values left and right of the channel value are smoothed around the average.
In order to have a common endpoint at 0 N, the curves were shifted. Since the
transport wire of the FD touches the bottom of the container at the beginning of each
experiment, the starting force is not equal to 0 N. After the FD has been completely
pulled through the tube, both the FD and the transport wire are suspended in the air, so
that a force equal to 0 N can be assumed. For the same reason, the tube-entering phase
(compare Fig. 8 and 9) is not relevant for the evaluation of results. Here the FD and the
transport wire must first align themselves until they are completely inside the tube. Due
to the elasticity of Elastic and Flexible material, the tubes are not perfectly straight.
Therefore, also the introducer may also be in contact with the tube wall. This leads to
force peaks, which are dependent on the orientation of the AM model. This is proven
by turning one tube between two tests - the force peak changes considerably (Fig. 9).
Since the force peaks are mainly caused by the test set-up, it is not appropriate to
evaluate the tests on the basis of the force peak.
330 J. Kuhl et al.
According to the considerations in Sect. 4.1, an evaluation of the measured results

should be made according to the breakaway forces. These can be seen in the oscillation
around the average force value, to which the force approximates during the uniform
movement in the tube. After developing the test setup, the suitability of the test rig was
first examined. For this purpose, a FD was pulled through a tube made of elastic
material. This test was initially carried out without a liquid bath and then repeated in a
tempered water bath. Since no lubricant is present in the dry state, a more pronounced
oscillation around a significantly higher average force value is expected due to a strong
slick-slip effect. These expectations could be verified in the test setup, as shown in the
load-stroke respectively load-time diagram in Fig. 8, where one curve shows a test run
with one tube. Therefore, the test setup seems to be fundamentally suitable for mea-
suring the interaction between the FD and the test tube.
Fig. 8. Validation of the test setup by tests in dry and wet environment
4.4 Outlook on First Results of AM Material Tests for Blood

Vessel Models
With the developed test environment different AM materials can be investigated. In this
chapter first qualitative results will be shown. In a first part of the AM material
investigation, the materials Clear, Flexible and Elastic of Formlabs are compared. For
this purpose, tubes with a printing layer thickness of 100 lm and a post-processing
(washing and curing process) according to the manufacturer’s specifications are
examined in the test setup. Figure 9 shows qualitatively the measured results in a load-
stroke respectively load-time diagram, where one curve shows a test run with one tube.
The tube designations in the legend show which tube is used in the test run.
The oscillation around the average force value in the sliding friction phase is the
lowest with Clear, slightly higher with Flexible and most pronounced with Elastic
material. This underlines that the breakaway force is the highest in the elastic material.
This can also be observed visually with the transparent Elastic - the FD gets stuck in the
tube for a short time and slides along jerkily. However, as stated before, in order to
finally interpret the AM material test results concerning the suitability for simulating
human vessels haptics the subjective opinions of physicians are needed, who rate the
AM materials according to their experience in treating human vessels. By combining
the values measured in the newly developed test and the subjective opinions of
physicians, a suitable material can be selected for the blood vessel models to be used in
a FD design validation at HANNES, without the physicians having to evaluate every
material and post processing procedure.
Fig. 9. Comparison of the Results of Clear (orange), Elastic (blue) and Flexible (grey) According
to the Position of the FD Within the Test Tube
Besides a realistic behavior, physicians prefer elastic and transparent properties for
the AM vessel models. Since the Elastic material fulfils both properties, the influence of
the manufacturing process is investigated. In first tests, it could be determined that the
332 J. Kuhl et al.
oscillation amplitude seems to be slightly reduced by soaking as well as by a shorter

and less frequent washing process and thus has a positive effect on the interaction
between the AM elastic model and the FD. During a curing process with a duration of
4 h or a with mineral oil wetted surface, the oscillation amplitude decreases only
minimally. Further investigations are ongoing.
5 Discussion of Limitations
The experiment has shown that different additive materials interact differently in
contact with a moving Flow Diverter. The magnitude of the measured forces is very
small, which is to be expected when measuring a friction force for this friction surface.
Nevertheless, qualitative differences in the oscillation of the force are visible. Apart
from that, the test set-up can be improved by finding a way to prevent the transport wire
from touching the ground at the beginning of the test in order to avoid force peaks
during initial FD alignment. A holder can also be added to straighten tubes made of
Elastic, so that dependencies in potentially non-ideal positioning of the tubes can be
minimized. Furthermore, the test setup could be validated a lubricating material, such
as hydrogel, which is regarded as particularly suitable for blood vessel models by
medical experts.
In addition, the test procedure can be extended by pulling the FD into the tube,
stop-ping the sliding movement and then measuring the breakaway force when starting
up again. Not for each test run, a new FD was used. Any material residues from the
tube may have influenced subsequent results. As the FD is elongated, it is difficult to
estimate influences.
The smoothing of the curves for the evaluation was necessary to be able to compare
curves. By documenting the performed smoothing, the evaluation of results can be
considered legitimate.
In this paper, an objective test setup for AM material evaluation is presented in order to
show how AM models become applicable for usability validation of new developments
of medical devices, here FD design innovations. The focus of this paper is to develop a
test setup, a test procedure and data analysis procedure to investigate which of selected
Formlabs printer materials are suitable for the reproduction of blood vessel models for
Flow Diverter designs tests. For this purpose, a test setup measuring the interaction
between the AM model and the FD was developed. Taking into account the limitations,
different AM materials as well as AM post processing procedures can be investigated
with the test setup, as has been shown in the first tests. In the future, different coatings
and materials will be examined with the test setup.
This paper shows an example on how AM can be used for certain aspects of the
design validation of medical devices. However, the interaction between the AM mate-
rial and the medical device has to be investigated in suitable test setups to be able to
reproduce an interaction as close to reality as possible.
Acknowledgment. The authors would like to thank the German Federal Ministry of Education
and Research for founding this work within the project BELUCCI (13GW0274D) belonging to
the program “Individualisierte Medizintechnik 2”.
References
1. European Parliament: Regulation (EU) 2017/745, Council of 5 April 2017 (2017)
2. Bundesverband Medizintechnologie (BVMed): Der lange Weg eines Medizinprodukts von
der Idee bis zur Anwendung am Patienten. Article, 20.01.2016 (2016)
3. Alexander, K., Clarkson, P.J.: Good design practice for medical devices and equipment, Part
I: a review of current literature. J. Med. Eng. Technol. 24(1), 5–13 (2000)
4. Kuhl, J., Sankowski, O., Krause, D: Investigation on methods and characteristics in medical
device development. In: Proceedings of the DESIGN (2020)
5. Pierot, L., Wakhloo, A.K.: Endovascular treatment of intracranial aneurysms: current status.
Stroke 44(7), 2046–2054 (2013)
6. Fiorella, D., Lylyk, P., Szikora, I., Kelly, M.E., Albuquerque, F.C., McDougall, C.G.,
Nelson, P.K.: Curative cerebrovascular reconstruction with the Pipeline embolization device:
the emergence of definitive endovascular therapy for intracranial aneurysms. J. Neurointer-
ventional Surgery 10(Suppl 1), i9–i18 (2018)
7. Bonney, P.A., Connor, M., Fujii, T., Singh, P., Koch, M.J., Stapleton, C.J., Mack, W.J.,
Walcott, B.P.: Failure of flow diverter therapy: predictors and management strategies.
Neurosurgery 86, 64–S73 (2020)
8. D’Urso, Pietro I., Lanzino, Giuseppe, Cloft, Harry J., Kallmes, David F.: Flow diversion for
intracranial aneurysms: a review. Stroke 42(8), 2363–2368 (2011)
9. Nawka, M.T., Spallek, J., Kuhl, J., Krause, D., Buhk, J.-H., Fiehler, J., Frölich, A.:
Evaluation of a modular in vitro neurovascular procedure simulation for intracranial
aneurysm embolization. J. Neurointerventional Surgery 12, 015073 (2019)
10. Spallek, J., Kuhl, J., Wortmann, N., Buhk, J.-H., Frölich, A.M., Nawka, M.T., Kyselyova,
A., Fiehler, J., Krause, D.: Design for mass adaptation of the neurointerventional training
model HANNES with patient-specific aneurysm models. In: Proceedings of the Design
Society: 22nd International Conference on Engineering Design, vol. 1, pp. 897–906 (2019)
11. Frölich, A.M.J., Spallek, J., Brehmer, L., Buhk, J.-H., Krause, D., Fiehler, J., Kemmling, A.:
3D printing of intracranial aneurysms using fused deposition modeling offers highly accurate
replications. AJNR Am. J. Neuroradiol. 37, 120–124 (2015)
12. Spallek, J., Sankowski, O., Krause, D.: Influences of additive manufacturing on design
processes for customised products. In: 14th International Design Conference - DESIGN
2016, Cavtat, Dubrovnik, Croatia, pp. 513–522 (2016)
13. Bártolo, P.J.: Stereolithographic Processes. In: Bártolo, P.J. (ed.) Stereolithography -
Materials, Processes and Applications. Springer, Heidelberg (2011)
14. Pahl, G., Beitz, W., Feldhusen, J., Grote, K.-H.: Engineering design. In: Wallace, K.,
Blessing, L. (eds.) A Systematic Approach, vol. 3, Springer, Berlin (2007)
15. Desai, C., Laube, N.: Development of a technical approach to modify the internal surface of
biomedical tubes and other elongated small lumen macrodevices with parylene coating.
J. Coatings Technol. Res. 16(1), 103–111 (2018). https://doi.org/10.1007/s11998-018-0104-1
16. Uyama, Y., Tadokoro, H., Ikada, Y.: Low-frictional catheter materials by photoinduced graft
Polymerization. J. Biomater. 22, 71–75 (1991)
17. ZwickRoell: Testing of Catheters (2020). https://www.zwickroell.com/en/medical/catheters-
stents/catheters. Accessed 02 Feb 2020
Innovative Use Case
Additive Manufactured and Topology
Optimized Flexpin for Planetary Gears
Anton Höller1(&) , Frank Huber1, Livia Zumofen1,

Andreas Kirchheim1, Hanspeter Dinner2, and Hans-Jörg Dennig1
1
Zurich University of Applied Sciences – School of Engineering – Centre
for Product and Process Development, Winterthur, Switzerland
anton.hoeller@zhaw.ch
2
KISSsoft AG, Bubikon, Switzerland
Abstract. This paper shows how the interdisciplinary use of additive manu-
facturing (AM) and topology optimization (TO) lead to a better load distribution
and a related increase in performance within a planetary gearbox. For this, it has
been investigated how to use TO for designing the flexible support of the planets
pin and to reduce thereby the load sharing factor (Kc). 18 digital experiments
were performed to evaluate the best setting for this TO. A completely new shape
for a “flexpin” design was found which offers significant advantages in terms of
compliance and misalignment for the flexible support structure of planet gears.
Because of AM, the focus can stay on the main function of the flexpin including
the component reduction. Critical interfaces are no longer necessary. This allows
a much better definition of the compliance, an improved assembling, and a safer
operation. The best TO-design concept was selected and edited for additive
manufacturing. The final validation by FEM showed an increase in load capacity
by 13% and an additional reduction of the misalignment of 77%.
Keywords: Additive manufacturing Structural optimization Flexpin Load

distribution Planetary gearbox
1 Introduction
Additive manufacturing (AM) processes such as Laser Powder Bed Fusion (LPBF)
open up unique possibilities in the design and manufacture of mechanical components.
Due to the layer-wise build-up of the generative manufacturing processes, structures
are possible which cannot be produced by any other manufacturing process, or only at
enormous effort and expense. This enables the component function to be brought much
more into focus during design [1]. The popularity of additive manufacturing processes
is also seen in the sales growth of the 3D printing and additive manufacturing industry.
In the period from 2014 to 2019, revenue was more than doubled [2]. The possibilities
of the new design freedoms for components have been presented in various scientific
papers [3–5]. In order to achieve economical solutions for additively manufactured
mass-production components, an additional functional integration must also be targeted
in addition to the production-oriented redesign.

https://doi.org/10.1007/978-3-030-54334-1_24
338 A. Höller et al.
These novel design freedoms through additive manufacturing offer the possibility
to better fit the topology of the component to the requirements within its application.
This paper investigates the possibilities of AM by use of TO for an improved design of
the flexpin within a planetary gear train.
By using planetary gears, large transmission ratios can be achieved in a space-
saving manner, with high efficiency and comparatively low self-weight. Planetary gears
have a very wide range of applications in industry, automotive engineering, aerospace,
and aviation.
Due to the arrangement of the planet gears, special attention must be paid to the
load distribution among the planets. Manufacturing and assembly tolerances lead to
uneven load distribution to the planet gears. This is considered by the load distribution
factor Kc as used in gear rating along e.g. ISO 6336. It is known that a significantly
better load distribution can be achieved by a specific compliant design of the planet
supports, often also called flexpin [6]. There are different geometrical shapes of the
flexpins. However, so far only the possibilities of subtractive manufacturing were
considered during the development. Thereby a notched pin is usually inserted into the
planet carrier, which in turn carries a sleeve at the other end [7]. Thus the flexpin is
usually built up from three individual parts, which are often press-fitted, Fig. 1.
Fig. 1. A common design of flexpin
2 State of the Art
2.1 Planetary Gear Trains

Gears are used in mechanical engineering where rotational speed and torque must be
converted. A common type of gearbox is the planetary gearbox. Thereby the central
wheel (sun) drives one or more rotating gear wheels (planets). A ring gear with internal
teeth encloses the planets. The planets are usually evenly distributed around the
Additive Manufactured and Topology Optimized Flexpin for Planetary Gears 339
perimeter. They are linked by the planet carrier. In the basic operating mode, the sun
(Fig. 2, 1) drives the planets (Fig. 2, p), which are coupled to the output shaft through
the planet carrier (Fig. 2, s). The ring gear (Fig. 2, 2) remains stationary and is fixed to
the housing.
Fig. 2. Planetary gear train (1: Sun gear, 2: Ring gear, p: Planet gear, s: Planet carrier) [8]
With planetary gear trains, large transmission ratios can be achieved in a space-
saving manner, with comparatively low dead weight and low moments of inertia [6, 8].
This leads to higher efficiencies and increased power densities. The compact design
also raises certain challenges. Thus, planetary gears impose higher demands on the
manufacturing accuracy. Furthermore, when using two or more planetary gears, the
system is statically overdetermined.
A number of different standards and methods are known for the dimensioning of
gears made of steel. Well established are ISO 6336:2006 method B, DIN 3990 method
B and AGMA 2001-C95. For this work, the calculations were made according to ISO
6336. The fail-safe design of gearboxes has to take several different influencing factors
into account, which are mainly selected according to the design and application of the
gearbox. The load sharing factor Kc and the face load factor KHb represent two main
factors in the sizing process [9]. The load sharing factor Kc considers the uneven load
distribution from the sun to the planets. With the face load factor KHb the uneven load
distribution over the tooth width can be taken into account. In accordance with [10], the
load sharing factor is defined as follows:
Ftimax
Kc ¼ [ 1; ð1Þ
Ft
Kc: Load Sharing Factor

Fti max: Maximum Tangential Force for Planet i of n
Ft: Nominal Tangential Force
In accordance with [10], the face load factor is defined as the maximum load
intensity compared to the average load intensity:

Ftj =bj max
KHb ¼ [ 1; ð2Þ
ðFtm =bÞ
KHb: Face Load Factor

Ftj: Amount of Tooth Force in Section bj
bj: Section of the Tooth Width
Ftm: Average Tooth Force
b: Tooth Width
Furthermore, the internal dynamic factor (KV) enables the additional torque caused
by the tooth quality, speed and load to be considered [10]. The application factor (KA)
additionally increases the nominal load additionally. These additional loads depend on
the characteristics of driving and driven machines as well as on the mass and stiffness
of the system. The factors KA and KV are given by the application and cannot be
influenced by the design of the planetary mounting. Therefore, they are no further
relevant for this work.
The tangential force is used in several cases during design. Thus, in addition to the
stress in the planet pin, it is also crucial for the stress in the tooth root and the tooth
flank of the gears as well as for the dimensioning of the planet bearings [10].
For single-stage planetary gearboxes (Fig. 2) the nominal tangential force Ft ; acting
on the planet axis, can be calculated based on the nominal output torque (Mn Þ:
Mn
Ft ¼ ð3Þ
nP r P
Mn: Output Torque

nP: Number of Planets
rP: Effective Radius of Planetary Axis
0
The dimensioning tangential force Ft on the planet can be calculated according to
Eq. 3 [10]:
0
Ft ¼ Ft KA KV KHb Kc ; ð4Þ
Unavoidable manufacturing or assembly inaccuracies always lead to an uneven

load sharing among the planet gears. It could be shown that an improved load sharing
can be achieved by a specific compliant design of the supporting structure of the
gearing components [6]. The improved load sharing is achieved by deformations
caused by the operating load being significantly larger than the manufacturing inac-
curacies to be compensated. One solution is the flexible design of the planetary support
structure. Therefore, Raymond Hicks patented the flexpin design in 1967 [11]. Hicks
placed the planetary bearings on a sleeve, which generates a counter torque under load
due to its one-sided attachment to the pin. This design concept allows parallel dis-
placement of the gear wheel under load with additional flexibility. In 2006 Fox and
Jallet extended Hicks’ idea with an additional notch in the pin, which further reduces
the stiffness [7]. The following are named as negative influencing factors for load
sharing [6]: Eccentricity of the gears, position deviation of the planet pin bore in the
carrier and uneven radial clearance distribution of the planet bearings. As positive
influences, Arnaudov names the flexibility of the transmission elements as well as
radial bearing clearance of the central shafts (of sun and carrier). It was also found that
the load distribution depends on the torque, the bearing clearance and the number of
planets. Based on Kahraman’s research, Montestruc investigated the load distribution
in planetary gears numerically [12, 13]. The developed model considers the three most
influential errors: the position error of the planet pin and the errors caused by the tooth
thickness tolerance and the runout of the gear. Montestruc [13] concludes that the flex
pin has a positive effect on the load sharing, even with a higher number of planets (up
to 11), the reduction of stiffness also reduces the resonance frequency and the load
sharing depends on the load case. In order to represent the compliance comprehen-
sively, the stiffness of the component interfaces, the carrier, the roller bearings and the
tooth contacts must be considered in addition to the compliance of the flexpin [6, 14].
A scientific study on the shape of the flexpin has not been found. Dinner analyses
different forms of notches. He concludes that the notch in the shape of a double cone
offers the best compromise between compliance and stress. Further, he points out that
the position of the notch is usually force-symmetrical in order to reduce a misalignment
of the gear [9].
Neubauer and Dinner show the relevance of the stiffness of the interfaces. [9, 14]
The component interfaces are also crucial for an exact design of the elasticity of the
planet pin. There has been no scientific investigation found for this problem in the
design of the flexpin.
2.2 Topology Optimization

AM offers completely new possibilities for lightweight designs. Both Emmelmann and
Brackett state that AM is redefining lightweight design, especially for aviation appli-
cations [15, 16]. Plocher offers a comprehensive report on the state of research and
industry. He shows, among other aspects, various approaches to lightweight con-
struction through topology optimization (TO) [17]. The TO has been significantly
further improved in recent years [15, 18].
A common TO approach is Solid Isotropic Material with Penalization (SIMP),
which is also applied by the software used in this study. Thereby the design space is
divided into small areas, the design variables x (DV). A relative material density (qne ð xÞ)
is assigned to each DV, which is used to optimize the component. The material density
can adopt values between 0 (void) and 1 (solid). Starting from a FEM calculation, new
design proposals are generated by the TO. The specifications such as maximum design
space, deformations, loads, remaining volume and compliance for a defined target
function are fulfilled in an optimum manner [15, 19]. In a simple TO, the mean com-
pliance is minimized while maintaining a weight restriction. In TO, manufacturing
restrictions such as the maximum overhang angle for a given print direction, component
symmetries, direction of demolding can be introduced by constraints [18, 20].
The generated design concepts are non-parametric models and need to be processed
for further use after optimization. Therefore, the obtained density distribution of the DV
must be interpreted and converted into a component by a Computer Aided Design
(CAD) program [19].
2.3 Additive Manufacturing in the Context of Gear Components

Thanks to the layer-wise build-up of components and the subsequent possibilities, AM
is called a disruptive technology for manufacturing [21]. The research in the field of
AM has made significant achievements in recent years. Thus, more complex compo-
nents with higher mechanical requirements can be additively produced. Thanks to the
layer-wise structure of additive manufacturing processes, structures are possible which
cannot be produced by other manufacturing processes, or only at great expense. This
enables the component function to be brought much more into focus during design [1].
For mechanically highly loaded metal components, the Laser Power Bed Fusion
(LPBF) process is of major relevance.
Rogers shows the potential for the gear industry through AM [21]. He identifies
several fields of action: Production of complex geometries such as internal cooling and
lubrication channels, reduction of mass inertia, improvement of fatigue strength,
reduction of manufacturing costs, reduction of development and time to market. The
number of studies that explore the design potential of AM and TO in gear train design
is limited. Kamps has developed a new method to develop an innovative gear design
which can be manufactured by AM [22]. Ramadani calculated a vibration- and weight-
optimized gear made of Ti-6Al-4 V by LPBF [23]. Thereby he uses lattice structures to
reduce noise emissions. Barreiro analyzes the potential and the restrictions due to AM
using the example of a transmission housing TO [24]. He highlights the importance of
integrating additional functions, such as lubrication channels.
Focused on AM and TO, the potential of AM in combination with TO has already
been investigated in several scientific studies. [18, 25–30] Almost all designs were
topology optimized in terms of stiffness at mass constraint. Scientific research in terms
of Gearbox Development and TO, has also been accomplished many times [21, 32–36].
Kolakowski [32] for example has optimized a planet carrier with the aim of improving
torsional stiffness. Jeevanantham [35], Zhuang [34] and Dietzel [37] developed new
gearbox cases. Some authors [36, 38, 39] address the topic of the optimal shape of a
gear wheel. Both Shah [38] and Heiselbetz [36] reduce the weight with respect to
symmetry and stress constraints. Furthermore, also the modal behavior was investi-
gated [36]. TO has been successfully used in the development of gear trains and has
improved some components. However, no paper has been found about the flexible
support of the planets by using TO nor AM.
3 Development Process and Methodology
The present work has the goal of an optimized flexpin design, which uses the freedom
of AM for better performance. Therefore, a reference gearbox was defined (Fig. 3,
No. 1). To obtain the load distribution, a simplified static FEM model of the gearbox
was created alike the model by Montestruc [13]. This model contains the influences of
tangential differences in clearance at the tooth contact of the planets and thus makes it
possible to show the load-dependent load sharing for this gearbox. With this model, the
load distribution was analyzed, the potential for increased compliance of the planetary
support was demonstrated and a target stiffness value was defined (Fig. 3, No. 2).
Using a submodel, the structure of the planetary pin was optimized in such a way that
the stiffness specification is met but the tilting of the sleeve is minimized and the stress
on the pin remains within the allowable limits. The static qualification was performed
against yielding with a safety factor SF = 2.0 (Fig. 3, No. 3). After successful opti-
mization, the results were transferred to a design suitable for AM (Fig. 3, No. 4)
Finally, a final validation of the solution by a FEM calculation and qualitative com-
parison of the new solution to the existing flex pin design has been carried out. It was
clarified whether further performance increases are possible thanks to the use of AM.
Fig. 3. Development process
The planetary gear was designed with the KISSsys 2019a, by KISSsoft. For the
FEM calculations ANSYS 2019 R3 has been used and for the optimization Tosca
Structure 2018 (Hotfix 5) from Dassault Systems has been used.
3.1 Development of Reference Gear Train

The starting point is a newly developed reference gear train with a total ratio of 4.33:1 and
a nominal power of 60 kW (Fig. 4). The output shaft has a rotational speed of 1’200 rpm
at a torque of 477.5 Nm. The drive shaft is connected by bolts to the sun wheel. The five
planet wheels are placed in a regular 72° pitch. They are supported by needle bearings
through a hollow pin and a sleeve in the planet carrier (Fig. 5). The planet pins are only
one sided supported in the planet carrier. The planet carrier is connected by bolts to the
output shaft. The output shaft is bearing mounted in the housing. The normal module of
the spur gearing is 1.5 mm. All structural components are made by titanium (Ti-6Al-4 V)
and all gears made of case-hardening steel (16MnCr5). Further information about the
gearing can be found in the following table (Table 1).
Fig. 4. Reference gear train (a: Isometric view with hidden gear box, b: Section view of planet
gear)
Table 1. Most important gearing data of the reference gear

Sun Gear Planet Gears Ring Gear
Normal module mn mm 1.5 1.5 1.5
Pressure angle at normal section an ° 20 20 20
Qualitya Q – 6 6 6
Tooth width b mm 17 17 15
Pitch circle diameter d mm 45 52.5 150
Addendum modification coefficient x 0 0 0
a
According ISO1328:1995.
The planet gears (Fig. 4b) are mounted on the sleeve with needle bearing SKF NA
4905 and are guided by lateral stops. The sleeve is press-fitted onto the and additionally
secured by a circlip (not shown). The pin is press-fitted into the planet carrier.
3.2 Calculation of Load Distribution

In order to calculate the load sharing in the reference gear, it is crucial to combine the
stiffness of the five flexpin assemblies (all five assemblies have the same stiffness) with
the different backlashes in the gear meshes (backlashes between the five planet gears and
the mating gears are not the same for all five planets due to manufacturing errors).
A FEM model (Fig. 5) was developed, which shows the stiffness along the force flow
from the sun to the output shaft, over the planet carrier, planet pin, sleeve by finite
elements. The radial bearing stiffness and the bearing clearance of the planetary bearings
are implemented in the calculation by spring elements with gap option. The bearings of
the input and output shaft are assumed to be clearance-free. The tooth contact stiffness
for the contacts between sun to planets and planets to ring gear is also implemented
using spring elements with a constant value according to ISO 6336.
For the analysis, three errors for different tooth contacts were considered. [6, 12] These
are the different radial bearing clearances limited by the bearing clearance tolerances, the
radial position of the finished flexpin sleeves and the pitch deviation of the toothing of the
planet gears determined by the gear quality. SKF specifies a tolerance width of 25 µm for
the internal clearance for the bearings selected. [40] The positioning accuracy of the
flexpin axis with respect to the gearbox axis is assumed to be ±10 µm due to IT6. The gear
tooth to tooth pitch deviation of 7 µm follows from Q-ISO1328:1995 and gear quality Q6.
For the calculation, the most conservative combination has been considered. This results
when the tolerances for one planet gear are at the minimal level and the tolerances for all
other planet gears are at the maximal level [13].
In accordance with ISO/TS 16281, KISSsoft 2019 calculates the radial stiffness of
the journal rolling element bearings at the operating point with 438’560 N/mm and the
tooth contact spring stiffness according to ISO 6336 with 52’346 N/mm.
Fig. 5. FEM model of reference gear train (a: Isometric view, b: Detail view of planet gear and
flexpin)
If a real gear train is loaded, at small loads one planet transmits the complete torque,
due to the uneven load distribution. For a gearbox with n planets this results in a Kc =
n. Due to the compliance of the loaded components the load distribution is further
improved by increasing torque. This is coherent with the results of [6, 12, 14, 31]
(Fig. 6). By use of the FEM Model and an applied nominal torque Mn a maximum
0
tangential force Ft;max of 2791.4 N, the nominal tangential force follows from (1) with
Ft of 1956.92 N and therefore Kc:
2791:4N
Kc ¼ ¼ 1:43; ð5Þ
1956:9N
The flexpin provides primarily additional tangential compliance. Behavior with

flexpin is investigated using an increased overall compliance CR, as the sum of radial
bearing stiffness and radial flexpin stiffness. As expected, this resulted in a better load
distribution.
2
Nominal Torque
1.9
Reference
1.8
Flexpin, CR = 70 kN/mm
Load Sharing Factor Kγ [1]
1.7 Flexpin, CR = 20 kN/mm

1.6
1.5
1.4
1.3
1.2
1.1
1
0% 25% 50% 75% 100% 125% 150% 175% 200%
% of Nominal Torque
Fig. 6. Mesh load factor for reference gear train
For different radial stiffness CR the improved load distribution can be computed
(Table 2). For the optimization, the target stiffness of the flexpin is assumed to be CR =
20 kN/mm. At a constant tangential load, (3) follows a potential increase in torque of
19%.
Table 2. Load sharing factor for different radial stiffness

Radial Stiffness CR [kN/mm] Load Sharing Factor Kc [-]
6 1.095
15 1.170
20 1.200
30 1.245
70 1.330

4.1 Topology Optimization of Flexpin
For the TO of the flexpin, a reduced model was created. It contains the essential
geometric elements to represent the force flow over the flexpin. A coordinate system
C0 was defined (Fig. 7, No. 2). The central axis of the flexpin is coincident with the Z-
axis. From Fig. 6 with Mn and CR = 20 kN/mm follows Kc = 1.2. This leads to a
0
maximum tangential force per flex pin of Ft;max 2’300 N. It is applied to the contact
surfaces of the planetary bearing (Fig. 7, No. 3). The material behavior was assumed to
be linear-isotropic. For the LPBF manufactured Ti-6Al-4 V an elastic modulus of 115
GPa was defined [41].
Fig. 7. Basis model for the TO of the flexpin
Two primary criteria are defined for the application of the flexpin: the local radial
stiffness CR for the flexpin, for a defined load distribution and the minimum
misalignment a of the sleeve, which leads to an uneven load distribution across the
width. For the TO an additional node N is defined at the position (-100/0/-13). The
radial stiffness of the solution is derived from the deformation of node N in X-direction
uxN. The tilting of the sleeve corresponds to a rotation around the Y-axis. This is
measured by the axial displacement of node N uzN.
The TO was performed with the target function and constraints specified in
Table 3. The overhang constraint was defined with the print direction -Z and the
overhang angle of 45°.
Table 3. TO of flexpin for AM

Target Function Minimize (uzN)
Constraint C1 uxmin < uxN < uxmax
with uxmin = 0.2185 mm and
uxmax = 0.2415 mm
Constraint C2 Overhang 45°, Print Direction -Z
Constraint C3 Volume reduction = 50%
Constraint C4 SIG_TOPO_MISESa = 75 MPa
Constraint C5 Plane Symmetry XZ
a
SIG_TOPO_MISES is defined by Tosca Structural [42].
To further understand the influence of the constraints C2 and C3, 18 digital TO

experiments are defined (Table 4). The design concepts must be processed for evaluation.
For this purpose, the density distribution must be interpreted using a limit value ISO,
which represents the hard border between void (no material present) and solid (material
present). This is done with the program Tosca Smooth. Space Claim was used to refine the
CAD geometry. The qualification of the prepared design concepts was accomplished on
the criteria of radial stiffness CReff, maximum equivalent stress according to von Mises
SvM and the misalignment a of the bearing surfaces on the sleeve by a subsequent sim-
ulation. The allowable equivalent stress was determined on the basis of the yield stress
(stress relieved, 1 h at 650 °C in vacuum) [41] and SF to Sallow = 542.8 MPa.
Table 4. Overview of TO experiments

No. C3 C4 ISO CReff SvM a
[-] [MPa] [-] [kN/mm] [MPa] %
1 0.30 50 0.75 47.9 422.0 0.02%
2 0.50 25 0.75 108.6 151.0 −0.02%
3 0.50 25 0.90 79.4 214.5 0.00%
4 0.50 50 0.60 70.5 247.3 −0.02%
5 0.50 50 0.75 31.0 440.0 0.07%
6 0.50 50 0.78 25.1 569.4 0.14%
7 0.50 50 0.83 15.8 901.5 0.28%
8 0.50 50 0.90 6.1 1925.5 0.94%
9 0.50 75 0.70 33.4 553.9 0.01%
10 0.50 75 0.75 22.1 724.4 0.06%
11 0.50 n. a. 0.50 22.8 732.6 −0.05%
12 0.50 n. a. 0.75 5.0 2661.1 −0.15%
13 0.65 50 0.75 41.7 494.4 0.03%
14 0.70 50 0.60 76.6 283.1 −0.02%
15 0.70 50 0.75 39.7 469.0 0.04%
16 0.70 50 0.80 27.5 643.0 0.09%
17 0.70 50 0.85 13.5 1164.0 0.30%
18 0.70 50 0.90 4.1 5308.4 1.49%
Almost all design concepts are found with a design similar to a parallelogram,
whereby four of each are connected by a crossbar (Fig. 8b, 8c and 8d). The form is
comprehensible due to the given boundary conditions.
The cross-section of the flexural joints is strongly dependent on C3 and ISO, which
therefore have a significant influence on the stiffness (Fig. 8b, Fig. 8d). The desired
compliance cannot be obtained directly but must be approximated iteratively. C4 limits
the maximum stress, there is a positive effect on the result, but the influence seems to be
subordinate compared to other constraints. With a low value of C4 a design, like the
known flexpin can be achieved. The volume and the plane symmetry constraint lead to
a two “leave spring” design (Fig. 8a).
Fig. 8. Results of topology optimized flexpin, a. Design concept 3, b. Design concept 13, c.
Design concept 10, d. Design concept 18
The influence of the manufacturing restrictions (C2) is getting more and more lost
during processing as a result of the dense design. The overhang angle does not go
directly into the processing and must be considered in the subsequent finishing in CAD
manually.
Fig. 9. Flexpin design (No. 9) after TO

No. 5, No. 9, No. 13 and No. 15 from Table 4 show a good balance between stress,
compliance and misalignment. The configuration of experiment 9 is selected for further
use (Fig. 9).
4.2 AM Design and Final Validation

In order to prepare the received design concept for production by LPBF, the geometry
must be edited considering the following aspects:
Smoothing. The structure which is available as output of the TO must be further
improved. The CAD tool Space Claim is used for this purpose. Further imperfections in
the geometry will be removed. Finally, it has to be taken care that the free form surfaces
are smoothed with a high resolution.
Merge the volumes. The pin and the sleeve can be merged. Special attention must be
paid to the sharp transition zone of the components.
Overhang. For overhanging structures in Ti-6Al-4 V, the surface roughness increases
significantly at an angle of < 40° to the horizontal [43], which was taken into account in
the component design. To ensure process reliability, overhanging structures at a critical
angle of 30° to the horizontal should be avoided [43, 44]. Horizontally overhanging
structures up to 1 mm wide and horizontally aligned holes up to 12 mm in diameter can
be produced without support structures [43]. These factors have been considered in the
component design, resulting in a geometry that can be reliably manufactured with
LPBF.
Gap. In order to avoid undesired melting or sintering, it must be ensured that adjacent
component walls have a minimum gap of > 0.2 mm [43, 44].
Interface. The interface contour to the planet carrier was adapted regarding a down-
stream test series.
Post-processing. During processing by LPBF, thermally induced internal stresses are
introduced. These must be released again by suitable heat treatment (stress relieving).
All functional surfaces must be machined after printing. Appropriate oversizing of the
functional surfaces has been taken into account. Furthermore, the powder extraction
after printing must be possible. Supports are required between the interface to the
planet carrier and the sleeve. These also give additional support to the sleeve during the
post-processing of the bearing surfaces. As a last step, the supports must be removed by
a side milling cutter all around the circumference (Fig. 10).
Fig. 10. Printed samples of flexpin
The solution was subjected to a FE calculation in the post-processed state. The

0
operating tangential load Ft must be recalculated according to (4) by the load distri-
bution factor and the radial stiffness. The new design will reduce Kc from 1.43 to 1.24,
the fewer misalignment leads according to a first estimation by use of KISSsoft to a KHb
reduction from 2.03 to 1.29. The total deformation is shown in Fig. 11 and Fig. 12 and
the displacement in x direction of the path on top of the sleeve shown in Fig. 13 illustrate
the reduced stiffness behavior and the improved misalignment for the load of 2’300 N of
the non-optimized (Full Cross Section Design) and the optimized (AM & TO Design)
flexpin. As a result of the lower stiffness, the optimized flexpin will improve the
equalization of the load distribution in planetary gears. Furthermore, its reduced
misalignment of the sleeve will improve the load distribution over the tooth width.
Fig. 11. Total deformation of full cross Fig. 12. Total deformation of AM &
section design TO design
Fig. 13. Displacement in X direction of Path 1 and Path 2 on sleeve
The structural stresses increase but stay at an acceptable level for static loads. For
the optimized structure, a maximal value of 494.7 MPa for Von Mises stress was
calculated (Fig. 14). The development of new flexpin requires TO. A heat treatment
(stress relieving) is for the printed version mandatory. A suitable surface finishing
method may positively influence fatigue properties, despite the initial high surface
roughness. However, due to the entangled shape of the TO flexpin, the accessibility of
the relevant surface area is limited. Therefore, primarily flow grinding is potentially
applicable.
Fig. 14. Von Mises stress of AM & TO design

