Professional Documents
Culture Documents
AM79 Cotter D. Implementing LFAM in Aerospace Tooling Via Process Integration and FEM of Print Performance. MIT 2021.
AM79 Cotter D. Implementing LFAM in Aerospace Tooling Via Process Integration and FEM of Print Performance. MIT 2021.
AM79 Cotter D. Implementing LFAM in Aerospace Tooling Via Process Integration and FEM of Print Performance. MIT 2021.
Author . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Philip D. Cotter
Department of Aeronautics and Astronautics, MIT Sloan School of Management
May 14, 2021
Certified by . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
R. John Hansman, Thesis Supervisor
Professor of Aeronautics and Astronautics
Certified by . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
John F. Carrier, Thesis Supervisor
Senior Lecturer, MIT Sloan School of Management
Accepted by . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Zoltán Spakovszky, Chair, Graduate Program Committee
Professor of Aeronautics and Astronautics
Accepted by . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Maura Herson, Assistant Dean
MBA Program, MIT Sloan School of Management
This page intentionally left blank.
2
Implementing Large Format Additive Manufacturing in Aerospace Tooling via
Process Integration and Finite Element Analysis of Print Performance
by Philip D. Cotter
Submitted to the MIT School of Engineering and MIT School of Management on May 14, 2021
in partial fulfillment of the requirements for the degrees of
Master of Science in Aeronautics and Astronautics and Master of Business Administration
in conjunction with the Leaders for Global Operations program
Abstract
Ascent Aerospace (Ascent) designs and manufactures a diverse array of customized aerospace tool-
ing, creating a low-volume/high-mix production environment where precision is critical. As a re-
sult, Additive Manufacturing (AM)—and, more specifically, Large Format Additive Manufacturing
(LFAM)—stands to provide Ascent a significant competitive advantage by reducing lead time, cut-
ting costs, and enabling the rapid production of novel tooling solutions. This project explores the
integration of the Large Scale Additive Manufacturing Machine (LSAM) into Ascent’s production
processes with the goal of maximizing the technology’s value impact. To this end, it focuses on two
components: understanding, controlling, and planning the production of LSAM-printed tools, and
simulating the behavior of LSAM-printed tools to better predict their performance.
First, a framework for the operational integration of the LSAM is developed. Comparison of tradi-
tional (current state) and LSAM-specific (future state) process maps provides a means to identify
and address probable bottlenecks. Next, a test plan is described enabling a clear understanding
of the LSAM’s capabilities and limitations. From these findings, Design for LSAM (DfLSAM)
Guidelines and various other continuous improvement initiatives are motivated and codified.
Next, this project develops an approach to the Finite Element Analysis (FEA) of LSAM-printed
objects. Current design principles are largely rooted in empirically calibrated processes which require
extensive trial and error. Due to the size of LSAM prints, this approach can be expensive and
unscalable. The FEA approach presented herein begins with characterization of material properties
of a common carbon fiber reinforced ABS feedstock. Based on these inputs, various modelling
approaches are explored for this anisotropic, composite material. Model outputs are then validated
against the results of physical experiments. An orthotropic solid modelling approach is shown to
compare best with physical reality, suggesting a promising direction for further development.
Organizational impacts and change management are considered throughout this document. Future
directions of both the integration and modelling work are also discussed. These findings, abstracted
from Ascent, comprise a framework for the implementation of LFAM in manufacturing operations.
3
This page intentionally left blank.
4
Acknowledgments
Thank you to the teams at AIP and Ascent Aerospace. Their willingness to rapidly on-board and
mentor me greatly accelerated this project. It directly enabled any useful findings this work has
produced. Their patience was particularly critical given the Covid-related impacts to travel. In ad-
dition to the broader Ascent organization, I would like to extend particular acknowledgement to the
LSAM Team at Ascent: Sean Henson, Mario Ruiz, Randy Moruzzi, and Dani David. Additionally,
thank you to Danny Davis, Tom Mangan, the AIP AM Team, and the additional Associates and
Partners who pushed and coached me along the way.