The comparison in Table 5 between non-optimized (Fig. 15) and optimized

(Fig. 16) pin provides an overview of the added value, but also the drawbacks of the
new solution.
Table 5. Comparison of Full Cross Section Design and Optimized Pin
Full Cross Section Design AM & TO Design
Fig. 15. Pin without Flexpin Fig. 16. Pin with AM and TO
Design
Design Load
for Pin
Result of • CR = 117.6 kN/mm • CR = 28.7 kN/mm (-75.6%)
Simulation • SvM = 145 MPa • SvM = 494.7 MPa (+341%)
• |α| = 0.068% • |α| = 0.016% (-77.3%)
Development • Average complexity • Increased complexity
Manufactur- • Heat treatment optional • Heat treatment mandatory
ing • Post-processing of press fit • Post-processing (machining,
faces perhaps flow grinding)
• Well established procedures • Increased precision due to
part integration
Assembling • Press fit (Pin/Sleeve) • Lower number of parts and
interfaces
Cost • Higher effort for assembling • Higher production cost ex-
pected for flexpin
• Smaller loads could lead to
smaller design and monetary
savings of gear train
Fig. 15. Pin without flexpin Fig. 16. Pin with AM and TO design
5 Conclusion
The present work aims at an optimized flexpin design, which uses the freedom of
additive manufacturing for improved performance of a planetary gear. For this purpose,
a planetary gear train was developed and analyzed regarding the uneven load sharing.
On this basis, topology optimization was used to search for new solutions that take
advantage of the freedom of additive manufacturing. 18 digital experiments were
carried out. A new, unique approach of a ‘flexible’ pin was found which significantly
improves the torque distribution to the planetary gears due to radial compliance.
Currently, the flexpin is designed with a notched pin [7, 11]. The parallelogram-like
design promises not only a significantly lower radial compliance, but also a lower
misalignment of the sleeve while respecting the stress limits. The newly discovered
design features an optimized geometry that takes full advantage of the AM design
freedom. The stiffness can be adjusted according to the application. Thanks to the part
integration of sleeve and pin, this critical interface is no longer present for the first time.
This enables a more precise design, manufacturing, assembling and finally operation.
The further improved load distribution allows to keep the dimensioning tangential load
on the planet pins on a lower level. Due to its influence on the gear dimensioning and
bearing selection, this is crucial for the costs generated in gear design. In order to
ensure safe operation, an additional proof of fatigue strength must be provided in future
work. Within this context, suitable surface treatments for the TO surface need to be
evaluated and the accessibility of relevant surface areas needs to be addressed in future
work.
The transfer from the design concept proposed by the TO to a design suitable for
production offers further potential for improvement. During smoothing, valuable
information of the geometry data is lost. Criteria for the overhang angle need to be
considered again during the preparation of the 3D data for AM.
It is intended to use the gained knowledge to realize the flexible support of the
planet as an integrated part of the planet carrier. Thanks to the elimination of the
interfaces between planet carrier, pin and sleeve, the performance of the gearbox can be
further improved.
References
1. Wörner, S., et al.: Rapid Prototyping im Maschinen- und Automobilbau – Ermüdung-
seigenschaften additiv gefertigter Bauteile in Additive Fertigung von Bauteilen und
Strukturen, 1st edn. Springer, Wiesbaden (2017)
2. Wohlers, T., et al.: “Wohlers Report 2018”, 3D-Printing and Additive Manufacturing State
of the Industry – Annual Worldwide Progress Report. Wohlers Associates, Fort Collins/US
(2018)
3. Spiegel, A., et al.: Lightweight Design: Wege zum wirtschaftlichen Einsatz der laseraddi-
tiven Fertigung. Springer, Wiesbaden (2015)
4. Grienitz, V., et al.: ATZ - Automobiltechnische Zeitschrift: Technikevaluation für die
generative Fertigung eines Serien-Radträgers. Springer, Wiesbaden (2016)
5. Wartzack, S., et al.: Besonderheiten bei der Auslegung und Gestaltung lasergesinterter
Bauteile, RTejournal - Forum für Rapid Technologie, vol. 7 (2010)
6. Arnaudov, K., Karaivanov, D.: Planetary Gear Trains, 1st edn. CRC Press, Boca Raton
(2019)
7. Fox, G., et al.: Epicyclic Gear System, US6994651 B2, United States Patent Office (2006)
8. Wittel, H., et al.: Roloff/Matek Maschinenelemente, 24th edn. Springer, Wiesbaden (2019)
9. Dinner, H., et al.: FlexPin Design for Maximized Power Density. EES KISSSOFT GmbH,
Menzingen (2014)
10. ISO 6336-1-2006: Calculation of load capacity of spur and helical gears (2006)
11. Hicks, R.J.: Load Equalizing Means for Planetary Pinions, US3303713. United States Patent
Office (1967)
12. Kahraman, A.: Static Load Sharing Characteristics of Transmission Planetary Gear Sets:
Model and Experiment. SAE Technical Paper 1999-01-1050 (1999)
13. Montestruc, A.: A Numerical Approach to Calculation of Load Sharing in Planetary Gear
Drives. Journal of Mechanical Design, ANSI/AGMA Paper No. 6123-B06 (2009)
14. Neubauer, B.: Lastverteilung und Anregungsverhalten in Planetengetriebensystemen,
Dissertation. Technischen Universität München (2016)
15. Bendsoe, M.P., et al.: Topology Optimization: Theory, Methods, and Applications. Springer,
Berlin (2013)
16. Emmelmann, C., et al.: Laser additive manufacturing and bionics: redefining lightweight
design. Phys. Procedia, 12(part 1), 364–368 (2011)
17. Plocher, J., et al.: Review on design and structural optimisation in additive manufacturing:
towards next-generation lightweight structures. Mater. Des. 183, 108164 (2019)
18. Zhu, J.-H., et al.: Topology optimization in aircraft and aerospace structures design. Arch.
Computat. Methods Eng. 23, 595–622 (2016)
19. Schuhmacher, A.: Optimierung Mechanischer Strukturen, 2nd edn. Springer, Wiesbaden
(2013)
20. Brackett, D., et al.: Topology optimization for additive manufacturing. Annual International
Solid Freeform Fabrication Symposium. Austin (Texas) (2011)
21. Rogers, K.: Additive Manufacturing Technologies for Gears. AGMA (2019)
22. Kamps, T., et al.: Systematic biomimetic part design for additive manufacturing. In: 3rd CIRP
Conference on BioManufacturing, Procedia, vol. 65, pp. 259–266. ScienceDirect (2017)
23. Ramadani, R., et al.: Topology optimization based design of lightweight and low vibration
gear bodies. Int. J. Simulat. Modell. 17(1), 92–104 (2017)
24. Barreiro, P., et al.: New improvement opportunities through applying topology optimization
combined with 3D printing to the construction of gearbox housings. Forsch Ingenieurwes,
vol. 83, pp. 669–681. Springer, Wiesbaden (2019)
25. Ghungarde, V., et al.: Design optimization of steering knuckle by adopting bionic design
approach. IOP Conf. Ser.: Mater. Sci. Eng. 624, 012023 (2019)
26. Ohlsen, J., et al.: ATZ - Automobiltechnische Zeitschrift: Function Integrated. Bionic
Optimised Vehicle Lightweight Structure in Flexible Production. Springer, Wiesbaden
(2015)
27. Seabra, M., et al.: Selective laser melting (SLM) and topology optimization for lighter
aerospace components. In: Conference 15th PCF, Procedia Structural Integrity, vol. 1,
pp. 289–296. Paço de Arcos (2016)
28. Sugavaneswaran, M., et al.: Design of robot gripper with topology optimization and its
fabrication using additive manufacturing. In: Shunmugam, M.S. (ed.) Advances in Additive
Manufacturing and Joining 2018. Springer, Wiesbaden (2020)
29. Smith, C.J., et al: Utilizing additive manufacturing techniques to fabricate weight optimized
components designed using structural optimization methods. Solid Free. Fabr. Symp. 879–
894 (2013)
30. Morgen, D., et al.: GE jet engine bracket challenge: a case study in sustainable design.
J. Innov. Impact 7, 95–107 (2016)
31. Yi, P., et al.: The multi-objective optimization of the planet carrier in wind turbine gearbox.
Appl. Mech. Mater. 1662-7482, 184–185, 565–569 (2012)
32. Kolakowski, Z.: Application of structural topology optimisation for planetary carrier design.
Selected problems of continuum mechanics (2016)
33. FE-SESIGN GmbH: “Weight reduction of a planet carrier with stiffness requirements”.
https://support.ansys.com/staticassets/ANSYS/staticassets/partner/FEDesign/Application_
TOSCA_ANSYS_Optimization%20of%20a%20Planet-Carrier.pdf. Accessed 15 Feb 2020
(2010)
34. Zhuang, S.: Gearbox housing topology optimization with respect to gear misalignment.
Master Thesis. Department of Management and Engineering, Linköpings universitet (2012)
35. Jeevanantham, A.K., et al.: Topology and shape optimization of gear casing using finite
element and taguchi based statistical analyses. ARPN J. Eng. Appl. Sci. 11(23), 13721–
13734 (2016)
36. Casiello, M.: Static and Dynamic Topology Optimization for Aeronautical Gears. Master
Thesis. Politecnico di Torino (2018)
37. Dietzel, M.: Topology optimization methods applied to automotive transmission housings.
In: European HyperWorks Technology Conference (2011)
38. Shah, C., et al.: Optimizing weight of a gear using topology optimization. Int. J. Sci., Eng.
Technol. Res. (IJSETR) 7(6), 403–406 (2018)
39. Heiselbetz, U., et al.: Weight optimization of a gear wheel considering the manufacturing
process and cyclic symmetry. In: 9th Stuttgart International Symposium “Automotive and
Engine Technology” (2009)
40. SKF, Table of Internal Radial Clearance for Needle Roller Bearings. https://www.skf.com/
de/products/bearings-units-housings/roller-bearings/needle-roller-bearings/needle-roller-
bearings-with-machined-rings/with-machined-rings-w-inner-ring/index.html. Accessed 19
02 2020
41. Fischer, M., Zumofen, L.: Tensile properties of laser melted Ti6AL4V. ZHAW/ZPP.
unpublished
42. Tosca Structure User Guide. SIMULIA User Assistance 2018. Dassault Systems (2018)
43. Kranz, J.: Methodik und Richtlinien für die Konstruktion von laseradditiv gefertigten
Leichtbaustrukturen, 1st edn. Springer, Berlin (2017)
44. Van Bael, S., et al.: Micro-CTbased improvement of geometrical and mechanical
controllability of selective la-ser melted Ti6Al4V porous structures. Mater. Sci. Eng., A
528(24), 7423–7431 (2011)
A Review of Optimised Additively
Manufactured Steel Connections for Modular
Building Systems
Zhengyao Li1, Konstantinos Daniel Tsavdaridis1(&) ,

and Leroy Gardner2
1
School of Civil Engineering, Faculty of Engineering and Physical Sciences,
University of Leeds, Leeds LS2 9JT, UK
k.tsavdaridis@leeds.ac.uk
2
Department of Civil and Environmental Engineering,
Imperial College London, London SW7 2AZ, UK
Abstract. This paper is presenting various types of innovative material tech-

nology that can be achieved by 3D printing and their potential to advance the
connection design in modular building systems (MBSs). Connections can
embrace flexibility, adaptability and resilience in the design of modular systems
enabling dismantling, repair and reuse – towards faster transition to autonomous
construction (e.g., with robotics). Rapid developments in additive manufacturing
methods play a key role in the design such innovative mechanical systems.
Connections in modular buildings are the most critical parts of the system
performance and integrity, but are currently receiving less attention compared
with other structural elements (e.g., joists and studs). Optimised 3D printed
connections can substantially improve the design of modular building systems
with the ability to connect complex geometries and components that would be
time and cost-prohibitive or even impossible to produce with traditional man-
ufacturing methods. Therefore, this study reviews the latest connection designs
of MBS units which can be replaced with 3D printed elements. In addition,
advancements in additive manufacturing (AM) materials are included in this
paper, exploring how AM can improve connection design in terms of both
material properties and geometry.
Keywords: Structural topology optimisation Additive manufacturing

Structural connections 3D printing Modular building systems Flexible
connections Adaptable connections Material composition
1 Introduction
Modular construction has become increasingly popular as an alternative to traditional

construction, because it is highly efficient with less input of time and labour, and less
on-site construction waste production. While many studies have focused on the
structural system of modular buildings, connections have been poorly investigated. As
one of the main contributors to carbon emissions, the construction sector is now
focusing on reducing the environmental impact of building designs and the connection

https://doi.org/10.1007/978-3-030-54334-1_25
358 Z. Li et al.
design is a key challenge in the realisation of flexible and adaptable modular structures,
featuring flat-pack systems, composite systems, or hybrid systems. The application of
‘plug & play’ modules, which are manufactured and tested off-site and then slotted into
position on-site, is fast gaining favour in the building sector, offering shorter con-
struction times, improved safety, reduced waste and higher performance. During the
construction stage, lifting and handling of modular units are still challenging tasks, thus
reducing module weight via a sophisticated connection design is a priority. In con-
ventional modular buildings, connections are simply designed to integrate adjacent
modules. Modules in MBSs are normally welded as well as bolted together to form a
rigid inter-module connection that is strong to carry vertical loads, but can fail in a non-
ductile manner under lateral loads due to wind and earthquake, and can be hard to
disassemble without damaging the structure [1].
Additive manufacture (AM) has already been successfully used in medical, aero-
space and manufacturing sectors, and there is an increasing interest in AM solutions for
the construction industry. Compared to conventional manufacturing methods, AM
offers numerous benefits including geometry freedom, better structural efficiency, as
well as the possibility of functionally graded material and prestressing [2]. AM has
already been utilised in a number of innovative construction projects [3–5], demon-
strating its potential for more widespread use. AM is often used in association with
structural topology optimisation (STO), as well as other forms of design optimisation,
to improve structural efficiency and has achieved positive outcomes when developing
innovative buildings, elements and systems. Limited work has been carried out yet on
AM-related connection designs which in turn can have great impact on the mechanical
performance of the overall structural system providing stability, robustness and
resilience.
This paper presents an overview of recent studies toward innovative demountable
connections in MBSs as well as existing studies on the application of AM for optimised
structural elements, and the ideas of how to integrate them in connection designs. In the
following section, future opportunities and challenges of additive manufactured con-
nections for modular buildings are investigated.
2 Connections in Modular Building Systems (MBSs)
2.1 Inter-module Connection

One form of modular construction is by using volumetric building modules which are
prefabricated in the factory, then transported to the construction site and assembled to
form the overall MBS. Connections in MBSs are essential part of the structure as
modules need to be connected horizontally and vertically to form the overall building
system, so that they strongly affect the overall stability and robustness [6]. There are
various connection systems now available for modular buildings (Fig. 1). Most of them
consist of bolts, plates, and welds to form rigid joints between columns, beams, and
modules [7–9].
A Review of Optimised Additively Manufactured Steel Connections for MBSs 359
Fig. 1. Examples of existing inter-module connections [10] (a) bolted connection with plug-in
device [9] (b) bolted connection with rocket-shaped tenon and gusset plate [11] (c) bolted end
plate [12] (d) bolted beam-to-beam module connection [13].
Despite the rigidity provided by bolted and welded connections, their installation is
still somewhat problematic [6]. The welding of connections is made on-site and external
access to the modules is required via mobile access platforms, which increases the
difficulty and risks of construction; such operations also have complex and expensive
requirements. The access for installation of the inter-module connection is more difficult
than the module to foundation connection as it is often not straightforward. For example,
case shown in Fig. 2; when the fourth module is to be placed, it is somewhat difficult to
connect it to the other three modules since the access space is limited [9].
Fig. 2. Access difficulty for installation of conventional inter-module connection [9].

360 Z. Li et al.
In addition to the installation difficulties, the inter-module connections (connections

between stacked modules) are more critical to the mechanical performance of the entire
building compared to the intra-module connections (connections within a module).
Apart from providing integration of discrete modules, inter-module connections are
also designed to transfer horizontal and uplift forces as a result of wind and seismic
actions, while gravity load is typically transferred by other structural elements such as
walls and columns. Under the worldwide tendency of building high-rise buildings to
accommodate increasing population, modular connection design is facing new chal-
lenges with relation to lateral stability and dynamic performance, which makes the
ductility of connection an important factor while withstanding shear (horizontal) and
tension (uplifting) forces as well as dissipating energy to prevent brittle and sudden
failures. Ductility and strength redundancy in inter-module connection design is also
important to allow load redistribution in an accidental case when part of the structure
gets damaged to avoid progressive collapse of the entire structure [14].
Overall, connection systems for MBS are still under development with scope to
improve the ease of installation while retaining structural performance for the full
development and exploitation of modular construction.
2.2 Demountable Modular Building

Besides the installation issues, another challenge brought by using non-demountable
welding in connection systems is the recycling and reuse of structural materials. Even
modular buildings are mostly disposable; currently available connectors in construction
cannot be efficiently removed without damaging the structure, thus the reuse of
structural materials and components is still a relatively new concept. With demountable
connections, structural elements can be reused at the end-of-life of buildings, so that
less waste will be produced and less energy is used for their recycling. In this way, the
environmental impact of the modular construction sector can be significantly reduced,
ahead of any other construction method in line with urgent demands to decrease carbon
emissions in the construction sector. Thus, the development of demountable connec-
tions in MBSs is becoming a new challenge.
2.3 Interlocking Connection for Modular Building

For reducing time and risks of modular assembly, recent studies have proposed novel
connection designs using fewer welds, key to achieve the abovementioned demount-
ability, with side benefits such as improvement in MBS seismic performance [1].
Integral mechanical jointing is a traditional connection that can integrate elements
without any additional connectors. Structural elements are connected via the inter-
locking of their geometric features rather than bolts and welds. It was widely used in
traditional timber buildings but then became less popular in the construction sector due
to manufacturing difficulties. Now, with the development of advanced manufacturing
technologies, such as additive manufacturing (aka 3D printing), integral mechanical
jointing offers the construction industry a new capability of connecting elements to
encourage the development of structures with easier assembly as well as demount-
ability i.e., the capacity to fully dismantle undamaged structural components and realise
reuse [15].
Generally, integral mechanical connection offers several advantages over conven-

tional mechanical connections [14, 15], such as:
• Easier and faster assembly;
• No additional adhesive or connectors like bolts and welds are needed (no additional
weight);
• Demountability and reusability;
• Potential for automatic assembly (reduce risk of operation);
• Reduced stress concentrations.
While some shortcomings should also be noted:
• Risks of coming apart;
• Manufacture cost increase;
• Difficulty in manufacture.
Simple interlocking designs are used in some existing integral connection systems
for modular buildings. Modular Housing System (MHS) is a traditional post and beam
system but with great demountability and adaptability benefited from the interlocking
technology. The aluminium beams and columns in the system are interlocked via their
special geometric design and strengthened by concealed bolt-and-clamp components
(Fig. 3) [16]. In addition to the ease of installation, maintenance and modification
offered by modular construction, the interlocking mechanism of connection in MHS
enables 100% recycle and renewability of the structure while providing sufficient
robustness and superior seismic performance [16].
Fig. 3. Structure of MHS connection [17].
Sharafi et al. [14] proposed a Modular Integrating System (MIS) using interlocking
connection strips that have high flexibility and are readily demountable (Fig. 4).
362 Z. Li et al.
Modules with this connection can be easily assembled to adjacent modules by pushing
and sliding moves. In numerical and experimental analyses, a structure using MIS
showed sufficient stability under notional static and dynamic horizontal forces and
partial structure removal simulations, indicating that this type of integral connection is
a potential alternative to modular building connection with easier assembly and suf-
ficient stiffness. Another benefit provided by MIS is its flexibility, which is important
for connection design but is often neglected. Modules can be manufactured with dif-
ferent properties and dimensions of structural components from different manufactur-
ers. Dimensional flexibility allows the connection to fit different module units, thus
simplifies the connection design process. In addition, dimensional tolerance can help
address unpredictable constructional issues such as incorrect foundation alignment and
variations in space between modules [14].
Fig. 4. Structure of MIS connection (a) Interlocking connection strip with tongues and grooves
(b) Attachment of strips to modules [14].
Jenett et al. designed a modular, reconfigurable unit element at meso-scale using

discrete lattice material and an interlocking system (Fig. 5). As the unit elements can
connect to each other without additional connectors, they can freely form different
structures and can be dissembled easily [18]. The geometric freedom offered by the
design demonstrates the potential of reconfigurable connection systems. That means, it
can be possible for the structural engineer and architect to use bespoke unit designs to
build inter-module connections, intra-module connections and module-foundation
connections, and connections for linking modules in special orientations. Such designs
can certainly save significant amount of time in design and construction and allow
freedom in architecture.
Fig. 5. Example unit element and dimensions [18].
Interlocking is a promising solution for demountable, flexible and resilient modular

construction, however is restricted by conventional manufacturing methods. There are
some commonly used types of interlocking designs, such as pin and cavity, dovetail,
cantilever snap-fits and annular nap-fits, each of them offers different kinds of con-
straints [19]. The aforementioned connection adopts simple interlocking designs for the
ease of manufacture, so may not be adequate for large constructions. With the freedom
in geometric design, more complex interlocking design, alike the one shown in Fig. 5,
can be possible using AM, which offers multi-directional constraints while satisfying
the structural properties.
3 Application of Additive Manufacturing in Construction
3.1 Structural Topology Optimisation of Structural Elements

Exploring more innovative connection designs is still constrained by the limitations of
material properties and conventional manufacturing methods. AM as one of the most
advanced manufacturing methods, allows new ideas to develop in terms of material and
geometry while designing high-performance connections for MBSs. Geometric/
architectural freedom is a key benefit offered by AM. Complex features which are
not normally used in conventionally manufactured elements due to manufacturing
limitations, such as those including curve surfaces, can easily be introduced in 3D
printed designs [20, 21]. Therefore, AM is an ideal manufacturing method for complex
designs such those developed by using design optimisation (e.g., structural topology
optimisation) techniques resulting in better efficiency in structural elements by satis-
fying the mechanical demand with minimum weight [22]. However, experimental
testing on the behaviour of the 3D printed products, and certification, is much needed
[23].
In structural topology optimisation specifically, structural efficiency is achieved by
removing unloaded or less loaded areas under given load and boundary conditions [24].
Optimised outcomes should be combined together and refined to obtain the final
design, after some post-processing. The stability of the optimised design should be
verified by comparing the critical stress in all parts with the stress limits in all the load
cases [16].
364 Z. Li et al.
Many studies have presented the significant benefits that can be derived from
design optimisation, one of which is the direct reduction in material use. Arup [25]
applied STO to the joints of a tensegrity structure and achieved a 75% reduction in
connection weight and 50% reduction in height, which resulted in an overall weight
decrease of 40% (see Fig. 6). Smith [26] optimised a benchmark beam with different
nodal spacing and boundary conditions, achieving better strength-to-weight ratios
(maximum 18% higher) while remaining similar ultimate strength. Similar trends have
been demonstrated in other experiments including the design of node-connections for
lattice [18] and reticulated [27] structures (Fig. 7), indicating that STO is an effective
tool for reducing material usage while retaining the same structural capacity. More
importantly, the stress flow through the structure is optimised by the spatial reticulated
shape with alternative load path, which enhances the flexibility and resilience of
structure without adding more stress into the connection. Therefore, the optimised
node-connection enables lighter and fewer members to support the structure, leading to
significant reduction in the overall weight. This characteristic solution by STO is
advantageous to connection design when weight reduction is becoming more signifi-
cant in modular building construction, for the increase in seismic and wind perfor-
mance and the ease of assembly and lifting.
Fig. 6. Original joint design (left), interim version of optimised joint (middle), final optimised
design (right) [25].
Fig. 7. Optimised and 3D printed steel node-connection design [27].

Besides substantial weight reduction, STO can also be used to tailor geometrical
properties such as moments of inertia to achieve better stiffness in structural elements
[28], as seen in Fig. 8.
Fig. 8. STO on a fixed-ended beam (a) Optimisation input (b) Variable cross sections in
optimised beam [28].
3.2 Additively Manufactured Functionally Graded Materials

Functionally graded materials (FGM) are advanced materials characterised by the
gradual change of properties within the material, which enables multi-functional per-
formances within a part. Such materials are composed of a series of self-repeating base
cells [29]. Mechanical properties of FGM vary according to the structure of unit cell
and the direction in which the base units are assembled. Its application in engineering,
especially in the construction sector, is still rare due to the limited knowledge as well as
the difficulty and cost to manufacture using traditional processes. The excellent
adaptability in properties of FGM offers potential advantages in structural engineering,
especially for the structures subjected to complex loading conditions. Engineers can
tailor the FGM based on the performance requirements. AM with high geometric
freedom is well-suited to the production of such materials and this topic needs further
exploration.
It has been indicated in a previous study that, just like macro-level structures, the
unit cell of FGM can also be modified via STO to achieve better material properties
with minimum weight [29]. As shown in Fig. 9, cross-sections of the unit cells were
designed to change gradually to achieve continuously varying moduli of elasticity and
linearly varying shear moduli.
Fig. 9. Optimised cross-section of FGM with (a) nonlinearly varying modulus of elasticity
(b) linearly varying shear modulus of elasticity [29].
366 Z. Li et al.
Lomiento and Valdevit [30] proposed an innovative seismic insulation device using
functionally graded cellular materials which exhibited substantial improvements in
flexibility. A special unit cell structure was used at the millimetre scale to tailor the
mechanical properties of the device. The structure was composed of two plates, at the
top and bottom, an internal core that is free to roll, and a cylindrical shell in the middle
which connects them together (Fig. 10). The design of the internal roll enables the
structure with strong vertical stiffness but relatively lower lateral stiffness, which results
in improved lateral flexibility. The unit cell was then assembled layer-by-layer to form
the material of the isolator [30].
Fig. 10. Cellar periodic material for seismic insulator (a) Structure of unit cell (b) Assembly of
unit cells in layers [30].
Unlike conventional cellular material which absorbs energy by plastic deformation,

Izard et al. [31] introduced negative stiffness elements in the unit cell to dissipate energy
in a recoverable way (see Fig. 11). Similarity, Correa [32] proposed negative stiffness
Fig. 11. Unit cell with two positive stiffness elements and one negative stiffness element
(a) configuration of unit cell (b) numerical model of unit cell (c) three-dimensional lattice
material (d) 3D printed model of unit cell [31].
honeycombs with high initial stiffness and recoverable energy dissipation. This design
(see Fig. 12) achieved about 65% more energy absorption under displacement-driven
cyclic loading compared with conventional hexagonal honeycombs.
Fig. 12. Negative stiffness honeycombs (a) prototype of unit cell with rigid central beam
providing horizontal reinforcement (b) 3D printed negative stiffness honeycombs (c) mechanism
of negative stiffness honeycombs [32].
Carstensen et al. [33] proposed formulation for STO considering the material and
geometric nonlinear elastic designs, which can maximise the energy absorbing capacity
of cellular material. Compared with the conventional honeycomb structure, the opti-
mised cellular structure generated from the formulation improved the energy absorption
of brittle material bulk metallic glasses by 38%. This study shows tremendous
opportunities of exploring new capacities of existing materials (Fig. 13).
368 Z. Li et al.
Fig. 13. STO on bulk metallic glasses with nonlinear mechanics (a) honeycomb cellular
structure (b) optimised cellular structure (c) experimental stress-strain response [33].
The FGM designs mentioned above all hold great potential for connection designs
in modular building systems, especially for those designed for high-rise buildings since
it is dominated by lateral resistance. FGM allows directional variety in connection
property, and those with negative stiffness elements can increase the damping of the
overall structure for achieving better seismic and wind performance. Apart from the
weight reduction, the highly customised property of FGM provides the structural
engineer with more freedom in connection design and the opportunity to solve more
complicated structural problems.
3.3 Adaptive Structures

Adaptive structure is another innovative method to improve material property that is
characterised by the adaptive stiffness and adaptive geometry [34]; the stiffness
adaption of which is normally achieved by connecting stiff components and flexible
components via joints with transformation mechanisms as seen in Fig. 14 [35]. The
inherent properties of these structures can passively or actively adjust beneficially in
response to external stimulations [36]. Active control systems have been shown to be
effective in controlling the dynamic response of structures such as displacement and
acceleration [37–39] and Fig. 15. The concept of adaptive structures can also be the
inspiration for 3D printed smart material, e.g., the joint between unit elements, to
achieve positive adeptness in material properties. Smart materials such as shape
memory polymers (SMPs) and phase change materials (PCMs) are normally used to
form adaptive structures. The adaptive structure composed of SMPs can withstand
significant stiffness reduction under excitation and recover afterwards (Fig. 14a). PCM
normally have two phases and its stiffness changes during the phase transition due to
the change of temperature.
Fig. 14. Examples of transformable joints (a) shape memory polymer (SMP) joints at original
solid state (left), deformed soft state (middle) and recovered state (right) (b) phase change
material (PCM) joints at solid state (left) and soft state (right) [35].
370 Z. Li et al.
Fig. 15. 3D printed shape memory joint with adaptive thermomechanical properties [39].
3.4 Challenges of Applying Addictively Manufactured Materials

in Connection Design
Despite the benefits, constraints still exist in the practical application of the afore-
mentioned materials and structures. FGM is still a new emerging technology so most of
the analyses have been carried out numerically. Experimental evidence on the efficacy
of FGM material is currently insufficient. Moreover, the mass production of FGM is
still problematic. It has been demonstrated that FGM is highly sensitive to dimensional
changes to the unit cell and the process parameter sets of AM, such as laser power, scan
speed, layer thickness, and more [21, 25, 37]. Therefore, high accuracy is required in
FGM manufacturing, so Selective Laser Melting (SLM) and Electron Beam Melting
(EBM) could be better AM options [37]. Currently, no specification is available to
assess these impacts during manufacture [25, 21]. So further studies are necessary to
optimise the AM process for more accurate geometry and property control to achieve a
stable outcome in material properties. Moreover, for the practical application, com-
prehensive databases for addictively manufactured FGM needs to be set up [40].
Another main restriction of the large-scale application of AM in the construction
sector is the cost. AM is now mainly used in prototype production. With the constant
improving technology and the elapse of patents, the market of the 3D printing will open
up and the price will gradually reduce, making it more feasible for general manufac-
turing process. Frankly, AM is highly independent to the geometry so it can manu-
facture different products without requiring additional assembly line setup which is
expensive and time-consuming [41]. It is therefore well justified to believe that with the
development of AM, 3D printed functionally graded material can become reality in the
near future, which means that highly efficient connection designs with customised
geometries and tailored properties will be possible.
4 Conclusions
To further improve the eco-friendliness and efficiency of modular construction and in

particular modular building systems, the flexibility as well as the demountability of
modular components and thus their connection systems need to be re-considered. The
concept of interlocking system has been adopted in some innovative connection
designs to reduce the use of additional connectors, while they are still constrained by
the limitation of conventional manufacturing methods, cost, and understanding
performance. additive manufacturing (AM) is a rapidly developing technology which

can bring new possibilities to complex connection design. The geometric freedom
offered by AM together with innovative materials can be achieved with the use of
powerful optimisation tools fostering the development of resilient and adaptable pro-
totype connection designs. This paper provides ideas and thoughts of how interlocking
connections, structural topology optimisation (STO), and functionally graded materials
(FGM) can be all integrated to improve the adaptability of modular building connec-
tions via exploiting AM potentials according to existing studies. Future work involves
the geometric design of interlocking systems as well as unit cells of FGM that are
suitable for modular joints and connections with specifically optimised mechanical
performance.
References
1. Ghassemieh, M., Kiani, J.: Seismic evaluation of reduced beam section frames considering
connection flexibility. Struct. Des. Tall Spec. Build. 22(16), 1248–1269 (2013)
2. Buchanan, C., Gardner, L.: Metal 3D printing in construction: A review of methods,
research, applications, opportunities and challenges. Eng. Struct. 180, 332–348 (2019)
3. Chuisoli, A.: The first 3D printed House with earth-Gaia. https://www.3dwasp.com/en/3d-
printed-house-gaia/. Assessed 18 Feb 2020
4. Apis, C.: Collaborative Project with Dubai Municipality. https://www.apis-cor.com/dubai-
project. Assessed 18 Feb 2020
5. Boissonneault, T. SQ4D completes construction of 1,900-sq-ft 3D printed home. https://
www.3dprintingmedia.network/sq4d-1900-sq-ft-3d-printed-home. Assessed 18 Feb 2020
6. Lawson, P.M., Byfield, M.P., Popo-Ola. S.O., Grubb, P.J.: Robustness of light steel frames
and modular construction. Proc. ICE – Struct. Build. 161(1), 3–16 (2008)
7. Crawley, S.W., Dillon, R.M.: Steel Buildings: Analysis and Design, 3rd edn. Wiley, New
York (1984)
8. Lacey, A., Chen, W., Hao, H., Bi, K.: Structural response of modular buildings – an
overview. J. Build. Eng. 16, 45–56 (2018)
9. Chen, Z., Liu, J., Yu, Y.: Experimental study on interior connections in modular steel
buildings. Eng. Struct. 147, 625–638 (2017)
10. Lacey, A., Chen, W., Hao, H., Bi, K.: Review of bolted inter-module connections in modular
steel buildings. J. Build. Eng. 23, 207–219 (2019)
11. Deng, E.H., Yan, J.B., Ding, Y., Zong, L., Li, Z.X., Dai, X.M.: Analytical and numerical
studies on steel columns with novel connections in modular construction. Int. J. Steel Struct.
17(4), 1613–1626 (2017)
12. Styles, A.J., Luo, F.J., Bai, Y., Murray-Parkes, J.B.: Effects of joint rotational stiffness on
structural responses of multi-story modular buildings. In: Transforming the Future of
Infrastructure through Smarter Information: Proceedings of the International Conference on
Smart Infrastructure and Construction. ICE publishing, pp. 457–462 (2016)
13. Gunawardena, T.: Behaviour of prefabricated modular buildings subjected to lateral loads
[Ph.D. Thesis]. The University of Melbourne (2016)
14. Sharafi, P., Mortazavi, M., Samali, B., Ronagh, H.: Interlocking systems for enhancing the
integrity of multi-storey modular buildings. Automat. Construct. 85, 263–272 (2018)
15. Messler, R.: Integral Mechanical Attachment a Resurgence of the Oldest Method of Joining.
Butterworth-Heinenmann, Burlington (2006)
372 Z. Li et al.
16. MHS Buildings System Homepage. http://www.modularhousingsystems.com/tp2.html.

Assessed 07 Feb 2020
17. MHS Buildings Systems Homepages. About MHS Building Systems 2020. http://www.
modularhousingsystems.com/download.html. Assessed 06 May 2020
18. Jenett, B., Cellucci, D., Gregg, C., Cheung, K.: Meso-Scale digital materials: modular,
reconfigurable, lattice-based structures. In: International Manufacturing Science and
Engineering Conference (2016)
19. Flynt, J. How to design interlocking parts for 3D printing. https://3dinsider.com/3d-printing-
interlocking-parts/. Accessed 09 Feb 2020
20. Strauß, H., Knaack, U.: Additive manufacturing for future facades: the potential of 3D
printed parts for the building envelope. J. Facade Des. Eng. 3(3–4), 225–235 (2016)
21. Feucht, T., Lange, J., Erven, M.: 3-D-printing with steel: additive manufacturing of
connection elements and beam reinforcements. CE/Papers 3(3–4), 343–348 (2019)
22. Pradhan, K.K., Chakraverty, S.: Computational structural mechanics. Matthew Deans,
California (2019)
23. Buchanan, C., Matilainen, V., Salminen, A., Gardner, L.: Structural performance of additive
manufactured metallic material and cross-sections. J. Construct. Steel Res. 136, 35–48
(2017)
24. Kazakis, G., Kanellopoulos, I., Sotiropoulos, S., Lagaros, N.: Topology optimization aided
structural design: Interpretation, computational aspects and 3D printing. Heliyon 3(10),
e00431 (2017)
25. Galjaard, S., Hofman, S., Perry, N., Ren, S.: Optimizing structural building elements in
metal by using additive manufacturing. In: Proceedings of the International Association for
Shell and Spatial Structures (2015)
26. Smith, C.: Application of layout optimisation to the design of additively manufactured
metallic components. University of Sheffield (2016)
27. Abdelwahab, M., Tsavdaridis, K.D.: Optimised 3D-Printed metallic node-connections for
reticulated structures. In: The 9th International Conference on Steel and Aluminium
Structures, 3–5 July, Bradford, UK (2019)
28. Tsavdaridis, K.D., Hughes, J.A., Grekavicius, L., Efthymiou, E.: Novel optimised structural
aluminium cross-sections towards 3D printing. In: Meboldt, M., Klahn, C. (eds.)
Industrializing Additive Manufacturing - Proceedings of Additive Manufacturing in
Products and Applications - AMPA2017. AMPA 2017. Springer, Cham (2018)
29. Radman, A., Huang, X., Xie, Y.: Topology optimization of functionally graded cellular
materials. J. Mater. Sci. 48(4), 1503–1510 (2013)
30. Lomiento, G., Valdevit, L.: Cellular periodic material design for enhanced seismic protection
of low-rise buildings. In: 16th World Conference on Earthquake Engineering, Santiago,
Chile (2017)
31. Guell Izard, A., Fabian Alfonso, R., Mcknight, G., Valdevit, L.: Optimal design of a cellular
material encompassing negative stiffness elements for unique combinations of stiffness and
elastic hysteresis. Mater. Des. 135, 37–50 (2017)
32. Correa, D., Klatt, T., Cortes, S., Haberman, M., Kovar, D., Seepersad, C.: Negative stiffness
honeycombs for recoverable shock isolation. Rapid Prototyp. J. 21(2), 193–200 (2015)
33. Carstensen, J., Reza, L., Chen, W., Schroers, J.: Topology optimization of cellular materials
with maximized energy absorption. In: ASME 2015 International Design Engineering
Technical Conferences, Boston, USA (2015)
34. Sobek, W., Teuffel, P.: Adaptive systems in architecture and structural engineering. In:
SPIE’s 8th Annual International Symposium on Smart Structures and Materials, pp. 36–45
(2001)
35. Wang, Q., Teuffel, P.: Adaptive structures and design concept of transformable joints. Int.
J. Des. Nat. Ecodyn. 12, 235–245 (2017)
36. McGraw-Hill Dictionary of Scientific & Technical Terms., Adaptive Structure. https://
encyclopedia2.thefreedictionary.com/adaptive+structure. Accessed 14 Feb 2020
37. Mirfakhraei, S.F., Andalib, G., ChanNumerical, R.: Investigation on toggled actuator forces
in active vibration control system. Adv. Res. Civil Eng. 1(2), 16–35 (2019)
38. Zhao, Y.L., Xu, Z.D., Cheng, W.: Vibration control of stay cables using magnetorheological
dampers under optimal equivalent control algorithm. J. Sound Vibrat. 443, 732–747 (2019)
39. Wang, Q., Teuffel, P., Habraken, A.P.H.W., Jansen, K., Senatore, G.: Characterization of
variable stiffness joints for adaptive structures. In: Proceedings of the 7th International
Conference on Structural Engineering, Mechanics and Computation, pp. 1046–1050, Taylor
& Francis, London (2019)
40. Mahamood, R.M., Akinlabi, E.I., Shukla, M., Pityna, S.: Functionally graded material: an
overview. In: Proceedings of the World Congress on Engineering, vol. 3, London (2012)
41. Marlin S.: 3D Printing Vs Traditional Manufacturing, https://www.marlinwire.com/blog/3d-
printing-vs-traditional-manufacturing. Accessed 06 May 2020
Application of Topology Optimisation to Steel
Node-Connections and Additive
Manufacturing
Moustafa Mahmoud Abdelwahab

and Konstantinos Daniel Tsavdaridis(&)
School of Civil Engineering, Faculty of Engineering and Physical Sciences,

University of Leeds, Leeds LS2 9JU, UK
k.tsavdaridis@leeds.ac.uk
Abstract. Structural Topology Optimisation (STO) is a prevalent optimisation

technique used nowadays to reach desired weight-to-stiffness ratios via highly
complex and efficient designs unable to achieve otherwise. Additive manufac-
turing (AM) is widely known in the manufacturing industry and provides
designers with a higher degree of freedom in realising highly optimised designs
through a layer-based fabrication process. This paper focuses on reticulated
structures and proposes using STO and AM to design and fabricate alternative
connection designs with outstanding bespoke performance and drastically
reduced weight. It studies the optimisation of a conventional node-connection
found in reticulated timber structures under four loading cases, to producing
state-of-the-art optimised connection designs, each capable of withstanding one
of the four selected loading cases. The results are compared with the conven-
tional node-connection, and the optimised configurations achieved up to 46.9%
weight reduction. A selection of the highly bespoke scaled-down designs was
additively manufactured in two different materials (metallic and polymer) as a
proof of concept for the capacity of the technologies available for future testing.
Keywords: Structural Topology Optimisation Additive Manufacturing

Structural connections 3D printing Reticulated structures Advanced
manufacturing
1 Introduction
1.1 Background and Critical Review

Reticulated roofs are highly versatile structures and are widely used for large and
landmark projects. Otherwise known as spatial structures, such structures achieve a
cost-effective span-weight ratio for roof constructions while reducing the number of
intermediate column supports required. Their numerous applications range from sta-
diums to warehouses to aircraft hangers [1]. Fundamental changes and improvements
have been undertaken on such structures over the years to increase their structural
capacity, simplify their fabrication and erection techniques, and improve the perfor-
mance of their node-connections. Yet, the fundamental worldwide shift in architectural

https://doi.org/10.1007/978-3-030-54334-1_26
Application of Topology Optimisation to Steel Node-Connections and AM 375
perspectives from modular and repetitive to free-form and unique is imposing critical
challenges and limitations, most of which are related to the connections of such
structural systems.
Recent attempts by researchers aimed at exploring the potential of using Structural
Topology Optimisation (STO) as a mean of creating bespoke node-connection designs
which can accommodate the variability of the tie member directions and angles in free-
form shapes [2–4]. The node-connection designs achieved had the major benefit of
attaining significant weight reductions which on a large scale can result in an
improvement to the overall structural performance. Such node-connection designs were
observed to have geometrical complexities and curvature patterns which were highly
challenging to construct using traditional fabrication methods such as milling, casting
or using subtractive manufacturing. Additive manufacturing emerged as the ideal
fabrication solution for this challenge.
1.2 Background and Critical Review

Additive Manufacturing (AM) allows designers to create physical models through
depositing and joining thin layers of a material together based on a three dimensional
computer model [5]. Research institutions and the industry worldwide are investing in
AM due to its high precision and minimal fabrication limitations when compared to
traditional manufacturing. The technology has been industrialised since the 90 s
through improving the fabrication processes and allowing fabrication using a wider
range of materials which include metals, ceramics, and polymers [6]. From the liter-
ature, it is apparent that manufacturing industries related to aerospace, automotive, and
medical (e.g., prosthetics) have declared strong interest in attaining the full potential of
AM, unlike the construction industry which seems to be moving at a much slower pace
for a number of reasons (e.g., size of printed elements, high load bearing, long lifespan,
etc.) [6]. This paper aims to support the research in bridging the gap between AM and
its application in the construction industry through the adoption of additively manu-
factured node-connections for reticulated structures.
1.3 Research Objectives

The aim is to build upon previous research attempts through optimising single-layer
node-connections, and testing against several loading cases that allow for a deeper
understanding and interpretation of the optimised designs. The need for creating several
connections is to allow for a fair comparison between a range of results and a con-
ventional connection that provides inherent robustness previously associated with the
uncertainly of loading cases and simplistic designs. Meticulous optimisation studies
and numerical analyses with specialised and sophisticated software tools are employed
to carry out this research.
376 M. M. Abdelwahab and K. D. Tsavdaridis
2 Selection Criteria of the Connection Design
A comprehensive study was undertaken to identify practical node-connections com-

monly used in the industry and choose the potential candidate which would represent
the control sample. Simplicity was the key part of the research, therefore, a list of
questions was prepared which helped decide upon the joint type which is likely to be
the simplest to model and optimise. Yet, that will yield significant weight reductions
which then translate to reduced carbon emissions and a more sustainable design. The
questions considered include the following:
• What is the total volume occupied by the connection?
• What is the material percentage in the total volume occupied by the connection?
• How simple would the connection be to create, model, and test?
• How simple would the connection be to predict its behaviour upon modelling?
• Does the connection need to have any internal bolting or welding?
• How are the tie members attached to the connection?
• How much commercial knowledge is available to judge upon the typical loading
capacity and dimensions of the connection?
3 Conventional Node-Connection Design Concept
Given the above questions and after reviewing numerous connection types, the Splice
Node-Connection “POLO-1” shown in Fig. 1 has been chosen as it represented the
ideal candidate for this research study [7, 8]. The applications employ this connection
range from geodesic domes to reticulated structures. The connection consists of a
central thick RHS core with several splice plates welded to it. The tie members attached
to this connection are fabricated with a fork-shaped ending which fits in between the
plates of the connection. The central RHS region provided the large volume needed for
the freedom to apply the shape formation with the optimisation algorithm and easily
interpret the formed designs and structural behaviour. Furthermore, eliminating any
bolts from the core part of the connection and keeping the bolting to the external splice
steel plates, provide further simplicity while running the analysis.
Fig. 1. Splice node-connection “POLO-1” [9, 10].

3.1 Connection Anatomy

The structure of the connection was divided into two regions, the design domain and
non-design domain. The design domain region as shown in red in Fig. 2, is the region
which was optimised by the software. The non-design domain region, not optimised by
the software, consists of the connecting splice steel plates indicated in grey, and the
semi-cylindrical members indicated in green which were added to allow for improved
stress flow and shape development within the design domain area. Similar model
division has been undertaken in the conventional connection design to ensure a fair
comparison between the stress and strain level of the results of both the conventional
and optimised connections.
Non-Design Domain
Fig. 2. Design anatomy of the conventional node-connection and “Connection A”.
3.2 Design Dimensions

The dimensions and geometries of the connection of reticulated structures are com-
mercial design ‘secrets’ which are usually developed through years of development and
testing. As a result, only limited design information concerning these connections are
provided by the manufacturers. The research team herein adopted case studies from the
industry to select representative connection dimensions which are suitable for con-
necting to available structural strut options. A practical case study is the Eden Project
dome which was designed using a bespoke type of semi-spherical nodes known as the
top chord “bowl” node [11]. The node had a diameter of 400 mm, a plate thickness of
40 mm, a weight of about 80 kg and a chord tube diameter of 193.7 mm. Following
the same design indications provided in the case study, a POLO-1 node with similar
dimensions was produced as shown in Fig. 3. The node-connection can practically
connect a 300 mm deep timber joist or a steel circular/rectangular hollow section using
fork-like fin-plate connection as indicated below in Fig. 4 below. The conventional
node-connection design weighted around 127 kg per joint in full scale.
Fig. 3. Dimensions of the conventional node-connection.
Fig. 4. Visualisation of the conventional node-connection with tie members.
3.3 Load Cases

The four connection test samples were analysed each against one of the four loading
cases as shown in Fig. 5. These loading cases have the benefit of simulating a spectrum
of asymmetrical loading scenarios and provide the variations required in the loading
types (tensile/compressive) and magnitudes. Even though this connection type has
some inherent moment capacity to it, industry practices were followed in this research
through only applying axial loads in the optimisation process. The resultant of all the
applied loads in each of the four loading cases sums up to zero, thus in equilibrium.
Fig. 5. Loading cases applied to the node-connections.
3.4 Methods and Tools

Three software packages were used throughout the various stages of this research. Fusion
360 was the main design tool which was used to create the connection designs and
undertake shape improvements. ABAQUS/CAE was used for the FE analysis testing and
the STO stages. MeshMixer was then used to post-process and refine the rough and low-
quality connection designs extracted from ABAQUS to more printable structures.
4 Optimisation Process
The optimisation process used in this research, as shown in Fig. 6, contains the core
elements from the processes established by previous relevant research [12, 13]. Steps 1
and 2 in the design process started by creating the conventional connection model in
Fusion 360 and visually optimise its shape [14]. A modified version of the conventional
connection shown in step 3 was created with the central void filled with material. This
modified version which is referred to as “Connection A” was utilised during the
optimisation task. “Connection A” was imported into ABAQUS, meshed and analysed
separately against each of the four loading cases considered, as shown in steps 4 and 5.
“Connection A” was then optimised in step 6 at a specified weight percentage against
one loading case at a time to obtain the final optimised results after roughly three
hundred trials. To obtain “Connection B” which was optimised simultaneously against
the abovementioned four loading cases, step 6 was re-undertaken with the four loading
cases input into the software rather than one single case. Designing one joint against
one loading case early on, is a step in the process of attaining and analysing the resulted
“Connection B”. The structural integrity and robustness along with the manufactura-
bility of the produced optimised connections in step 7 were visually assessed using
engineering intuition and an understanding of the AM processes involved. As a result, a
continous trial and improvement cycle was undertaken between stages 3 and 7 to
indentify the optimal design parameters which produces some highly robust and pro-
ducable connections. In steps 7 and 8, the produced designs were then exported from
ABAQUS to MeshMixer for post-processing and smoothening to prepare for the AM
process. Finally, in step 10, the modified conventional connection design was imported
into ABAQUS and applied the same conditions, and tested against the aforesaid
loading cases in order to act as the control sample in this research.
Fig. 6. The design process.
4.1 Design Modifications

Initial modifications to the design of the splice steel plates were introduced through
adding the semi-circular shape shown in Fig. 7 around the bolt holes in all the con-
sidered models and filleting the sharp edges of the holes. Such modifications allowed
for a further reduction in the overall weight of the connections, allowed for a better
stress flow through the structure and utilised the full fabrication potential of AM
through introducing complex geometry.
Fig. 7. Splice Node-Connection “Polo-1” before and after design modifications.
4.2 Material Properties and Permissible Stress Levels

For all the node-connections in this project, a steel grade of S355 was used to define the
material properties; 210GPA was chosen as the Modulus of Elasticity and 0.3 as the
Poisson’s ratio. An early assumption was made regarding the permissible stress levels
in the node-connections (permissible strain energy is used in the design of free-form
reticulated structures). A linear-elastic analysis was undertaken; therefore, it was
assumed the stress levels in the design-domain region of the connections shall not
exceed 355 N/mm2. The loading cases and percentage weight reduction considered
were achieved through trying a number of iterations while monitoring the maximum
stress levels observed.
4.3 Meshing
“Connection A” was imported in ABAQUS and the properties were assigned as well as
the loads and boundary conditions were applied. The FE mesh of the connection is
shown in Fig. 8. A free meshing technique using quadratic tetrahedron element (Ele-
ment type: C3D10) was employed due to the model’s complex geometry. The mesh
size was chosen as 12 (unit less) and remained constant across all the models under-
taken in this research to avoid accuracy-related variations. The number of elements and
nodes obtained in the model are 148,438 and 212,127, respectively.
Fig. 8. FE mesh of “Connection A” for the optimisation
4.4 Weight Percentage

The weight percentage target of “Connection A” to be used in the optimisation tasks
was set in the analysis tool as 53.1% (reduce weight by 46.9%), which produced
optimal designs that consistently remained within the linear elastic range. Increasing
the weight reduction percentage in most of the optimisation tasks resulted in the failure
of the resulted optimised joint designs. The optimised real-size connections weighted
approximately 67.5 kg on average, achieving a weight reduction of 59.5 kg compared
to the conventional connection. Studying the stress flow within the connection and
applying further modifications could help reduce the weight even further.
The few initial loading cases considered when optimising “Connection A” were
excluded either due to excessively high stress levels or minimal material formation in
certain areas. This minimal material formation in the optimisation study was later
linked to low applied loads, which resulted in low straining, and since the STO tool is
driven by increasing strain, minimal or no material was formed throughout the design-
domain region. Based on that, loading cases with high enough magnitudes were
selected. These loading cases utilised a significant capacity of the splice steel plates and
resulted in a reasonable material formation throughout the design-domain region.
4.5 Optimisation Approach

A stiffness-based structural topology optimisation process has been adopted in this
research which allows the addition of material only in the regions displaying noticeable
displacement in the structure. Thus, the design-domain region in “Connection A” has
been filled with material as shown in Fig. 8 to undertake the optimisation work.
Following the model preparation, the design-domain region was defined in ABA-
QUS and “Minimising the connection volume” of the target volume was introduced as
the optimisation objective, while “Minimising the maximum design response values” of
the strain energy was taken as the optimisation constraint. For each of the loading cases,
an optimised design has been obtained as shown in Figs. 9, 10, 11 and 12. In order to
obtain “Connection B” which is shown in Fig. 13, the optimisation objective remained
the same, while the optimisation constraint was changed to account for the maximum
strain value associated with each loading case. The maximum number of cycles allowed
in each optimisation trial was taken as 200 cycles, which was proven adequate to allow
the software to reach the optimal design, achieving the highest stiffness at the target
volume specified. The actual number of cycles ranged from 28 to 50.
Load Case 2 (LC2) Legend
Load Case 1 (LC1)
Fig. 9. LC1 optimised design. Fig. 10. LC2 optimised design.

Load Case 3 (LC3) Load Case 4 (LC4)
Fig. 11. LC3 optimised design. Fig. 12. LC4 optimised design.
Fig. 13. FEA structural testing of “Connection B” against LC1
4.6 Comparative Study

Once all the optimisation work has been undertaken, the same load cases used in the
optimisation tasks were applied to the reference connection. Figure 14 shows the Von
Misses stress contour plot of the conventional connection against load case 1. Identical
dimensions, features, and meshing techniques were used to avoid accuracy-related
variations. Moreover, the stress and strain levels were measured at the design-domain
region of the conventional connection to allow for a fair comparison.
Fig. 14. FEA structural testing of the conventional node-connection against LC1.
5 Optimisation Process
Figures 15 and 16 demonstrate a comparison between the recorded stress and strain
levels in the design-domain area in the reference conventional connection, the opti-
mised designs, and “Connection B”. The non-design domain stress and strain levels are
not important for this research study; therefore, their values were not presented.
In loading cases 2 and 3, which consisted of only tensile or compressive forces in
one case, with different magnitudes acting on the splice plates, the conventional con-
nection displayed the lowest stress levels and significantly low strain levels. As for
loading cases 1 and 4, which consisted of a combination of tensile and compressive
Fig. 15. Recorded stress levels.

loads on the splice plates, the optimised models had better performance with lower
stresses than both the reference connection and “Connection B” while displayed rel-
atively low strain levels. As “Connection B” was optimised to all of the four loading
cases, the ideal shape formed was not specific towards a single loading case, rather, all
of them. Therefore, the resulted “Connection B” contains the highest stress levels in
three out of four of the cases and some varying strain levels.
Fig. 16. Recorded strain levels.
6 Manufacturing Process
The optimised designs extracts from ABAQUS inherently experienced low mesh
quality and required further refinement in preparation for the AM process. In total, four
scaled-down optimised designs with bespoke and complex geometries were manu-
factured. The fabrication of these prototypes showcases the capability of AM in pro-
ducing intricate shapes and its advantage through eliminating many of the traditional
manufacturing constraints.
6.1 Post-processing and Preparation for the Additive Manufacturing

The post-processing steps undertaken on the optimised designs ‘prepared’ them
structurally and geometrically for the AM process. These steps involved creating
smoother curves, removing any weak regions likely to develop during the printing, and
reducing the number of sacrificial structural supports required. Moreover, the modifi-
cations undertaken improved the mesh quality to allow for better fabrication as shown
in Fig. 17. Care was taken while using MeshMixer to maintain the original design-
domain geometry of the connections, while allowing for improved overall curvatures in
the shape.
Fig. 17. An optimised node-connection before and after post-processing.
6.2 Additively Manufactured Designs and Material Properties

Two different materials were used in constructing the designs as the desired material is
not commonly adopted in the AM markets. The first material is a metallic alloy formed
through a combination of 420 Stainless Steel (60%) with Bronze (40%), while the
Bronze component consists of 90% Copper and 10% Tin. According to the manu-
facturing company (i.materialise based in Leuven, Belgium), this alloy with a density
of 7.86 g/cm3 has a Modulus of Elasticity of 147GPa, a 0.2% offset Yield Strength of
455 MPa, along with an elongation of 2.3%, and an ultimate tensile strength of
682 MPa [15]. Thus, the research team adopted this material in the three geometrically
most complex designs to showcase the limitless fabrications capabilities of AM and the
aesthetical benefits gained, as shown in Figs. 18 and 19. These connections could be
involved in future lab-based testing and validation studies for analytical models.
Ongoing research focusing on refining the geometry of the connections. Moreover, the
true metallic properties and any anisotropic effects will be characterised after coupon
tests to allow for more reliable and accurate FEM validation.
The second choice is a polymer-based material, trademarked under the name of
“nylons” and referred to as “polyamide” by the manufacturer. This material was
selected as a showcase (demo) material for the fabrication of the reference connection
and the complex design generated with a weight reduction of 46.9%.
The designs printed in the metallic alloy were manufactured through the “Binder
Jetting” Technology while the polyamide designs were printed through trademarked
process referred to as “Multi Jet Fusion Technology” (MJF).
Fig. 18. Additively manufactured node-connection designs.
Fig. 19. Visualisation of the assembled optimised node-connection with strut members.
6.3 Cost Comparisons

The three most pressing factors for the introduction of AM in the construction sector
are the limiting fabrication dimensions, printing time, and high cost. The first factor can
be considered early on in a project by controlling the maximum object/connection size
to meet the print volume of the available metallic printer. With 3D printing patents
lapsing, more companies will develop newer and bigger printers for the construction
industry needs. The second factor can be met by allowing enough time in the con-
struction programme for fabricating the connections ahead of erecting the super-
structure. Since structural components can be prefabricated and stored in the ware-
house, the exposure time on-site can be significantly minimised and well managed. The
third factor requires more research on to present in-depth statistics. AM can be con-
sidered an expensive process for the typical mass-production but an ideal solution for
highly-optimised and bespoke designs. The main reason for considering AM connec-
tions is that they are typically labour intensive and may require a heavy and expensive
fabrication either using injection moulding or welding tailored to each connection
specifically. As the AM market will open after the lapsing of 3D printing patents,
competition will be increased, also leading to lowering costs.
An exact cost for each connection type could not be obtained from the suppliers, as
the design of such joints is yet considered a commercial secret by its manufacturers and
is very much dependent on the complexity, quality, and number of the designs.
Therefore, any figures and cost estimations presented are solely based on case studies
attained from the literature and industrial experience of the researchers. The cost of the
manufacturing the scaled down metallic connections (scaled down by 75%) ranged
between $635–680 per node. Scaling up the metallic connections would result in an
approximate fabrication cost of 2.4 times higher (0.753 = 0.42, i.e., 42% of the vol-
ume), thus ranging between $1,510–1,620 per node (£1,210–1,300). The two slightly
smaller demo Polyamide connections (as in Fig. 18) were fabricated with a much lower
price tag of £35–47 per node.
Adopting the case study of the New Trade Fair freeform reticulated structure in
Milan [16], the total roof weight is around 77,000 tonnes and for an average steel price
of around £900 per tonne, the total steel tonnage cost of the roof is £69,300,000
(excluding design, manufacturing, and transportation costs). The connections typically
account for 20% of the frame from experience and as a result, the connections in the
project would cost £13,860,000 (£770 per each of the 18,000 nodes). As anticipated,
the optimised AM connections remain higher in cost; however, considering the weight
reduction of the entire structure, not only due to the connection weight reduction, but
also the cost reduction as a result of the shape optimisation on the roof structure, could
account for 46.9% of the connections cost (i.e., 53.1% of the initial cost). This would
reduce the cost per connection to £410 per node (£70 per connection times 0.531),
assuming direct correlation between steel tonnage and cost. No definitive conclusions
can be drawn from this, since more detailed cost-related studies are required to assess
the feasibility of this design and fabrication method for the joints of large-scale
reticulated structures, but they present descent indicative costs.
7 Conclusions and Ongoing Research
Recently, structural topology optimisation (STO) together with additive manufacturing

(AM) have been employed for various structural applications in the Civil Engineering
sector [17, 18]. This study enhances this knowledge by investigating the application of
STO and AM for the design optimisation of node-connections found in reticulated
structures. A conventional node-connection is optimised against individual four load-
ing cases at a percentage weight reduction of 46.9%. In addition, an optimised con-
nection design “Connection B” is achieved, being capable of withstanding each of the
four loading cases. A comparison of the design-domain stress and strain levels between
the reference conventional connection and the optimised connections is also presented
in this paper. Finally, a selection of the generated highly bespoke and complex opti-
mised designs has been manufactured in metal and polyamide.
Ongoing research is focusing on improving the complexity of the node-connection
model by using the multi-material components, validating the results through experi-
mental testing as well as considering a wider range of adverse loading cases in
designing “Connection B” and comparing its performance to the conventional con-
nection. Research also focuses on studying the cost-effectiveness and environmental
impact of the additively manufactured connections to the entire reticulated structure.
Acknowledgements. The authors of this paper would like to thank Lord Laidlaw for the
generous financial contribution (Laidlaw Scholarship) and the Leadership training provided by
the Laidlaw Undergraduate Research and Leadership Scholarship programme. The authors would
also like to acknowledge the contribution to this project by Dr Osvaldo M. Querin from the
School of Mechanical Engineering at the University of Leeds for his guidance on the structural
optimisation process.
References
1. Ramaswamy, G., Eekhout, M., Suresh, G., Papadrakakis, M.: Analysis, Design and
Construction of Steel Space Frames. Thomas Telford, London (2002)
2. Galjaard, S., Hofman, S., Perry, N., Ren, S.: Optimizing Structural Building Elements in
Metal by using Additive Manufacturing (2015)
3. Galjaard, S., Hofman, S., Ren, S.: New opportunities to optimize structural designs in metal
by using additive manufacturing. Adv. Archi. Geometry 2014, 79–93 (2014)
4. Crolla, K., Williams, N., Muehlbauer, M., Burry, J.: Smartnodes pavilion towards custom-
optimized nodes applications in construction. In: RMIT Research Repository, pp. 467–477
(2015)
5. Ngo, T., Kashani, A., Imbalzano, G., Nguyen, K., Hui, D.: Additive manufacturing (3D
printing): a review of materials, methods, applications and challenges. Composites Part B:
Eng. 143, 172–196 (2018)
6. Wang, X., Jiang, M., Zhou, Z., Gou, J., Hui, D.: 3D printing of polymer matrix composites:
a review and prospective. Composites Part B: Eng. 110, 442–458 (2017)
7. Hwang, K.: Advanced investigations of grid spatial structures considering various
connection systems, OPUS - Publication Server of the University of Stuttgart. (2010)
8. Stephan, S., Sánchez-Alvarez, J., Knebel, K.: Reticulated Structures on Free-Form Surfaces
(2010)
9. EA European Architecture, 120 EA-Wood Structures, Pinterest (2018)
10. Design Biennial Boston, Archdaily (2015)
11. Mero, Eden Project: Construction, Assembling and Manufacturing, mero.de (No date)
12. Seifi, H., Rezaee Javan, A., Xu, S., Zhao, Y., Xie, Y.: Design optimization and additive
manufacturing of nodes in gridshell structures, Eng. Struct. 160, 161–170 (2018)
13. Seifi, H., Xie, M., O’Donnell, J., Williams, N.: Design and fabrication of structural
connections using bi-directional evolutionary structural optimization and additive manufac-
turing. Appl. Mechan. Mater. 846, 571–576 (2016)
14. O’Donnell, J., Seifi, H., Sitler, B., Williams, N., Crolla, K., Xie, Y.: Smart nodes pavilion:
Bi-directional evolutionary structural optimization and additive manufacturing. In: RMIT
Research Repository, pp. 1–12 (2015)
15. i.materialise, Technical Specifications (420 Stainless Steel), i.materialise.com (2019)
16. Glass Magazine, New Milan Trade Fair, Tabpi.org (2019)
17. Tsavdaridis, K.D., Hughes, J.A., Grekavicius, L., Efthymiou, E.: Novel optimised structural
aluminium cross-sections towards 3D printing. In: Meboldt, M., Klahn, C. (eds.) AMPA
2017, pp. 34–46. Springer, Cham (2018). https://doi.org/10.1007/978-3-319-66866-6_4
18. Abdelwahab, M., Tsavdaridis, K.D.: Optimised 3D-printed metallic node-connections for
reticulated structures. In: The 9th International Conference on Steel and Aluminium
Structures. 3–5 July, 2019, Bradford (2019)
The AM Dowel – A Novel Insert
for the Integration of Threads into Additive
Manufactured Polymer Components
Daniel Omidvarkarjan1(&), Peter Balicki2, Harry Baumgartner2,

Ralph Rosenbauer3, Filippo Fontana4, and Mirko Meboldt2
1
inspire AG, Leonhardstrasse 21, 8092 Zurich, Switzerland
omidvarkarjan@inspire.ethz.ch
2
Product Development Group Zurich pd|z, ETH Zurich,
Leonhardstrasse 21, 8092 Zurich, Switzerland
3
Fährstrasse 34, 3004 Bern, Switzerland
4
Güterstrasse 23, 8952 Schlieren, Switzerland
Abstract. Within this paper, an insert for the integration of female threads into
additive manufactured (AM) polymer parts is presented. The fastener consists of
three components: a metallic threaded insert, an AM optimized female cutout
and an AM polymer clip. By exploiting AM’s geometrical complexity, the
design has several advantages compared to conventional solutions including a
fast and reversible assembly process without the use of specialized tools. The
insert’s pull-out strength was evaluated on a tensile testing machine. The results
showed that the fastener can withstand loads of up to 5600 N (for 3/8″-16 UNC
thread size). Within a case study, the insert is integrated into a Selective Laser
Sintered (SLS) top handle for videography applications. The fasteners were fully
functional and endured extensive real-life operation. The results of the experi-
mental tests and case study demonstrate that the novel insert design creates new
potentials for AM products, as additional applications for polymer-based AM
processes are enabled. Customization and flexibility are established within the
hardware part as designers can choose their required insert from ready-to-use
building blocks & even change it throughout use as the connection is stan-
dardized and reversible.
Keywords: Additive Manufacturing Threaded insert Design for AM

Hybrid design
1 Introduction
In recent years, Additive Manufacturing (AM) has consistently been more established
in industry. Starting from rapid prototyping, the technology is increasingly more used
for the production of end-user parts and products [1]. Especially polymer-based pro-
cesses are highly suitable for end-user applications, as they tend to be significantly
more affordable compared to metal-based AM technologies, such as Selective Laser
Melting (SLM). Polymer AM processes include for instance Selective Laser Sintering
(SLS), Fused Deposition Modeling (FDM) and Stereolithography (SL) [2]. A wide

https://doi.org/10.1007/978-3-030-54334-1_27
392 D. Omidvarkarjan et al.
range of different polymers is offered by numerous suppliers. Furthermore, there are

various post processing and finishing options to choose from (like coating or dyeing).
As numerous success stories have demonstrated [3, 4], the geometrical freedom of
AM offers new opportunities in product design. For example, to join multiple AM
parts, fasteners optimized for the specific characteristics of the different AM processes
and materials can be used [5]. However, at the interfaces between conventionally
manufactured parts and AM plastic components, these design potentials are not
exploited yet, as seen with the integration of female threads into AM polymer parts.
State of the art approaches include the tapping of plastic threads [6], embedding of nuts
during the print process or the integration of conventional threaded inserts (through
heat sinking, ultrasound, screwing via self-tapping threads or pressing) [7]. The
assembly process for these inserts is time-consuming, unprecise and may require
special tools. Furthermore, the resulting connections tend to have low loading capac-
ities and are non-reversible, meaning that in case of a damaged thread, the often costly
AM part has to be replaced as a whole.
In this paper, we present a novel insert design for the integration of threads into AM
polymer components. After introducing the functional working principle (Sect. 2 and
3), an experimental validation of the design concept on a tensile testbench is shown
(Sect. 4). We show a demonstration of the insert in a case study (Sect. 5) and discuss
its relevance on the process of developing AM products (Sect. 6).
2 AM Dowel Concept
As depicted in Fig. 1, the fastening feature consists of three components: a metallic

threaded insert, an AM polymer clip and a female cutout within the main AM polymer
part. To install the insert, the metallic cylinder is placed into the AM polymer part
through slots of the female cutout. An Allen key recess in the bottom allows the
rotation of the insert. The three conical fins now create a form fit in the axial direction.
To prevent tangential movement, an AM polymer clip is inserted into the according slot
of the cutout. The clip is prevented from falling out or unintentional disassembling
through small hooks that create a snap fit connection. The insert is now fixed from
moving in any direction.
Fig. 1. AM dowel assembly procedure.
Special focus has been put on minimizing mechanical play between the compo-
nents of the assembly (see Fig. 2), as rattling and perceptible movement of the insert
The AM Dowel – A Novel Insert for the Integration of Threads 393
would harm the functional integrity and would lead to a bad impression of quality. The
clip thickness as a whole can be varied to accommodate different manufacturing tol-
erances. The manufacturing costs are minimized since the variation is limited to the clip
as small, sacrificial part instead of the much more expensive main part. Furthermore,
small knobs are placed on the bottom surface of the female counterpart, creating an
undersized slot. When the insert is placed into the AM polymer part, the knobs will
deform locally, removing the clearance through a small pretension.
Fig. 2. a) Section view of female cutout b) tolerance compensating features (encircled).
By limiting the geometrical complexity to the AM parts, the metallic insert has a
simple, symmetrical shape which can be manufactured conventionally and therefore
economically in large quantities (in this case on a turn-mill machine with brass alloy
CuZn39Pb2). Since the clips consist of the same material as the surrounding AM part
they may be printed in the same batch or even be connected to the polymer parts,
minimizing additional effort for stock-keeping and part logistics: On parts for industrial
use the clips may be joined directly with fillets so that they can be broken away before
assembly. By doing so, even multiple clips (for later repairs or for different tolerances)
may be produced together with very little cost involved for production, finishing and
logistics. For end-user parts where break-away solutions are not suited due to the
resulting traces on the surface, the corresponding clips may be connected via chain
links to the polymer parts.
Compared to state of the art solutions, the following benefits can be identified:
1. Contrary to existing inserts, the proposed design is locked through a form fit by
exploiting the freedom of design on the AM part. The polymer is therefore not
deformed under the effect of force, which avoids unattractive deformations.
2. The installation process is quick and does not require specialized tools (like for
pressing or heat sinking). A simple Allen key is sufficient.
3. The connection is reversible. The insert can therefore be disassembled through
small provisions in the case of a damaged thread or if another thread size is
demanded by the user.
4. The AM polymer clip acts as a cheap, sacrificial part. Under excessive torque the
clip fails first, protecting the much more expensive surrounding polymer part.
5. The closed design of the metal cylinder prevents accidental pulling of the insert if a
bolt is screwed in too deeply.
3 Analytical Model
To investigate the functional performance of the insert, the load bearing characteristics
were modelled analytically (see Fig. 3). The most relevant load case for this fastening
feature is the out of plane tensile load F (pull-out), as screwed connections are primarily
designed for this load case. The shearing of the AM counterpart (red area in Fig. 3) was
assumed as critical failure mode (compared to thread ripping, failure of brass insert
wings, bolt failure), since the material strength of PA2200 with 48 MPa [8] is consid-
erably lower compared to the other materials in the connection (steel bolt: min. 800 MPa
for 8.8 bolt [9], brass insert: 360 MPa [10]). The radial force V and the resulting
cylindrical stress were neglected, as their impact is significantly lower than the one from
the axial force F. The maximum shear stress smax within the AM polymer part is given
by the shear stress formula for rectangular beams [11], as depicted in Eq. 1.
3F
smax ¼ ð1Þ
2A
The shear strength sz of the AM polymer material can be approximated with the yield
strength rz through the von Mises yield criterion [6], as seen in Eq. 2.
rz
sz pffiffiffi ð2Þ
3
By doing so, a linear dependency between the maximum tensile load Fmax and cross
section of the ledges A can be established (see Eq. 3). Using this model, the insert
design was parametrized for different thread sizes.
2 rz
Fmax ¼ A p ð3Þ
3 3
Fig. 3. Simplified shear stress distribution within SLS ledge.

4 Experimental Validation
Experimental tests were performed to validate the analytical model of the fastening
design. The test setup is similar to other experiments in the field of insert development
[12, 13]. For this case, a Zwick Roell Retroline Z007 tensile testing machine is used.
As seen in Fig. 4a, a SLS test body is clamped to the bottom of the testbench with a
metal sheet to prevent bending. The upper fixture consists of another metal sheet with a
hole to fix the bolt. Three different sized inserts (M4, M6 & 3/8″-16 UNC) were tested
within this study. They were manufactured with SLM (stainless steel 1.4404) for
prototyping reasons. The tests are performed quasi-statically with a loading rate of
2 mm/min in accordance to test protocols from prior studies [13].
Fig. 4. a) Test body with retaining sheet metal, b) pull-out test setup, c) insert with sheared
ledges of test body.
Table 1 shows the experimental results for the three different thread sizes. The
measured pull-out forces are significantly higher (up to 81% more) than the analytically
computed values, indicating that the theoretical model provides a very conservative
prediction. As anticipated, the connection fails due to the shearing of the SLS ledges
(Fig. 4c). No visible deformations were observed on the metal insert.
Table 1. Experimental results for the different thread sizes.

Thread size M4 M6 3/8″-16 UNC
Shear area [mm2] 61.2 130.2 180.5
Fmax analytical [N] 1130 2406 3334
Fmax measured [N] 2050 3810 5540
Deviation [%] 81 58 66
5 Case Study
The inserts were tested in real life operation within the following case study. The
demonstrator part is shown in Fig. 5: a SLS top handle for videography camera sys-
tems. These are especially used in professional film making for handling or trans-
porting the camera. Furthermore, they act as mounting point for a large variety of
additional camera accessories (e.g. screens, microphones, wireless transmitters etc.),
that are screwed to the handle structure. As a result, conventional top handles are
mostly made of metallic materials, such as aluminum or titanium.
AM is highly suitable for this product field as it offers the opportunity to customize
the ergonomics of the handle and interfaces to the user’s need. The depicted SLS top
handle houses three 3/8-in. inserts in different orientations. The process of integrating
the complex female cutout into the CAD design was straightforward. The designer
simply needed to import the fully parametrized design into the handle model and
subtract the cutout from the main part. The inserts were fully functional and endured
extensive real-life operation. The AM dowel therefore enabled the use of SLS for an
application, where previously mostly metal based processes have been employed due to
metallic functional surfaces.
Fig. 5. a) SLS top handle with three inserts, b) complete camera setup in real-life operation.
The results of the experimental tests and case study indicate that the novel fastener
creates not only new potentials for AM polymer parts but also for the AM product
development process itself. As metallic functional surfaces are only applied where it is
required, material is more efficiently used. Hybrid design between different AM
technologies (in this case SLS and SLM) and conventional techniques is promoted as
the most suitable technology (e.g. in terms of cost, weight or other properties) can be
used where it is required. Especially polymer-based AM processes (i.e. SLS, FDM &
SL) benefit from that, as they can now be used for far more applications when com-
bined with the new fastener concept. Furthermore, customization and flexibility are
established within the hardware part as designers can choose their required feature
from ready-to-use building blocks & even change it throughout use as the connection is
standardized and reversible. In addition, new value creation models can be maintained
as the integration of the insert could be executed by an AM supplier based on speci-

fications of an OEM (e.g. through a marker in the CAD design).
Overall, the AM dowel acts as a good example how added value for products and
processes can be generated through the use of AM specific design elements. The
concept of reusable building blocks is highly relevant to the field of product devel-
opment with AM as prior studies have shown [5]. They can be rapidly recombined
without time consuming & costly re-iteration to enable an Agile, fast paced develop-
ment process. Furthermore, they provide inspirational input to designers and therefore
lead to a wider use of the design freedom offered by AM.
In the future, additional tests (i.e. torque out) should be conducted to further
investigate the characteristics of the fastening solution. These experiments should also
incorporate different materials, building and mounting orientations etc. Moreover, the
insert could be integrated in other applications to demonstrate its benefits in additional
application fields outside of camera equipment (e.g. robotics, prosthetics).
References
1. Bindl, S., Csiky-Strauss, C.: Developing applications for additive manufacturing series
production. In: 2017 IEEE European Technology and Engineering Management Summit (E-
TEMS), pp. 1–7. IEEE, Munich (2017)
2. Gebhardt, A.: Understanding Additive Manufacturing. Rapid Prototyping, Rapid Tooling,
Rapid Manufacturing. Hanser Publishers, Munich (2012)
3. Klahn, C., Omidvarkarjan, D., Meboldt, M.: Evolution of design guidelines for additive
manufacturing - highlighting achievements and open issues by revisiting an early SLM
aircraft bracket. In: Meboldt, M., Klahn, C. (eds.) Industrializing Additive Manufacturing -
Proceedings of Additive Manufacturing in Products and Applications - AMPA2017, pp. 3–
13. Springer, Cham (2018)
4. Rosenbauer, R., Fontana, F., Hastedt, H., Luhmann, T., Ochsner, D., Rieke-Zapp, D.,
Rofallski, R.: Advantages in additive manufacturing for a medium format metrology camera.
In: Meboldt, M., Klahn, C. (eds.) Industrializing Additive Manufacturing - Proceedings of
Additive Manufacturing in Products and Applications - AMPA2017, pp. 296–307. Springer
International Publishing, Cham (2018)
5. Omidvarkarjan, D., Cipriano, D., Rosenbauer, R., Biedermann, M., Meboldt, M.:
Implementation of a design support tool for additive manufacturing using a feature
database: an industrial case study. Prog. Addit. Manuf. 12(1), 364 (2020)
6. Coy, J.A., Miyashita, K., Entsfellner, K., Gumprecht, J.D.J., Luth, T.C.: First pullout
strength measurements of threads produced by selective laser sintering. In: 2013
IEEE/ASME International Conference on Advanced Intelligent Mechatronics, pp. 1381–
1386. IEEE, Wollongong (2013)
7. Kajzer, W., Gieracka, K., Pawlik, M., Basiaga, M., Kajzer, A., Szewczenko, J.: Tests of
threaded connections made by additive manufacturing technologies. In: Tkacz, E., Gzik, M.,
Paszenda, Z., Piętka, E. (eds.) Innovations in Biomedical Engineering, vol. 925, pp. 329–
337. Springer International Publishing, Cham (2019)
8. EOS GmbH: PA 2200 material datasheet. https://eos.materialdatacenter.com/eo/material/pdf/
431791/PA2200Balance1.0?sLg=de&rnd=1584106932001. Accessed 16 Mar 2020
9. DIN EN ISO 898-1:2013-05. Mechanische Eigenschaften von Verbindungselementen aus

Kohlenstoffstahl und legiertem Stahl - Teil 1: Schrauben mit festgelegten Festigkeitsklassen
- Regelgewinde und Feingewinde (ISO 898-1:2013); Deutsche Fassung EN ISO 898-1:2013.
Beuth Verlag GmbH, Berlin (2013)
10. Spittel, M., Spittel, T.: Flow stress, mechanical and physical properties of CuZn39Pb2. In:
Spittel, M., Spittel, T., Warlimont, H. (eds.) Part 3: Non-ferrous Alloys - Heavy Metals,
pp. 228–232. Springer, Berlin Heidelberg (2016)
11. Hagedorn, P.: Technische Mechanik: Festigkeitslehre, vol. 2. Harri Deutsch Verlag,
Frankfurt am Main (2006)
12. ECSS European Cooperation for Space Standardization: Space Engineering Insert Design
Handbook. ECSS-E-HB32-22A. ECSS, Noordwijk (2011)
13. Oltmann, J., Seemann, R., Spallek, J., Krause, D.: Additively manufactured components for
structural applications in aircraft interior. In: DFX 2016: Proceedings of the 27th Symposium
Design for X, pp. 147–156. TuTech Verlag, Hamburg (2016)
Novel Pressure Swirl Nozzle Design
Enabled by Additive Manufacturing
Michael Umbricht1 , Kaspar Löffel1(B) , Marc Huber1 , Patrick Lüscher2 ,

Janine Bochsler2 , Daniel Weiss2 , and Tom Duda3
1
Institute of Product and Production Engineering, FHNW, Windisch, Switzerland
kaspar.loeffel@fhnw.ch
2
Institute of Thermal and Fluid Engineering, FHNW, Windisch, Switzerland
3
Emerson Automation Solutions, Baar, Switzerland
Abstract. Conventionally produced pressure-swirl nozzles suffer from

the disadvantage that the geometry is restricted by the manufactur-
ing technique. Modern technologies like metal 3D printing by Selective
Laser Melting allow for more complex geometries with reduced flow resis-
tance, which can improve overall efficiency. The performance of a nozzle
is directly related to the droplet size produced by the nozzle: a fast evap-
oration of the droplets is the goal. In this paper, two conventional nozzle
designs and one design newly developed for additive manufacturing are
compared with respect to performance. The droplet distributions gener-
ated by the nozzles were measured at different pressures on a test stand
using laser diffraction analysis.
To rate the performance of the different nozzles, the Sauter Mean
Diameters of multiple measurements are compared to each other at two
different water pressures. At the lower pressure, the best conventional
nozzle and the novel nozzle design perform around the same. At an
increased pressure level (three times the lower pressure), the performance
of the newly designed nozzles is around 10% better.
Keywords: Additive manufacturing (AM) · Selective Laser Melting

(SLM) · Pressure-swirl nozzle · Atomization
1 Introduction
Since the inception of different additive manufacturing processes there has been
a steadily growing interest of these processes for various application areas [1–3].
Additive manufacturing of metals, their processes and the structure and prop-
erties of the produced metallic parts are an important topic in research [4].
The diversity of materials, which can be produced through additive manufactur-
ing, lead to fitting materials also for highly loaded applications [5]. According
to a study of Ampower GmbH & Co. KG [6], the metallic powder-bed fusion
processes, like Selective Laser Melting (SLM) of metals, are suitable for highly
https://doi.org/10.1007/978-3-030-54334-1_28
400 M. Umbricht et al.
loaded, additively manufactured metal parts. The enormous advantages of such

parts are achieved through the design freedom of this manufacturing process,
which can be best achieved through a complete redesign of a product. Advan-
tages with additively manufactured parts can be achieved in a lot of different
applications, as shown by Ngo et al. [1].
Changing the manufacturing technique of a highly loaded metal part from
conventional methods to SLM will bring the chief benefit of design freedom [7].
This design freedom can be used to drastically increase the performance of the
part. This can e.g. be the performance of a nozzle, as investigated in this project.
Currently produced nozzles for the atomization of water suffer from the disad-
vantage that their design is limited by the manufacturing process. If they are
designed for SLM-manufacturing, smoother water channels are possible, which
leads to less pressure drop and therefore to smaller droplets. Smaller droplets
are the key element in engineering of desuperheater spreayheads of the next
generation.
A desuperheater is a device that injects a controlled amount of cooling water
into a superheated steam flow in an effort to reduce or control the steam temper-
ature. Desuperheaters come in various physical configurations and spray types
that optimize performance within specified control parameters. Figure 1 shows
a sketch of a desuperheater in a steam cycle. For this type of desuperheater, the
control unit is outside of the steam cycle, and the sprayhead injecting water into
the steam cycle is located underneath it. Such a sprayhead consists of multiple
pressure swirl nozzles, which form a hollow cone of water that will atomize to
single droplets after a certain distance. The atomization principle used is the
most promising among a lot of different possibilities, for the according require-
ments. Different atomization and spray techniques are shown by Lefebvre and
McDonell [8]. The droplets in the steam cycle will evaporate after some time.
The performance of a sprayhead is among other things measured with means of
the evaporation distance. If there is a bend in the steam cycle downstream of
the sprayhead and not all droplets are evaporated at this bend, corrosion and
Fig. 1. Sketch of a steam cycle

Novel Pressure Swirl Nozzle Design Enabled by Additive Manufacturing 401
cracking can occur at the wall of the pipe. Therefore, a faster evaporation is
better since it will drastically decrease the cost of a steam cycle. To get a fast
evaporation the key factor is to produce small droplets with the used nozzles.
Investigation of sprays from a nozzle can be done experimentally, as shown
by Laryea and No [9], Xie et al. [10] and Zhao and Yang [11]. Therefore, the
performance of a new nozzle design can be measured and compared to a previ-
ous design. Achieving nozzle improvements through changing the manufacturing
technology to an additive one is currently hardly covered in research literature.
However, it is currently being investigated by different companies, and prod-
ucts with this improvement already are available e.g. on the gas turbine market,
which is shown by Stytsenko et al. [12] and Appleyard [13].
The existing sprayhead consists of multiple parts and is shown in Fig. 2. The
main body of the sprayhead is a monolithic part, whereas the nozzles (one or
two additional parts for each nozzle) are threaded into it and then brazed to seal
it. Therefore, a further improvement of the SLM-manufacturing will be that the
assembly time and brazing is not needed anymore. The new sprayhead can be
printed as a single monolithic part.
Fig. 2. Existing sprayhead consisting of Fig. 3. Negative of the SLM sprayhead

multiple parts which consists of one solid part
A negative of such a sprayhead designed for additive manufacturing and

including the novel pressure swirl nozzle design is shown in Fig. 3. It can be
seen that the waterpaths of this design will be smoother in comparison to the
existing sprayhead design. A further advantage of this monolithically produced
sprayhead will be that the nozzles and their inlet channels can be arranged with
less restrictions than before.
This paper investigates the improvement of the nozzles enabled by SLM man-
ufacturing. This improvement is measured in droplet size and will be compared
to existing, conventionally manufactured nozzles.
2 Initial Design, and Design of Experiments (DoE)

The current pressure swirl nozzle (conventional nozzle) is manufactured with
conventional manufacturing methods. It does not only consist of multiple parts,
it is also brazed together to seal it. The main problem with this nozzle is that
there are a lot of sharp edges in the water path, which leads to a large pressure
loss and therefore to bigger droplets. The conventional nozzle, shown in Fig. 4, is
a pressure swirl nozzle with two chambers, a feed chamber and a swirl chamber.
It is visible that this nozzle consists of multiple inlet ports, which are the entry
of the nozzle and the feed chamber. From the feed chamber the water is guided
through two channels to the swirl chamber. Because of this guidance a swirl
is generated in the swirl chamber. As soon as the water leaves the nozzle, a
hollow cone of water is formed. After a certain distance from the nozzle exit, the
breakup from this circular sheet of water to single droplets occurs.
Fig. 4. Conventional nozzle design Fig. 5. Novel nozzle design
The novel nozzle design works in a similar way to the conventional design
and is also a pressure swirl nozzle. For the novel nozzle, the design freedom of
additive manufacturing is used. In a heuristic procedure, this new nozzle design
was created with water paths as smooth as possible, leading to less pressure loss
and therefore to smaller droplets. The novel nozzle design is shown in Fig. 5. The
four inlet channels are guided directly to the swirl chamber and no feed chamber
is needed. This printable design has the advantage that a sprayhead engineered
of multiple such nozzles can be printed by SLM in one part, and therefore the
assembly and braze time is not needed anymore.
Since not only smooth water paths are needed for the smallest droplets pos-
sible, the right proportions of the dimensions were investigated. A Design of
Experiment (DoE) approach is used to statistically vary the significant parame-
ters in contrast to just trial and error. A model can be fitted into the measured
data, by the carefully defined measurement points and the output variable (in
this case the Sauter Mean Diameter (SMD)1 ) can be predicted and thus opti-
mized. The dimensions displayed in Fig. 6, which shows a section of the novel
nozzle design, were defined and varied for the DoE. In Table 1 the setup for the
1
Diameter of a droplet whose surface-to-volume ratio is equal to that of the entire
spray. The SMD is about 15%–25% smaller than the mass median diameter [8].
Fig. 6. Sketch of the swirl chamber with the parameters of the nozzle for the DoE
different parameters are shown in coded units2 . The nozzle “Initial design” is
the initially designed nozzle, with no changes in the parameters. All the other
nozzles, named DoEXX, are used to gain different data points to fit the data
to afterwards. All these nozzles were manufactured and measured to create a
model for the prediction of the SMD. The measurement procedure is described
in Sect. 3.
Table 1. Input variables of DoE
Nozzle Name ds /d0 dp /d0 l/d0

Initial design 0 0 0
DoE00 −1 −1 −1
DoE01 −1 −1 +1
DoE02 −1 +1 −1
DoE03 −1 +1 +1
DoE04 +1 −1 −1
DoE05 +1 −1 +1
DoE06 +1 +1 −1
DoE07 +1 +1 +1
√
DoE08 − 2 0 0
√
DoE09 + 2 0 0
√
DoE10 0 − 2 0
√
DoE11 0 + 2 0
The measurements of these different nozzles showed that it is possible to fit

the data with different models. The model which fits the best was chosen and
the global minimum of the SMD in this model was searched. This led to the
2
In the plus-minus coding scheme +1 is used to denote one level of a factor, often
called the high level, whereas −1 is used to denote the low level.
optimized nozzle design SE6. Again, this nozzle was measured and compared to
the other nozzles. The measurement setup and procedure is described in Sect. 3,
the results are shown in detail in Sect. 4, and discussion takes place in Sect. 5.
3 Measurement Setup and Procedure