This thesis benefited greatly from the expertise and mentorship of my advisors, Prof. Hansman and
Dr. Carrier. Their commitment to translating academic research into real-world application was
immensely beneficial for myself, Ascent, and AIP. Without their wisdom I may still be lost down a
rabbit hole of dubious value.
To the Leaders for Global Operations Program, many thanks for this unique opportunity at the
intersection of engineering, management, and industry. The students, faculty and staff that this
program attracts are the source of its core value. I have learned much and look forward to engaging
with the community into the future.
Finally, thank you to my wife, Sarah. Without her love, support, patience, and counsel, none of
the work captured within this document would have been possible. Given the stress of long days,
frequent travel, Covid-19, and our first pregnancy, this was truly a team effort. As I perform the
final editing of this document in the labor and delivery ward of Brigham and Women’s Hospital, I
am immensely grateful and excited for our future.
5
This page intentionally left blank.
6
Note on Proprietary Information
Due to the proprietary nature of Ascent’s production processes (both traditional and additive),
some of the data in this document may be redacted, modified, or otherwise disguised so as to not
violate Non-Disclosure Agreements. Data, if modified, will be identified and treated in such a way
that conclusions presented in the document remain valid and logically follow.
7
This page intentionally left blank.
8
Contents
List of Figures 11
List of Tables 12
List of Acronyms 13
1 Introduction 14
1.1 Project Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
1.2 Problem Statement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
1.3 Project Approach and Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
1.3.1 Process Integration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
1.3.2 Print Performance Modelling . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
1.4 Thesis Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2 Background 23
2.1 Aerospace Tooling Industry Dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.2 Ascent Aerospace Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.2.1 Corporate History and Structure . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.2.2 Relevant Tooling Types . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.2.3 Traditional Production Process Summary . . . . . . . . . . . . . . . . . . . . 28
2.3 Large Format Additive Manufacturing . . . . . . . . . . . . . . . . . . . . . . . . . . 30
2.3.1 Foundations of Additive Manufacturing . . . . . . . . . . . . . . . . . . . . . 30
2.3.2 Thermwood’s Large Scale Additive Manufacturing Machine . . . . . . . . . . 33
2.4 Chapter Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
9
4.3.1 Tensile Tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
4.3.2 Shear Tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
4.3.3 Physical Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
4.3.4 Thermal Tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
4.4 Chapter Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
10
List of Figures
11
5-14 Modelled Deflection of 9ft Sub-Assembly in Z and X at Each Load Step . . . . . . . 82
5-15 Absolute Error for Sub-Assembly Experiments . . . . . . . . . . . . . . . . . . . . . . 82
5-16 Comparison of Various Modelling Approaches on 9ft and 6ft Sub-Assemblies via RMSE 83
List of Tables
List of Equations
12
List of Acronyms
3D . . . . . . . . . . . . . . . . . . . Three Dimensional
AM . . . . . . . . . . . . . . . . . . Additive Manufacturing
CPs . . . . . . . . . . . . . . . . . . Checkpoints
PC . . . . . . . . . . . . . . . . . . . Polycarbonate
PEI . . . . . . . . . . . . . . . . . . Polyetherimide
PESU . . . . . . . . . . . . . . . . . Polyethersulfone
PM . . . . . . . . . . . . . . . . . . Program Manager
13
Chapter 1
Introduction
Chapter 1 frames the reasoning for Ascent’s investment in Large Format Additive Manufacturing
(LFAM) and articulates the challenges this project aims to address. First, the high-level investment
thesis for LFAM and the Large Scale Additive Manufacturing Machine (LSAM) at Ascent is pre-
sented. Next, some of the core challenges of the LSAM’s integration are described, culminating in
a Problem Statement. Two primary Project Objectives, Process Integration and Print Performance
Modelling, are described to address this Problem Statement. Finally, this chapter concludes with
commercial, defense, and space sectors. The bespoke nature of its tooling offerings creates a low-
volume/high-mix production environment (a more thorough overview of Ascent and its products is
presented in Chapter 2). Ascent designs and builds large, complex, metallic tools that are frequently
manufactured as single units for single customers. SpaceX’s main body tool for the BFR (Figure
The monolithic nature of this tooling, its high complexity and requisite precision, and its
one-off nature generate a significant amount of variability in Ascent’s design and manufacturing
14
Figure 1-1: SpaceX BFR Main Body Tool (Tesla for scale) exemplifies large scale aerospace tooling
processes. This variability can make it challenging to implement traditional Lean Manufacturing or
Six Sigma approaches to reduce cost and increase operational efficiency. Further compounding this
issue, the commercial environment surrounding the aerospace tooling market — including increased
competition from other firms and the recent downturn in commercial aerospace more generally
— has increased the economic incentive to differentiate and innovate. Thus, Ascent’s challenge
is to determine how to remain the industry leader in aerospace tooling despite these headwinds.
LFAM, with its ability to directly print the complex geometries aerospace tooling requires, provides
LFAM enables Ascent to produce tooling with reduced lead time, lower cost, and potentially
opens up more direct access to alternative markets. Exposure to these other markets also creates a
synergistic effect, wherein lessons learned from automotive, maritime, or construction applications
could enhance Ascent’s ability to produce aerospace tooling as well—allowing Ascent to capitalize
once the commercial aerospace market returns to more traditional conditions. After an exhaustive
survey of the LFAM processes and machines available, led in part by the work of Jeffery Chu (MIT
LGO, ’20), Ascent made the decision to purchase Thermwood’s LSAM in late 2019 (Figure 1-2). The
machine was subsequently installed in March-April of 2020. Throughout this document "LFAM"
refers to the more general concept of additive manufacturing beyond the size of more conventional
desktop-sized machines. The term "LSAM" is used when specifically referring to the Thermwood
machine.
15
Figure 1-2: Thermwood’s Large Scale Additive Manufacturing Machine (10ft x 40ft x 5ft)
The LSAM is capable of printing with a variety of Carbon Fiber Reinforced Thermoplastic
(CFRP) feedstocks. Additionally, at 40ft long, 10ft wide, and 5ft tall, the LSAM has one of the
largest print volumes of any commercially available 3D printers. The pelletized CFRP materials
and their additively-manufactured process are both new introductions to Ascent. The material and
manufacturing process combine to create complex anisotropic properties in the printed materials.
In order for Ascent to fully realize the value of the LSAM’s introduction, the performance of LSAM-
printed parts must be fully understood. This will enable designs to be optimized and reduce the
From a process perspective, it is important to note the LSAM utilizes a hybrid manufacturing
approach: it prints at near net shape and requires subsequent machining to achieve final contours
and tolerance. Thus, the LSAM is reliant on existing resources within Ascent that are shared by
the traditional manufacturing process. These resources include: sales and marketing effort, project
management, design time and software, machining capacity, assembly, shipping, and more. Even if
Ascent fully understands the print process parameters and can accurately predict the performance of
printed parts, the challenge of process integration still remains. Ascent must determine the best way
to allocate these shared resources such that LSAM production is enabled without any degradation
16
1.2 Problem Statement
itself. Especially in times of crisis, a company may latch on to a novel product or service with the
hope that it will resolve or ameliorate its problems. However, this naive approach clearly neglects
the deliberate work necessary to understand and integrate a new technology into existing processes
and culture. As a result, AM and transformative technologies like it, frequently get lost in what
the Gartner Hype Cycle terms the "Trough of Disillusionment" [15] (Figure 1-3). While the Hype
Cycle has been criticized for overly simplifying the process of technology adoption, acknowledging
this phenomenon and wrestling with it directly are critical for lasting and meaningful adoption
Figure 1-3: Poorly Adopted AM can Lead to Stagnation in the Trough of Disillusionment [15]
LFAM integration at Ascent, with its new materials, new manufacturing process, and new
design considerations, must be adopted and assimilated with care, accounting for its effects on
process, product, and people. The core problem is relatively straightforward: how does Ascent
transform plastic pellets into products that customers demand (Figure 1-4). However, given the
novelty of this manufacturing process and its utilization of shared resources, this is a multi-faceted
Viewed from a high level, this project in many ways represents a New Product Development
(NPD) initiative, including all of the commercial implications: marketing, sales, pricing, revenue