The main goal of the project is to design a nozzle with a better performance.
This can be quantified with the droplet size, since smaller droplets result in a
faster atomization of the droplets, which is identical to a better performance. The
distribution of the droplet size can be measured with laser diffraction. Thereby,
droplets which get hit by a laser beam scatter this laser light. The scattering
angle can be measured and used for a calculation of the droplet diameter. During
the measurement, the size distribution for all particles along the laser path is
measured. To measure the droplet size of the nozzles, the Malvern Spraytec
device from Malvern Panalytical GmbH, Malvern, United Kingdom, which works
with this principle, is used. This device can measure droplet sizes between 0.1
and 2000 µm.
Fig. 7. Sketch of the measurement setup
A sketch of the whole measurement setup is shown in Fig. 7. There it is visible

that only half of the rotationally symmetrical spray is measured at once with
the Malvern Spraytec and the rest of the spray is excluded with a metal sheet.
This is done to improve transmission of the laser light at higher pressures (higher
mass flow, a lot of droplets in the laser path of the measurement device). During
the measurements the mass flow and the pressure are measured and recorded.
To get comparable measurements a water pump is used to measure at different
pressure levels of the feeding water. Since the nozzles should perform well at
a broad pressure range two pressure levels are measured: the lower pressure
level p0 and the higher pressure level 3p0 , a three-fold increase of p0 . A further
decrease in pressure level could lead to a bad swirl in the swirl chamber and
therefore to big droplets. A further increase in pressure level will not be possible
to be measured with the current setup since the transmission of the laser will
not be high enough for a reliable measurement. The whole measurement setup is
designed in a way that it is easy to measure the droplet size at different angular
positions of a nozzle. For every measured nozzle an insert is produced, which
can easily be changed in the test stand.
The 3D view of an insert is shown in Fig. 8. The SLM-manufacturing of this
insert is done in the same orientation as the orientation of the nozzles inside
a sprayhead will be for the manufacturing of the whole sprayhead. Therefore,
the nozzle axis is orthogonal to the build direction (as shown in Fig. 8). The
complete nozzle is made without support structure, since it will be hard or even
impossible to completely remove these in the inside. On the bottom side of the
insert, support structure is needed for the connection to the build plate. After
the SLM manufacturing, the outlet diameter is milled to a completely circular
hole, which leads to a more homogeneous spray. The backsides of the inserts are
also milled to get a sealing surface.
Fig. 8. Insert for the measurement of the droplet size
A single measurement of a nozzle insert always includes eight different angu-

lar positions with an offset angle of 45◦ . Every position is measured for 30 s.
The droplet measurement device software generates one SMD for every mea-
sured second. This leads to a graph as shown in Fig. 9 for every nozzle measure-
ment. To ensure that the results are reliable, a multitude of these measurements
(repeatability measurements) with the same insert and the same procedure are
performed. To ensure that the measurement setup is trustworthy, it is necessary
that these repeatability measurements do not vary too much.
It is important that the nozzles being compared are in the same mass flow
range since smaller nozzles will automatically reduce the size of the droplets. The
problem with smaller nozzles is that their massflow will not be high enough. The
novel nozzle was designed so that it has a similar mass flow to the conventional
nozzle. Variation of the mass flow at a certain pressure can be achieved through
130
120
Normalized SMD [%]
110
100
90
80
0 45 90 135 180 225 270 315

Angle position [°]
Fig. 9. Example of a nozzle measurement at pressure p0 , with a massflow of

11.1 kg/min (SE6-Eco a), (On each box, the central mark indicates the median, and
the bottom and top edges of the box indicate the 25th and 75th percentiles, respec-
tively. The whiskers extend to the most extreme data points not considered outliers,
and the outliers are plotted individually using the ‘+’ symbol. Outliers are all points
which are 1.5 times the interquartile range smaller or greater than the median. The
notch displays the 5% significance level of the median [14].) (Normalization due to
confidentiality, which does not allow to show any discrete values. Median SMD of the
YE nozzle at pressure p0 belongs to 100%.)
a uniform scaling of the nozzle. This can not only be done through a physical
scaling of the nozzle and therefore a new manufacturing, it is also possible to
scale the measured SMD to a certain mass flow. Experiments with different sizes
of nozzles led to the following formulas at the two pressure levels. Scaling at
pressure p0 can be done with Eq. 1 and at pressure 3p0 with Eq. 2 [15]. This
scaling is always done if multiple nozzles are compared to each other.
0.359
ṁ(2)
SMD(2) = SMD(1) · (1)
ṁ(1)
0.204
ṁ(2)
SMD (2)
= SMD (1)
· (2)
ṁ(1)
4 Measurement Results
In this section the measurement results of the different nozzles are compared
to each other. From the initial design (before DoE) to the optimized nozzle
design (after DoE, SE6) an improvement of around 20% could be achieved for
both pressure levels (p0 and 3p0 ). With this optimized nozzle design, different
comparisons are made in this chapter.
Figure 10 shows the comparison of different SE6 nozzle inserts at pressure

p0 . Two of these inserts (SE6a and SE6b) are SLM-produced in-house, like all
the SLM nozzles for the DoE. This is done with a Realizer SLM 125 machine
from Realizer GmbH, Borchen, Germany. Two nominally identical nozzles were
produced to evaluate repeatability between different parts. The SE6-Eco a and
the SE6-Eco b are produced at Ecoparts AG, Hinwil, Switzerland (external
supplier). All these nozzles named SE6 have the same nozzle geometry. How-
ever, the nozzles produced by Ecoparts are made of Inconel 718, whereas the
internally manufactured nozzles are all made of Inconel 625 (SLM powder with
standard grain size distribution). Every boxplot in Fig. 10 belongs to one SLM-
manufactured insert with the same nozzle geometry (SE6 nozzle design) and
every insert is measured five times over all eight angular positions (every 45◦ )
to include the repeatability and the angular variation.
140
130
Normalized and scaled SMD [%]
120
110
100
90
80
70
SE6a SE6b SE6-Eco_a SE6-Eco_b
Fig. 10. Comparison of the SE6 nozzle manufactured in-house and at an external
supplier at pressure p0 , mass flow scaled to 11.2 kg/min
Comparing the SE6a to the SE6b one can see that there is a significant
variation between these two inserts, even though they have the same nozzle
geometry and are produced with the same machine and the same material. It is
likely that this lack of reproducibility originates from turbulence in the cross-flow
of inert gas across the build platform of the in-house SLM machine. Comparing
the inserts from the external supplier (SE6-Eco a and SE6-Eco b) it is visible
that these nozzle inserts are more reproducible. They are also reproducible for a
higher quantity of nozzle inserts (for simplicity, these results are omitted here)3 .
The externally manufactured SE6 inserts have around the same or smaller
median droplets than the internally manufactured inserts. However, the variation
3
With this information it is conceivable that the DoE performed with externally
manufactured nozzles would have led to a different optimized nozzle design.
of the droplet sizes of a single externally manufactured insert is bigger. This

bigger variation can be explained with a look at the measurements of specific
angular positions. This is shown in Fig. 11 for an in-house manufactured nozzle
(SE6 a) and in Fig. 12 for an externally manufactured nozzle (SE6-Eco a), where
again all the five repeatability measurements of the respective insert are included,
but shown at the different angles. One of the five repeatability measurements
included in Fig. 12 was already shown in Fig. 9. In Figs. 11 and 12 it is possible to
see that the large variation of the boxplots from Fig. 10 are due to different SMDs
at different angular positions. If these different angular positions of both inserts
are compared to each other, it is noticeable that the worst angular positions
(90◦ ) are only slightly better in the externally manufactured inserts. However,
the better performing angles are clearly better in the externally manufactured
inserts than in the in-house produced inserts. This leads to the somewhat larger
variation in the case of the externally manufactured nozzle (see Fig. 10). The
worst SMD occurs at 90◦ , which is 45◦ (in swirl direction) after the top location
of the SLM manufacturing. The higher surface roughness at the top position
therefore has an influence on the droplet size.
140 140
130 130
120 120
110 110
100 100
90 90
80 80
70 70
0 45 90 135 180 225 270 315 0 45 90 135 180 225 270 315
Angle position [°] Angle position [°]
Fig. 11. Angular variation of the SE6 a Fig. 12. Angular variation of the SE6-
nozzle, SLM manufactured in-house, Eco a nozzle from an external supplier
at pressure p0 , mass flow scaled to at pressure p0 , mass flow scaled to
11.2 kg/min (sum of five repeatability 11.2 kg/min (sum of five repeatability
measurements) measurements)
The most important comparison is the one between the novel nozzle design
and the conventionally manufactured nozzles. For this comparison the externally
manufactured nozzle inserts are used, since these results are more reproducible.
The comparison between the novel nozzle design and the conventional nozzle is
shown in Fig. 13 at pressure p0 . The conventionally manufactured nozzles (NE
and YE) are measured eight times over all eight angular positions in compari-
son to the five measurements of the SLM nozzles. For all the five repeatability
measurements of the SLM nozzle inserts nothing was changed. But for the eight
measurements of the conventional nozzles, the two main parts of the nozzle,
which are not rotationally symmetric (inlet ports to the feed chamber and chan-
nels to the swirl chamber) are also rotated relatively to each other in 45◦ steps
(eight different positions). This showed, that this relative angle to each other
is also very important, whereas it is random in the existing sprayhead. These
positions are not further investigated in this project. It is visible that the novel
nozzle design can perform slightly better with regard to the median droplet size
than the best conventional nozzle. The variation of the droplet size is larger in
case of the SLM-manufactured inserts.
140
130
120
110
100
90
80
70
SE6-Eco_a SE6-Eco_b NE YE
Fig. 13. Comparison of the best SLM nozzles with the conventional nozzles at pressure
p0 , mass flow scaled to 11.2 kg/min
In Table 2 the measured droplet size values of the optimized SLM nozzles and
the conventional nozzles, at pressure p0 , normalized and scaled to a mass flow
of 11.2 kg/min, are shown. For this table the two inserts with the same nozzle
geometry and manufacturing on the same SLM machine are added together. For
the complete comparison of the different nozzles it is not only important to com-
pare the median values of the SMD, it is also necessary to have a measure for the
droplets larger than the median. This measure is the 90%-Quantile. Comparing
all these values the same conclusion as before can be made. The performance
of the novel nozzle design and the best conventional nozzle is around the same.
The median droplet size of the novel nozzle design is slightly better than the
best conventional nozzle, but the 90%-Quantile and the standard deviation of
the best conventional nozzle are slightly better.
Table 2. Comparison of different nozzle types at pressure p0 , mass flow scaled to

11.2 kg/min
Nozzle Median [%] Standard 90%-Quantile [%]

deviation [%]
SE6 in-house 97.9 7.4 110.6
(2 nozzle inserts)
SE6 Ecoparts AG 96.7 10.4 113.8
(external supplier)
(2 nozzle inserts)
NE 106.4 8.1 116.1
(1 nozzle insert)
YE 100.0 6.5 109.7
(1 nozzle insert)
The same comparison as for pressure p0 can be made for other pressure lev-
els. This is shown for pressure 3p0 in Fig. 14, for the same nozzle inserts like
before. The main message does slightly change with the pressure rising. Again,
the externally manufactured nozzles (SE6-Eco a and SE6-Eco b) are more repro-
ducible than the same geometry produced in-house (SE6 a and SE6 b). Unlike
for the lower pressure level, it is possible to see here that the novel nozzle design
is clearly better than the best conventional nozzle. The overall performance of
the novel nozzle design is around 10% better than the best conventional nozzle.
70
65
60
55
50
45
40
SE6a SE6b SE6-Eco_a SE6-Eco_b NE YE
Fig. 14. Comparison of all nozzles at pressure 3p0 , mass flow scaled to 20 kg/min (note
that two outliers around 130 µm for the nozzle SE6-Eco a are not shown)
In Table 3 the droplet size values of the comparison at pressure 3p0 , nor-
malized and scaled to a mass flow of 20 kg/min are shown. For this table the
two inserts with the same nozzle geometry and manufacturing on the same SLM
machine are added together. The droplet sizes of this higher pressure level (3p0 )
are clearly smaller than the droplet sizes of the lower pressure level (p0 ). The
novel nozzle design (SE6) is around 10% better in case of the median than the
best conventionally manufactured nozzle (NE).
Furthermore, out of the two conventional designs we note that for the higher
pressure level, a different conventional nozzle performed better than at the lower
(YE at pressure p0 and NE at pressure 3p0 ).
Table 3. Comparison of different nozzle types at pressure 3p0 , mass flow scaled to
20 kg/min
Nozzle Median [%] Standard 90%-Quantile [%]

deviation [%]
SE6 in-house 48.8 2.4 53.1
(2 nozzle inserts)
SE6 Ecoparts AG 49.1 3.4 53.2
(external supplier)(2
nozzle inserts)
NE 55.0 2.6 57.5
(1 nozzle insert)
YE 55.8 3.3 59.7
(1 nozzle insert)
5 Discussion
In this section the measurement results presented in Sect. 4 are discussed. For
both pressure levels (p0 and 3p0 ) it is visible that the in-house manufactured
SLM nozzles are sometimes significantly different with regard to the SMD, even
if they are completely the same geometry (SE6a and SE6b). These differences in
the nozzles can also be measured in the surface roughness of the nozzle inserts.
The Sa (arithmetical mean height)4 values at a specific position of the internally
manufactured inserts varies between 6.3 µm and 10.3 µm between different noz-
zles of the nominally same geometry. The surface roughness values are measured
with a Laser Scanning Microscope (LSM), from Keyence, Osaka, Japan. As men-
tioned previously, the unfavorable gas cross-flow of the in-house SLM machine
is suspected to be the dominant cause of this variability.
Since the final nozzle will be manufactured with a state-of-the-art SLM
machine, some inserts were produced at Ecoparts AG (external supplier). These
inserts have a smaller variation of SMD and surface roughness, which is due to
4
Sa is the extension of Ra (arithmetical mean height of a line) to a surface. It
expresses, as an absolute value, the difference in height of each point compared
to the arithmetical mean of the surface [16].
a more stable SLM-process. Measured in the same way and position as for the
in-house inserts, for the external nozzles the Sa values vary between 5.0 µm and
5.9 µm. The lower surface roughness of these externally produced nozzles leads
to the assumption that a lower surface roughness will generate less pressure loss
and therefore smaller droplets. This could be achieved with an additional surface
finishing process, which will be further investigated.
The cause of the angular variation of the SMD of a single nozzle is the depen-
dence of surface roughness on overhang angle commonly seen in SLM manufac-
tured parts [17]. Figure 12 shows that the biggest droplets are produced at an
angle of 90◦ . Figure 15 shows a sketch of the nozzle outlet, where it is visible that
the biggest overhang angle is at measurement angle 45◦ and the swirl in the swirl
chamber is going in the direction of increasing angular positions. The droplet
sizes of the angles after the largest overhang angle (90◦ to 180◦ ) are somewhat
higher than the rest, which is explained with the higher surface roughness of
the angles before. To significantly improve the performance of the novel noz-
zle design, it is necessary to reduce the surface roughness at the top surface of
the nozzle. If this can be achieved, the variation of the SMD will be smaller
and therefore the overall performance better. This will be the topic for further
investigations.
Fig. 15. Sketch of nozzle outlet.

A novel pressure swirl nozzle design with smooth transitions from inlet channels
to the swirl chamber was designed. The initial design was further developed with
a design of experiment (DoE) and finally compared to the existing (conventional)
nozzle design. It was possible to show that the performance of the novel nozzle
design is around the same at a low pressure level (p0 ), and an improvement of
around 10% at a higher pressure level (3p0 ) could be achieved. Together with the
ability to freely arrange the nozzle positions and the inlet positions in a sprayhead
and the ability to manufacture a whole sprayhead at once, without the need of
assembly time, this shows the clear potential that SLM manufacturing has for
desuperheater sprayheads.
The measurement of the same nozzle geometry, once manufactured in-house
and once at Ecoparts AG, showed a strong dependency of the SMD to the surface
roughness of the SLM manufacturing. This leads to the assumption that a further
improvement of the surface roughness will lead to even smaller droplets. For
the improvement of the surface roughness of SLM-manufactured parts, different
processes are known in the industry. The improvement of the nozzles after the
treatment with these different processes will be further investigated.
Acknowledgment. This work was financially supported by the Innosuisse - Swiss

Innovation Agency and Emerson Automation Solutions, Baar, Switzerland.
References
1. Ngo, T.D., Kashani, A., Imbalzano, G., Nguyen, K.T.Q., Hui, D.: Additive manu-
facturing (3D printing): a review of materials methods applications and challenges.
Composites Part B-Eng. 143, 172–196 (2018)
2. Bikas, H., Stavropoulos, P., Chryssolouris, G.: Additive manufacturing methods
and modelling approaches: a critical review. Int. J. Adv. Manuf. Technol. 83(14),
389–405 (2016)
3. Stavropoulos, P., Foteinopoulos, P.: Modelling of additive manufacturing processes:
a review and classification. Manuf. Rev. 5, 26 (2018)
4. DebRoy, T., Wei, H.L., Zuback, J.S., Mukherjee, T., Elmer, J.W., Milewski, J.O.,
Beese, A.M., Wilson-Heid, A., De, A., Zhang, W.: Additive manufacturing of metal-
lic components - process, structure and properties. Progress Mater. Sci 92, 112–224
(2018)
5. Bourell, D., Kruth, J.P., Leu, M., Levy, G., Rosen, D., Beese, A.M., Clare, A.:
Materials for additive manufacturing. CIRP Annals-Manuf. Technol. 66(2), 659–
681 (2017)
6. Ampower GmbH & Co. KG: Metal Additive Manufacutring with sinter-based tech-
nologies (2018)
7. Herzog, D., Seyda, V., Wycisk, E., Emmelmann, C.: Additive manufacturing of
metals. Acta Materialia 117, 371–392 (2016)
8. Lefebvre, A.H., McDonell, V.G.: Atomization and Sprays. CRC Press, Boca Raton
(2017)
9. Laryea, G., No, S.: Spray angle and breakup length of charge-injected electrostatic
pressure-swirl nozzle. J. Electrostat. 60(1), 37–47 (2004)
10. Xie, J.L., Gan, Z.W., Duan, F., Wong, T.N., Yu, S.C.M., Zhao, R.: Character-
ization of spray atomization and heat transfer of pressure swirl nozzles. Int. J.
Thermal Sci. 68, 94–102 (2013)
11. Zhao, J.: Simulation and experimental study on the atomization character of the
pressure-swirl nozzle. Open J. Fluid Dyn. 2, 271–277 (2012)
12. Stytsenko, A., Mylnikov, S., Baibuzenko, I., Maurer, M.: Additive manufactured
gas turbine fuel injector with a nested article having a non-removable internal
supporting structure. No. EP 3 324 120 A1, Ansaldo Energia Switzerland AG
(2018)
13. Appleyard, D.: Powering up on powder technology. Metal Powder Report 70, 285–
289 (2015)
14. The MathWorks Inc.: Matlab 9.7.0.1190202 (R2019b). Natick, Massachusetts
(2019)
15. Lüscher, P., Bochsler, J., Weiss, D., Huber, M., Löffel, K., Nieulande, R., Duda,
T.: Numerical and experimental investigation of pressure-swirl nozzles produced
by additive manufacturing. In: ILASS-Europe 2019, 2–4 September 2019, Paris,
France (2019)
16. Keyence Corporation: Area Roughness Parameters. Osaka, Japan (2020). https://
www.keyence.com/ss/products/microscope/roughness/surface/parameters.jsp
17. Xiang, Z., Wang, L., Yang, C., Yin, M., Yin, G.: Analysis of the quality of slope
surface in selective laser melting process by simulation and experiments. OPTIK
176, 68–77 (2019)
Design of an Additively Manufactured
Customized Gripper System for Human Robot
Collaboration
Nikolai Hangst1 , Stefan Junk2(&) , and Thomas Wendt1

1
Laboratory for Human Robot Collaboration, Offenburg University,
Klosterstr. 14, 77723 Gengenbach, Germany
2
Laboratory for Rapid Prototyping, Offenburg University,
Brückenhäuser Str. 26, 77723 Gengenbach, Germany
stefan.junk@hs-offenburg.de
Abstract. The Human-Robot-Collaboration (HRC) has developed rapidly in

recent years with the help of collaborative lightweight robots. An important
prerequisite for HRC is a safe gripper system. This results in a new field of
application in robotics, which spreads mainly in supporting activities in the
assembly and in the care. Currently, there are a variety of grippers that show
recognizable weaknesses in terms of flexibility, weight, safety and price.
By means of Additive manufacturing (AM) gripper systems can be developed
which can be used multifunctionally, manufactured quickly and customized. In
addition, the subsequent assembly effort can be reduced due to the integration of
several components to a complex component. An important advantage of AM is
the new freedom in designing products. Thus, components using lightweight
design can be produced. Another advantage is the use of 3D multi-material
printing, wherein a component with different material properties and also
functions can be realized.
This contribution presents the possibilities of AM considering HRC
requirements. First of all, the topic of Human-Robot-Interaction with regard to
additive manufacturing will be explained on the basis of a literature review. In
addition, the development steps of the HRI gripper through to assembly are
explained. The acquired knowledge regarding the AM are especially empha-
sized here. Furthermore, an application example of the HRC gripper is con-
sidered in detail and the gripper and its components are evaluated and optimized
with respect to their function. Finally, a technical and economic evaluation is
carried out. As a result, it is possible to additively manufacture a multifunctional
and customized human-robot collaboration gripping system. Both the costs and
the weight were significantly reduced. Due to the low weight of the gripping
system only a small amount of about 13% of the load of the robot used is
utilized.
Keywords: Additive manufacturing 3D multi-material-printing

Human-robot-collaboration Lightweight design

https://doi.org/10.1007/978-3-030-54334-1_29
416 N. Hangst et al.
1 Introduction
Human-robot collaboration (HRC) refers to the collaboration between humans and

robots in the same workspace [1]. HRC is expected to increase the quality rate and
performance efficiency. Studies show that the market for collaborative robots (cobots)
will increase rapidly in the next years. By 2025, about a third of all sold robots are
expected to be cobots [2].
To ensure error-free and risk-free human-robot collaboration, it is necessary that
both the robot and the gripping system are safe within the meaning of IEC 61508 [1].
This means that appropriately safe grippers are required for the robots [3, 4]. Today,
dangerous movements and collisions can be detected and corrected using extensive
simulations before commissioning [5]. A large number of grippers already exist, but
they show recognizable weaknesses in terms of flexibility, weight and price. Today, the
majority of the grippers on the market are inflexible 2-finger parallel gripping systems.
Using additive manufacturing, also called 3D printing, gripping systems can be
developed that are both multifunctional, quickly manufactured and can be customized.
The subsequent assembly effort can also be reduced due to the combination of several
components into one complex component [6, 7]. In view of these advantages, it would
be possible to develop new business models in relation to Industry 4.0, in that grippers
can be manufactured in a single process. First approaches for the additive manufacture
of grippers are ring-shaped grippers that can implement simple gripping processes with
the aid of a silicon ring with an air channel. The advantage is quick adjustment for a
small variation of the components to be gripped. But such simple grippers are only
suitable for a very limited range of applications [8].
1.1 Challenges in Human-Robot Collaboration

In HRC, weaknesses can be derived from both the robot and the gripper due to high
security requirements. Today, a disadvantage of the robots is, that the detection of
obstacles is often only possible after a direct contact. This detection is realized in the
joints of the robot using torque and/or motor current sensors. If an employee works
closely with the robot, the person may be impaired due to the constant contact with the
robot. Another example can be taken from medicine: if a robot guides a scalpel, contact
with humans must be avoided, since even the slightest contact can cause major damage
from cuts [9, 10]. There are currently robots available that can use a capacitive skin to
recognize objects up to 5 cm away. If an object is in the immediate vicinity of the
robot, the dialectric around the robot skin changes and its capacity changes too [1].
In addition, there are already external security devices such as a camera system for
3D room surveillance, a tactile floor and laser scanners for two-dimensional room
surveillance [11–13]. Since these are external safety devices, the flexibility of the
robots decreases so that they can no longer be moved easily. A large number of
solutions already exist with regard to the grippers. The grippers initially differ in their
mode of operation. A distinction is made between a pneumatic and an electrical con-
nection. The best type of connection cannot be determined in general and always
depends on the respective application.
Design of an Additively Manufactured Customized Gripper System 417
1.2 Challenges in Grippers for HRC

The grippers also differ in terms of their grip options. These range from simple parallel
grippers to complex replicas of a human hand. Accordingly, the price varies in a range
from € 1,000 to over € 50,000. In addition, the price depends on the integrated
intelligence of the gripper. For example, there is an approach with a 3-finger gripping
hand available, that uses a tactile sensor system, so that confusion of the component to
be gripped can be excluded with the help of pressure and surface detection [14]. This is
done using a two-dimensional contact force profile, which is defined by the structure of
the component to be gripped. When a component is gripped, the profile is evaluated
and compared with the data stored in the program. The principle of the 3-finger gripper
has proven itself since it can perform a variety of tasks. Further developments using a
reconfigurable gripper enables up to eight degrees of freedom to be achieved [15].
There are also differences with regard to the gripping force and the intrinsic weight
of the gripper. This weight is not directly dependent on the gripping force. When using
a gripper, care must be taken that a robot may already be able to achieve a large amount
of its maximum load when the gripper is attached. With regard to the gripping force, it
depends on the type of application. Today, there are considerable differences between
the properties shown above. The manufacturers have so far not been able to resolve
these conflicts. Either the grippers are light and inexpensive, but inflexible, or they can
be used multifunctionally, which in turn causes a high weight and a high price.
2 Development of an Additively Manufactured HRC Gripper
The framework of this contribution builds on the following steps. The examination of
various gripping principles serves as the basis. After selecting a suitable principle, a
rough concept for the gripping system is developed. A suitable method for additive
manufacturing the gripper is then determined. When implementing the concept, par-
ticular attention is paid to lightweight design. Finally, the implementation of the
concept in the laboratory environment is demonstrated. The gripping system is to be
used for demonstrations in the human-robot collaboration laboratory. The robot is
supposed to hold a pen with the aid of a gripper. Individual words or text passages as
well as a simple sketch are to be transferred onto a whiteboard using different pens.
Words can be entered on the control panel. The drawing of a simple sketch should be
able to be selected using a predefined template on the control panel.
2.1 Selection of the Functional Principle and the Connection

Preliminary tests were carried out to hold a pen. It turned out that there are several grip
options. It can be gripped successfully with both two and three fingers. The best
solution in connection with a robot turned out to be three fingers with a stop due to the
safe handling. Furthermore, three fingers are advantageous in terms of the flexibility of
the gripper. With two fingers, only parallel and/or angle grips can be used. With three
fingers, on the other hand, additional grips in the form of pincer grip, scissors grip or
spreading grip can be used.
To implement the additional grip options, it is necessary that the fingers can be
rotated through 360°. To use parallel and angled handles, a finger must be folded away
or removed. According to our own research, the best possible flexibility can be
achieved using this functional principle. With an additional spreading handle, com-
ponents can be gripped from the inside. In terms of function, the handle is based on the
opposite of the tweezer handle. Another way to improve the gripping behavior of a pen
is to implement a guide notch. The area for gripping is increased by a rubber-like
material nestling around the pen. With a pair of tweezers, this prevents the pen from
slipping out. Basically, electrical or pneumatic solutions are available for the selection
of the connection. Due to long holding processes with a few gripping cycles, the
pneumatic solution is better suited for this application and is therefore aimed at.
2.2 Overall Concept for the HRC Gripper System

The overall concept of the system to be developed is designed as shown in Fig. 1.
Since this is still a rough concept, all assemblies and components are only shown in an
exemplary and simplified manner. The gripper consists of the connections, the control
unit and the pneumatic fingers, which can be quickly and easily removed or replaced.
The control unit is the mechanical link between the gripper fingers and the robot
and also uses valves to establish the pneumatic connection between the gripper fingers
and the robot. Each finger should be provided with its own actuator in order to ensure
that the fingers can be twisted and quickly removed and converted. The gripper is
equipped with three identical fingers with an adaptive functional principle. 3D printing
should make it possible to adapt the connection of the gripping system to the KUKA
iiwa robot. The gripper is programmed using the KUKA Sunrise Workbench in the
Java programming language.
2.3 Selection of the Additive Manufacturing Process

For the additive manufacturing of the gripper system, two different additive manu-
facturing devices were used. While the HP Designjet Color 3D uses Material Extrusion
(Fused Deposition Modeling, FDM), the Stratasys J750 is based on Material Jetting
(PolyJet Modelling, PJM). The PJM process was selected for the production of the
fingers. The reasons for this are the better resolution with regard to a smaller layer
thickness of up to 14 lm and the possibility of printing several materials at the same
time. It is possible to embed grip areas and covers made of rubber-like materials in the
print. Another reason is the investigation of the manufacture of pneumatic actuators in
the form of muscles or cylinders. This would not be possible without rework due to the
processability of only one model material and the generation of a rough surface with
the FDM process [16]. A disadvantage of the PJM process, however, is the increased
costs in terms of material and hourly machine rates. The FDM process was used to
implement the housings, as they are large and easy-to-implement components. The
layer thicknesses and the rough surface are sufficient compared to the PJM process.
This can reduce the printing time and the associated costs. Furthermore, the material is
much cheaper with this process [17, 18].
Fig. 1. Overall concept of the gripper with connection to the robot.
3 Development and Testing of Gripper and Housing
The design of the gripping system included several stages of development. At the
beginning there were preliminary material and printer tests. These were tests regarding
the printing of threads and the tightness and resilience of various materials, so that
actuators of the gripping system and their compressed air connections can be embedded
directly in the gripper. In the next step, different types of fingers were developed, using
the CAD system CATIA V5. The task was to evaluate the right finger type in con-
nection with AM. This was followed by the development of suitable actuators and their
control. Finally, the fingers and actuators were combined and integrated in a corre-
sponding housing containing the control unit. Functionality tests have been carried out
after all development stages.
3.1 Lightweight Design of the Gripper

After the actuator was completed, the gripper finger was optimized with regard to the
load and its design. Using bionic forms in the form of spaceframes, rods and struts, it
was possible to realize lightweight and yet resilient components. The fully filled areas
only serve as the skin of the finger (see Fig. 2a). Cavities and small guidance surfaces
were also created (see Fig. 2b). The small guiding surfaces reduce the friction of the
boom and thus ensure better smoothness and less wear on the finger. “Pure Tango
BlackPlus” with Shore hardness A27 is used as the material for the gripping surfaces
and Bellows. This ensures high friction, the possibility of deformation and no damage
to the parts to be gripped. All other components are made of digital ABS.
After the gripper finger was completed, an adjustment was made so that the gripper
finger can be installed in a corresponding housing. Furthermore, the compressed air
connections were routed to the rear end of the gripper finger so that no air hose could
get stuck in the actuator during operation. The peculiarity of the finger is that, due to
the specially developed flange, it can be rotated through 360° and tilted forwards and
backwards by 7.5°. This design allows the gripping of components of different sizes.
Fig. 2. 3D-CAD model of the finger: (a) 3D view of the assembly, (b) 3D sectional view to
visualize the lightweight design.
With small components, the speed of the gripping process can also be accelerated
due to the shorter closing path. The setting is done manually using screws or pins.
When turning the finger, predefined angles can be set with two centering pins. The
finger is attached to the housing using a pressure plate. The tilting movement takes
place using a centered cylindrical pin and two screw connections. Due to the possibility
of turning the fingers, all grip options, as described in Sect. 2.1, can be realized.
3.2 Development of the Housing

After the development of the gripper fingers, it was necessary to develop a housing for
it. It was important that the actuators and the control unit, consisting of valves, hoses
and connection connections, be integrated into this housing. Due to the size, the low
complexity and the low demands on the surface quality of the components, the Material
Extrusion (FDM) method was used. With regard to the control components, only the
actuators and the directional control valve are attached to a fixed location using screw
connections. All other control components are loosely located in the control box and
are protected from the moving parts of the actuators by means of a partition.
The gripper fingers can be rotated at 30° intervals using the centering pins on the
finger flange and the holes on the housing. This means that twelve different rotary
positions can be realized per finger. The control box is intended for the integration of
the control technology. As for the gripper fingers, bionic shapes in the form of trusses
and struts were used for the lightweight design. The housing only serves as protection
and is 1 mm thick, so that it hardly contributes to stability. The reason for the size of
the housing is the size of the pneumatic components. The directional valve alone, with
the largest dimension of 140 mm, requires a lot of space. Furthermore, enough space
for the actuators to the edge area is required for the inclination of the gripper fingers.
After the successful development of all components, the gripping system could be
completely assembled with the control unit. The integration of the loose control
components is difficult due to the limited space. Nevertheless, all components could be
integrated without mutual interference.
Fig. 3. (a) Laboratory set-up, (b) UI for writing commands and (c) lettering written by the robot.
3.3 Application of the Developed Gripper

With the help of the developed gripping system, the existing KUKA iiwa 7 robot was
able to write lettering and paint a sketch with different pens on whiteboard. All required
components of the laboratory set-up (robot, gripper, board with pens and control panel)
are shown in Fig. 3a. After installation, the gripper including a pen must be measured
in terms of its dimensions, center of gravity and weight. At the beginning, the tool
center point (TCP) is measured. In this example, this is at the tip of the pen.
The weight and the center of gravity of the gripper are measured by the robot itself
using a stored program. This data is important so that the force and moment sensors can
work reliably in later use. Due to the weight of the gripper, the robot would otherwise
issue an error message at the start of the application that it has detected a collision.
After measuring the gripper, this weight is calculated out accordingly. So that the
original world coordinate system of the robot can be placed on the board, it must first
be taught using three points. Text passages are written individually by the user entering
various lettering on the control panel of the robot using a process variable can (see
Fig. 3c).
The programming of the letters was realized with multi-dimensional arrays in
which the coordinates of individual letters are stored. A user interface (UI) as shown in
Fig. 3b was developed to facilitate the programming of the text. The color of the
lettering and the size of the letters can also be individually selected. The alternative way
of writing is to draw a sketch. In this case, only the color of the pen can be selected
individually. A force of 2 N is applied each time the pen touches the board. This
ensures an even and good contact pressure of the pen.
4 Technical Evaluation
After the finger system was finished, a technical evaluation was carried out. Both a
conventional and an additively manufactured gripping system from the market served
as a comparison object. In this work it was possible to create a gripping system with a
weight of 998 g, which is a satisfactory result. Due to the low weight of the gripping
system, only a minimal part of approx. 13% of the load of the robot used is used.
A commercial additively manufactured gripper is a good comparison option, both
technically and economically (see Table 1).
Technically, this is subject to the developed gripper, but is in a similar price range.
The electrical gripper is many times more expensive due to an electrical connection and
the equipped sensors and cannot be compared directly. From a technical point of view
alone, this turns out to be the best of the three grippers. While the rotary movements of
the finger take place automatically with this gripper, this has to be done manually with
the gripping system presented here. Depending on the application, it depends on the
extent to which this manual mode of operation makes sense or is disruptive. If the
fingers only need to be turned in different applications and rarely, the developed
solution is recommended. The situation is similar with the size of the components to be
gripped. With a large variation per application, the system developed here is also
unsuitable due to the manual adjustment. The advantage of mechanical adjustment is
that the gripper is easier to use when programming.
Table 1. Technical and economical characteristics of different grippers.

Criteria HRC gripper system Pneumatic gripper Electric gripper
Manufacturing Additively Additively Conventional
Finger length [mm] 130 >31,2 155
Weight [g] 998 >300 1950
Closing angle [°] +18 0 (Parallel gripper) ±90
Additional inclination [°] ±7.5 N.A. N.A.
Degrees of freedom 2 1 2
Multifunctionality High Low Very high
Price range [€] 1.450 >1.000 >45.000
5 Economic Evaluation
Furthermore, an economic evaluation was carried out. To determine the material

consumption and the related costs, all components had to be evaluated using the printer
software of both the Stratasys J750 and the HP Designjet Color 3D. The material costs
could then be determined using the cost rates, which were calculated on the basis of the
manufacturer’s price information.
The gripper fingers are made of the materials RGD515, RGD531 and Tango-
BlackPlus. The two RGD components result in digital ABS when mixed. All other
materials with the exception of the SUP706 support material are not required for the
gripper fingers. With material extrusion (FDM), care is taken to cause as little support
material as possible and thus rework. This is achieved by the fact that the support
material can be removed manually by breaking it away. After adding up all material
costs for printing with both 3D printers, the gripping system incurred material costs of
€ 366.17.
In order to determine the printing time and the corresponding costs, all components
first had to be evaluated using software programs. Furthermore, it was necessary to
determine the hourly machine rates of the devices used. Most of the costs are due to the
high hourly rate for printing with the Stratasys J750. Despite the triple print time of the
HP Designjet, the cost is less than 50% compared to the J750. The auxiliary compo-
nents for post-processing the components are very small compared to the pressure.
All control components, with the exception of the printed actuators and the elec-
trical connection line from the main valve to the robot, were purchased from a leading
supplier. The connecting line was used by manufacturer of the robot. Small parts such
as cable ties, solder, screws, etc. were not considered in the calculation.
At the end of the cost calculation, the total costs of the gripping system were
determined (see Table 2). The biggest cost driver is the printing and cleaning costs.
This is due to the high machine hourly rate of the Stratasys J750. For this reason, it is
advisable to only use it to print components that have high quality requirements.
Table 2. Comparison of cost types and cost shares.

Costs Absolute costs [€] Relative costs [%]
Material costs 366.17 25.3
Printing and post-processing costs 748.41 51.6
Additional parts 335.04 23.1
Total costs 1,449.62 100.0
When planning the gripping system, the costs were estimated at around € 1,500.
With total costs of € 1,449.62, this value was even undercut by 3.36%. That represents
a satisfactory result. It should be added that no personnel costs have been factored into
the developed gripping system. For this reason, the comparison with other gripping
systems should be treated with caution, since other components such as research and
development costs, overhead costs, personnel costs and a profit have been included.
6 Conclusion
With the help of this contribution it was possible to develop and to implement a
multifunctional and customer-specific human-robot collaboration gripping system
using additive manufacturing. A satisfactory result was achieved at a cost of € 1,449.62
and a weight of 998 g. In particular, the use of lightweight design made it possible to
realize the low weight of the gripping system, which only uses a small portion of
approx. 13% of the load capacity of the robot used. By using a multi-material printer,
complex components in different materials and functions (e.g. rigid and solid structure
and flexible bellows) could be combined in one component. This approach has also
contributed significantly to weight loss.
It has been shown that in addition to the material costs, the printing costs also have a
high impact on the overall costs. It was demonstrated that a reasonable combination of
demanding and therefore expensive printing processes (e.g. for the manufacturing of the
fingers) and simple and inexpensive processes (e.g. for the manufacturing of the
housing) enables compliance with the specified costs. The budget was met with a saving
of 3.36% compared to the calculated costs. It should be noted that no personnel costs
and small parts with minimal and therefore negligible costs have been included. It could
be shown that with the approach presented here, a practical example in which a gripper
is used to hold a pen when writing and painting, can be successfully implemented.
References
1. Wendt, T.M., Himmelsbach, U.B., Lai, M., Waßmer, M.: Time-of-flight cameras enabling
collaborative robots for improved safety in medical applications. Int. J. Interdiscipl.
Telecommun. Networking 9(4), 10–17 (2017)
2. Oberc, H., Prinz, C., Glogowski, P., Lemmerz, K., Kuhlenkötter, B.: Human robot
interaction – learning how to integrate collaborative robots into manual assembly lines.
Procedia Manuf. 31, 26–31 (2019)
3. Vogel, C., Elkmann, N.: Novel safety concept for safeguarding and supporting humans in
human-robot shared workplaces with high-payload robots in industrial applications. In:
Mutlu, B., Tscheligi, M., Weiss, A., Young, J.E. (eds.) Proceedings of the Companion of the
2017 ACM/IEEE International Conference on HRI, pp. 315–316. ACM Press, New York,
USA (2017)
4. Ranz, F., Komenda, T., Reisinger, G., Hold, P., Hummel, V., Sihn, W.: A morphology of
human robot collaboration systems for industrial assembly. Procedia CIRP 72, 99–104
(2018)
5. Mourtzis, D.: Simulation in the design and operation of manufacturing systems: state of the
art and new trends. Int. J. Prod. Res. 58(7), 1927–1949 (2020)
6. Lachmayer, R., Lippert, R.B., Fahlbusch, T. (eds.): 3D-Druck Beleuchtet. Additive
Manufacturing auf dem Weg in die Anwendung. Springer, Berlin (2016)
7. Knapp, M.: Roboter bauen mit Arduino, 1st edn. Galileo Press, Bonn (2015)
8. Schreiber, F., Manns, M., Morales, J.: Design of an additively manufactured soft ring-
gripper. Procedia Manuf. 28, 142–147 (2019)
9. Sanchez-Tamayo, N., Wachs, J.P.: Collaborative robots in surgical research. In: Kanda, T.,
Ŝabanović, S., Hoffman, G., Tapus, A. (eds.) Companion of the 2018 ACM/IEEE
International Conference on Human-Robot Interaction, pp. 231–232. ACM Press, New York
(2018)
10. Hatscher, B., Luz, M., Nacke, L.E., Elkmann, N., Müller, V., Hansen, C.: GazeTap: towards
hands-free interaction in the operating room. In: Lank, E., Vinciarelli, A., Hoggan, E.,
Subramanian, S., Brewster, S.A. (eds.) Proceedings of the 19th ACM International
Conference on Multimodal Interaction, pp. 243–251. ACM Press, New York (2017)
11. Safeea, M., Neto, P.: Minimum distance calculation using laser scanner and IMUs for safe
human-robot interaction. Robot. Comput.-Integr. Manuf. 58, 33–42 (2019)
12. Strobel, M., Döttling, D.: High dynamic range CMOS (HDRC) imagers for safety systems.
Adv. Optical Technol. 2(2), 147–157 (2013)
13. Zanella, A., Cisi, A., Costantino, M., Di Pardo, M., Pasquettaz, G., Vivo, G.: Criteria
definition for the identification of HRC use cases in automotive manufacturing. Procedia
Manuf. 11, 372–379 (2017)
14. Haase, T. Wörn, H., Nahrstaedt, H.: Shared memory in RTAI Simulink for kernel and user-
space communication at the example of the SDH-2 - ICINCO 2010. In: Proceedings of the
7th International Conference on Informatics in Control, Automation and Robotics, vol. 3,
pp. 160–165 (2010)
15. Spiliotopoulos, J., Michalos, G., Makris, S.: A Reconfigurable gripper for dexterous
manipulation in flexible assembly. Inventions 3(1), 1–4 (2018)
16. Ngo, T.D., Kashani, A., Imbalzano, G., Nguyen, K.T.Q., Hui, D.: Additive manufacturing
(3D printing): a review of materials, methods, applications and challenges. Compos. Part B
Eng. (2018)
17. Kadkhoda-Ahmadi, S., Hassan, A., Asadollahi-Yazdi, E.: Process and resource selection
methodology in design for additive manufacturing. Int. J. Adv. Manuf. Technol. (2019)
18. Bourell, D., et al.: Materials for additive manufacturing. CIRP Ann. 66(2), 659–681 (2017)
Enhanced Cooling Design in Wire Drawing
Tooling Using Additive Manufacturing
Joakim Larsson(&), Patrik Karlsson, Jens Ekengren, and Lars Pejryd
School of Science and Technology, Örebro University,

Fakultetsgatan 1, 701 82 Örebro, Sweden
joakim.larsson@oru.se
Abstract. Wire drawing is a manufacturing process in which metal rods or

wires are drawn through a single or a series of dies, reducing the wire cross-
section and enhancing the mechanical properties of the wire. The tribological
conditions in wire drawing are quite extreme and high friction between the wire
and the die results in an increased die temperature. Previous studies have shown
that by reducing the die temperature the lifetime of the die increases and thus
efficient cooling of the die is of high importance.
Additive manufacturing enables fabrication of tools with advanced conformal
cooling channels with high cooling efficiency. This technique may, therefore, be
of high importance in the design of the cooling system of drawing dies. In the
present study, the effect of conformal cooling design of die holder on the die
temperature, and thus die performance, was investigated. A die holder was
manufactured by means of laser powder bed fusion (LPBF) in an EOS M290
machine using atomized corrosion resistant steel (Corrax). The cooling effi-
ciency of the manufactured tool holder was evaluated in an industrial wire
drawing process and further analysed using FEM modelling. This study shows
promising results on improved cooling efficiency for die holder designed and
manufactured by additive manufacturing.
Keywords: Wire drawing Cooling Additive manufacturing
1 Introduction
Wire drawing is a manufacturing process in which metal rods or wires are drawn
through a single or a series of dies, reducing the wire cross-section and enhancing the
mechanical properties of the wire. Even though the process is defined as a cold working
method, heat is generated in the wire and the tool as the wire is deformed. Studies have
shown that up to 15% of the total power used for the reduction stays in the drawing die,
Enghag [1]. The part of the drawing tool that is in contact with the wire (drawing die) is
usually made of either cemented carbide or diamond and can either be integrated in a
steel/carbide die or used as nibs in an interchangeable die core system. The cemented
carbide used for the dies is sensitive to high temperatures. At 800 °C already 50% of
the hardness is lost, according to a producer’s data sheet [2].
Finite element studies of an industrial wire drawing situation have previously been
reported by Larsson and Jarl [3]. The simulations were verified against the actual

https://doi.org/10.1007/978-3-030-54334-1_30
Enhanced Cooling Design in Wire Drawing Tooling 427
industrial process and the result showed that temperatures can exceed 800 °C in the
drawing die. Improving the cooling of dies is, therefore, of highest importance in order
to ensure long life time of the drawing die. In another study, Larsson and Jarl [4]
showed that more efficient cooling may indeed extend the life of the die. However, the
tool holder used in that particular experiment had a short life time due to the hardness
of the used material, making it unsuitable for industrial use. In order to increase the
understanding of the factors influencing the performance and for further development
of the process to extend the life of the tools, a revisiting of the cooling of drawing dies
and nibs was therefore deemed interesting.
Another limiting factor is the lubricants used. The most commonly used in dry
drawing wire drawing processes are calcium or sodium soaps that oxidize when
exposed to high temperatures. If the lubricant oxidizes, it loses its lubricating abilities
[5]. If the cooling of the inlet cone of the die could be improved, increased productivity
may be achieved.
The idea to use additive manufacturing (AM) as a production method for metallic
tools in order to allow for more freedom in the design of cooling systems is not new.
The AM methods have advanced in maturity lately and examples of applications of
conformal cooling channels through the use of AM can be found in literature, e.g.
Hölker et al. [6] and Jahan and El-Mounayri [7]. Hölker et al. [6] investigated a hot
extrusion die for aluminium using AM to produce the tool with conformal cooling
channels, claiming a 300% productivity increase. Jahan and El-Mounayri [7] reported
on a design process for optimizing the cooling of a die for plastic injection moulding.
Based on the potential seen in the increased life of the drawing tools by improved
cooling and the identified possibilities in using AM to accomplish this, the current
project was directed towards investigating the cooling capabilities of a tool holder for
wire drawing nibs with conformal cooling channels. This was done both by experi-
mental investigations of the cooling process in a controlled laboratory situation and
experiments in an industrial wire drawing environment. The data from the experiments
was used to verify the finite element model that was developed. The verified model was
then used for simulation to give further understanding of the cooling in the wire
drawing process.
2.1 Wire Drawing Theory

The force required to pull the wire through the drawing die can be calculated theo-
retically. This is commonly done using the formula derived by Siebel and Kobitzsch [8]

A0 2a l A0
F ¼ A1 Rem ln þ þ ln ; ð1Þ
A1 3 a A1
where F is the total drawing force, A0 and A1 are the area of the wires cross section
before and after the reduction, Rem is the mean flow tension for the material before and
after the reduction, 2a is the semi-die angle of the die and l the coefficient of friction
428 J. Larsson et al.
between the wire and the die. According to literature, the friction coefficient between
the wire and the die lies between 0.01 and 0.07 for a well functional dry lubricated wire
drawing process using soap powder as lubrication [9].
The power needed for a pass in a wire drawing process can be calculated as
P ¼ VF; ð2Þ
where P is the needed power and V is the drawing speed.

When the wire is drawn through the die, heat is generated due to the plastic
deformation of the wire and friction between the wire and the die. The temperature
increase in the drawn wire during one reduction step can be estimated using the
following equation [1],
F=A1
DT ¼ k ; ð3Þ
q Cp
where DT is the temperature increase, q is the density of the wire, Cp is the specific
heat capacity of the drawn material and k is a correction factor. The loss of energy from
the wire that the constant k represents is mostly due to the cooling of the wire in the
drawing die, which is depending on die design, drawing speed, die cooling system,
drawing tools material, the thermal conductivity of the wire and die material and other
parameters.
Literature states that up to 15% of the energy produced in the wire during the
reduction process is removed by energy transport to the drawing die and subsequent
cooling [1]. Many parameters influence the exact number, but the drawing speed is one
of the most important factors. Although the heat generated by the deformation of the
wire is independent of the drawing speed, an increase in drawing speed increases the
energy released per time unit. This results in an increased die temperature due to
insufficient removal of heat from the drawing die. As the die temperature increases this
leads to a reduction of the fraction of the total energy that goes to the drawing die. In a
previously study, it was shown that as a carbon steel wire was drawn at 0.33 m/s, 7.5%
of the total energy went to the drawing die [10].
2.2 Die Holder

There are many different designs of cooling systems for the drawing tool in a con-
ventional wire drawing setup. The most efficient setup is a so called direct cooled die
holder. The drawing tool is placed in a die holder and the holder is placed in a cooling
body. As the process runs, the cooling body is filled with coolant, meaning that the die
holder is surrounded by coolant. Figure 1a and b shows a conventional wire drawing
die holder that is made for a directly cooled system. The die is positioned in the conical
surface that can be seen inside of the cooling flanges illustrated in Fig. 1 b).
Fig. 1. CAD models of the two tool holders a) the conventional direct cooled b) a schematic cut
view of the conventional tool holder c) the new printed design with conformal cooling channels.
As the die holder itself, already in the conventional setup, is submerged in turbu-
lently flowing coolant, the idea to be able to increase the cooling rate of the die is to
move the coolant closer to the die. Thus, minimizing the distance between the die and
the coolant should result in enhanced cooling of the die. In this study, the die was
designed with a cooling channel placed as close to the drawing tool as possible. The
cooling channel was designed to revolve around the drawing tool in a helical geometry.
The coolant channel cross-section has the geometry of a teardrop to avoid problems
when the roof of the channel was being printed [11]. The top of the teardrop was
oriented in the building direction, along the positive Z-axis direction during printing.
The cross-sectional area of the channel was designed to be 8.9 mm2. A more detailed
description of the design process for the cooling body can be found in [12]. The
resulting cooling body used in this study is shown in Fig. 1c.
2.3 Laboratory Experiments

A previous study reports an experimental trial performed in order to evaluate and
compare the cooling capacity between the conventional die holder and the additive
manufactured holder in a laboratory setup. The results in form of temperature differ-
ences between the coolant and the cooled object placed at the position were the die
would sit in an industrial setup is shown in Fig. 2. The graph shows that the die holder
with conformal cooling gives roughly 25% lower temperature difference between the
coolant and the cooled object, thus increasing the cooling efficiency. During the lab-
oratory experiments, other parameters such as temperature of coolant and flow of
coolant was also measured to ensure that the experiments with the different die holders
were performed under the same conditions. The results were also used to obtain
suitable boundary conditions for finite element simulations (FEM) [12].
ΔTemperature measuring probe/coolant

20
15
∆T (°C)
10
Conventional
5
AM
0
0 50 100 150
Time (s)
Fig. 2. Results from laboratory experiments using conventional die holder and AM die holder
with conformal cooling channels. The graphs show temperature differences between the cooled
object and the coolant as a function of running time.
2.4 Finite Element Study

FEM analyses were made using Ansys 2019 R3 [13], using a standard implicit transient
heat solver. The models consisted of approximately 19000 elements. Boundary con-
ditions, convection and heat flow, were taken from the laboratory experimental con-
ditions [12]. The convection boundary condition was used to simulate the heat removal,
which is done by the coolant. A cut of the mesh (without the die) used for the analysis
of the die holder with conformal cooling channels is shown in Fig. 3.
Fig. 3. Sectioned image of the FEM model used in this study

The heat generated by the wire drawing process was added to the model as a heat
flow on the inside surface of the drawing die. The heat flow was set to represent the
measured results from the industrial experiments.
2.5 Industrial Experiments

In order to verify that the results found in the laboratory experiments are valid for the
real application, industrial experiments were made. The die holder was used in the last
draw of an industrial drawing machine with 9 steps. In the process, a high carbon steel
wire was produced and at the finishing block the wire had a diameter of 4.24 mm and a
maximum speed of 7.5 m/s (multiple drawing speeds were tested during the experi-
ments). To be able to compare the conventional die holder with the additive manu-
factured tool holder, data for the conventional tool holder first needed to be acquired.
The same experimental procedure was used for the two die holders.
To study the differences in cooling efficiency between the two die holders, the
drawing dies used in the experiments were equipped with type k thermocouples. The
thermocouples were spot welded to the exit side of the dies as shown in Fig. 4.
Fig. 4. Drawing die equipped with a thermocouple on the exit side.
The thermocouples were sampled during the experiments using a Testo 176T4. The
sampling frequency was 1 Hz and the reported measuring error is ±0.5% of measured
value.

3.1 Wire Drawing Experiments
In the industrial wire drawing tests, the additively manufactured die holder with con-
formal cooling channels was mounted in the same manner as a conventional die holder.
The coolant was feed to the die holder through an 8 mm (28 mm2) hose. The coolant
outlet of the die holder was equipped with a short hose in order to steer the coolant into
the standard outlet of the coolant container. An image of the die holder mounted in the
cooling body can be seen in Fig. 5.
Fig. 5. Additively manufactured die holder mounted in a conventional wire drawing machine.
During the experiment, five different drawing speeds were used in order to study if
any differences could be found depending on the drawing speed. The highest drawing
speed used in the experiments, 7.5 m/s, is the normally used speed for the specific
product. To ensure a good lubrication in the start, the highest speed was used in the
beginning, as the lubricant used was chosen for that speed. The highest drawing speed
400
Tool temperature (°C)
350
300
250
200
150 3d-printed
100
Conventional
50
0
0 1000 2000 3000 4000 5000 6000
Time (s)
Fig. 6. Measured temperatures from the two different experiments. Five different drawing
speeds were used (one specific was used twice, resulting in six distinct levels).
was also used for an extended time at the end of the experiments. Results from the
experiments, in form of temperature measured at the exit side of the drawing die, is
shown in Fig. 6.
The temperature curves have been analysed and mean temperature values for the
different drawing speeds have been extracted, as presented in Table 1.
Table 1. Mean temperatures from the performed industrial experiments.

Drawing speed Conventional Additive manufactured Difference
(m/s) (°C) (°C) (K)
3 253 206 47
4 265 216 49
5 285 231 54
6 305 250 55
7,5 329 269 60
As shown, both the temperature itself and the temperature difference between the
conventional die holder and the additive manufactured die holder increase with
increased drawing speed. This is explained by the higher power added to the system at
a higher drawing speed. However, if the percentage difference is studied for the dif-
ferent drawing speed almost no difference can be seen, the result show that the die
holder with conformal cooling channels gives roughly 18% lower temperatures than the
conventional die holder for all drawing speeds.
3.2 Finite Element Analysis

To be able to evaluate the cooling process in the industrial wire drawing process by
FEM, the amount of energy that goes into the die needs to be evaluated. This was done
by iterating the value of the boundary condition (heat flow) for the ingoing heat until
the surface where the thermocouple was mounted during the experiments showed the
same value as from the experiments. This calibration was performed for all different
drawing speeds for the conventional die holder case. The heat flows used in the
simulations of the additively manufactured die holder by FEM were those obtained by
calibrating the simulations of the conventional die holder to the measured temperatures,
for each drawing speed. The rest of the boundary conditions used were the same as in
the FEM study reported earlier [12].
For the normal production speed (7.5 m/s), it was estimated that around 485 W is
going into the drawing die. Using Eq. 1 and 2, an approximate power of 33.25 kW was
needed for the specific case, meaning that around 1.5% of the total power that was used
in the forming process went to the drawing die as heat. These parameters were put into
the FEM models for the conventional case and for the case with the additive manu-
factured die holder. The resulting temperature plots from the end (steady state) of two
of the FEM simulations can be seen in Fig. 7.
Fig. 7. Temperature result plot from FEM simulations. Left: conventional die holder. Right:
AM die holder with conformal cooling channels.
As can be seen, the estimated maximum temperature of the die is much lower when
using the additively manufactured die holder. In the real process, there will be much
higher temperature in local spots where the highest pressures occur during the forming
process. To be able to catch these high temperature peaks, a different type of simulation
needs to be performed. However, to reach a steady state situation regarding tempera-
ture, with the cooling body modelled, would with that type of simulation take immense
time and computer power. To reach steady state approximately 15 min of process
needs to be simulated, each second 7.5 m of wire is drawn. This means that around
6 750 m of wire drawing needs to be simulated to reach a steady state using a transient
mode simulation. Researchers at Örebro University have simulated wire drawing
processes including wire deformation. These simulations only handled a few cen-
timetres of wire being drawn, and still the simulations took hours.
The results from the simulations with the other drawing speeds from the experi-
ments are presented in Table 2. PowerT is the total power needed for the reduction pass
calculated using Eq. 1 and 2, PowerD is the power that was found going to the drawing
die, AM represents the new die holder with enchanted cooling, C represents the con-
ventional die holder, Tm is the maximum temperature from the simulation, Tmp is the
temperature from the simulation in the point where the temperature was measured
during the industrial experiments and Δ%EvFEM is the percentage temperature dif-
ference between the simulations and the experiments in the measuring point.
Although a larger power is needed to perform the drawing at higher drawing
speeds, a lower fraction of that power is transferred as heat to the die holder. Figure 8
graphically show the results regarding the percentage of the energy that goes to the
drawing die.
Table 2. Results from the FEM studies

Speed PowerT PowerD PowerD AM C TM AM C Tmp C Δ% AM Δ%
(m/s) (W) (W) (%) TM (°C) Tmp (°C) EvFEM EvFEM
(°C) (°C) (%) (%)
3 13000 370 2,8 264 307 212 254 0,5 2,8
4 17300 390 2,3 278 323 223 267 0,7 3,2
5 21700 420 1,9 297 346 238 286 0,5 3,1
6 26000 450 1,7 317 369 254 304 0,3 1,7
7,5 32500 485 1,5 340 396 272 327 0,6 1,0
5
Percentage of total energy that goes to
4.5
4 y = 5.9867x-0.693
3.5
drawing die (%)
3
2.5
2
1.5
1
0.5
0
2 3 4 5 6 7 8
Drawing speed (m/s)
Fig. 8. Percentage of the total energy needed for the drawing pass that ends up in the drawing
die.
4 Conclusions and Future Work
In the present study, a conventional tool holder for the wire drawing process was
compared to a tool holder with conformal cooling channels manufactured in an anti-
corrosive tool steel using additive manufacturing. The additive manufactured tool
holder had cooling channels placed as close as possible to the drawing die. The goal of
the project was to increase the cooling of the drawing die, and thereby increase the
lifetime of the drawing die.
The cooling efficiency was evaluated in an industrial wire drawing process. Mul-
tiple production speeds were used to study the influence of the drawing speed. To be
able to compare the two die holders, the dies used in the experiments were equipped
with thermocouples. The results from the experiments show that the die cooled using
the additive manufactured tool holder had roughly 18% lower temperature for all
drawing speeds.
Finite element calculations were performed for the different experimental situations
and the result from the simulations were in good agreement with experimental results.
The models can be used to further understand and develop cooling systems for the wire
drawing process.
The present study has shown the potential of additively manufactured tool holder
for wire drawing processes. New design of the holder resulted in enhanced cooling of
the die, which may enable longer tool life or increased production speed. At a set
speed, there will be a lower temperature in the die and thus also in the lubricant. The
cooling may be used to extend the life of the tool, and may also be used to increase
drawing speed while still working in a temperature range where the lubricant is
working properly. Wire drawing lubricants are designed for optimal properties in
different temperature regions and increased cooling would extend the useable pro-
duction speed window for the specific lubricant.
However, to be able to verify these benefits, further studies need to be conducted.
To study the effect of the increased cooling capacity on the lifetime of the drawing dies,
longer industrial experiments need to be performed, where the wear is measured over
time.
References
1. Enghag, P.: Steel Wire Technology. Materialteknik HB, Örebro (2009)
2. Sandvik: Understanding Cemented Carbide The material with staying power, p. 20 (2011)
3. Larsson, J., Jarl, M.: Högre draghastigheter/temperaturens inverkan. NTTF, pp. 51–58
(2011)
4. Larsson, J., Jarl, M.: Högre draghastighet. NTTF Årsb., pp. 63–70 (2012)
5. Haglund, B.O., Enghag, P.: Characterization of lubricants used in the metalworking industry
by thermoanalytical methods. Thermochim. Acta. 282–283(SPEC. ISS.), 493–499 (1996)
6. Hölker, R., Haase, M., Ben Khalifa, N., Tekkaya, A.E.: Hot extrusion dies with conformal
cooling channels produced by additive manufacturing. Mater. Today Proc. 2(10), 4838–4846
(2015)
7. Jahan, S.A., El-Mounayri, H.: Optimal conformal cooling channels in 3D printed dies for
plastic injection molding. Procedia Manuf. 5, 888–900 (2016)
8. Siebel, E., Kobitzsch, R.: Die Erwärmung des Ziehgutes biem Drahtzienhen. Stahl Eisen 63
(6), 110–114 (1942)
9. Shemenski, R.: Ferrous Wire Handbook. Wire Association International, Guildford (2008)
10. Larsson, J., Jansson, A., Karlsson, P.: Monitoring and evaluation of the wire drawing process
using thermal imaging. Int. J. Adv. Manuf. Technol. 101(5), 2121–2134 (2019)
11. Adam, G.A.O., Zimmer, D.: On design for additive manufacturing: evaluating geometrical
limitations. Rapid Prototyp. J. 21(6), 662–670 (2015)
12. Pejryd, L., Larsson, J.: Additively manufactured tool holder for wire drawing processes. In:
EURO PM2018 Congress Proceedings (2018)
13. Ansys: 2019 version 19.0 R3
Aortic Model in a Neurointerventional
Training Model – Modular Design
and Additive Manufacturing
Nadine Wortmann1(&) , Andreas M. Frölich2 ,

Anna A. Kyselyova2 , Helena I. De Sousa Guerreiro2 ,
Jens Fiehler2 , and Dieter Krause1
1
Institute of Product Development and Mechanical Engineering Design,
Hamburg University of Technology, 21073 Hamburg, Germany
nadine.wortmann@tuhh.de
2
Department of Diagnostic and Interventional Neuroradiology,
University Medical Center Hamburg-Eppendorf (UKE),
20246 Hamburg, Germany
Abstract. For training physicians in endovascular techniques such as

mechanical thrombectomy in acute stroke, synthetic in-vitro models may replace
animal models. A neurointerventional training model was developed in previous
works using additive manufacturing (AM) for the reproduction of patient
specific anatomy. Different patient anatomies, such as curvatures, can compli-
cate the pathway of treatment. For this reason, realistic training requires a
simulation of the entire access path from the femoral artery to the affected vessel
in the brain, which includes the simulation of the aorta.
The training model currently uses a commercially available silicone aorta,
which has several disadvantages, including high cost and unrealistic surface
friction. Furthermore, the aortic model is not modular and therefore does not
allow changes in configuration of the aortic arch, which is a strong factor
influencing procedural difficulty and therefore an important variable for training.
In this study, a modular aortic model is designed and manufactured according
to the requirements for training endovascular stroke treatment. AM offers many
advantages in the production of anatomical models. Therefore, different man-
ufacturing alternatives are tested based on a modular concept, using both direct
and indirect manufacturing. Criteria for an evaluation of the production pro-
cesses and the resulting models are defined and the test set-up is described. In
this study, the procedures are first evaluated under cost and time aspect and a
first assessment of the qualitative criteria is given.
Keywords: Synthetic aortic model Modular design Additive manufacturing
1 Introduction
1.1 Motivation
The training of catheter-based interventions for endovascular treatment of vascular
diseases (e.g. thrombectomy for treatment of acute stroke) is mainly performed on

https://doi.org/10.1007/978-3-030-54334-1_31
438 N. Wortmann et al.
animal models [1]. For example, an anesthetized pig is injected with previously col-
lected and clotted blood and the mechanical removal of the blood clot via the femoral
artery is trained [2]. However, training on animal models has many disadvantages. In
addition to the general ethical aspects of animal tests, the vascular anatomy of pigs does
not correspond to that of humans [1], which reduce the training effect. In particular,
vascular curves that occur in elderly patients and challenge the intervention cannot be
trained in the animal model [3].
In order to avoid animal models and achieve better training possibilities, various
endovascular training models were developed and partly marketed. The training model
of the company Vascular Simulations, Inc. (New York, USA) allows the training
endovascular techniques, such as aneurysm or stroke treatment. The company produces
patient-specific vascular models [4]. The model EVE (EndoVascular Evaluator) from
FAIN-Biomedical Inc. (Nagoya, Japan) is also a holistic model for training endovas-
cular diseases. It allows the exchange of different modules to allow training on different
vascular diseases [5]. Spallek et al. pointed out the advantages and disadvantages of
these models and justified the need for a new neurointerventional training model by
stating that a simple and cost-efficient exchange of patient-specific models is not
possible even during training with the commercial models [6].
In previous works the training model HANNES (Hamburg ANatomical NEuroin-
terventional Simulator) was developed for training of aneurysm treatment [6].
HANNES is characterized by its high modularity, which allows for easy change of
vessel models to represent a wide range of anatomies. Additive Manufacturing (AM) is
used for the production of the vessel replicas because it offers a high degree of geo-
metric freedom and enables fast production in small quantities [7].
Essential adaptations to HANNES for use in the training of stroke treatments were
shown in Wortmann et al. [8]. This includes the possibility to replace the aortic arch
with different models to achieve different levels of training difficulty. Currently,
HANNES has a commercial silicon aorta (United Biologics, Inc., Santa Ana, USA),
which is not modular and therefore does not allow the replacement of the aortic arch.
The aim of this study is to design a modular aortic model, utilizing AM to replicate
patient-specific anatomy. Three different manufacturing processes are compared. Both
direct and indirect AM is taken into account. A comprehensive evaluation is being
prepared to assess cost, time and quality aspects. Criteria will be defined for this
purpose. The different processes are evaluated in this study under the focus of cost and
time aspects.
1.2 Medical Background

The common femoral artery often serves as the access point for endovascular treatment.
The catheters and treatment devices are advanced via the aorta, the cervical arteries to
the cerebral arteries where the treatment takes place.
The aorta is the central artery of the human body and transports the blood from the
heart into the large blood circulation. The aorta is an elastic artery which, like the other
arteries of the body, is made up of three layers of walls [9]. Anatomically, the aorta can
be divided into the five segments aortic root, ascending aorta, aortic arch, descending
thoracic aorta and abdominal aorta (Fig. 1).
Aortic Model in a Neurointerventional Training Model 439
Fig. 1. Division of the aorta into segments (based on [10, 11])
The aortic root, shown on the left in the figure, connects to the aortic valves and,
together with the ascending aorta, forms the transition to the aortic arch up to the outlet
of the brachiocephalic trunc (the first large branching vessel). The supraaortic vessels
brachiocephalic trunc dexter, carotis communis sinistra and subclavia sinistra arise
from the aortic arch, which in turn ensure blood flow to the arm and the cervical and
cerebral arteries [9].
Anatomically, three types of aortic arches can be classified [9, 12, 13]. These differ
mainly in the position of their outlets to the cervical arteries (see Fig. 2) and thus
represent different curves for the treatment path, resulting in different levels of difficulty
in treatments [9, 12].
Fig. 2. Classification of the aortic arch by the location of the brachiocephalic trunc into three
types (based on [9])
Type I aortic arch is characterized by the brachiocephalic trunc lying on the same
horizontal plane that describes the curvature of the outer aortic arch contour (see Fig. 2,
(a)). In type II aortic arch, the vessel outlet lies deeper between the outer and inner
aortic arch curvature (see Fig. 2, (b)). An aortic arch is categorized as type III if the
outlet of the brachiocephalic trunc is below the inner aortic arch curvature (see Fig. 2,
(c)) [9]. Thus, a type III arch results in more severe curvature to overcome during
catheter delivery, making the intervention more difficult [3].
The aortic arch is further bordered by the descending thoracic aorta, which extends to
the diaphragm, and then the abdominal aorta, which extends to the aortic bifurcation [9].
In this study the focus is on the reconstruction and manufacturing of the aortic arch
and the possibility of exchangeability of different aortic arch types in the training
model.
1.3 Hamburg ANatomical NEurointerventional Simulator (HANNES)

The study is based on HANNES (Hamburg ANatomical NEurointerventional Simu-
lator). HANNES is an endovascular training model for aneurysm treatment and has
completely replaced animal-based training in the rabbit model at the University
Medical Center Hamburg-Eppendorf (UKE) since 2016. HANNES was developed in a
collaborative project between the Hamburg University of Technology (TUHH) and the
Department of Diagnostic and Interventional Neuroradiology at UKE. HANNES
consists of a base frame, electronic and control unit, fluid system, the purchased aorta, a
head module with skull base and interchangeable cerebral and cervical vessel models.
In-house developed adapters allow an easy change of vessel models even during
training without creating inner edges [6]. Figure 3 shows HANNES in the angio suite
environment.
Fig. 3. HANNES in the experimental angio suite at the Medical Center Hamburg-Eppendorf
(UKE)
Wortmann et al. show the extensions of the HANNES platform for stroke treat-
ment. Besides the integration of synthetic blood clots and stenosis models, the different
types of aortic arch will be integrated into the training [8].
HANNES’ current aorta is a commercially available model (United Biologics, Inc.,

U.S.A.). As shown in the Fig. 4, the aorta is not modular and an exchange of different
aortic arches is not possible. To enable connection to the HANNES model, the
supraaortic connections were replaced by the adapters typical for HANNES.
Fig. 4. Silicone aorta of the company United Biologics (a) [14] and the aortic model integrated
into HANNES (b)
2 Modular Design and Manufacturing Process Selection
First, the requirements for the aortic model were determined together with the neuro-
radiologists of the UKE. The model should be transparent so that the catheter guide is
visible even without fluoroscopy. Furthermore, the aorta should be elastic so that it
behaves similar to reality and it should provide realistic friction between catheter and
vessel material. An interchangeability of the aortic arch should be given so that training
on the different arch types is possible. At the same time, compatibility with the adapters
previously used in the model should be ensured.
The variety required by the customer (UKE) in relation to the aortic model was
included in the form of a variety tree. The variety driving properties are especially the
anatomy of the aortic arch.
Based on anonymized CT imaging data of a type II aortic arch, a model was
designed with Meshmixer (Autodesk, U.S.A.) and reconstructed in CAD with CATIA
V5 (Dassault Systemes SA, France), resulting in an hollow vessel model (Fig. 5). The
wall thickness was set to 2 mm based on experience with the cerebral vessel models.
Fig. 5. CT-scan of an aorta of aortic arch type II (a) and the generated CAD model in CATIA
V5 (b)
HANNES adapters were added to the CAD model at the supraaortic outlets. Due to
the larger diameters at the transition between the aortic arch and the descending tho-
racic aorta, a new adapter was developed, which also allows an edge-free connection of
the models (see Fig. 6).
Fig. 6. Reconstruction of the aortic arch in CATIA V5 (a) and subsequent generation of the
STL-file (b)
Spallek et al. compared different AM procedures and materials for the direct
manufacturing of cerebral vessel models. It was shown that the procedures Material
Jetting (MJ) and Stereolithography (SLA) are well suited for the fabrication of cerebral
vessel models with aneurysms. For the MJ, the materials TangoPlus FLX930 and
HeartPrint Flex (Materialise GmbH, Munich) on the Objet printer proved to be
promising [7]. With the HeartPrint Flex material, Materialise is able to produce models
such as vessels with elasticity similar to the real vessel. [15]. No elastic material was
available on the Form 1+ from Formlabs (U.S.A.) at the time of the study, making MJ
the preferred procedure.
Since 2019 Formlabs has been offering the material Elastic Resin. The properties
can be taken from the material data sheet [16]. Due to its promising properties and to
the fact, that this printer is available to the research partners at the university as well as
at the university medical center, this material is included in this study.
Also the direct manufacturing out of HeartPrint Flex (Materialise) is to be
compared.
Indirect manufacturing is chosen as another manufacturing alternative. In her work,
Heidemanns produced a silicone model of the aorta [17]. First, she used the CT data to
create native segments from a modelling compound in order to make an impression
using wax, silicone and gypsum. The wax models were then poured into the prepared
mould and served as a positive model to apply the silicone in several layers with a
brush [17]. Heidemanns did not use the possibilities of AM in her approach. Macroni
et al. produce a parameterized aortic model based on literature data. The model was
produced by casting the silicone in a 3D-printed mould. Inner and outer shells were
used and the model was cast under vacuum [18]. In both described studies the aortic
arches are not interchangeable. In this work, a mould printed by means of Stere-
olithography is to be produced, which is then used to create a wax model. This in turn
forms the core for the layered application of silicone.
3 Manufacturing
3.1 Stereolithography (SLA) with Formlabs Form 3
Method and Material: Stereolithography (SLA) with the Form 3 from Formlabs,
Elastic Resin (209.25 ml), Form Wash (IPA (90%)), Form Cure, in-house production.
Production-Specific Preparation of the Model: The interfaces on the aortic arch model
had to be modified in the CAD model so that the model fits into the permissible
installation space of Form 3 145 145 185 mm. In the PreForm software by
Formlabs, the model is virtually orientated on the building platform and support
structures can be generated (see Fig. 7, (a)). The model almost fills the permissible
installation space. From PreForm the model can be transferred directly to Form 3 and is
ready for printing. The process of preparation is calculated at about 1.5 h.
Production of the Model: The Elastic Resin material is inserted at the printer and
printing is started. The printing time for the model is 30 h. Figure 7, (b) shows the
model after printing on the building platform and in (c) the finished model. The
postprocessing time is about 1 h.
Fig. 7. Aortic arch model in the PreForm software (Formlabs) for preparing the print (a), model
after printing on the building platform of the printer Form 3 (b), model after post processing (c)
3.2 Material Jetting (MJ) with Materialise
Method and Material: Material Jetting (MJ), HeartPrint Flex, order production.
Production-Specific Preparation of the Model: In a telephone conversation the
requirements for the model were clarified and a decision was made to print it. The
finished STL file is sent to Materialise for printability testing. With this printing method
it is possible to have the model printed in places with different Shore hardnesses. For
this model a uniform Shore hardness is chosen first. The material properties of the
aortic arch model have a Shore hardness of 30 A and correspond to a tensile strength of
1.04 ± 0.04 MPa with the wall thickness of 2 mm used [19].
Production of the Model: The model is produced by Materialise after the STL file has
been sent. From receipt of order the delivery time is 14 days. The total price including
tax and shipping is just above the three-figure range. The model is shown in Fig. 8.
Fig. 8. Heartprint aortic arch model ordered from Materialise

3.3 Silicone Cast with a Wax Model
Method and Material: Paraffin pastilles (idee. Creativmarkt, Germany) (200 g), coo-
ker, melting pot, silicone shore hardness 33 (250 g), thickener (Thixotropic additives)
(silikonfabrik.de, Germany) (2 ml), mixing bowl, mixing paddle, brushes, Formlabs
Form 3, Clear Resin (547,53 ml), Elastic (Formlabs, USA) (103,9 ml).
Production-Specific Preparation of the Model: To produce the wax model, a casting
mould needs to be created from the CAD file first. For this purpose, a block is created
around the model in CATIA and the model is removed using Boolean operations. The
mould created in CAD then has to be further divided to allow casting and wax removal.
Holes are provided for fixing the mould parts (see Fig. 9 (a)).
Fig. 9. Casting mould generated in CAD for the wax model, failed attempt with one casting
mould (a) and several moulds for the individual vessel models (b)
The first attempt to create the wax model failed because the relatively thin branches of
the aortic arch broke off when the mold was removed. The arch itself could be pro-
duced well by the mould. It was therefore decided to produce separate casting moulds
for the individual branches (see Fig. 9, (b)). The construction of the mould in CATIA
and the production is calculated with about 75 h (65 h printing time). The adapters
were printed separately in Elastic. An exemplary form is shown in Fig. 10, (a). The
individual wax patterns were casted into the pre-warmed casting moulds and cooled
down completely. Afterwards the individual wax models were melted at the interfaces
and connected to form a uniform model (see Fig. 10, (b)). The process of creating the
wax model is calculated at about 11 h.
Fig. 10. Casting mould (Clear Resin) for the creation of the wax model (a), the assembled wax
model with attached adapters/Elastic Resin) (b), application of the silicone layers (c)
Production of the Model: The silicone was mixed in a ratio of 1:1 base: catalyst and
thickened with one percent by weight thixotropic additives. The model was coated with
two additional layers at intervals of 2 h, with the third layer dispensing with the
thickener in order to produce a smoother surface (see Fig. 10, (c)). The process of
silicone application is calculated with 2 h. The model is then melted out in a water bath
(Fig. 11). The process step is calculated with 1.5 h.
Fig. 11. Melting of the wax in a water bath (a), silicone model after loss of wax with detached
adapters and wax layer outside and inside (b)
3.4 Production of the Other Aortic Model Parts

The remaining aortic model sections are divided into thoracic aortic section, abdominal
aortic section and femoral arteries. These models are made with Formlabs 2/3 and
connected with adapters with outer shells (Tough Resin, Formlabs). The complete aorta
is shown in Fig. 12. It was decided to use Elastic Resin for the rest of the aorta because
the catheter-vessel wall contact is not as high as in the aortic arch and the models can
be produced at low cost by the authors themselves.
Fig. 12. Total aortic model consisting of (a) femoral arteries, (b) abdominal aorta, (c) thoracic
aorta, (d) aortic arch
4 Evaluation of the Production Processes and Materials
4.1 Evaluation Criteria

The evaluation criteria are stored based on the production requirements and defined as
follows: The production process must not restrict the accurate reproduction of the inner
contour of the aortic arch model. The model should be made of an elastic material. The
model should have a certain degree of transparency to facilitate catheter positioning.
The production must guarantee a tightness of the model wall. The aim is to produce the
model as quickly as possible so that it can be put back into use as soon as possible after
any damage (Wish). The material should be robust so that it can be used for several
training sessions (Wish).
On the first level, this results in the criteria of cost, time and quality for the
evaluation of the models from the various production processes. Figure 13 shows the
sub-criteria for evaluating the finished models in terms of time, cost and quality.
Fig. 13. Criteria for the evaluation of manufacturing processes and the resulting models
A quantitative testing of criteria would go beyond the stress limit of the materials
(e.g. testing of elasticity). Due to the requirement to keep the models non-destructive,
which results from the high costs and the unit of 1, the criteria are mainly tested
qualitatively. The existing aortic model from United Biologics serves as a reference for
the evaluation. The sub-criteria are described in more detail below.
The production time is considered from the point of production initiation to the
finished and usable vessel model and includes the necessary preparation and follow-up
procedures. The basis for the production of the aortic arch is the completed CAD
construction, which was exported as an STL file.
For the aortic arch from contract manufacturing, the delivery time and costs are
included in the evaluation.
The production costs are the sum of the manufacturing and material costs. The
manufacturing costs include the labor costs for production. Production overheads are
calculated under assumptions.
Statements on reproducibility can be made qualitatively on the basis of production.
The elasticity is assessed by the physician on the basis of experience with real aortic
vessels and is put into practice with the existing aortic model.
An evaluation of the geometrical correctness of the models by overlaying the
scanned models with the CAD model is not carried out for two reasons. Firstly, the
geometric correctness of the aortic model is only partially relevant for the intended
intracanial treatment simulation and the CAD model was reconstructed on a patient-
based level (no patient-specific model). On the other hand, the elasticity of the model
allows a certain deviation.
The tightness of the aortic arch is tested in itself and at the junctions to the cervical
vessels and brachial vessels in the existing neurointerventional training model.
The transparency of the model is assessed during operation with the blood sur-
rogate (water and soap). For this purpose, it is assessed whether the catheter is ade-
quately visible.
The surface quality is qualitatively evaluated after production and in tests with the
physicians. Therefore the behavior of the catheter on the vessel wall is evaluated
qualitatively.
For the reasons mentioned above, a stress test is not carried out. The robustness of
the model is qualitatively assessed in the application at HANNES. For this purpose, it
is assessed to what extent the model shows material stress in the application, e.g. during
assembly at the interface and during pressurization in the system.
The quality criteria will be tested by means of qualitative testing in HANNES
together with experienced neuroradiologists. The behavior in angiography and in
interaction with the treatment devices will be tested. The focus in this study is on the
evaluation of the different production processes in terms of cost and time and a first
assessment of the qualitative criteria resulting from it. The test setup in HANNES is
planned.
4.2 Evaluation of the Production Processes