modelling, etc. While all of these aspects were extensively explored during the author’s time with
17
Figure 1-4: Core Problem: Transforming Plastic Pellets into High-Performance Products
Ascent—and they are absolutely necessary to truly generate business value from the machine—
these commercial aspects have been largely excluded from the discussion in this document due to
their proprietary nature. The focus, instead, is on the operational integration of the machine into
the production process. Still, even excluding explicit discussion of the commercial impacts and
the broader AM-adoption dynamic, much can be learned from Ascent’s experience integrating the
Ultimately, this project aims to define the currently vague process for producing LSAM-
printed tools and operationally integrating this new capability into an existing manufacturing en-
vironment. As part of this wider integration initiative, it also explores developing a quantitative
model for the behavior of printed parts. These initiatives, taken together, will enable operational
excellence and the production of high-quality tooling via this novel additive technology, maximiz-
ing value for Ascent. Further, it will serve as a detailed case study providing a framework for the
integration of LFAM into other manufacturing environments, demystifying the process for potential
adopters.
This project began immediately prior to the physical installation of the LSAM at Ascent’s
facility. This provided a unique opportunity to witness the installation, initial testing and validation,
and first production tools made on the LSAM during the course of this project. It also enabled clean-
sheet process development. While the project began with certain assumptions about the future-state
process for producing LSAM tools, these assumptions evolved as the team’s understanding of the
machine’s and the organization’s capabilities and limitations developed. Consequently, the scope
18
Initially, the project was focused on defining the manufacturing process for the LSAM and
what changes to the traditional, metallic process would be required. However, in developing and up-
dating this future-state process map, multiple potential bottlenecks and issues emerged, generating
altogether new avenues of exploration. Thus, this thesis first explores the Process Integration and
selected operational improvement initiatives that were developed during the LSAM’s introduction.
Next, motivated by the identification of the shortcomings in the design process, the thesis explores
As with most new technologies or capabilities, the LSAM does not exist in isolation. Rather,
it exists within a pre-existing production system at Ascent. This pre-existing system was optimized
for a traditional manufacturing process and thus needs to be adapted to accommodate the LSAM.
The goal of this portion of the investigation is to characterize the operational environment of the
LSAM as a holistic process—from tool design to tool delivery—as well as understand, control, and
In order to accomplish this, first, a future-state process map was developed and explored.
This process map—with a clear understanding of the upstream and downstream inputs and outputs—
was used as the foundation for the identification of potential process bottlenecks. Importantly, it
represents a hypothesis of the future-state LSAM-production process and was continually revisited
throughout the project. Next, focusing more centrally on the LSAM as an individual machine, a
series of operational improvement initiatives are discussed to better define the impact of its print
This portion of the thesis goes from a high-level, factory-centric perspective to a lower-level,
machine-centric perspective to provide the full context. The objective of the Process Integration
initiative is to characterize the process and set develop systems for operational improvement such
that the process can be optimized once the machine is operating at production rate.