The evaluation of production time and costs is based on certain assumptions: The costs
are calculated excluding all taxes. The labor cost rate for one hour is estimated at 40 €.
The printing time on the AM printers is not included in the working time, as they can
run unattended. The starting point for the calculation is the finished STL file of the
aortic arch.
Production overheads are assigned to the manufacturing costs using the machine
hour rate [20]:
machine dependent costs

machine hour rate ¼ ð1Þ
running hours
Machine-dependent costs represent cost-accounting depreciation, accounting interest,

costs for maintenance & repair, space costs and energy costs [20].
The cost-accounting depreciation of the machine is calculated as follows [21]:
replacement value residual value

cost accounting depreciation ¼ ð2Þ
useful life
The replacement value is calculated as follows [21]:
replacement value ¼ acquisition cost ð1 þ inflationÞn ð3Þ
The acquisition cost of the Form 3 was 3299 € without taxes and shipping costs.
The average of the inflation rates for the years 2015 to 2019 was used as the calculated
inflation rate [22], resulting in a value of 1.14%. The residual value and the useful life
of the printer were estimated. It is assumed that the printer has a useful life of 5 years
and a residual value of 500 €. This results in the replacement value of Form 3 at
3491.38 € and the cost-accounting depreciation at 598.276 € per year.
The accounting interest and space costs are not included in the calculation. The
repair costs are estimated at 100 € per year.
The energy costs are calculated on the basis of the energy requirement of 220 W of
the Form 3 [16], the machine running time and an electricity price of 0.29 € per kWh.
The machine running time results from the assumption that the printer is used two
days a week with an average printing time of 6 h. With 230 working days per calendar
year, the machine running time is 552 h/year.
These assumptions result in an annual electricity price of 40 € for the Form 3 [23].
The calculation of the machine hour rate is shown in Table 1. Based on the assump-
tions made, this results in a machine hour rate of 1.52 h/€.
Table 1. Calculation of the machine hour rate of the Formlab Form 3 (based on assumptions)
Machine hours per year: 552 h
Cost-accounting depreciation 698.28 €
Accounting interest −€
Maintenance and repair cost 100 €
Space cost –
Energy cost 40 €
Summe 838.28 €
Machine hour rate 1.52 €/h
Evaluation of the Current Aorta Model of the Company United Biologics. The
purchase price of the entire Aorta model was in the four-figure € range at that point in
time. The model was characterized by its good and constant transparency and high
robustness. The elasticity appears good. The connections to the HANNES model
cannot be made immediately. The friction of the catheter on the model vessel wall is
also considered by the physicians to be too high. It is not possible to change the aortic
arch type.
Evaluation of the SLA Print in Elastic Resin. Production time: The process of
preparing the model for the printer consisted of the creation of the PreForm file with
1 h and the machine preparation with about 0.5 h. The printing time of the aortic arch
was 30 h. For the post-processing of the model 1 h was needed.
Production Costs: The material costs are calculated on the basis of the material con-
sumption calculated in PreForm and the cost of a tank of Elastic Resin (1 l). This
results in material costs of 39.76 € for the aortic arch. The labor costs for a total
working time of 2.5 h at an hourly rate of 40 € results in 100 €.
With a machine hour rate of 1.52 €/h, the production overheads for printing the
aortic arch are 1.52 €/h 30 h = 45.6 €. The production costs are calculated in total at
about 186 € (exclusive taxes).
The production of the model is subject to high reproducibility due to the settings on
the printer. The first impression of the model in terms of elasticity, geometrical cor-
rectness and tightness appears good. At first the model shows a high transparency,
which however decreases over time and appears rather milky. The inner surfaces of the
models are initially sticky after post-treatment, but this also subsides over time. In
terms of robustness, the models appear to be relatively sensitive.
Evaluation of the Heartprint Flex Model by Materialise. Production and delivery
time: With the aortic arch model manufactured by Materialise, it is not possible to
divide production and delivery time. The total time is the period from order confir-
mation by Materialise until delivery of the aortic arch model. This results in duration of
14 days.
Purchase Price: The purchase price includes the total cost price of the product, which
cannot be further broken down. The purchase price without taxes and transport is in the
upper three-figure € range.
The production of the model is subject to high reproducibility due to the settings on
the printer. The first impression of the model in terms of elasticity, geometrical cor-
rectness and tightness appears good. The model initially shows good transparency and
seems to maintain this over time. The inner surfaces seem to be smooth and the whole
model robust.
Evaluation of Silicon Casting. Production time: Since moulds have to be developed,
the design time for the initial production is taken into account (95.5 h).
Production Costs: For the calculation of the labor costs, the construction of the casting
moulds, the production of the wax pattern, the application of the silicone and the
melting of the wax are considered. This results in labor costs of 980 € at a calculated
40 €/h. The material costs include the proportionate costs for Clear Resin (mould),
paraffin wax, 2-component silicone, thickener and Elastic Resin (adapter). This results
in material costs of about 105 €. The production costs are calculated on the basis of the
machine hour rate of Form 3. With a printing time of 65 h for the mould and 4 h for the
adapters, the manufacturing overheads are approximately 105 €. In total the production
costs amount to approximate 1190 €.
The reproducibility of the model is estimated to be low, as no reproducibility can be
guaranteed, especially by manual application of the silicone. The first impression of the
model in terms of elasticity and geometrical correctness appears good. The tightness of
the model is not ensured due to many defects caused by an irregular wall thickness.
There is no transparency of the model due to wax residues inside the model. The inner
surface of the model has adopted the structure of the wax core and is therefore rough. In
terms of robustness, the model appears to be very sensitive, especially due to the
insufficient wall thickness.
4.3 Test Set-Up for the Evaluation of Quality Criteria

For a further evaluation of the qualitative criteria, the aortic arch models must be
connected in HANNES and tested with the experienced physicians at the UKE. For this
purpose, the connections must be designed for the new aortic arch model. Figure 14
shows a first test for the geometric requirements of the aortic arch model.
Questionnaires were created to evaluate the criteria, which should first give an
assessment of the current aorta model. Based on this, the criteria are to be evaluated in
real terms in relation to the current aortic model. The focus of the evaluation is on the
aortic arch model.
Fig. 14. Geometric comparison of the aorta model from United Biologics (above) with the aorta
model made of Elastic Resin, manufactured on Formlabs Form 3 (below). (a) Comparison of the
aortic arch with outlets, (b) comparison of the remaining aorta
5 Discussion of the Results and Outlook
This study described the design of an interchangeable aortic arch and its fabrication
using the SLA fabrication procedure with Formlabs Form 3, the contract fabrication by
Materialise with HeartPrint Flex and the fabrication of a silicone model by applying it
to a cast wax model. The evaluation of the different manufacturing processes was
quantitatively based on production costs and time. An initial assessment was given with
regard to a qualitative evaluation of the models resulting from the processes.
Some disadvantages have occurred when fabricating the silicone model using a
wax model. The production of the casting mould based on the positive model is very
time-consuming. The wax model must cure for several hours. To create the smoothest
possible surface inside the silicone model, the wax model must be finished. Between
the applications of the different layers, the silicone must be cross-linked for at least 2 h,
which prolongs the whole process. When applying the silicone to the wax model it is
difficult to create a constant wall thickness. In addition, the wax could not be com-
pletely removed from the silicone model when melting it in a water bath, the surfaces
were covered with wax and no good surface properties or transparency was produced.
The process is far more expensive than direct printing and the reproducibility is low.
Direct production with Formlabs in Elastic Resin has several advantages. It is
particularly convincing due to its short production time and low production costs in
comparison to the procedures compared. The good availability of the printing process
at both project partners (TUHH and UKE) plays an important role. In addition to these
criteria, the model scores well in terms of the qualitative criteria in the first estimation.
The HeartPrint model seems to be the best in terms of quality, although it is much more
expensive than the manufacturing process in Elastic Resin. A comparison of the two
models in HANNES with the medical professionals is necessary to make a final
selection.
Further work consists in the integration of other production processes and materials
in the evaluation. Based on the currently purchased aorta, it could be observed that
silicone is very well suited for long-term use, while e.g. the Elastic Resin shows
material changes in the long run. Due to the complex manufacturing process of the
silicon casting, a silicone print is to be included in order to be able to evaluate the
materials silicone, Elastic and HeartPrint in comparison and in realization to the current
aorta model of the company United Biologics. The models are to be tested qualitatively
regarding their suitability in the training model HANNES with the experienced
physicians of the UKE.
Acknowledgment. The authors would like to thank the German Federal Ministry of Education
and Research for founding this work within the project COSY-SMILE (031L0154A).
References
1. Mehra, M., Henninger, N., Hirsch, J.A., Chueh, J., Wakhloo, A.K., Gounis, M.J.: Preclinical
acute ischemic stroke modeling. J. Neurointerventional Surg. 4(4), 307–313 (2012)
2. Gralla, J., Schroth, G., Remonda, L., Fleischmann, A., Fandino, J., Slotboom, J., Brekenfeld,
C.: A Dedicated animal model for mechanical thrombectomy in acute stroke. AJNR Am.
J. Neuroradiol. 27, 1357–1361 (2006)
3. Leischner, H., Flottmann, F., Hanning, U., Broocks, G., Faizy, T.D., Deb-Chatterji, M.,
et al.: Reasons for failed endovascular recanalization attempts in stroke patients.
J. Neurointerventional Surg. 11(5), 439–442 (2019)
4. Vascular Simulations. https://vascularsimulations.com/endovascular-simulators/. Accessed
29 Jan 2020
5. FAIN-Biomedical. http://fain-biomedical.com/. Accessed 29 Jan 2020
6. Spallek, J., Kuhl, J., Wortmann, N., Buhk, J.-H., Frölich, A.M., Nawka, M.T., Kyselyova,
A., Fiehler, J., Krause, D.: Design for mass adaptation of the neurointerventional training
model HANNES with patient-specific aneurysm models. In: Proceedings of the 22nd
International Conference on Engineering Design (ICED19), pp. 897–906, Delft (2019).
https://doi.org/10.1017/dsi.2019.94
7. Spallek, J., Frölich, A., Buhk, J.-H., Fiehler, J., Krause, D.: Comparing technologies of
additive manufacturing for the development of vascular models. In: Fraunhofer Direct
Digital Manufacturing Conference, Berlin (2016)
8. Wortmann, N., Spallek, J., Kyselyova, A.A., Frölich, A.M., Fiehler, J., Krause, D.: Concept
of an in-vitro model for endovascular stroke treatment using additive manufacturing. In:
Buzug, T.M., Seitz, H. (eds.), Additive Manufacturing Meets Medicine (AMMM) 2019, vol.
1, pp. 101–102. Infinite Science Publishing, Luebeck (2019). https://doi.org/10.18416/
ammm.2019.1909s04t03
9. Mahnken, A.H.: 5 Große Gefäße. In: Krombach, G.A., Mahnken, A.H. (eds.) Radiologische
Diagnostik Abdomen und Thorax: Bildinterpretation unter Berücksichtigung anatom.
Landmarken u. klin. Symptome. Georg Thieme Verlag, Stuttgart (2015)
10. Schünke, M., Schulte, E., Schumacher, U.: Prometheus. LernAtlas der Anatomie. Bd. Innere
Organe, vol. 2, Thieme, Stuttgart (2009)
11. Ladicha, E., Butanyb, J., Virmania, R.: Aneurysms of the aorta: ascending, thoracic and
abdominal and their management. In: Buja, L.M., Butany, J. (eds.) Cardiovascular Pathology
(Fourth Edition), pp. 169–211 (2016)
12. Liapis, C.D., Avgerinos, E.D., Chatziioannou, A.: The aortic arch: markers, imaging, and
procedure planning for carotid intervention. In: VDM Vascular Disease Management, vol. 6,
p. 1 (2019)
13. Lin, L.-M., Colby, G.P., Jiang, B., Uwandu, C., Huang, J., Tamargo, R.J., Coon, A.L.:
Classification of cavernous internal carotid artery tortuosity: a predictor of procedural
complexity in pipeline embolization. J. Neurointerventional Surg. 7, 628–633 (2015)
14. United Biologics. https://unitedbiologics.com/product/neuro-testing-model/. Accessed 29

Jan 2020
15. Baeck, K., Lopes, P., Verschueren, P.: Material Characterization of HeartPrint® Models and
Comparison with Arterial Tissue Properties. Materialise, Leuven (2020)
16. Formlabs. https://formlabs.com/de. Accessed 17 Feb 2020
17. Heidemanns, S.: Konzeption, entwicklung und evalution eines kostengünstigen reproduzier-
baren Gefäßmodells für die simulation und das training endovaskulärer interventioneller
Prozeduren an der Aorta anhand anatomischer Vorlagen eines realen Patienten. Dissertation,
Ludwig-Maximilians-University München, Faculty of Medicine, München (2015)
18. Marconi, S., Lanzarone, E., van Bogerijen, G.H.W., Conti, M., Secchi, F., Trimarchi, S.,
Auricchio, F.: A compliant aortic model for in vitro simulations: design and manufacturing
process. Med. Eng. Phys. 59, 21–29 (2018)
19. Schickel, M., Townsend, K., Farotto, D.: Material Characterization of 3D Printed HeartPrint
Flex Plus Models and Comparison with Arterial Tissue Properties. Materialise NV, Leuven
(2019)
20. Kaesler, C.: Kosten- und Leistungsrechnung der Bilanzbuchhalter IHK. Mit Übungsklau-
suren für die Abschlussprüfung, 6th edn., Springer Fachmedien Wiesbaden GmbH (2018).
https://doi.org/10.1007/978-3-658-18305-9
21. reimus.NET GmbH. https://www.controllingportal.de/Fachinfo/Kostenrechnung/kalkulatori-
sche-Abschreibungen.html. Accessed 20 May 2020
22. finanz-tools.de. https://www.finanz-tools.de/inflation/inflationsraten-deutschland. Accessed
20 May 2020
23. qmedia GmbH. https://www.rechner.pro/stromkostenrechner/. Accessed 20 May 2020
Integration of Additive Manufacturing
into Process Chain of Porcelain Preservation
Bingjian Liu1(&), Fangjin Zhang2, Xu Sun1, and Adam Rushworth1

1
University of Nottingham, Ningbo, China
Bingjian.liu@nottingham.edu.cn
2
Design School, Loughborough University, Loughborough, UK
Abstract. Relic restoration and preservation is a huge market. Antique Chinese

Porcelain, which is regarded as a symbolic type of artefacts representing Chi-
nese art, craft and culture, has attracted significant study into its preservation,
crossing the fields of policy-making, science and emerging technologies. In
recent years, Additive Manufacturing (AM) has demonstrated its advantages in
the restoration of relics made of a variety of materials. However, the study of its
implementation with antique porcelain, which not only has different shapes, but
also glossy surface and rich colours, still remains in its infancy.
This paper presents the case studies into the application of AM of antique
porcelain preservations and creation. The studies include one practice convert-
ing the image of a vase on a Chinese painting to a 3D object with AM and
another practice that applied AM to replicate an antique Famille-rose porcelain
piece featuring rich colour in order to restore the missing pieces on a window in
the Palace Museum, Beijing. The aim is to use AM polymer to simulate the
visual features of antique porcelain. It was found that with the proper set-up of
parameters and integration of other technologies and skills, AM with polymer
materials also can support the replication of the features of antique porcelain to a
significant extent. However, the glossy surface of porcelain made it difficult to
acquire the surface details. In addition, the rich colours of Famille-rose porcelain
were not only presented challenges with regard to directly obtain its colour
features through 3D scanning, but also limited the application of AM owing to
the limited colour series of 3D printing materials. In this paper, integrated
methods were proposed and tested to address the above challenges which could
impact on the application of AM in imitating porcelain features while attempting
to contribute to its preservation and creation.
Keywords: Additive manufacturing Antique porcelain Integrated

Archaeological preservation
1 Introduction
As the origin of the English name of the country, porcelain reserves an irreplaceable
position in Chinese cultural heritage and played top roles in connecting China and
Europe through trade [1]. However, due to the natural calamities and man-made mis-
fortunes, numerous pieces of antique porcelain work were missing or damaged. On the

https://doi.org/10.1007/978-3-030-54334-1_32
456 B. Liu et al.
other hand, there is a big shortage of the number of craftsmen who are able to restore or
duplicate the porcelain with the conventional manual craftsmanship.
Additive Manufacturing (AM) is a well-acclaimed emerging and disruptive tech-
nology [2, 3]. Thanks to its advantages in producing intricate and customized parts, the
application of AM in heritage preservation has attracted continuous research attentions.
This can be proved by the ample number of publications in the application of AM to
the restoration/reproduction of different types of historical artefacts [4–6]. In these
studies, the application of digital fabrication has been investigated for the restoration of
a variety of types of antiques such as wood furniture, bronze and stone sculpture,
enamel, and so on. In summary, AM has shown its benefits in these applications in
several key aspects, such as:
1. AM is able to produce objects with intricate geometry [7]. This is an important
feature since most antiques were made by crafting and thus have diverse shapes that
are usually difficult to be reproduced directly with traditional manufacturing
processes.
2. As a type of digital fabrication, AM has the strength in its precision compared to
manual work. In addition, the digital model of the AM is more editable than the
manual work handling the real materials in the physical world, which can further
save cost and time.
Although AM has already shown its advantages in the aforementioned applications,
it is found that the related work on one important relic material, porcelain ware, is still
in its infancy. Within the limited publications, it is realized that some desired
requirements are difficult to meet if AM is used alone, which indicates that the
integration/hybrid way that involves AM and other emerging or traditional technolo-
gies could be a valuable approach to problem-solving in such practices.
This paper will review the related work that could influence the application of AM
in these practices, then introduce and analyze two relevant projects completed by the
research team and conclude with discussions and suggestions for the future research.
2 Related Work
In this section, the Chinese porcelain is briefly reviewed to understand its key features
that should be reproduced through additive manufacturing; the relevant technologies
including colour 3D printing and hybrid technologies are also investigated to clarify the
advantages and disadvantages of AM.
2.1 Chinese Antique Porcelain

Porcelain is a type of translucent ceramic material made by heating, generally including
kaolin (a fine white clay), in a kiln to high temperatures between 1,200 and 1,400 °C.
The materials and the heating process give porcelain the glossy surface.
Chinese porcelain has a long history of around 2,000 years and became well known
in Europe through trade from the Ming Dynasty (AD 1368-1644) onwards [8]. By then
Integration of Additive Manufacturing 457
there were two types of porcelain-ware popular in the market: 1, the well-known white
and blue porcelain; 2, single colour porcelain ware, such as, white, black, or celadon.
From Qing Dynasty, another type of Chinese antique porcelain, Famille-rose,
became popular as well. Different from other types of porcelains, Famille-rose is
distinctive for its great range of colours. In addition, different from underglaze blue-
and-white porcelain, which has smooth surface, Famille-rose porcelain’s colours are
overglazed and the colour pigments cause textures on the surface after heating.
The application of AM to porcelain can be for preservation/restoration, historical
and educational purposes. Currently, the study of this type of application still remains
limited. One possible reason is: as a type of historical relic, porcelain-ware are not as
popular in the rest of the world as in China, and hence have not attracted global
attention. For instance, within the limited relevant publications, the project completed
by Miller 3D in the United States was able to create the replica of a Chinese artefact, a
blue and white antique Chinese porcelain vessel from Qing Dynasty [9].
2.2 Colour Additive Manufacturing

Currently, Additive Manufacturing is still in the research era of monochrome with a
focus on materials [10–13], printing precision [14, 15] and speed [16]. As an emerging
and advanced technology of prototyping and manufacturing, AM has been applied in
multiple fields [2, 17], particularly with its strength introducing complex components
[7]. However, in some applications, other important attributes have challenged the
development of this technology. For instance, to simulate the rich colours as well as the
glossy and translucent surface of Chinese antique porcelain wares, it is necessary to
investigate how to add colour and required surface effect to the 3D printed objects.
Basically, there are two ways to have colours on 3D printed objects, one is direct 3D
printing with colour materials and another is to add colours to the surface of the print
through post-processing.
Direct colour 3D printing here refers to that the printing substrates have colour
property so the printed objects can have two or more colours [3, 16]. Compared to
monochrome, colour 3D printing has its advantages in some applications, such as
customized decorative accessories and teaching models [18], which make it attract
many investments to explore the methods for colour 3D printing [2, 14]. Among these
explorations, the researchers try to produce colour printing through different perspec-
tives, for instances, colour 3D printing based on UV ink technology [19]; single and
multiple nozzle colour 3D printing [2] (Yang et al. 2018); new algorithms [20] or even
the integration of Artificial Intelligence (AI) into colour 3d printing technology [21].
Although the research into the direct colour 3D printing has become quite popular,
there are still lots of limitations to apply this technology to practice. Firstly, the variety
of colours that be 3D printed are very limited and hard to print more than two colours
[2] and usually have poor colour feature quality [22, 6]. Secondly, when compared to
2D printing, there is a lack of standards related to the evaluation of the colour quality of
3D printing [3]. The last but might be the most important, despite vast academic
research into different colour 3D printing methods, only a few are commercialized to
market and can be used in practice, such as Connex 3 from Stratasys [23], Zprinter850
from 3Dsystem [24] and Mcor IRIS from Mcor Company [25]. Therefore, in many
458 B. Liu et al.
practices, it is hard to use direct colour 3D printing to obtain the required colour
properties in terms of colour variety and quality.
2.3 Ceramic Additive Manufacturing

The study of AM on the material of ceramics has started since 1990’s [26, 27].
Attempts to apply AM of ceramic have been made in various areas, such as the
construction industry [28], prothesis [29] and fine art [30]. However, compared to state-
of-the-art polymer and increasingly developed metal, the application of AM ceramics is
much more limited [31].
Although ceramics are an ideal material for use in the production of porcelain, the
high cost of the technology and the brittleness (which is the same weakness as antique
porcelain) of materials [31, 32] limit its application. In addition, porcelain is the
combination of ceramic and glaze. Therefore, the AM of ceramic cannot solely create
the features of porcelains and conventional post processing is still needed and afore-
mentioned problems still remain. Therefore, given the advantages in accessibility, cost
performance and mechanical toughness (not as brittle as ceramics) of polymer AM
[33], it is worthwhile to investigate the application of polymer in simulating the fea-
tures of porcelain.
2.4 Hybrid Technology with Additive Manufacturing

Like most technologies, Additive Manufacturing is not all-powerful but unavoidably
has both strengths and limitations. To obtain optimal results in applications, attempts
have been made to create hybrid technologies that combine the advantages of AM and
other technologies to solve practical problems, e.g. CNC with AM or 2D with 3D
printing [34–38].
These studies acknowledged the disadvantageous side of the Additive Manufac-
turing and showed the necessity of the hybrid approach in solving practical problems.
Although these findings from these studies cannot be directly deployed in this study, it
provides an inspiring methodology that in the application of AM, integration with other
technologies can be proactively considered to explore a better solution than using AM
alone.
3 Practices
To investigate the integration method on Chinese porcelain, two projects collaborated

with museums were conducted. In both projects, AM was integrated into process chain
that include AM, 3D scanning, 2D printing and Manual work, in order to complete the
tasks in a proper manner of time, cost and quality.
3.1 3D Creation from 2D Chinese Painting

The first project is 2D-to-3D creation to produce a physical model for exhibition
purpose, based on the Emperor Qianlong’s painting: ‘Shi Yi Shi Er Image’ in Qing
Dynasty (Fig. 1). To understand the size, colour and material of the objects on the
painting, especially the porcelain vase, comprehensive literature reviews were con-
ducted and experts were consulted. It is believed that, the vase on the painting is a type
of ‘Ru Yao’ porcelain which is from the Song Dynasty with blue colour.
Regarding the technologies for the project, although direct coloured ceramic or
plaster 3D printing is possible, several constraints were discussed with stakeholders,
such as the fact that texture base on powdered printing is not good enough, the choice
of colour 3D printers is limited and needs unexpected experiments to adjust, technical
difficulties of finishing and the overall cost and time scale. Given these limitations, the
integrative method with other technologies was decided as the best compromised
solution for this project at this current stage.
The Integrative process was adopted using forward engineering - digital modelling
to produce a virtual vase, reverse engineering - 3D scanning data of real lotus followed
by data manipulation for the lotus, after the virtual testing, high resolution AM pho-
topolymer physical model with traditional finishing techniques, such as manual
assembly and refining with modelling knife, to create the physical model of the vase, as
shown in Fig. 2. After that, the colour and gloss were added with hand painting.
Fig. 1. Chinese painting from the Qing Fig. 2. The porcelain and lotus produced
Dynasty through AM
460 B. Liu et al.
Fig. 3. The finished creation of the 3D prints for exhibition
The final model (Fig. 3) was accepted in the ‘Painting and Calligraphy Exhibition
of Emperor Qianlong’s Meeting with Ministers’ after it passed the experts group
evaluation, displayed as the first exhibits at the entrance to welcome up to 80,000
visitors daily views for three months.
3.2 Replication of the Antique Famille-Rose Porcelain Piece (More

Detailed Process)
Palace Museum is one of the biggest and most famous imperial palaces around the
world and also has the biggest collection of Chinese historical relics. The second
project is to replicate a missing Famille-rose porcelain piece, which used to be a
decorative part on the window of the museum.
As aforementioned, the features of Famille-rose porcelain cause several challenges
to the application of AM, which can be summarized as follows:
1. The glossy glaze layer of the porcelain makes it difficult to conduct 3D scanning to
acquire surface geometry. In addition, the use of contact 3D scanning and contrast
intensifying agent is prohibited on the precious royal porcelain relic.
2. The painting on the Famille-rose has very rich colour including gradient change
colours. In addition, different from the modern artificial paint, the painting on the
antique piece was made with natural mineral dye, which gives Famille-rose 3D
embossment-like texture.
With similar considerations to the first project, the project team decided to use the
integration approach that, adopt AM to build the monochrome shape and 2D printing to
create the colour painting.
The 3D data of the piece was collected through a non-contact 3D scanner and then
printed with Stratasys objet24 with rigid white opaque material (VeroWhitePlus) as
printing materials. The colour of the material is suitable since the white is suitable as
base to add other colours later on. To get the best texture quality, the model was placed
in four orientations to compare the results, as shown in Fig. 4. It was found that the
prints from No. 1 and No. 4 can show the texture more clearly than No. 2 and No. 3,
mainly due to the staircase effect.
Fig. 4. 3D printing set-ups and printed samples with VeroWhitePlus
Figure 5 shows the acquisition process of the porcelain piece 2D image. From left
to right, first of all, the photograph needs to be taken in a professional environment with
proper set-up on lighting condition and DSLR camera to get high-resolution picture.
After further editing, the picture was printed on the AM piece surface with printer UJF-
3042HG which can provide over-coating on transparent and coloured materials as well
as glossy finish, which is exactly needed in replicating Famille-rose porcelain. UJF-
3042HG is a type of 2D printer produced by the company Mimaki. There are no official
specifications from the product booklet stating that it can print on surface with uneven
textures and depth. However, through experiments in this study and later expert
evaluation, it has acceptable performance in printing pictures on uneven surfaces and
matches the texture pattern. Further study is needed to investigate the maximum depth
of texture the printer can work on. Figure 6 shows the results of the project at the
moment.
Fig. 5. The acquisition of high-resolution image of the porcelain piece

462 B. Liu et al.
Fig. 6. Final result after the combination of 2D and 3D printing process
The results were then presented to the staff the from Palace Museum to evaluate
and give feedbacks on the quality. There were five people invited including three experts
from Architecture, Technology, and Exhibition Department respectively and two
non-specialists, from Sales Department and Administration. The evaluation is presented
in Table 1. The numbers refer to the number of participants in that category. The data
showed that in general the reviewers gave positive feedbacks to the results. Overall
speaking, non-experts gave higher scores than experts who have higher expectations on
some specific parameters, such as the colour and texture quality. However, both experts
and non-experts agreed that the results can be beneficial for temporary exhibition, tourism
and product design, with comments, such as: ‘Overall, the range of experiments and
efficiency are impressive, digital technologies are very useful for relic preservation’
Table 1. The feedbacks from the participants on the quality of the prints (the numbers refer to
the number of participants in that category)
Quality parameter Excellent Good Acceptable Poor
Participants (E: expert; N: Non-expert)
E N E N E N E N
Measurements 3 2
Visual effects 1 3 1
Resolution 2 2 1
Colour 2 2 1
Texture 2 1 2
Surface roughness 1 2 2
4 Discussions
From the application view of AM, the case studies explored integration approaches to
create Chinese antique porcelains. These approaches set additive manufacturing as a
core technology and the materials, machines and printing orientation should be prop-
erly chosen and tested. In addition, a process chain that flexibly used pre- or post-
processing methods, such as 2D and 3D image acquisition, manual work, 2D and 3D
printing, will be employed to complete the task and meet the customer’s requirement.
The feedback from experts and non-specialists showed the advantages of the integra-
tion in terms of flexibility, quality, cost and time. However, although exploratory in
nature, the studies still possess many limitations.
Due to the constraints on budget and time, the projects did not systematically test
the quality of direct colour 3D printing to reproduce the porcelains and make com-
prehensive comparison with the proposed hybrid methods. Therefore, these studies
only proved the effectiveness of the methods on the two projects, but cannot indicate
they have advantages over direct 3D printing on all dimensions.
Chinese antique porcelain has a great diversity in terms of age, shape, colour,
texture and surface features. The presented practices indicate that more work needs to
be done to simultaneously solve various visual and geometrical attributes when using
Additive Manufacturing. In the future, it is expected that, through more practices in this
area, some guidelines can be generated which will provide a positive impact on the
development of Additive Manufacturing.
5 Conclusions
The exploration of application is essential to the sustainable development of Additive

Manufacturing. Although the research on aspects such as materials, precision and speed
are fundamentally important to the technology, critical thinking is also needed when
considering the research investment. For instance, in some applications, the current
precision and printing speed have well met the requirement; behind the vast researches
on colour 3D printing, only a few color 3D printers are successfully commercialized.
Continuous exploration of the application of AM will open more markets to this
technology and could inspire research directions for the above fundamental aspects. For
example, the industry can develop AM materials with particular glossy effect to meet
customer’s requirement on the surface.
In recent years, apart from industrial manufacturing, Additive Manufacturing has
been applied into different fields, such as medical science, education, fine art and
culture heritage preservation. The practices in these areas showed that, to complete the
tasks more efficiently and effectively, more in-depth studies into the integration
approach are needed to combine the advantages of Additive Manufacturing and other
methods or technologies.
The study presented in the paper indicated that Additive manufacturing is an
effective means to bridge the ‘old’ and ‘new’, to work on one of the oldest objects with
the most modern technology. In addition, it also proved that polymer, which is the most
popular and state of the art material of AM, can be used to simulate other material, such
as porcelain.
464 B. Liu et al.
The study presented in the paper also indicates that some applications may not have
global significance but under specific cultural background, Additive Manufacturing still
can play an important role, e.g. in the cultural heritage of Chinese porcelain in this
study. In the field of cultural heritage preservation, different countries could have
completely different historical relics to be preserved that have different materials and
features, therefore, the integration of local resources, such as craftsmanship and other
technologies, into Additive Manufacturing could contribute to the problem-solving.
In addition, the successful application of AM in this area will not just benefit the
archiving and repairing, but also can develop a new medium for the creativity in
product design, for instance, the derivative souvenir products, which will create eco-
nomic benefits to financially support heritage preservation and the development of AM
in turn.
References
1. Wang, G.: Chinese porcelain in the manila galleon trade. In: Archaeology of Manila Galleon
Seaports and Early Maritime Globalization (2019)
2. Yang, M.H., et al.: Research on colour 3D printing based on colour adherence. Rapid
Prototyp. J. 24(1), 37–45 (2018)
3. Yuan, J., et al.: Review on processes and colour quality evaluation of colour 3D printing.
Rapid Prototyp. J. 24(2), 409–415 (2018)
4. Zhang, F., et al.: Application of additive manufacturing to the digital restoration of
archaeological artifacts. Procedia Technol. 20, 249–257 (2015)
5. Thompson, M.K., et al.: Design for additive manufacturing: trends, opportunities,
considerations, and constraints. CIRP Ann. 65(2), 737–760 (2016)
6. Scopigno, R., et al.: Digital fabrication techniques for cultural heritage: a survey. Comput.
Graphics Forum 36(1), 6–21 (2017)
7. Fullerton, J.N., Frodsham, G.C., Day, R.M.: 3D printing for the may not the few. Nat.
Biotechnol. 32(11), 1086–1087 (2014)
8. Chinese porcelain history. https://www.chinahighlights.com/travelguide/culture/porcelain-
history.htm. Accessed 13 Feb 2020
9. Miller 3D Replicates Smithsonian Artifacts with 3D Printing. https://miller3dprinting.com/
3d-printing-replication-smithsonian-artifacts/. Accessed 13 Feb 2020
10. Gibson, I., Shi, D.: Material properties and fabrication parameters in selective laser sintering
process. Rapid Prototyp. J. 3(4), 129–136 (1997)
11. Ivanova, O., Williams, C., Campbell, T.: Additive manufacturing (AM) and nanotechnol-
ogy: promises and challenges. Rapid Prototyp. J. 19(5), 353–364 (2013)
12. Schubert, C., van Langeveld, M.C., Donoso, L.A.: Innovations in 3D printing: a 3D
overview from optics to organs. Br. J. Ophthalmol. 98(2), 159–161 (2014)
13. Roberson, D.A., et al.: Fracture surface analysis of 3D-printed tensile specimens of novel
ABS-based materials. Rapid Prototyp. J. 14(3), 343–353 (2014)
14. Hergel, J., Lefebvre, S.: Clean colour: improving multi-filament 3D prints. Computer
Graphics Forum. 33(2), 469–478 (2014)
15. Melenka, G.W., Eujin Pei, D., Schofield, J.S., Dawson, M.R., Carey, J.P.: Evaluation of
dimensional accuracy and material properties of the MakerBot 3D desktop printer. Rapid
Prototyp. J. 21(5), 618–627 (2015)
16. Chen, G., Chen, C., Yu, Z., Yin, H., He, L., Yuan, J.: Color 3D printing theory, method, and
application. New Trends in 3D Printing (2016)
17. Mertz, L.: New world of 3-D printing offers completely new ways of thinking. IEEE Pulse 4
(6), 12–14 (2013)
18. Shi, Y.S., et al.: The development of 3D printing technology and its software implemen-
tation. Sci. Sinica 45(2), 197–203 (2015)
19. He, L.X., Chen, G.X.: 3D Colour printing based on UV ink-jet technology. Packaging Eng.
(2013)
20. Brunton, A., et al.: Pushing the limits of 3D colour printing: error diffusion with translucent
materials. ACM Trans. Graphics, 35(1) (2016). Article 4
21. Chen, C., Chen, G.X., Yu, Z.H., Wang, Z.H.: A new method for reproducing oil paintings
based on 3D printing. Appl. Mech. Mater. 644–650, 2386–2389 (2014)
22. Balletti, C., Ballarin, M., Guerra, F.: 3D printing: State of the art and future perspectives.
J. Cult. Heritage 26, 172–182 (2017)
23. Stratasys, https://www.stratasys.com/resources/search/white-papers/connex3. Accessed 9
Feb 2020
24. Dsystems. http://infocenter.3dsystems.com/product-library/cjp/z-printer/zprinter-850.
Accessed 9 Feb 2020
25. Mcor company. https://cleangreen3d.com/#hp-2. Accessed 9 Feb 2020
26. Lakshminarayan, U., Ogrydiziak, S., Marcus, H.L.: Selective laser sintering of ceramic
materials. In: Proceedings of Solid Free-Form Symposium, pp. 16–26 (1990)
27. Lauder, A., Cima, M.J., Sachs, E., Fan, T.: Three-dimensional printing: surface finishing and
microstructure of rapid prototyped components. Materials Research Society Symposium
Proceedings, vol. 249, pp. 331–336 (1992)
28. Bhardwaj, A., Kalantar, N., Molina, E., Zou, N., Pei, Z. Extrusion-based 3D printing of
porcelain: feasible regions. In: ASME 2019 14th International Manufacturing Science and
Engineering Conference (2019)
29. Buj-Corral, I., Petit-Rojo, O., Bagheri, A., Minguella-Canela, J.: Modelling of porosity of
3D printed ceramic prostheses with grid structure. Procedia Manuf. 13, 770–777 (2017)
30. Huson, D., Hoskins, S.: 3D printed ceramics for tableware, artists/designers and specialist
applications. Key Eng. Mater. 608, 351–357 (2014)
31. Scheithauer, U., Slawik, T., Schwarzer, E., Richter, H.-J., Moritz, T., Michaelis, A.: Additive
manufacturing of metal-ceramic-composites by thermoplastic 3D-printing (3DTP).
J. Ceramic Sci. Technol. 06(2), 125–132 (2015)
32. Chen, Z., Li, Z., Li, J., Liu, C., Lao, C., Fu, Y., Liu, C., Li, Y., Wang, P., He, Yi.: 3D
printing of ceramics: a review. J. Eur. Ceramic Soc. 39, 661–687 (2019)
33. Cha, C. Polymer meets ceramic: polymer-driven advancement of ceramic 3D printing
technology. Ceramist 23(1), 4–15 (2020). https://doi.org/10.31613/ceramist.2020.23.1.02
34. Pandey, P.M., Reddy, N.V., Dhande, S.G.: Improvement of surface finish by staircase
machining in fused deposition modeling. J. Mater. Process. Technol. 132(1–3), 323–331
(2003)
35. Townsend, V., Urbanic, R.: A systems approach to hybrid design: fused deposition modeling
and CNC machining. In: Bernard, A. (ed.) Global Product Development: Proceedings of the
20th CIRP Design Conference, Ecole Centrale de Nantes, Nantes, France, 19th–21st April
2010, pp. 711–720. Springer, Berlin (2011)
36. Kabaldin, Y.G., Kolchin, P.V., Shatagin, D.A., et al.: Digital twin for 3D printing on CNC
machines. Russ. Engin. Res. 39, 848–851 (2019)
466 B. Liu et al.
37. Müller, M., Wings, E.: An architecture for hybrid manufacturing combining 3D printing and
CNC machining. Int. J. Manuf. Eng. 2016 (2019), http://dx.doi.org/10.1155/2016/8609108
Article ID 8609108. Hindawi Publishing Corporation
38. Seleznev, V.A., Prinz, V.Y.: Hybrid 3D-2D printing for bone scaffolds fabrication.
Nanotechnology 28, 064004 (2017)
Business Cases
Decision Support System for a Metal Additive
Manufacturing Process Chain Design
for the Automotive Industry
Markus Johannes Kratzer1,2(&), Julian Mayer1,3, Florian Höfler1,

and Nikolaus Urban3
1
BMW Group, 80788 Munich, Germany
markus.kratzer@bmw.de
2
Institute of Industrial Manufacturing and Management, University of Stuttgart,
70569 Stuttgart, Germany
3
Institute of Factory Automation and Production Systems, FAU Erlangen-
Nuremberg, 90429 Nuremberg, Germany
Abstract. Additive Manufacturing (AM) is becoming increasingly important in

various industries, particularly due to its freedom of design, functional inte-
gration and faster product development cycles. In many applications, additively
manufactured components cannot be used directly after the printing process, but
require subsequent process steps such as heat and surface treatment. A large
number of process alternatives are available both for the execution of the
printing process and for the subsequent steps. Hence, there are numerous pos-
sible combinations for the design of the entire production process chain.
In order to simplify process selection in the area of AM, approaches and
systems for decision support have already been developed in research. However,
they do not consider the entire process chain including post-processing. This
extended perspective is necessary to make full use of the technical potential of
AM components and to optimize production with regard to economic criteria.
Furthermore, automotive specifics regarding selection criteria and an underlying
database of materials and processes are not taken into account in most cases.
Thus, the following article presents a decision support system for the design
of the entire production process chain in the conceptual planning phase. It will
be useful for storage and retrieval of knowledge about process alternatives. The
work focuses on the use of powder-bed-based metal AM in automotive appli-
cations. For this purpose, the process chains of Binder Jetting and Laser Powder
Bed Fusion including their alternatives in post-processing are considered.
Evaluation criteria and general conditions for automotive production are iden-
tified. Subsequently, the individual process steps and their properties are logi-
cally linked based on the defined criteria to support the selection of the optimal
process chain. Finally, the methodology is demonstrated with an automotive
component.
Keywords: Additive Manufacturing Decision support Process chain

selection Automotive industry

https://doi.org/10.1007/978-3-030-54334-1_33
470 M. J. Kratzer et al.
1 Introduction
Additive Manufacturing (AM) is increasingly applied in industrial production [1]. It is

defined as a layer by layer building-up process from formless substances, without the
need for any tools [2, 3]. By triangulation, a CAD model is transferred into an STL-
format, which is then sliced into single layers, ready for the build-up process [4]. The
production technology AM has enormous potential due to its flexibility in design [2].
Even though the roots of AM in the 1990 s are found in polymer-based processes [5],
nowadays ceramic-, foundry sand- and especially metal-based processes gain impor-
tance [6]. Today, metal AM represents around 16,2% of the whole market volume [3].
Beside the medical and aerospace industries, metal AM technologies are increasingly
used in automotive applications [3, 7]. There is enormous potential, considering that
modern cars mostly consist of metal components. With a ratio of 60%, steel is the
dominant material in car production [7, 8]. The automotive industry utilizes metal AM
mostly for tooling, motorsport applications, prototyping and also series production [1].
In 2018, the German car company BMW Group released its first metal additively
manufactured component in series production [9]
To unfold the potential of industrialized AM and widespread the technology in the
automotive field, academia and industry recognizes the need to consider not only the
AM process, but the entire process chain from CAD to the final part [1, 10]. In order to
derive the respective production system for a newly designed part, process planning
builds the link between R&D and production [11]. Here, it is specified, which mate-
rials, process steps and operating resources are required to produce the part. On this
basis, processing times and part costs can be estimated [11]. To the author’s knowledge
until today, there is no applicable methodology for metal AM to link the component’s
technical requirements to the selection process of appropriate materials, combined with
suitable printing processes and respective post processing steps. The goal of this paper
is to close this research gap by representing an approach for deriving feasible technical
combinations.
2 State of the Art in AM Process Selection
2.1 Metal AM Process Steps

This paper focuses on Laser Powder Bed Fusion (LPBF) and Binder Jetting (BJT), two
powder bed based additive manufacturing processes. Due to the good mechanical
properties, accuracy and material variety, LPBF has the highest industrial relevance
today [10]. However, BJT might have future potential in series production, due to its
advantages in productivity and surface quality [12, 13]. As opposed to LPBF, the
powder particles are not melted by laser energy. Instead, the powder cohesion is first
generated by a binder and then in a second step, by sintering [14] In order to produce a
final part, not only the fully automated printing process is required. There is also a need
for pre-processing, which consists of build job preparation (digital), powder supply
(physical) and preparation of the build platform (physical) [10]. More crucial for
achieving technical requirements are post-processing steps after the AM process and
Decision Support System for a Metal Additive Manufacturing 471
therefore considered as more important in the present paper [10, 15] Based on a
literature review of existing post-processing steps for LPBF and BJT, a classification
for the main groups is gathered (Fig. 1) providing the basis for the developed method
introduced in Sect. 3. There are obligatory process steps for each printed part and
optional process steps to improve properties. In addition, there must be a distinction
between “process specific steps”, which cannot be transferred to the process chain of
the other printing technology, and “synergetic steps”, which can be used in the same
manner for post-processing of LPBF and BJT parts.
LPBF
Separation Removal of
De- Heat
of parts and support
powdering treatment
build plate structures
Further
improvement Surface
of mechanical treatment
BJT properties
Debinding Removal of
De- Heat
Curing and support
powdering treatment
Sintering structures
Color Printing process Shape-

LPBF-specific BJT-specific Obligatory Optional
synergetic
Coding: process step process step Coding: process step process step
process step
Fig. 1. Post-Processing Steps for LPBF and BJT parts
Directly after the LPBF printing process follows the depowdering of the build job
with vacuum cleaners. In areas that are difficult to reach it can also be done with
ultrasonic vibrations, air pressure, brushes and alternatively by vibration and rotation in
automated depowdering systems [1, 3]. To reduce thermal stress and control
mechanical properties, heat treatment processes, depending on the material and the
expected mechanical properties, are applied [3, 10]. Part separation from the build plate
is typically realized with wire-cut EDM machines, band saws or by manual separation
using hammer and chisel [3, 10]. Also the required support structure for the LPBF
process is either removed manually, or by using chemical or electrochemical removal
processes [10, 16]. Still in research stage is an automated support removal with multi
axis flexible cutting tools [17].
Output of the BJT process are green parts embedded in a build box, which are then
transported to an oven for the curing process. The goal is to raise the hardness and
consequently the manageability of the parts [14]. Depowdering is carried out by air
pressure and brushes, since the parts are still fragile. The next step is thermal, thermal-
catalytic or fluid based debindering, followed by a sintering process [2, 18]. Depending
on geometrical conditions of the part and sinter profile, material shrinkage rates are
already considered in the printing process [18]. For complex parts, support structures
generated by the printing process to control shrinkage during sintering have to be
removed manually afterwards. Refractory barriers between part and support structure
simplify the removal process [14]. Subsequent bronze infiltration is optional and not
common in automotive applications [2] To improve mechanical properties, heat
treatment is used similarly to the LPBF process chain.
For both LPBF and BJT parts further improvements of mechanical properties are
possible. In order to generate a positive impact on fatigue strength and breaking
elongation, hot isostatic pressing (HIP) is used in aerospace applications [3, 10, 19].
Laser peening and shot peening are also employed to optimize fatigue behavior and
mechanical strength of AM components. Furthermore, mechanical surface treatment
methods like abrasive sandblasting, vibratory grinding and machining are carried out
[6]. Alternatively, there are physical methods like plasma cleaning, ion beam cleaning
and laser polishing, as well as different chemical and electrochemical methods [3, 20].
The collection of post-processing alternatives in this paper should not be seen as
complete. On principle all production technologies listed in DIN 8580 can be con-
sidered to fulfill part specific requirements [21].
2.2 Process Selection in AM

The number of AM processes as well as the variety of manufacturing techniques
described in the previous section offer a large solution space for producing a part. In
order to support the determination of the optimal process chain, numerous methodical
approaches for process selection in the field of AM have been developed [22, 23]. In
principle, all methodologies have the following essential elements:
In the beginning, selection criteria are defined, which have an influence on the
decision-making process [23]. The extensive literature review by Wang et al. [23] on
different methodologies of AM process selection shows that the criteria of costs, sur-
face quality, time, mechanical properties and dimensional accuracy are considered most
frequently.
The second element of a decision support system (DSS) is the creation of a data-
base [22]. Here, various AM processes are included and evaluated with regards to the
defined selection criteria [22, 24, 25]. While some methods consider the material to be
used as predefined [26, 27], other methods also build up a database with respect to
different material alternatives [22, 24, 25, 28]. This supports the user of the system not
only in the selection of the AM technology but also in material selection. In addition,
some methodologies also provide data of market available machines in order to suggest
a suitable machine for the user’s application [22, 24, 27].
Finally, an evaluation method is established in each of the cited works in order to
select the most appropriate process, material or machine based on the user’s require-
ments and by using the established database. In most cases, methods such as Simple
Additive Weighting (SAW) or Analytical Hierarchy Processes (AHP) are used.
With SAW, a result is calculated for each possible process, based on the stored data and
the weighting of the defined criteria [25] An integer scale is typically used for the
weighting. With AHP the weighting is done by comparing all criteria in pairs [29]. For
each pair, the user indicates which criterion is preferred over the other and by what
extend, using a quantitative scale. Both evaluation methods result in a ranking of
processes, materials or machines and thus provide the user a foundation for making
decisions on the best possible design of the production process.
3 Methodology for AM Process Chain Selection
Even though numerous methodologies for the selection of the optimal AM process are
already available, they are not sufficiently suitable for industrial application in auto-
motive series production. Therefore, a methodology with the following requirements
was developed to support the decision process in conceptual production planning:
A. Focus on automotive applications and its requirements
B. Selection of specific material alternatives
C. Selection of specific post-processing steps for AM technologies
D. Use of an appropriate evaluation system
E. Consideration of the entire process chain in the decision making process
In order to select the optimal production process chain, the two-stage approach
shown in Fig. 2 is proposed, taking into account the criteria listed above. In the first
stage, all process chains are examined with regard to the manufacturability of a specific
part. Based on this, in the second stage the optimal process chain can be selected using
additional criteria such as time, costs or sustainability. This work concentrates exclu-
sively on the first stage.
A. When defining the relevant selection criteria and building up a solution space,
specifics of automotive series production are not taken into account in any of the
literature investigated. Therefore, relevant criteria have to be identified, depending
on the company and the different applications. First, the company’s conventional
product portfolio for suitable AM parts should be screened, based on simple and
practical criteria like part size and lot size. After having enough example parts, the
characteristics and requirements of the parts (which are manufactured conven-
tionally today), as well as the further processing steps of the parts in the automotive
value chain should be examined by interviewing product developers, reading
specification sheets and collecting the requirements mentioned. Thus it becomes
obvious what kinds of requirements an AM part has to fulfill in the company and
area of application.
B. In order to select appropriate AM materials for the selection process, investigating
the company’s product portfolio for commonly used materials is proposed. This
can be done by screening the parts lists for different materials and their respective
properties. After that, an investigation of market-available AM materials and their
attainable properties should be made. Matching AM materials and materials used in
the company, based on its properties leads to a narrowed solution space, applicable
for the method.
C. Post-processing steps – as described in Sect. 2.1 – can have an enormous influence
on part properties, as well as time and costs. Therefore, they must be included in
the selection process of the AM technology in order to be able to make compre-
hensive process chain statements with regard to technical and economic criteria. To
reduce the amount of data to collect, only the two or three most promising post-
processing technologies for each process step should be included. Criteria for
whether or not considering a specific technology can be the amount of expert
knowledge or the technological maturity.
D. Almost all open literature approaches contain a structured evaluation system
which, based on defined criteria, leads to the selection of the best of various
alternative solutions. With the widespread methods SAW and AHP, a qualitative
assessment of the user is transformed into a quantitative assessment in the form of a
weighting. This procedure is particularly critical in the area of vehicle manufac-
turing, since for certain selection criteria, such as mechanical properties, fixed
specifications in the form of quantitative values already exist. If, for example, a part
is required to have a tensile strength of 250 MPa and a mean roughness value of
10 µm, this is a must-meet criterion. For this reason, weighting the mechanical
properties and surface quality is not appropriate in the present case. Rather, the
decisive factor for the user is whether or not the required values are achieved by
certain materials and process alternatives. Therefore, threshold conditions should
be used to identify technologically suitable process chains in order to examine
whether a technical requirement is fulfilled or not.
E. By including materials, AM processes and post-processing steps, an integrated
consideration of the entire process chain can be made. The necessary consideration
of dependencies between the individual processes is missing in previous research
approaches. In addition, the evaluation in regard to the selection criteria is also
carried out for individual process steps instead of combining several processes. The
solution space of potential process chains is already narrowed down to LPBF and
BJT, since interviews of automotive AM experts lead to the conclusion, that these
two technologies are currently most promising for Metal AM series production. For
the realization of the first stage of the methodology, it is essential to build up the
database shown on the left of Fig. 2, which contains all conceivable production
process chains, including materials. To do so, first, all unfeasible combinations of
materials, AM- and post processes should be excluded. This is done with a Design
Structure Matrix (DSM) [30, 31]. With this method, all elements are compared and
marked manually as compatible or not. This can be used to demonstrate, for
example, that aluminum alloys currently cannot be used in the binder jetting
process [18]. Second, individual process steps along the sequence must be linked to
“fragments of process chains”, enriched with information, such as the achievable
tensile strength, surface quality or dimensional deviation. For each combination of
process-steps, achievable properties have to be determined by investigating data
sheets, speaking with technologists or by performing tests. To reduce the amount of
work and generalize the findings, ranges should be specified instead of single
thresholds. For example, the combination of the LPBF process, the material
AlSi10Mg and abrasive blasting creates an average roughness of 10 to 30 µm. The
user’s part specific requirements and constraints, entered in the system (Fig. 2
middle), must be logically linked to the ranges specified for the different “frag-
ments of process chains”. It comes naturally, that different “fragments of process
chains” fulfill different requirements. As an example, there are “fragments of
process chains” needed to fulfill mechanical requirements and others to fulfill
surface requirements. Logically linking the value enriched “fragments” to entire

process chains allows the user of the method/ system to get all technical feasible
“entire process chains” to fulfill his requirements (Fig. 2 right). The result is that
process steps are never evaluated in isolation with regards to one criterion. Instead,
all corresponding elements of the process chain that influence the respective value
are taken into account.
The methodology is intended to structure the selection of suitable manufacturing
steps and provides transparency about the process sequence. The structure of the
system also offers the possibility to store and retrieve knowledge. By using the system,
planning times can be reduced due to the rapid generation of possible solutions.
optimal
Methodology for identifying feasible process chains
selection
User
Potential Technological feasible
Data Input
process chains process chains
€
Identification of
Pre In Post Pre In Post
optimal process
chain based on
further criteria
Part-specific (e.g. time, costs,
requirements and sustainability)
constraints
Creating a solution space Use of relevant

Output presentation of
by linking process steps selection criteria for
part-specific narrowed
and determine its automotive series
down solution space
combined values. production
Fig. 2. Methodical approach for selection of AM process chains
4 Implementation of a Method for AM Process Chain

Selection
For the implementation of the methodology in Microsoft Excel, specific selection

criteria were identified and based on further requirements the developed methodology
was transferred into a computer-aided decision support system.
4.1 Selection Criteria for Identifying Technical Feasible Process Chains

As described in the previous section, tangible technical requirements and general
conditions for the selection procedure must be defined for the use and implementation
of the methodology. For this purpose, literature on technical selection criteria for
production process chains and AM processes as well as automotive-specific part and
manufacturing characteristics were considered. The identified selection criteria are

shown in Fig. 3. The four superior selection criteria are mechanical requirements,
requirements on further processing and utilization of the part, geometrical require-
ments, and requirements on part quality. In particular regarding further processing and
utilizations, industry specifics play a role in defining requirements (e.g. automotive
might differ from medical applications).
Among the mechanical requirements, the yield strength, tensile strength, elongation
at break as well as the Youngs’s Module E and Vickers hardness are identified as the
most important criteria. The consideration of subsequent processing and utilization of a
part allows further implications with regards to mechanical and quality requirements.
As an example, for punch riveting a certain elongation at break or for cathodic elec-
trophoretic deposition (EPD) a certain surface quality is required. This has to be taken
into account when selecting the process chain. Further, geometrical requirements, like
information on component dimensions, edges, cylindrical elements, bores, cavities and
auxiliary geometries should be used to test the suitability of additive manufacturing
technologies and post-processing steps for AM part manufacturing. Other conditions
like the score of support structures request technological knowledge of the user.
Finally, process chains should also be assessed on quality requirements like surface
quality and dimensional accuracy.
Figure 3 does not claim to fully represent all relevant criteria for the selection of
process chains. Some of the relevant criteria might also be applicable in other indus-
tries, whereas others might matter less. Obviously, in specific cases, numerous other
criteria for the selection of materials and processes must be taken into account, such as
requirements for acoustics, fatigue strength as well as thermal and electrical conduc-
tivity. It should also be noted that each additional criterion requires a corresponding
Part-specific requirements and constraints (selection criteria for automotive AM series production)
a) Technical selection criteria for b) Specific criteria and constraints for c) Automotive specific part and
production processes AM-process selection manufacturing characteristics
r(1) Mechanical requirements (3) Geometrical requirements

Yield strength Rp0,2 (MPa) Component measures – x, y, z (mm)
Tensile strength Rm (MPa) Wall thickness min/max (mm)
Elongation at break A (%) Sharp edging (yes/no)
Young’s Modul E (GPa) Diameter cylindrical elements (mm)
Hardness (HV10) Bore diameter and length (mm)
(in case boreholes are existing)
(2) Requirements on further processing and utilization Cavities difficult to access, with high surface requirements
(yes/no)
Corrosion resistance required (yes/no)
Complexity of support structures (Score)
Need for cathodic EPD (yes/no)
Weldability (yes/no)
(4) Requirements on part quality
Need for punch riveting (yes/no)
Need for adhesive bonding (yes/no) Surface quality Rz, Ra (μm)
Need for further processing according to DIN 8580 (yes/no) Permissible dimensional variation (μm)
Fig. 3. Part specific requirements and constraints

acquisition of data and therefore increases the effort required for the development and
use of the system. The developed structure nevertheless covers the most important
criteria, which are particularly relevant for a large number of automotive AM
applications.
4.2 System Requirements

Key features of a decision support system are the easy handling, the simple execution
of alternative calculations and the consideration of different variants. Typically, the
results are presented in graphics or tables [32]. In combination with the goal of
applying the system within an industrial context on the one hand and the continuous
improvement of AM-performance on the other hand, following requirements for the
decision support system are formulated:
• Intuitive usability without specific software knowledge
• Updateability and expandability
• Barrier-free application within companies
• Simple subsequent use of the generated results
The system should be intuitively usable without specific software knowledge. Even
without the complex provision of a GUI, the user should be able to easily interact with
the system by entering data and obtaining results. Furthermore, the updatability and
expandability of the system must be ensured. Values stored in the database, which may
change over time, should be easily adjustable by a system administrator. An example
for the required update of the database is the availability of a new machine or optimized
parameter sets, which allow lower mean roughness values on the surface of a com-
ponent. Furthermore, the database should be easily expandable with new materials and
process step alternatives. In addition, a barrier-free application within the company
without the need for new installations must be ensured. Finally, the generated data
should be easy to use and further processing of data should be possible without new
interfaces to be programmed. Moreover, the overall results should be usable by other
stakeholders – for example for detailed cost calculations.
4.3 Overview, Structure and Description of the Implemented System

To meet the requirements defined in Sect. 4.2, the spreadsheet program Microsoft
Excel in combination with the programming language Visual Basic Applications
(VBA) was chosen to implement the decision support system. The structure of the data
basis, the input of data by the user and the presentation of results is done in the form of
rows and columns. Since Microsoft Excel is a standard software widely used in
industrial companies, intuitive usability can be ensured, if the system was designed
accordingly. Updatability and expandability is also given, if the system can be used to
modify or add new cells without having to adapt the program code. Due to access to
Microsoft Excel within many companies, the developed system can be quickly made
accessible to potential users. Figure 4 provides an overview and illustrates the structure
of the system.
In total, there are nine linked sheets, whereby the user’s interaction is limited to the
yellow marked input (6) and output sheet (8). In the input sheet the user can enter the
“part-specific requirements and constraints” shown in Fig. 3. In case one criterion does
not have any requirements, it is not considered in the decision making process. In
addition, the user can specify which part properties must be quality checked during the
production process. Finally, the user can indicate which process steps are already
available in the existing production system, so the system is supporting by making
transparent, whether or not the part can already be manufactured by the existing
operating equipment. After completing the data input, the user can trigger the gener-
ation of the output sheet by clicking a button.
In the output sheet (8), the user receives all the technologically feasible process
chains based on the input. Each process chain is represented in the form of a row with
each process step being represented by a cell. In addition to the recommended material
and printing processes, specific post-processing procedures and methods for quality
assurance are suggested to the user. Furthermore, process steps that already exist in the
production system are highlighted. To increase knowledge transfer from the system to
the user, by clicking on a specific process step, the user can obtain further information
on the selected process step in a text box. This knowledge is retrieved via the linked
knowledge sheet (9). Lastly, the user receives information about the values that the
different process chains achieve with regard to the selection criteria. This approach
creates transparency, which enables a simple comparison between different alternatives.
Process step alternatives 1 Compatibility table 2

User Interaction
PS 1 PS 2
PS 1 PS 2 PS 3 A1 A2 B1 B2
A1 B1 C1
A1 x x Administrator-
PS 1
A2 B2
B3
C2
A2 x - Interaction
B1 x x
PS 2
B2 x -
No Interaction
+
(Calculation-Sheet)
Combinatorics and compatibility check 3 Database: values for process steps 4
A1 B1 C1 C1
PS 1 PS 2 PS 3
C1 of To of to
B2 A1 - C2 400 500 30 38
C2 - - C1 - 20 30
- B3 C2 200 -
A2 B3 A2 B1 C2 - 50
Value enriched compatible process chains 5 Data entry by user 6 Quality methods and techniques 7
C1 C2
PS 1 PS 2 PS 3
Of to of to
A1 B1 C2 400 500 30 38 C1 (>) 300 QM 1 yes QM 1 PM 1
A1 B2 C1 - 20 30
A1 B2 C2 400 500 30 38
A1 B3 C2 200 250 - C2 (<) 40 QM 2 no QM 2 PM 2
A2 B1 C2 - 50
A2 B3 C2 200 250 -
Output for user 8 Knowledge 9

C1 C2
QM 1 PS 1 PS 2 PS 3
Of to of to PS 1 Details 1
Q1 A1 B1 C2 400 500 30 38 i Details 1
Q2 A1 B2 C2 400 500 30 38 PS 2 Details 2
A1 B2 C1 - - 20 30
Fig. 4. System overview and structure
In addition to the input (6) and knowledge sheet (9), further data from the
remaining six sheets is required to generate the output sheet (8). In order to build up the
data base of “potential process chains” (see Fig. 2), in sheet 1, all material and process
step alternatives are stored. To reduce the amount of data to be processed and to
exclude technical unfeasible combinations, table 2 uses a DSM to specify, which
materials and process steps are compatible with each other. As a result, in sheet 3, all
“consistently compatible process chains” are generated based on the upper two sheets.
In the next step, the “fragments of process chains” (see Sect. 3E) are to be evaluated
according to the technical selection criteria. The data required for this is stored in the
database of sheet 4. In this sheet, the combination of process chain elements (fragment
of process chain) is assigned a value range, with respect to a criterion. This is done by
the system administrator, as discussed in Sect. 3. The result is table 5, where the values
from the database (4) are assigned to the process chains generated by combinatorics (3).
This is done by an automated process, where the “consistently compatible process
chains” are accessing the different “fragments of process chains” and combining them
in that way.
By comparing the achievable values of the process chains with the user’s input
data, the output sheet (8) excludes all process chains that do not meet the requirements
of one or more criteria. As explained in Sect. 3, this is done by threshold conditions.
A criterion is not met if the input value is below the stored range. If the required value
lies within the range, the corresponding criterion is marked in the output sheet with the
text “Detailed inspection required”. In this case, a technological process check must be
carried out to determine whether the required value can actually be achieved. Alter-
natively, a communication loop back to the design team can be performed to check
whether the requirements or the part design can be modified to use the respective
process chain. Finally, the link between the output sheet and the sheet “Quality
Methods and Techniques” (7) should be mentioned, by which, on the basis of user
information, possible methods for component testing can be suggested. This additional
feature allows the user to select suitable quality checks, according to part requirements
and process steps.
5 Demonstration and Discussion
For demonstration, the system is tested on a fictive second version of the water pump
wheel, which is already produced in small series production [33]. As shown in Fig. 5,
after entering the requirements and general conditions of the component in the input
sheet, the possible solutions for the process chain design are generated.
With the use of the system, the solution space of the database was limited from a
five-digit number of theoretically possible process chains to 18 entire process chains
suitable for the application. The green highlighted cells represent processes which
already exist in the user’s AM production. For light weight reasons, the process planer
chooses to exclude steel for this motorsport component, before he starts the selection
process, so the system automatically excluded all BJT process chains. Other LPBF
process chains were also excluded by the system, because they do not satisfy all
requirements. An example is the exclusion of vibratory grinding because of sharp
edging and too high requirements on surface quality. Other process steps are set as
mandatory by the system to reach required properties. For example, stress relief
annealing is needed to meet mechanical requirements.
It should be noted that the benefit of the entire system is accompanied by the quality
of the database. The acquisition of the required data is difficult for several reasons:
Since the values to be fed in depend on several process steps, the effort for data
acquisition is high. Moreover, it is only possible to indicate ranges for the achievable
values at process level. Thus, the result space of the system also contains solutions that
may not be practicable, because data is missing or the stored range is not precise
enough.
Finally, it should be emphasized that the usability of the system is limited, due to
the large number of process chains. Some process chains could be quickly eliminated
by applying economic criteria. Therefore, the integration of further criteria in a sub-
sequent step is indispensable to determine the optimal process chain.
Fig. 5. Use Case: System input and output
6 Conclusion
The contribution presents an approach for the identification of technical feasible pro-
cess chains for metal AM components. By linking process steps and determining its
combined values, a solution space for entire process chains can be generated. The
solution space can be seen as an expandable knowledge retrieval which should be
updated regularly with new findings. This integrated view of process chains is different
from existing approaches for AM process selection, but essential for fulfilling high
requirements of automotive series components being manufactured with a focus not
only on the printing process but the whole production process. For the creation of
relevant selection criteria, it seems to be essential to combine general technical

selection criteria with requirements of automotive series production and AM specifics
and constraints. To create the desired result, compatibility checks and combinatorics of
value enriched process steps and different materials are logically connected with part-
specific values for selection criteria. The methodology supports process planning by
contributing to a more structured generation of technical feasible process chains and
builds the basis for selecting the optimal process chain in the next step, based on
criteria like time, costs and sustainability.
References
1. Zeyn, H.: Industrialisierung der Additiven Fertigung. Digitalisierte Prozesskette - von der
Entwicklung bis zum einsetzbaren Artikel. 1st edn. Beuth Verlag GmbH, Berlin, Wien,
Zürich (2017)
2. Gibson, I., Rosen, D., Stucker, B.: Additive Manufacturing Technologies. 3D Printing,
Rapid Prototyping and Direct Digital Manufacturing. 2nd edn. Springer, New York (2015)
3. Wohlers, T.T. (Hrsg.): Wohlers Report 2018. 3D Printing and Additive Manufacturing State
of the Industry. Annual Worldwide Progress Report Wohlers Associates, Fort Collins (2018)
4. Zäh, M.: Wirtschaftliche Fertigung mit Rapid-Technologien. Anwender-Leitfaden zur
Auswahl geeigneter Verfahren. 1st edn. Hanser, München (2006)
5. Gibson, I., Khorasani, A.: Metallic Additive Manufacturing: Design, Process, and Post-
Processing. Metals Open Access Metall. J. 9(2), 137 (2019)
6. Gebhardt, A.: Additive Fertigungsverfahren. Additive Manufacturing und 3D-Drucken für
Prototyping - Tooling – Produktion, 5th edn. Hanser, München (2016)
7. Meixlsperger, M.: Anwendungsspezifische Prozessführung des Selective Laser Melting am
Beispiel von AlSi-Legierungen im Automobilbau, 1st edn. Shaker, Herzogenrath (2018)
8. Evertz, T., et al.: Der Leichtbauwerkstoffe für den Fahrzeugbau. In: Friedrich, H.E. (ed.)
Leichtbau in der Fahrzeugtechnik, pp. 205–449. Springer Fachmedien, Wiesbaden (2017)
9. BMW GROUP. https://www.press.bmwgroup.com/global/article/detail/T0286895EN/a-millio
n-printed-components-in-just-ten-years:-bmw-group-makes-increasing-use-of-3d-printing?la
nguage=en. Accessed 19 Feb 2020
10. Möhrle, M.: Gestaltung von Fabrikstrukturen für die additive Fertigung. Springer,
Heidelberg (2018)
11. Westkämper, E., Decker, M.: Einführung in die Organisation der Produktion, 1st edn.
Springer, Heidelberg (2005)
12. Langefeld, B., Moehrle, M., Balzer, C., Schildbach, P.: Advancements in Metal 3D Printing.
Roland Berger Focus, München (2018)
13. Stavropoulos, P., Foteinopoulos, P.: Modelling of additive manufacturing processes: a
review and classification. Manufact. Rev. 5(2) (2018)
14. Ziaee, M., Crane, N.-B.: Binder jetting: A review of process, materials, and methods. Addit.
Manufact. 3(82), 781–801 (2019)
15. Kaynak, Y., Kitay, O.: The effect of post-processing operations on surface characteristics of
316L stainless steel produced by selective laser melting. Addit. Manufact. 26, 84–93 (2019)
16. Schmithüsen, T., Eibl, F., Meiners, W.: Untersuchungen zum automatisierten Entstützen
SLM-gefertigter Bauteile. RTejournal Forum für Rapid Technol. 14 (2017)
17. Nelaturi, S., Behandish, M., Mirzendehdel, A.M., de Kleer, J.: Automatic support removal
for additive manufacturing post processing. Comput. Aided Des. 115, 135–146 (2019)
18. Munsch, M., Schmidt-Lehr, M., Wycisk, E.: Additive manufacturing with sinter-based
technologies. AM Power (2018)
19. Milewski, J.O.: Additive Manufacturing of Metals. From Fundamental Technology to
Rocket Nozzelz, Medical Implants, and Custom Jewelry. 1st edn. Springer International
Publishing, Cham (2017)
20. Hassanin, H., Elshaer, A., Benhadj-Djilali, R., Modica, F., Fassi, I.: Surface finish
improvement of additive manufactured metal parts. In: Gupta, K. (ed.) Micro and Precision
Manufacturing, pp. 145–164. Springer, Cham (2018)
21. DIN Deutsches Institut für Normung e.V.: DIN 8580:2003-09, Fertigungsverfahren.
Begriffe, Einteilung. Beuth, Berlin
22. Ghazy, M.: Development of an additive manufacturing decision support system (AMDSS).
Newcastle (2012)
23. Wang, Y., Blache, R., Xu, X.: Selection of additive manufacturing processes. Rapid
Prototyping J. 23(2), 434–447 (2017)
24. Park, H.S., Tran, N.H.: A decision support system for selecting additive manufacturing
technologies. In: ICISDM 2017, pp. 151–155 (2017)
25. Uz Zaman, U.K., Rivette, M., Siadat, A., Baqai, A.A.: Integrated product-process design:
Material and manufacturing process selection for additive manufacturing using multi-criteria
decision making. Robot. Comput.-Integr. Manufact. 51, 169–180 (2018)
26. Bikas, H., Koutsoukos, S., Stavropoulos, P.: A decision support method for evaluation and
process selection of Additive Manufacturing. Procedia CIRP, 1107–1112 (2019)
27. Mançanares, C.G., S. Zancul, E., Cavalcante da Silva, J., Cauchick Miguel, P.A.: Additive
manufacturing process selection based on parts’ selection criteria. Int. J. Adv. Manufact.
Technol. 80(5–8), 1007–1014 (2015)
28. Wang, Y., Zhong, R.Y., Xu, X.: A decision support system for additive manufacturing
process selection using a hybrid multiple criteria decision-making method. Rapid
Prototyping J. 9, 1544–1553 (2018)
29. Saaty, T.L.: Decision making with the analytic hierarchy process. Int. J. Serv. Sci. 1(1), 83–
89 (2008)
30. Danilovic, M., Browning, T.R.: Managing complex product development projects with
design structure matrices and domain mapping matrices. Int. J. Project Manage. 25(3), 300–
314 (2007)
31. Maurer, M.: Complexity Management in Engineering Design – A Primer, 1st edn. Springer,
Heidelberg (2017)
32. Wirtschaftslexikon Gabler. https://wirtschaftslexikon.gabler.de/definition/decision-support-
system-dss-28299/version-251932. Accessed 19 Feb 2020
33. BMW Group. https://www.press.bmwgroup.com/deutschland/article/detail/T0215062DE/re
nntechnik-aus-dem-3d-drucker:-bmw-fertigt-wasserpumpenrad-fuer-dtm-rennwagen-in-addi
tivem-fertigungsverfahren?language=de. Accessed 19 Feb 2020
Business Models for Additive Manufacturing:
A Strategic View from a Procurement
Perspective
Andreas H. Glas, Matthias M. Meyer(&), and Michael Eßig
Procurement Research Group, Bundeswehr University Munich,

85579 Neubiberg, Germany
{andreas.glas,ma.meyer,michael.essig}@unibw.de
Abstract. Additive Manufacturing (AM) is an emerging manufacturing method

that possesses great potential for new product designs and improved product
performance. Thus, a production strategy has to assess how to use AM. However,
production requires supply; thus, this work adds a strategic view from a pro-
curement perspective to the AM discussion.
This work distinguishes several procurement scenarios for AM. These sce-
narios are based on the distinction of internal, hybrid or external production
designs. A change in the manufacturing method towards AM in pure in-house or
pure external production models will trigger existing procurement tasks within a
new context and with new content. For example, in-house production triggers
the procurement processes for investment goods (printers) and commodities
(powder). External production triggers the procurement process for finished
products.
The most complex scenarios are hybrid ones in which new procurement tasks
emerge, e.g., orchestrating the input or developing an AM-strategy and archi-
tecture. Furthermore, new or changed business models emerge, e.g., Data-Model
Generators, 3D-Model Marketplaces, 3D-Printing Capacity Sharing Platforms,
License Supervisors, Powder Marketplaces, 3D-Printing Integrators, etc.
Overall, this work conceptualizes the new procurement arena in terms of new
or existing tasks and new or existing business models. The findings show that
the role of procurement as the central intersection to upstream supply markets is
becoming even more important. Decisions relating to AM design structure/
architecture (centralization/decentralization, in-outsourcing, open/closed data
access, and capacity sharing) have strategic impacts. Thus, this research pro-
vides theoretical and practical guidance on how the procurement function may
develop concerning new AM business models.
1 Introduction
Business expectations for Additive Manufacturing (AM) in the context of Supply

Chain Management (SCM) are exorbitant [1]. Several management consultancies
predict astronomical impacts of this technology on the market, such as annual growth
rates of 35% [2] or a market value of 250 billion US$ by 2025 [3].

https://doi.org/10.1007/978-3-030-54334-1_34
484 A. H. Glas et al.
Technically, promising industrial use cases exist. In 2017, Renault Trucks SAS
introduced an improved truck engine, which was developed by using design and
manufacturing benefits of AM. As a result, the amount of parts was reduced by 25%
(200 total parts), which led to the single-engine having a weight reduction of 120 kg
and improved transportation capacities for logistic providers [4]. This use case provides
an indication for the high relevance of AM in companies for new product development.
Considering the business and technical potential, it is assumed that AM can change
how future production will look. A change in production will affect other functions of a
company, e.g., logistics, sales, after-sales services, infrastructure, etc. This article will
focus on how the procurement function is affected.
The procurement function supplies a company by sourcing required and not self-
produced goods. This article will take the perspective of the procurement function and
investigate the sourcing situation in the case of AM production. The research aim is to
conceptualize the AM sourcing situation. Thus, the following research question is
addressed: “What potential impact does AM has on procurement sourcing activities?”
For this purpose, the remainder of this paper is structured as follows. First, the
basics of procurement, business models and AM are briefly presented. This is followed
by the conceptualization of four AM sourcing scenarios. For each scenario, specific
observations/propositions are derived. This work concludes with a discussion of two
AM procurement models.
2 Literature Review
Since scientific research on procurement and AM is in an early stage, both topics will
be shortly introduced.
2.1 Strategic Procurement: Sourcing Decisions

Procurement represents a strategic function within an organization [5, 6], which is
responsible for managing the external resources of an organization. Input factors, which
are required for a subsequent transformation process within the company to obtain a
required output in form of a good or service, are bought [7]. Sourcing represents a
strategic element within procurement and refers to the sourcing decision [8–10]. In the
sourcing decision, it is decided whether to produce a good in-house (insourcing) or to
buy the good on the market (outsourcing).
Sourcing decisions are dynamic and situational. Thus, the in- or outsourcing
decision may change over time and depend on if and how the good has been sourced
before. An important systematization to structure sourcing tasks is the distinction
between straight rebuy, modified rebuy and new buy situations [11]. In a new buy
situation, a high amount of information and process effort is required; whereas in a
straight rebuy situation, procurement can accomplish its supply goals with minimum
amounts of information and process effort. A modified rebuy situation is in between
both extremes.
Business Models for Additive Manufacturing 485
2.2 From Products to Services: The Emergence of Business Models

The concept of business models became prominent with the technological inception of
the Internet during the 1990s [12]. According to Chesbrough [13], a technology has no
single objective value and arises only from its commercialization via business models.
In addition, servitization – the move from the manufacturing of tangible goods to
service delivery – led to a number of new or changed business models [14]. A business
model can be defined as an abstracted accumulated depiction of a company’s relevant
activities, describing how value for the customer is added due to products or services
and how profits are generated [15]. Since arising technologies describe drivers for new
business models [13], the appearing technology of AM [16] is already leading to new
business models, which are further outlined in the next section.
2.3 Additive Manufacturing and Its Specific Business Models

Manufacturing processes are classified into formative, subtractive and additive pro-
cesses [17]. AM describes a technology consisting of not a single but rather multiple
distinctive processes [18], of which the first one - stereolithography - was invented in
1983 by Chuck Hull [16]. Its synonyms are layer manufacturing, generative manu-
facturing and 3D printing [19].
By comparing AM´s definitions within the literature, all definitions state that raw
materials are positioned and joined layer by layer [20, 21] based on a digital repre-
sentation (CAD-data) of the object to be produced [21, 22] until its physical repro-
duction is formed [19, 23].
A generic AM process can be described as follows. At first, a Computer Aided
Design (CAD) file is created. Afterwards, it is converted into a sliced model readable
by a 3D printer [17, 24]. Next, several pre-processing steps take place such as adjusting
the printer settings or preheating the build platform [24]. During the production pro-
cess, layers of material are added iteratively until a physical part is created based on the
virtual drawing [17]. Subsequently, post-processing steps take place, such as the
removal of support structures, heat treatment processes or surface finishing [24].
AM differs from traditional manufacturing technologies such as CNC machining in
four dimensions, and these are given as follows.
Direct and Immediate Interaction of the Virtual and Physical World: “Direct
digital manufacturing” [20] aptly describes AM´s advantages in the transition of goods
from the virtual to the physical world. To modify components, only the CAD file needs
to be changed so that a new product can be produced [19].
Reduced Asset Specificity: This can be seen in the “tool-free production” [25] in
which different components can be produced synchronously in just one print job. This
reduced specificity enables decentralized manufacturing close to the end customer [22]
and allows one to blur the separation of the manufacturer and user of a product, e.g.,
such as in the maker movement [26].
Changed Cost Structure: AM also significantly changes the previous cost structure of
products. For example, when component complexity increases, there are no additional
costs (“Complexity for free”) [27] and there are no cost reductions due to the pro-
duction of larger quantities, contrary to traditional manufacturing processes [28].
Increased Resource Efficiency: Due to the application of material almost exclusively

at the designated places, higher material utilization can be guaranteed [20]. There is
also the opportunity to directly manufacture components as a functional (even mov-
able) assembly. In this case, assembly is reduced [29].
The specifics of AM technology do not only lead to a change from formative or
subtractive production technologies to additive ones. AM also allowed the emergence
of new business models. Some of these are already discussed in the literature.
User Innovation and Maker Community
Since AM parts can be easily designed and printed [30], markets for home manufac-
turing emerge [31]. One example is MakerBot. With it, end-users can use their cre-
ativity to create and individualize desired products by themselves, which founded the
“Maker Movement” [26].
Design Sharing Platforms and Marketplaces
CAD files are readable by all printers and can easily be exchanged and adapted [32].
Therefore, design sharing platforms such as Thingiverse developed where files can
easily be exchanged and traded [33].
Infrastructures for Secured IP and Data Exchange
Since design data can easily be shared with others, new concepts for securing virtual
property rights need to be created [34]. Therefore, new data infrastructures such as
decentralized databases e.g. Blockchains [35] are used.
Machine Sharing
Since distinct products containing the same material can be manufactured simultane-
ously by the same printer, the print capacity can be shared and costs can be reduced.
Therefore, companies collaborate to achieve a competitive advantage [36].
AM Service Provider and Supercentres
For AM, several service providers have established, which provide generative services
such as creating data files (e.g., by scanning), manufacturing services, or full-service
[37]. Hereby, AM-supercentres [38] arise, where AM is complemented by traditional
production methods such as heat treatment [39].
AM Printing Platforms
Via a digital platform [32], manufacturing providers with free capacity can be selected
for the pure execution of production [40].
Overall, it is of relevance if and how procurement can assess AM and AM business
models in order to supply a company. The next chapter will conceptualize the sourcing
situation in the form of scenarios.
3 How to Source (for) AM?
In the following, the reasoning in this chapter is anchored by a generic scenario in

which a buying organization wants to have a finished product (FP). That product can be
manufactured with AM. We are aware that AM is the enabling technology for
decentralized manufacturing, which impacts logistics. From a procurement perspective,
focus is set on the sourcing relations of the organizations and contracting, on which
decentralized manufacturing remains indifferent. Therefore, we illustrate how a supply
system between a buying organization and its supply base could look like in four
sourcing scenarios. These scenarios comprise the two extreme scenarios of full in-
house vs. full outsourced AM production. Next, two hybrid scenarios are presented,
one of which explicitly focuses on AM new business models. The scenarios are used to
investigate how sourcing will change in each scenario.
3.1 Scenario 1: Source Input Goods for In-House AM Production Facility

The first hypothetical scenario represents in-house production with AM (see Fig. 1).
Here, procurement is responsible for sourcing AM-input goods such as raw materials,
capital goods such as 3D-printers and equipment for pre- and post-processing activities.
Design and engineering activities in this scenario represent a core capacity of the
organization and therefore stay in-house.
Fig. 1. Sourcing of AM-input goods
The AM´s physical input goods can be observed to be highly unspecific. The term
raw material by itself shows that it can become easily standardized [41]. The same
applies for 3D-printers as standardized production machines. A 3D-printer allows the
printing of several designs with different levels of complexity within the same print job
[18] with a large variety of materials [37]. This leads to a reduction of the complexity
within supply chains and the cut of the intermediate tier-structures [34] as functional
modules can be directly printed [42]. Therefore, procurement obtains increased
negotiation power as AM´s physical input goods become an unspecific “commodity”.
Proposition 1a: Physical input goods are standardized and become “commod
ties”, which lead to increased price competition.
Raw materials represent the only variable input good (their consumption is
dependent on the production quantity). The obsolescence of raw materials within the
AM process hinders organizations’ production. Both types of raw materials, polymers
and metals, are characterized by high price volatilities [43]. To minimize the supply
risk and have fast responses to changing demand situations, companies might use a
multiple sourcing approach.
Proposition 1b: Multiple sourcing concepts should be used to minimize the raw
material supply risk.
The supply constellation of in-house AM shows that the addressed buying tasks
already exist for procurement. Many companies regularly source raw materials or
capital investment goods. Even as 3D-printers represent a new market, no task novelty
for the procurement function is given.
Proposition 1c: For an in-house AM constellation, procurement sources goods

from new markets using already established sourcing tasks.
3.2 Scenario 2: Lean AM Implementation

For the second scenario, the buying organization directly sources AM outputs, such as
prototypes [44], tools [45], spare parts [22] or customer individual products [31]. As
seen in Fig. 2, no AM input goods are required for the buying organization, and the
buying organization sources finished products from additive manufacturing service
providers (AMSP).
Fig. 2. Sourcing of AM-output goods
Since no tooling within AM is required, products can be manufactured with a

shorter manufacturing lead time and shorter time to market [45]. This leads to inten-
sified competition on time to market, which is already established within various
industries. Therefore, procurement must enable fast and flexible processes on which the
capacity of their supply base consisting of service providers is monitored to enable fast
reactions to sudden demands.
Proposition 2a: Procurement must enable agile sourcing processes in which the
printing capacity of their service provider supply base is monitored.
Since an organization using a lean AM approach might be interested in shorter lead

times and quicker availability of the finished product when required, it is likely that
contracts are used that tie payments to performance indicators such as performance-
based contract models [46].
Proposition 2b: Organizations using the lean AM approach likely apply perfor-
mance-based contract models for sourcing finished AM products.
Since the sourcing constellation of scenario 1 (sourcing of raw materials, 3D-printer

and equipment for pre-and post-processing) now applies for the supplier in a lean
scenario, the supply chain tier-level is shifted one step backwards. For procurement, it
does not care whether the product was produced additively or traditionally since it
solely needs a finished product. Therefore, the lean AM approach represents the
simplest constellation with the lowest effort for procurement to source AM.
Proposition 2c: The lean AM approach represents the easiest sourcing constell
tion for the procurement of AM produced finished goods.
3.3 Scenario 3: Selective, Hybrid AM Sourcing

The third scenario is a hybrid one and comprises elements of scenarios 1 and 2. Within
this scenario, the organization creates CAD-files in-house and executes the print-job
either in-house or at an AM-supplier (see Fig. 3). Therefore, procurement is respon-
sible for sourcing AM-input goods similar to in scenario 1 and print jobs from AM-
service providers (AMSP) to whom the CAD files are sent similar to scenario 2. While
a company might benefit from such a scenario due to enhanced production capabilities
(in-house plus external) that reduce risks (satisfaction of peak demand and product
availability), this scenario adds complexity to procurement and causes more sourcing
tasks.
Fig. 3. Hybrid sourcing AM
AM allows one to produce goods without any kind of extra tooling [47]. Due to the
appearance of demand fluctuations, peaks could be compensated by other organiza-
tions´ printers [36] such as AMSPs. Thus, the risk of being unable to manufacture an
urgent demand is minimized.
Proposition 3a: Organizations use a mix of AMSPs and in-house manufacturing

to minimize the risks caused by demand fluctuations.
Kraljic [48] mentions that new technology helps to minimize bottleneck risks. Due
to AM´s ability to create any shape [18] with a standardized set of input goods, AM
could be used to reduce the risk of material bottlenecks. Thus, an organization might
have its own AM capabilities for which AM would become a sourcing lever for risk
minimization.
Proposition 3b: Procurement uses a hybrid AM approach to decrease the risk of

supply bottlenecks with its own AM printing capabilities.
Data become the key input factor for AM [49]. It can be seen that CAD data can be
highly customized [50] or topologically optimized by bio-inspired algorithms con-
taining high engineering knowledge [21]. CAD data can easily be interchanged due to
the usage of a standardized format (STL) [32]. Therefore, procurement must focus on
achieving access to data to fulfil its own demands, to provide data to AMSPs and to
secure sensitive data handling.
Proposition 3c: Procurement must have a more prominent focus on the sourcing
and securing of data and property rights.
Since both AMSP markets and AM-input markets are required, sourcing tasks
increase since these must address both aspects. Increased sourcing and a more complex
AM supply imply increased procurement complexity.
Proposition 3d: Procurement complexity increases due to the doubling of the

sourcing tasks (input and AMSPs).
3.4 Scenario 4: Sourcing New AM Business Models