19
1.3.2 Print Performance Modelling
The Process Integration initiative identified the design process as being a potential bottleneck
and source of variability. A significant driver of this is not understanding the behavior of parts
printed on the LSAM. While Ascent has tremendous expertise using quantitative tools to understand
and model the behavior of their traditional, metallic tools, tools produced on the LSAM are not
well characterized. As a result, designers are forced to essentially "guess and check" whether their
rework, and an increased risk of tool failure after delivery to a customer. This iterative approach is
This portion of the thesis begins with first principles from materials science. While Ascent
has supplier-provided data sheets for the physical and mechanical properties of its various feedstocks,
these values need to be more fully understood—in fact, no standards currently exist for the testing
of LFAM materials because ASTM International (ASTM) tests are not appropriately scaled. Thus,
a testing regime is developed for a common feedstock to independently identify and compare these
properties. The same testing regime used for this feedstock could be used for other feedstocks
elling approach is developed leveraging Ascent’s existing FEA simulation software. Beginning with
simple tensile specimens and progressing toward more complex geometries, this Print Performance
Modelling initiative aims to create a software simulation of the performance of LSAM-printed tools
• Chapter 2: Background
This chapter provides the context necessary to understand the remainder of the document. It
begins with an overview of the structure and dynamics within the aerospace tooling industry.
20
Next, Ascent Aerospace, including its history, tooling offerings, and traditional manufacturing
greater AM industry, is described. Specific attention is given to Thermwood’s LSAM and its
capabilities.
This chapter is primarily concerned with the operational impacts of the LSAM from both
a literature review of existing frameworks for integrating additive manufacturing and NPD
is conducted, providing the academic foundation for the investigation. Next, a future state
process map is used to identify bottlenecks and prioritize integration efforts. Finally, a series
of continuous improvement initiatives are described which will help to reduce variability in
This chapter explores the development and execution of a material testing regime for LSAM-
printed objects. First, a literature review of existing attempts to characterize Three Dimen-
sional (3D)-printed materials is conducted. Based on these initiatives and the lack of ASTM
standards, modifications are suggested to the testing procedures. Tests are performed on
Acrylonitrile Butadiene Styrene (ABS), a common CFRP feedstock. After executing this test
This chapter presents an approach to modelling LSAM-printed objects in software using FEA.
It begins with a literature review of existing approaches to modelling 3D-printed materials with
a focus on fused deposition modelling (FDM) technologies utilizing CFRP. The simulation’s
output is compared against two empirical experiments to judge its validity and suggest a
This chapter suggests directions for further investigation and summarizes this project’s key
21
findings. It posits that, by using Ascent’s experience as a detailed case study, the processes
and approaches suggested in this thesis could be leveraged by other manufacturing operations
22
Chapter 2
Background
Chapter 2 provides background on the industry, company, and technology that form the context
for this thesis. The aerospace tooling industry in general, Ascent Aerospace specifically, and AM
technology are reviewed. Special attention is given to LFAM and Thermwood’s LSAM to set the
stage for detailed discussion of the print process and printed-object behavior in future chapters.
Aerospace tooling, in the most general sense, is any piece of equipment found on the man-
ufacturing floor of an aerospace Original Equipment Manufacturer (OEM) that is used to assist in
the production of an aircraft or spacecraft. Tooling is separate and distinct from "fly-away" parts,
which are those pieces of equipment or items intended to be a final component of the aircraft or
spacecraft. Tooling is typically used to produce fly-away parts more quickly and precisely than
would otherwise be possible without it. This tooling can take many forms: composite layup molds,
trim and drill fixtures, mill fixtures, shop aids and jigs, work stands, automated fastening systems,
robotic part manipulation solutions, and assembly line integration among others (Figure 2-1).