In the fourth scenario, AM´s new business models are used by the buying company.
Thus, the buying organization uses digital platforms, such as design sharing or man-
ufacturing platforms, as seen in Fig. 4.
Companies might take advantage of the creativity of others and source CAD data
from design sharing platforms such as Thingiverse [33]. Then, procurement can benefit
from the innovational strength of the platform as a means of “innovation sourcing”.
Proposition 4a: Procurement sources CAD files via design sharing platforms and
design marketplaces as a means to get access to innovation.
The manufacturing costs of the print job decrease when build plate utilization
increases [36]. Thus, an organization might use a collaborative strategy to minimize
these costs. Therefore, procurement needs to either find and select convenient partners
by itself or take advantage of platforms on which the printer’s capacity can be shared as
a competitive advantage.
Proposition 4b: Procurement must ensure that the organization is able to acquire
competitive advantages due to the sharing of the printer capacity.
Fig. 4. Sourcing of new AM business models
Since the execution of a print job itself becomes an automated standardized task
where supervision is merely required [51], procurement might adopt manufacturing
platforms to minimize their transaction costs, on which AMSPs are connected and
monitored by their free capacity on real time data.
Proposition 4c: Manufacturing platforms are used to source print jobs in AMSPs.
Since new business models in AM are often based on platforms, such as design
sharing, data marketplaces [33] or manufacturing platforms [32], procurement must
establish ways to execute its sourcing strategies, e.g., sourcing location or preferred
suppliers, digitally via these platforms.
Proposition 4d: Procurement executes sourcing strategies digitally via platforms.
Due to the creation of new business models due to AM [52] and new ways of
interacting with suppliers, e.g., sharing printer capacity [36], procurement has to find
new contractual models to enable these. In the case of data marketplaces, arrangements
such as on demand printing or unlimited printing from a CAD model have to be created
and monitored.
Proposition 4e: Procurement requires new contract models to effectively use

AM´s new business models.
The summarization of propositions for all four scenarios can be seen in Table 1.
Table 1. Summary of the propositions on changes in sourcing due to AM

Scenario 1: Source input goods for in-house AM production facility
1a Physical input goods are standardized and become “commodities”, which lead to
increased price competition
1b Multiple sourcing concepts should be used to minimize the raw material supply risk
1c For an in-house AM constellation, procurement sources goods from new markets using
already established sourcing tasks
Scenario 2: Lean AM implementation
2a Procurement must enable agile sourcing processes in which the printing capacity of their
service provider supply base is monitored
2b Organizations using the lean AM approach likely apply performance-based contract
models for sourcing finished AM products
2c The lean AM approach represents the easiest sourcing constellation for the procurement
of AM produced finished goods
Scenario 3: Selective, hybrid AM sourcing
3a Organizations use a mix of AMSPs and in-house manufacturing to minimize the risks
caused by demand fluctuations
3b Procurement uses the hybrid AM approach to decrease the risk of supply bottlenecks
using its own AM printing capabilities
3c Procurement must have a more prominent focus on the sourcing and securing of data
and property rights
3d Procurement complexity increases due to the doubling of the sourcing tasks
Scenario 4: Sourcing new AM business models
4a Procurement sources CAD files via design sharing platforms and design marketplaces as
a means to get access to innovation
4b Procurement must ensure that the organization is able to acquire competitive advantages
due to the sharing of the printer capacity
4c Manufacturing platforms are used to source print jobs in AMSPs
4d Procurement executes sourcing strategies digitally via platforms
4e Procurement requires new contract models to effectively use AM´s new business models
4 Discussion and Conclusion
Overall, the four scenarios have shown that AM affects sourcing decisions (see
Table 1). Surprisingly, the sourcing tasks in some scenarios are very close to existing
ones already known from conventional production methods. Thus, the following
explanations refer to the initial research question regarding the impacts that AM has on
sourcing activities. (see Fig. 5).
In the case of in-house production, sourcing focuses on the access to input goods,
which might lead to new or changed commodities (raw materials, printers, and pre- and
post-processing equipment). These markets differ from the input markets of subtractive
or formative production methods, but the sourcing methods may be very similar to
existing ones (sourcing strategies and sourcing levers).
Similarly, buying companies have huge experience in sourcing finished products.
Manufacturing service providers are widely used in the industrial context already in
conventional production methods. Thus, outsourcing production to AMSPs can be seen
as a new content for already existing sourcing activities. However, it is acknowledged
that the focus on risk mitigation and availability might lead to the management of
performance indicators (lead-time) together with product related performance (quality).
Even the hybrid scenarios are not per se a new and unique phenomenon. Addi-
tionally, for conventional production methods, hybrid in-house and external production
is possible. What is new to this setting is the emergence of new AM business models.
The development of new platforms might be an indicator that the cooperation with
upstream
Fig. 5. Impacts of Making or Buying in AM for procurement

AM markets is driven or at least enabled by new AM business models. Then,

sourcing must orchestrate new and existing business models, which might also change
sourcing methods (service and license contracts, new contracting approaches, etc.).
Finally, this work refers loosely to the distinction of new and existing sourcing
tasks (new buy/rebuy model [11]). It is intended to reduce the complexity of the
management of AM in procurement by showing that at least some tasks are not new
while others are (see Fig. 6).
In the first quadrant (existing procurement task and existing markets), sourcing
deals with already known methods and markets. The sourcing of energy, pre- or post-
processing equipment, and scanning software are good examples.
In the second quadrant (existing procurement task and new markets), sourcing must
deal with new market situations. The markets for 3D-printers and raw materials
(powder) are good examples. The sourcing techniques are similar to those for the
procurement of other investment goods and raw materials, but the technology, quality
criteria, market risks and negotiation power differ significantly. Proposition 1a shows
that 3D-printers and raw materials become commodities, which imply low asset
specificity [53]. Based on Williamson, these goods are sourced from a market, which
allows an easy changing of suppliers and decreases the effect of uncertainty within the
transaction.
However, this work also revealed that procurement must develop new activities and
sourcing tasks. The items in the third quadrant (new procurement tasks and existing
markets) refer to the digital content in the manufacturing process. The access to CAD
files and its secure exchange requires new sourcing instruments (e.g., contract clauses).
Another example for this quadrant is the task that orchestrates AM products and
methods with non-AM products and methods. The application of tracking and tracing,
risk analysis and quality assessment to the AM phenomenon could be mentioned.
However, even if these tasks are new or at least have enhanced relevance, they are used
in a market with experience in, e.g., data exchange and security or tracking and tracing.
Finally, this work could reveal that the emergence of new AM business models,
namely, platform driven ones, justifies the existence of quadrant four (new procurement
tasks and new markets). Here, the interplay of digital business models and market-
places with AMSPs and in-house capacities require active management and new or
enhanced forms of sourcing instruments (license management). In addition, many
platforms establish a business model that incorporates pay-per-use fees while others
have fixed fees for entering the platform or accessing a specified level of flat rate/full-
service offerings (“prime customers”). To develop a new sourcing strategy to cope with
these new and changed business model and market conditions is a sourcing task.
Overall, this work addresses the exaggerated expectations of AM as a technology.
AM will obviously affect production functions, but its potential for de-centralization
might also severely affect logistics and supply chain management functions [50]. This
work adds a strategic view from the procurement function to that discussion.
The conclusion contrasts assumptions that AM will also severely affect sourcing
tasks with the findings that only some aspects of sourcing will face new content or new
market conditions. The exception is a sourcing situation that incorporates new AM
business models. Such a scenario requires a quite new sourcing approach.
Fig. 6. Distinction by buying task and business model
Of course, this work faces several limitations since it is based on conceptual,

deductive reasoning. The presented scenarios are selected ones. Other situational set-
tings with the presented extremes might exist. Furthermore, this work merges
knowledge from the sourcing literature with characteristics of AM technology and its
impacts on production and logistics. The propositions and conceptual results can be
seen as the starting point for an empirical investigation of how AM affects sourcing in
managerial practice.
References
1. Gartner Inc.: Hype Cycle for 3D Printing (2018). https://www.gartner.com/en/documents/
3881825-hype-cycle-for-3d-printing-2018
2. Berger, R.: Additive Manufacturing in Aerospace and Defense (2017)
3. McKinsey & Company: Additive manufacturing: a long-term game changer for manufac-
turers (2017)
4. Renault Trucks SAS: 3d-Metalldruck: Eine Zukunftstechnologie für leichtere und kompak-
tere Motoren, Lyon (2017)
5. Carter, J.R., Narasimhan, R.: Is purchasing really strategic? Int. J. Purchas. Mater. Manage.
32(4), 20–28 (1996)
6. Roscoe, S., Cousins, P.D., Handfield, R.: The microfoundations of an operational capability
in digital manufacturing. J. Oper. Manage. 65(8), 774–793 (2019)
7. van Weele, A.J., Eßig, M.: Strategische Beschaffung. Springer Fachmedien Wiesbaden,
Wiesbaden (2017)
8. Nikolarakos, C., Georgopoulos, N.: Sourcing: issues to be considered for the make-or-buy
decisions. Oper. Res. 1(2), 161 (2001)
9. Tayles, M., Drury, C.: Moving from make/buy to strategic sourcing: the outsource decision
process. Long Range Plan. 34(5), 605–622 (2001)
10. Vitasek, K.: Strategic sourcing business models. Strateg. Outsourcing Int. J. 9(2), 126–138
(2016)
11. Robinson, P.J., Faris, C.W., Wind, Y.: Marketing Science Institute: Industrial Buying and
Creative Marketing. Marketing Science Institute. Series of books. Allyn and Bacon, New
York (1967)
12. Zott, C., Amit, R., Massa, L.: The business model: recent developments and future research.
J. Manag. 37(4), 1019–1040 (2011)
13. Chesbrough, H.: Business Model Innovation: Opportunities and Barriers. Long Range Plan.
43(2–3), 354–363 (2010)
14. Lindberg, N., Nordin, F.: From products to services and back again: Towards a new service
procurement logic. Ind. Mark. Manage. 37, 292–300 (2008)
15. Wirtz, B., Pistoia, A., Ullrich, S., Göttel, V.: Business Models: Origin, Development and
Future Research Perspectives. Long Range Plan. 49(1), 36–54 (2016)
16. Kietzmann, J., Pitt, L., Berthon, P.: Disruptions, decisions, and destinations: Enter the age of
3-D printing and additive manufacturing. Bus. Horiz. 58(2), 209–215 (2015)
17. Gebhardt, A., Kessler, J., Thurn, L.: Additive Fertigungsverfahren: Additive Manufacturing
und 3D-Drucken für Prototyping, Tooling, Produktion, 5th edn. Hanser, München (2016)
18. Durach, C.F., Kurpjuweit, S., Wagner, S.M.: The impact of additive manufacturing on
supply chains. Int. J. Phys. Distrib. Logist. Manage. 47(10), 954–971 (2017)
19. Huang, S.H., Liu, P., Mokasdar, A., Hou, L.: Additive manufacturing and its societal impact:
a literature review. Int. J. Adv. Manuf. Technol. 67(5), 1191–1203 (2013)
20. Berman, B.: 3-D printing: the new industrial revolution. Bus. Horiz. 55(2), 155–162 (2012).
https://doi.org/10.1016/j.bushor.2011.11.003
21. Beyer, C.: Strategic implications of current trends in additive manufacturing. J. Manuf. Sci.
Eng. Trans. ASME 136(6) (2014)
22. Khajavi, S.H., Partanen, J., Holmström, J.: Additive manufacturing in the spare parts supply
chain. Comput. Ind. 65(1), 50–63 (2014)
23. Knofius, N., van der Heijden, Matthieu C., Zijm, W.H.M.: Selecting parts for additive
manufacturing in service logistics. J. Manuf. Technol. Manage. 27(7), 915–931 (2016)
24. Gibson, I., Rosen, D., Stucker, B.: Additive Manufacturing Technologies: 3D Printing,
Rapid Prototyping and Direct Digital Manufacturing. Springer, New York, Heidelberg,
Dodrecht, London (2015)
25. Ruffo, M., Hague, R., Tuck, C.: Make or buy analysis for rapid manufacturing. Rapid
Prototyping J. 13, 23–29 (2007). https://doi.org/10.1108/13552540710719181
26. Waller, M.A., Fawcett, S.E.: Click here to print a maker movement supply chain: how
invention and entrepreneurship will disrupt supply chain design. J. Bus. Logist. 35(2), 99–
102 (2014)
27. Holmström, J., Partanen, J.: Rapid manufacturing in the spare parts supply chain. J. Manuf.
Technol. Manage. 21(6), 687–697 (2010)
28. Baldinger, M., Levy, G., Schönsleben, P., Wandfluh, M.: Additive manufacturing cost
estimation for buy scenarios. Rapid Prototyping J. 22(6), 871–877 (2016)
29. Essig, M., Kronthaler, S., Gutsmiedl, J., Glas, A.H.: Geschäftsmodelle additiver Ferti-
gungsverfahren. In: Schröder, M., Wegner, K. (eds.) Logistik im Wandel der Zeit – Von der
Produktionssteuerung zu vernetzten Supply Chains: Festschrift für Wolfgang Kersten zum
60. Geburtstag, pp. 207–234. Springer Fachmedien Wiesbaden, Wiesbaden (2019)
30. Afshari, H., Jaber, M.Y., Searcy, C.: Investigating the effects of learning and forgetting on
the feasibility of adopting additive manufacturing in supply chains. Comput. Ind. Eng. 128,
576–590 (2019)
31. Halassi, S., Semeijn, J., Kiratli, N.: From consumer to prosumer: a supply chain revolution in
3D printing. Int. J. Phys. Distrib. Logist. Manage. 49(2), 200–216 (2018)
32. Potter, A.T., Eyers, D.R.: E-commerce channels for additive manufacturing: an exploratory
study. J. Manuf. Technol. Manage. 26(3), 390–411 (2015)
33. Friesike, S., Flath, C.M., Wirth, M., Thiesse, F.: Creativity and productivity in product
design for additive manufacturing: mechanisms and platform outcomes of remixing. J. Oper.
Manage. (2018)
34. Mohr, S., Khan, O.: 3D printing and its disruptive impacts on supply chains of the future
technology. Innov. Manage. Rev. 5(11), 20–25 (2015)
35. Holland, M., Stjepandić, J., Nigischer, C.: Intellectual property protection of 3D print supply
chain with blockchain technology. In: 2018 IEEE International Conference on Engineering,
Technology and Innovation (ICE/ITMC), pp. 1–8 (2018)
36. Hedenstierna, C.P.T., Disney, S.M., Eyers, D.R., Holmström, J., Syntetos, A.A., Wang, X.:
Economies of collaboration in build-to-model operations. J. Oper. Manage. 65(8), 753–773
(2019)
37. Rogers, H., Baricz, N., Pawar, K.S.: 3D printing services: classification, supply chain
implications and research agenda. Int. J. Phys. Distrib. Logist. Manage. 46(10), 886–907
(2016)
38. Sasson, A., Johnson, J.C.: The 3D printing order: variability, supercenters and supply chain
reconfigurations. Int. J. Phys. Distrib. Logist. Manage. 46(1), 82–94 (2016)
39. Strong, D., Kay, M., Conner, B., Wakefield, T., Manogharan, G.: Hybrid manufacturing –
integrating traditional manufacturers with additive manufacturing (AM) supply chain. Addit.
Manuf. 21, 159–173 (2018)
40. Pahwa, D., Starly, B., Cohen, P.: Reverse auction mechanism design for the acquisition of
prototyping services in a manufacturing-as-a-service marketplace. J. Manuf. Syst. 48,
134–143 (2018)
41. Öberg, C.: Additive manufacturing – digitally changing the global business landscape. Eur.
J. Manage. Bus. Econ. 28(2), 174–188 (2019)
42. Ghadge, A., Karantoni, G., Chaudhuri, A., Srinivasan, A.: Impact of additive manufacturing
on aircraft supply chain performance: a system dynamics approach. J. Manuf. Technol.
Manage. 29(5), 846–865 (2018)
43. Pellegrino, R., Costantino, N., Tauro, D.: Supply chain finance: a supply chain-oriented
perspective to mitigate commodity risk and pricing volatility. J. Purchasing Supply Manage.
25(2), 118–133 (2019)
44. Munguía, J., Riba, C., de Ciurana, J.: Pursuing successful rapid manufacturing: a users’ best-
practices approach. Rapid Prototyping J. 14(3), 173–179 (2008)
45. Attaran, M.: The rise of 3-D printing: the advantages of additive manufacturing over
traditional manufacturing. Bus. Horiz. 60(5), 677–688 (2017)
46. Glas, A.H., Kleemann, F.C.: Performance-based contracting: contextual factors and the
degree of buyer-supplier integration. J. Bus. Ind. Market. 32(5), 677–692 (2017)
47. Khajavi, S., Holmström, J., Partanen, J.: Additive manufacturing in the spare parts supply
chain: hub configuration and technology maturity. Rapid Prototyping J. 24(7), 1178–1192
(2018)
48. Kraljic, P.: Purchasing must become supply management. Harvard Bus. Rev. 1983(61), 109–
117 (1983)
49. Tuck, C., Hague, R., Burns, N.: Rapid manufacturing: Impact on supply chain methodolo-
gies and practice. Int. J. Serv. Oper. Manage. 3, 1–22 (2007)
50. Oettmeier, K., Hofmann, E.: Impact of additive manufacturing technology adoption on
supply chain management processes and components. J. Manuf. Technol. Manage. 27(7),
944–968 (2016)
51. de La Torre, N., Espinosa, M.M., Domínguez, M.: Rapid prototyping in humanitarian aid to
manufacture last mile vehicles spare parts: an implementation plan. Hum. Factors Ergon.
Manuf. 26(5), 533–540 (2016)
52. Piller, F.T., Weller, C., Kleer, R.: Business models with additive manufacturing—
opportunities and challenges from the perspective of economics and management. In:
Brecher, C. (ed.) Advances in Production Technology, Cham, pp. 39–48 (2015)
53. Williamson, O.: The economics of organizations: the transaction cost approach. Am.
J. Sociol. 87(3), 548–577 (1981)
A Performance Upgrade of an Industrial Gas
Turbine Based on Additive Manufactured
Components
Pankaj Bajaj1(&), Fulvio Magni1, and Peter Flohr2

1
General Electric, Baden, Switzerland
pankaj.bajaj@ge.com
2
Zurich University of Applied Sciences (ZHAW), Winterthur, Switzerland
Abstract. The application of additive manufacturing to a unique performance

upgrade for the GT13E2 turbine fleet is discussed in this paper. Based on an
existing design for the turbine hot section, two components - a stator heat shield
and the first stage turbine vane - have been completely reimagined, in order to
fully exploit the potential of additive technology.
Both components contain cooling technology with intricate and small cooling
channels that are highly effective and allow highest combustion temperatures
with lowest cooling air consumption. These channels are however very costly to
manufacture with traditional casting methods. Additive manufacturing has no
such limitations on design complexity, and it is therefore a true enabler for this
application. The improvement in cooling effectiveness directly translates into a
performance upgrade for the entire turbine. For the stator heat shield, the overall
manufacturing process is substantially simplified by eliminating multiple
machining steps. For the first stage vane, a unique feature is the application of
the “coupon” concept, where only a segment of the traditional cast is replaced
by an additive part.
The technology has recently been validated in a customer powerplant, and
some of the key elements of the validation approach and findings are presented.
Keywords: Additive manufacturing GT13E2 Performance upgrade

Repair Selective laser melting (SLM)
1 Introduction
The use of 3D printing technology dates back to the 1980s for polymer applications,
but the ability to print functional components from metal alloys has spurred significant
interest and investment into AM over recent years.
Historically, polymer printing has controlled the majority of industry investment
and utilization for additive technologies. However, the future of AM is in production,
and the metals segment is already altering the balance between polymer and metal
printing.
Additive manufacturing is an exciting game-changer for many industries, including
the energy space, as it disrupts and accelerates the way products are engineered and

https://doi.org/10.1007/978-3-030-54334-1_35
A Performance Upgrade of an Industrial Gas Turbine 501
manufactured. Additive manufacturing will change the energy industry as we know

it—shifting workplace and skills requirements, evolving supply chains and business
models, speeding up development, and reducing costs.
The key feature of selective laser melting (SLM), also GE internally referred as
DMLM, is its capability to generate highly sophisticated part configurations based on
sliced CAD data. In contrast to investment casting even complex geometries can be
produced without any tooling or mold. This is facilitated by slicing the part into a series
of consecutive layers and converting the geometrical information of the cross-sections
into machine code. Based on this machine code a scanner guides the focused laser
beam across a 2D plane, where it selectively melts metal powder particles and fuses
them to previously generated layers. In this way this additive process can create
structures with a very high density >99.5% (Meiners 1999). After the melting step the
substrate plate moves down by an incremental distance corresponding to the nominal
powder layer thickness in a typical range from 20–50 µm. Thereafter the sequence is
finished with the deposition of a new powder layer. This cycle repeats, often several
thousand times until the last layer of powder has been fused, see Fig. 1. More details
can be found in Schurb et al. (2016).
Fig. 1. Principle of Selective laser melting (SLM)
One of the key advantages of additive manufacturing is the fact that manufacturing
cost is decoupled from part complexity, and that is a differentiator from standard casting
methods where cost scales typically with complexity. That is illustrated in Fig. 2. The
cost of a cast typically scales linearly with the weight of the part, and the gradient
increases for more complex components (curves (i) & (ii)). An identical part with
additive manufacturing has in most cases a penalty on cost due to the high equipment
cost etc., however the gradient is at first order independent from the complexity of the
part, curve (iii). This implies that an identical part of low complexity is typically more
expensive in additive (A), but this can often be mitigated by introducing weight
reduction measures in the additive part (C). Also, for more complex components, the
additive part actually becomes cheaper due to elimination for sophisticated casting
cores, or elimination of additional manufacturing steps like brazing or machining (B).
502 P. Bajaj et al.
ex
pl
m
co
cost
t,
as
)c
e
itiv
( ii
B
add
(iii)
A p le
sim
C ast,
(i) c
weight
Fig. 2. Schematic illustration for one the key features of additive– “complexity for free”.
With Additive Manufacturing, GE transformed how to design and manufacture

components and how to manage the supply chain. GE started exploring 3D printing
more than a decade ago, when its Aviation division tasked a group of engineers with
developing a highly efficient fuel nozzle for a new jet engine. Currently, GE has
already shipped more than 30.000 3D-printed parts for machines like the LEAP jet
engine, which uses the nozzle that started it all. Referring to the schematic above, for
the LEAP fuel nozzle 20 parts could be combined into one, five times the durability and
25% less part weight. In the Advanced Turboprop (ATP) engine for Cessna’s new
aircraft, GE’s design condensed 855 conventionally manufactured parts down to 12
additive parts. By now, the engineers are also 3D printing turbine blades for the GE9X,
the world’s largest jet engine.
GE Power is applying successfully the additive technology in several power gen-
eration turbines of its product portfolio. For example, the world-record efficiency of its
HA technology is driven by its advanced combustion system with additive technology,
see Goldmeer (2018).
This paper discusses the application of additive manufacturing to a unique per-
formance upgrade for the GT13E2 gas turbine model. Based on an existing, retro-
fittable solution for the turbine hot section, two components have been completely
reimagined, in order to fully exploit the potential of additive technology.
2 Additive Components in the GT13E2 Gas Turbine
The GT13E2 is one of the world’s most reliable gas turbines and also offers class-
leading efficiency for an E-class engine. Most recently, GE engineers were able to
increase the efficiency and performance of the engine by applying additive technolo-
gies. This upgrade is called “AMP” for Additive Manufactured Performance.
Fig. 3. GT13E2 MXL2 gas turbine
The baseline for this new performance improvement package is the GT13E2 MXL2
rating (see Fig. 3), designed and introduced in 2012 as a retrofit upgrade to the
GT13E2 (Magni et al. 2014). Currently 40 MXL2 GTs are in operation and have
accumulated >430.000 operating hours, >3.800 starts.
The new MXL2 with AMP includes two components produced by additive man-
ufacturing: the first-stage turbine vanes and the heat shields (see Fig. 4) at the outer
diameter of the first stage turbine blades. These parts are among the turbine’s hottest-
running parts, and the significant amount of cooling air they traditionally require
impacts the engine’s performance.
Fig. 4. (a) GT13E2 MXL2 turbine section; the arrow indicates the location of the additive parts
in the engine. (b) stator heat shield. (c) first stage vane with integrated additive coupon
Integrating advanced cooling technologies by means of additive manufacturing

results in an improved cooling effectiveness. This enables the engine operation at
higher firing temperature as well as a reduction of the required amount of cooling air to
the additive component. The result is an improved turbine performance which offers
our customers remarkable benefits.
504 P. Bajaj et al.
2.1 Stator Heat Shields

GE manufactured heat shields with near wall cooling scheme for the GT13E2 AMP gas
turbine. Near wall cooling as well as cooling improvement features were introduced
with the SLM printed design. The cooling effectiveness of the parts is increased and at
the same time the required cooling air mass flow reduced. An additional advantage of
the additive based design is that typically machined features such as seal slots, pockets
and cooling holes can now be directly integrated in the printing process. The design is
light weight and can achieve cost reduction via material savings.
The parts are produced in the GE additive factory in Greenville, USA. An example
of one set of prints is shown in Fig. 5; the optimization of the printing process, i.e.
making maximal use of the printing volume of the printer, is evident from the image.
The lead time for producing these components is reduced significantly compared to
conventional casting process, thereby supporting the customer with the part available in
a shorter time.
Fig. 5. GT13E2 gas turbine heat shield
2.2 First Stage Vane

For the first vane configuration the so called printed “coupon” concept was used. In
other words, the front part of the cast vane is removed and replaced by an additive
“coupon” with integrated cooling features. Comparable to the heat shield the near wall
cooling effectiveness will allow higher firing temperatures and therefore an improved
turbine performance. This technology can be applied to both, new components and as a
repair solution to increase the performance of the used component. These cooling
schemes could only be realized using additive manufacturing technology.
Figure 6 gives an overview of the cast vane body, printing coupon & the final
engine ready part. It is worth mentioning that this concept has been possible by
mastering not only the additive manufacturing itself, but also using a highly accurate
bonding of the coupon to the cast.
Fig. 6. GT13E2 MXL2 with AMP, first stage vane with coupon
3 Examples of Specific Design Challenges
The important step towards additive enabled designs is the substitution of today’s
conventionally manufactured parts and assemblies. Reaching this more advanced level
requires good knowledge of the conventional production process chain and operational
engine experience. Prior to replacing conventional manufacturing methods such as
investment casting by additive technology, an in-depth benchmarking and investigation
of relevant material properties is indispensable.
GE created design rules based on the experience of usage of additive manufacturing
in aviation industry and lessons learned during the development phase of additive
manufacturing of stator parts for the turbine. These rules determine the limits, con-
straints and opportunities of the new manufacturing approach.
The part functionality and interfaces with other components of the assembly defines
the possible build-up directions due to the following main aspects:
Overhangs. Printing the component using additive manufacturing, overhanging

regions should be avoided during the design phase for a specific build direction.
However, sometimes overhangs cannot be completely avoided due to functional sur-
faces or interfaces with other components. In these cases, supports need to be used.
These can either be included as part of the structure or are removed after the process.
Due to the geometrical complexity of many hot gas parts, it is usually not straight
forward to determine a perfect solution to avoid overhangs.
GE has therefore developed tailored CAD tools, which help designers to find an
optimum orientation. Despite these helpful SW tools, a good interaction between the
design teams and the SLM processing specialists remains essential to minimize process
iterations until a complex part can be built with adequate quality.
Morphing. Deformation can typically occur during the selective laser melting and
subsequent heat treatment process, having an impact on intricate design features and
potentially driving the need for additional machining stock. Due to high material cost
of nickel based super alloys and the requirement to minimize any additional machining
steps, the printed part is required to stay as close to the design intent as possible.
506 P. Bajaj et al.
In order to morph a part, a comparison between a 3D scan of printed hardware with

the CAD design model will reveal which sections need to be corrected. Based on the
deltas measured from the comparison, the design engineer will adjust the component
model for the DMLM process to morph in the opposite direction. Subsequent com-
ponents printed with the morphed model will deform during the process in such a way
that final part shape is towards the design intent.
For complex parts the trials showed, that the deformations are not necessarily linear
and also linked with each other so that morphing of one section alters the deformation
of another section. Due to this, multiple morphing iterations are deemed necessary until
the part fulfills the design specification and the process can be frozen. Figure 7 illus-
trates the schematic morphing process.
Fig. 7. Illustration of Morphing
In general, it is recommended to consult the additive manufacturing expert

immediately if uncertainties exist regarding a design to avoid working on unfeasible
designs.
In the specific case of the heat shield, the vertical build orientation of the com-
ponent has been varied with angles ranging from 0 to 15°. In fact, the left picture in
Fig. 5 shows an example of a trial build with different build angles. From these
variations the best orientation has been identified, first by establishing a build without
material cracking which may occur immediately during the material deposition or after
heat treatment, and secondly by verifying geometrical accuracy with minimal need for
morphing corrections. This has been an iterative process, and in practice, a substantial
experience of the additive process engineers is needed to make these iterations effective
and efficient. Figure 8 shows the outcome of the carried-out iteration to achieve the
desired dimensional results.
Fig. 8. Printed heat shield fulfilling the dimensional tolerances after several iterations
4 Engine Validation Results
The MXL2 AMP upgrade does not only rely on GE’s long-term experience in additive
manufacturing but in fact also on a substantial operation experience of additive parts,
specifically in the GT13E2 engine. A few additive parts were installed already in 2015
in the GT13E2 with the primary focus to validate the additive manufacturing tech-
nology. The components were produced with identical geometries as produced with the
precision cast process and the validation has been very successful (Hoebel 2018).
The validation of the new printed first stage components has been carried out in
several steps during the commissioning of the upgrade. Selected parts were equipped
with temperature and pressure measurement probes, to verify the temperature distri-
bution on the part surfaces to measure actual temperatures at the most critical locations.
Other parts have been prepared with thermal paint on the hot gas surface allowing a
detailed analysis of the upgrade performance through qualitative temperature maps, in
order to confirm that the distribution pattern corresponds to expectation. Figure 9
shows examples of the instrumented part.
Fig. 9. GT13E2 MXL2 with AMP, instrumented parts

508 P. Bajaj et al.
With the thermal paint tests the material temperature distribution of the new printed
parts have been confirmed. The cooling effectiveness of the brazed area also matches
with the models. Figure 10 shows the representative thermal paint results. The pressure
and temperatures measurements of the cooling air and material are used for the con-
firmation of the proprietary layout calculations and tools.
Fig. 10. GT13E2 MXL2 with AMP, thermal paint observed after operation
The validation campaign was successfully completed. The accumulated data and
results confirmed the validation of the additive components. The expected impact of the
additive components could be demonstrated with broad operation concept and lower
emissions. The performance targets of the upgraded gas turbine were achieved.
5 Summary
The GT13E2 AMP upgrade is a showcase example how additive manufacturing can
bring a new step-change in performance to a proven engine platform with large
operating experience. The improvements in cooling technologies could be successfully
implemented in two components of the turbine engine, to produce a substantial per-
formance increase for the gas turbine customer operational requirements in the new
complex power offering space. The additive technology was an enabler for this tech-
nology step and shows the way for smart solutions in the competitive environment that
our customers face today.
Acknowledgements. GE additive manufacturing team at GVL/Birr & Design Team at Baden.

Dietmar Kodim & Julian Ahlhoff, Vattenfall Wärme Berlin AG.
Matthias Hoebel, Professor at University of Applied Sciences and Arts Northwestern
Switzerland, for providing the technical expertise in the field of material properties and its
sensitivity on the DMLM process.
Andrew Passmore, Senior Product Manager, for supporting the project.
References
Meiners, W.: Direktes Selektives Laser-Sintern einkomponentiger metallischer Werkstoffe
(Doctoral thesis). Shaker Verlag GmbH, RWTH Aachen, Aachen (1999)
Schurb, J., Hoebel, M., Haehnle, H., Kissel, H., Bogdanic, L., Etter, T.: Additive manufacturing
of hot gas path component and engine validation in a heavy duty GT. In: Proceedings of
ASME Turbo Expo 2016, GT2016-57262, Seoul, South Korea (2016)
Magni, F., Zoli, R., Marx, P.: Betriebserfahrungen mit dem neuesten Upgrade der GT13E2
Gasturbinenfamilie; 9. VDI-Fachtagung – Stationäre Gasturbinen (2014)
Hoebel, M.: Direct Metal Laser Melting (DMLM) for Gas Turbine Applications; Stuttgart Laser
Technology Forum. University of Stuttgart, Stuttgart (2018)
Goldmeer, J.: Fuel flexible gas turbines as enablers for a low or reduced carbon energy
ecosystem. Presentation at Electrify Europe 2018 and GE white paper GEA33861 (2018)
Author Index
A E
Abdelwahab, Moustafa Mahmoud, 374 Eckhardt, Lukas, 223
Arvieu, Corinne, 177 Eichler, Fabian, 82
Eisenbarth, Daniel, 160
B Ekengren, Jens, 426
Bajaj, Pankaj, 500 Elspass, Wilfried J., 130
Balicki, Peter, 391 Engels, Gregor, 37
Baumgartner, Harry, 391 Eßig, Michael, 483
Bliedtner, Jens, 223
Bochsler, Janine, 399 F
Bos, Philip, 130 Favre, Sébastian, 96
Braschkat, Andrés, 321 Fiehler, Jens, 321, 437
Bremen, Sebastian, 82 Flohr, Peter, 500
Buhk, Jan-Hendrik, 321 Fontana, Filippo, 391
Frauchiger, Alex, 192
C Frölich, Andreas M., 437
Cao, Shuaishuai, 26
Clemens, Frank, 67, 293 G
Cloots, Michael, 192 Gardner, Leroy, 357
Comminal, Raphaël, 241, 251 Georgopoulou, Antonia, 67
Cui, Di, 96 Ghanbari, P. Gh., 268
Glas, Andreas H., 483
D Golab, Mark, 279
Dalmer, Christian, 112 Gorjan, Lovro, 293
De Sousa Guerreiro, Helena I., 437 Gutknecht, Kai, 192
de Souza Melo, Gustavo Menezes, 17
Dennig, Hans-Jörg, 304, 337 H
Ding, Andreas, 321 Hangst, Nikolai, 415
Dinner, Hanspeter, 337 Hirsch, Andre, 112
Duda, Tom, 399 Höfler, Florian, 469

https://doi.org/10.1007/978-3-030-54334-1
512 Author Index
Holdener, Simon, 130 Moritzer, Elmar, 112

Höller, Anton, 337 Moultrie, James, 279
Honigmann, Philipp, 26
Hopf, Andreas, 223 N
Hosseini, Ehsan, 268 Ngo, Ngoc Tuan, 321
Huber, Frank, 337 Nickchen, Tobias, 37
Huber, Marc, 399
O
J Omidvarkarjan, Daniel, 391
Jafarzadeh, Sina, 241 Ortona, Alberto, 52
Jahn, Simon, 82
Junk, Stefan, 415 P
Pandiyan, Vigneashwara, 205
K Pedersen, David Bue, 251
Karlsson, Patrik, 3, 426 Pejryd, Lars, 3, 426
Kasch, Susanne, 82 Pelanconi, Marco, 52
Keller, F., 268
Kirchheim, Andreas, 304, 337 Q
Kraenzler, Thomas, 146 Quang-Le, Tri, 205
Kratzer, Markus Johannes, 469
Krause, Dieter, 321, 437 R
Kuhl, Juliane, 321 Rettberg, Robin, 146
Kyselyova, Anna A., 437 Rigo, Olivier, 177
Rosenbauer, Ralph, 391
L Rushworth, Adam, 455
Lacoste, Eric, 177
Lanfant, Briac, 96 S
Larsson, Joakim, 426 Sarraf, Fateme, 293
Layher, Michel, 223 Schleifenbaum, Johannes Henrich, 17
Le Guen, Emilie, 177 Schmidt, Thomas, 82
Leparoux, Marc, 96 Schuh, Günther, 17
Li, Zhengyao, 357 Schulz, Josef, 293
Liersch, Antje, 293 Sebastian, Tutu, 293
Liu, Bingjian, 455 Serdeczny, Marcin, 241, 251
Löffel, Kaspar, 399 Sharma, Neha, 26
Lohn, Johannes, 37 Shevchik, Sergey A., 205
Lukas, Gerret, 17 Soffel, Fabian, 160
Lüscher, Patrick, 399 Spangenberg, Jon, 241, 251
Strömberg, Niclas, 3
M Sun, Xu, 455
Magni, Fulvio, 500
Marelli, S., 268 T
Masinelli, Giulio, 205 Thieringer, Florian, 26
Massey, Sam, 279 Thurn, Laura Katharina, 82
Mayer, Julian, 469 Tsavdaridis, Konstantinos Daniel, 357, 374
Mazza, Edoardo, 268
Meboldt, Mirko, 391 U
Menichelli, Alessandro, 160 Umbricht, Michael, 399
Meyer, Matthias M., 483 Urban, Nikolaus, 469
Author Index 513
V Wirth, Florian, 192

Vanderborght, Bram, 67 Wortmann, Nadine, 437
von Netzer, Barbara, 26 Wüthrich, Michael, 130
W Y
Wasmer, Kilian, 205 Yadav, Pinku, 177
Wegener, Konrad, 160
Weiss, Daniel, 399 Z
Welker, Dennis, 26 Zhang, Fangjin, 455
Wendt, Thomas, 415 Ziegler, Stephan, 17
Willkomm, Johannes, 17 Zumofen, Livia, 304, 337

Industrializing Additive Manufacturing: Mirko Meboldt Christoph Klahn Editors

Uploaded by

Copyright:

Available Formats

You might also like

Industrializing Additive Manufacturing: Mirko Meboldt Christoph Klahn Editors

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Industrializing Additive Manufacturing: Mirko Meboldt Christoph Klahn Editors

Uploaded by

Copyright:

Available Formats

Mirko Meboldt

Christoph Klahn Editors

ISBN 978-3-030-54333-4 ISBN 978-3-030-54334-1 (eBook)

The 2nd AMPA Conference (Additive Manufacturing in Products and

All scientiﬁc contributions are double-blind peer-reviewed by industrial com-

Christoph Klahn inspire AG, ipdz, Switzerland

Tobias Weber TRUMPF Laser- und Systemtechnik GmbH,

Anna Valente SUPSI, Automation, Robots and Machines

Solution Approaches and Process Concepts for Powder Bed-Based

Estimations of Interlayer Contacts in Extrusion Additive

Prototyping and Testing

Innovative Use Case

The AM Dowel – A Novel Insert for the Integration of Threads

Author Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 511

Patrik Karlsson, Lars Pejryd, and Niclas Strömberg(B)

Örebro University, 701 82 Örebro, Sweden

Abstract. In this work, generative design optimization and characteriza-

Keywords: Generative design · Lattice structures · AlSi10Mg

Fig. 1. The basic idea of the generative design optimization approach.

graphics operations, such as e.g. boolean, blending and morphing operations,

2 The Generative Design Optimization Approach

Gyroid f1 , G-prime f2 and Schwarz-D f3 lattices are considered by the fol-

f1 = sin(γx) cos(γy) + sin(γy) cos(γz) + sin(γz) cos(γx) ± κ,

Fig. 2. The triple periodic lattice structures.

The implicit surfaces in (1) representing our TPMS-based lattice structures

where the vector ρ contains density variables ≤ ρe ≤ 1, where is a small

where Ve represents the volume of element e and it is constrained by the upper

The topology optimization solution obtained by solving (2) using n = 4 is

where yi (x) represents a Kriging model, a radial basis function network or a

3 Validation of Tensile Tests

In order to perform design optimization using TPMS-based lattice structures as

Fig. 5. Tensile tests of lattice structures.

Fig. 6. Non-linear ﬁnite element analyses of tensile tests.

4 Characterization of Tensile Tests

Fig. 8. Fracture surfaces of samples with lattice structures observed in OM a) and

In Fig. 10, the results from 3D analysis of specimens by means of CT are

Gustavo Menezes de Souza Melo1(&), Gerret Lukas2,

Abstract. Additive Manufacturing (AM) has a great potential of disrupting

Keywords: Additive Manufacturing Design methods DfAM Product

© Springer Nature Switzerland AG 2021

The presented research utilizes qualitative research. AM mindset cannot be described

Table 1. Overview of interview participants.

proposed by Kitchenham, numerous literature studies were screened, with a focus on

Table 2. Examples from in-vivo codes, themes, and categories.

Throughout the literature review study, it is elaborated that an AM-adjusted design

4 Development and Validation of Framework

By means of a workshop, the above mentioned solution was presented to repre-

5 Conclusion and Outlook

Neha Sharma1,2 , Dennis Welker2,3, Shuaishuai Cao1,2,

Abstract. Sport-related injuries have an increased prevalence of maxillofacial

© Springer Nature Switzerland AG 2021

Keywords: Face-protective orthosis Additive manufacturing Face

2 Materials and Methods

In this section, an interactive digital design workflow for customized face-protective

Fig. 1. An overview of the schematic representation of the digital workflow.

2.1 Clinical Case Fracture Treatment

Fig. 2. Postoperative CBCT 3D volumetric reconstructions A: Bony 3D volumetric recon-

2.2 Computational Image Data Acquisition and Virtual Model