The primary customers of aerospace tooling manufacturers are aerospace OEMs. These
include any company that produces aircraft or spacecraft at any meaningful scale. These customers
can be further grouped into the commercial, defense, and space sectors. Each of these sectors
23
Figure 2-1: Ascent Tooling Examples (from L to R: work stand, drill fixture, layup mold) [5]
are driven by separate market dynamics that dictate things like rate, tolerance, price, and other
is generally a "lumpy" business, wherein business comes in somewhat unpredictably, and it can be
difficult to forecast demand on a sub-annual time scale. This is because OEMs typically "tool-up"
for a production run on a particular program or aircraft and then require minimal tooling support
afterwards. "Tooling-up" occurs when an OEM has finalized a design, knows that rate production is
imminent, and begins to accumulate the tools necessary to support their anticipated demand. The
"tool-up" can predate the production run by multiple years in some cases in part due to the lead
times required to produce and vet tooling. In this way, the aerospace tooling industry is a leading
indicator of aircraft and spacecraft production quantities. Because there simply are not that many
new models of aircraft and spacecraft being produced on a sub-annual basis requiring new tools,
the business can arrive in clusters of demand when new programs mature beyond a certain point.
However, even once tooling is required, aerospace tooling is often needed to support very
small (<10 unit) production runs. This means that the cost of all of the tooling’s material, design,
and manufacturing is amortized over very small quantity, making the cost of goods sold relatively
high for a tooling manufacturer. An increasing number of entrants into the aerospace tooling man-
ufacturing market capable of meeting OEM requirements has made competition for this tooling
business increasingly stiff. This competition has caused an erosion in margins across the aerospace
tooling industry, exacerbated as the commercial aerospace OEMs have been impacted by the com-
bined effects of the Boeing 737 MAX crisis and Covid-19. Thus, the cost of tooling is a primary
concern for aerospace OEMs who have multiple potential suppliers in many cases.
In addition to cost, aerospace OEMs value a number of other important characteristics in their
24
tooling, namely: lead time, design flexibility, dimensional tolerance, weight, and durability. Even
to produce. This is typically because the tooling must meet very tight dimensional tolerances on
complicated geometries over very large areas. These tight tolerances are necessary because the
tools are used to manufacture fly-away parts. With modern aerospace requirements, flight surfaces
and other critical parts are often required to meet dimensional specifications of less than a few
thousandths of an inch over a large surface. Aerospace tooling must match these requirements in
order to be useful. This translates into long lead times due to the need to cut, bend, and weld large
pieces of metal.
They are typically able to meet tolerances of a few thousandths of an inch over surfaces on the
order of hundreds of square feet in area. However, this precision comes at a cost: long lead times,
design needing to be locked in far in advance of delivery date, exceedingly heavy tools, and thin
AM, properly implemented, could meaningfully change the dynamic in the aerospace tooling
industry. AM’s most significant impact will likely be in lead time reduction. Aerospace OEMs
value lead time reduction because it enables them to simplify and accelerate entire program life-
cycles. Shortening tooling lead time enables OEMs to wait longer before locking in designs and thus
reduces wasteful rework when changes occur. Additionally, because AM significantly automates
large portions of the manufacturing process, a tooling manufacturer could realize more consistent
dimensional tolerancing, utilize lighter-weight non-metallic materials, and improve cost structure.
It should be noted that AM—and specifically the LSAM studied in this thesis—will not
replace or even address certain portions of the aerospace tooling market. LFAM is mostly relevant
to large scale (>1 m3 in bounding box volume) traditional tools. A more lengthy discussion of the
25
2.2 Ascent Aerospace Overview
Ascent Aerospace is a leading supplier of aerospace tooling and serves the commercial, space
and defense markets. Its corporate structure, in-house design expertise, and existing manufacturing
capabilities are important to understand when discussing the integration of LFAM. This section
provides context necessary to understanding the existing state of the business, prior to the LSAM’s
installation.
Ascent was purchased by the private equity firm American Industrial Partners (AIP) in 2012.
It has evolved over the last eight years through a series of acquisitions and exits and is comprised
of some legacy tooling manufacturers who have been in operation for multiple decades. This roll-up
strategy has created a vertically-integrated tooling solutions provider, essentially providing a "one
stop shop" for aerospace OEMs. While based in Michigan, Ascent has operations in California, New
The company has deep design expertise, which helps to differentiate it from competitors. As-
cent is able to consult with a customer, determine requirements, design, and build tooling solutions
in-house. This is a critical capability that can be leveraged in the integration of LFAM as well.
Ascent has over 750,000 ft2 of manufacturing space including more than 30 5-axis machines. The
Ascent is essentially broken into three divisions by product: tooling, automation, and inte-
gration. LFAM, due to the constraints of the process and the materials, is intended to primarily
serve the tooling segment. Within the tooling segment, there are four primary categories of tooling
which each have their own purpose, operating conditions, and materials:
• Mold Tooling: Mold tooling provides the surface upon which non-metallic parts—often carbon
fiber or fiber glass—are laid up to achieve their intended shape. The shape of the "facesheet"
26
of the tool is imparted on the shape of the final part. The facesheet is supported by a backup
structure, typically an "egg-crate" or a billet (Figure 2-2). These tools are often designed to
be placed in an autoclave or oven in order to go through a cure cycle with the part’s final
material. As a result of this thermal cycling, mold tooling is made from a variety of specialized
materials in order to match the Coefficient of Thermal Expansion (CTE) of the tool and part:
aluminum, steel, Invar, carbon fiber itself, and hybrids. Accounting for the CTE of the tool
and part is critical to achieving final part tolerances. Additionally, mold tooling must typically
• Trim and Drill Fixtures: Parts that come off of mold tooling typically have rough edges and
need to have secondary machining processes conducted in order to achieve their final useful
state. These additional machining processes can include trimming the part’s edges as well as
cutting grooves or drilling holes at predetermined locations. Trim and drill fixtures hold the
parts in place while these processes are conducted. Consequently, trim and drill fixtures have
very similar facesheet geometries to their respective mold tooling. In fact, mold tooling and
trim and drill fixtures are frequently sold together as part of a package. Because trim and
drill fixtures are typically only used at room temperature, they do not need to be made of
more exotic materials like Invar or carbon fiber. They are, therefore, generally constructed
from aluminum or steel. Trim and Drill fixtures are required to maintain vacuum integrity as
well in order to secure the part in place for these secondary machining processes.
• Assembly Fixtures: Assembly fixtures are used by OEMs to properly align and orient parts
while additional operations are performed. These additional operations are often joining
27
operations like welding or fastening. These are frequently large and complex tools comprised
of multiple joints, brackets, and clamps. These components are typically made from aluminum
or steel.
• Work Platforms or Facility Tooling: Because aerospace parts and the flight vehicles they
comprise can be so large, OEMs require custom-built structures in order to access parts of
the vehicles during manufacturing. These work platforms have unique geometries and must
support the weight of OEM personnel and their equipment. As a result, structural integrity
LFAM is initially targeted for application in the Trim and Drill Fixtures and Mold Tooling
portions of the tooling business. These are typically monolithic metallic tools with complicated
geometries that lend themselves well to the advantages of LFAM. However, it should be noted
that LFAM also stands to create value in the automation segment of Ascent’s business as well.
For example, light-weighting robotic end effectors, frames, and fixtures could produce significant
performance benefits. As Ascent continues to refine its understanding of the design space, more
tools and applications will become relevant and achievable, increasing the utility of LFAM over
time.
In this section, the traditional production process for Trim and Drill Fixtures and Mold
Tooling is summarized from initial customer interface through tool delivery. As these are the initial
primary use-cases for LFAM application, it is important to define and understand the current-state
manufacturing process. The manufacturing process for both types of tools is functionally identical
for the purposes of this discussion. The details of the process have been obscured and simplified to
• Inbound: These steps capture the front-of-house activities necessary to transform a request
for quote from a customer into usable inputs for engineers to begin work. This process is
28
Figure 2-3: Process Flow Diagram, Traditional Metallic Tooling (Current State)
relatively streamlined in traditional tooling, because customers know what to expect, account
managers understand the product, estimators know how to price it, and project managers
know how to package the necessary information for downstream work centers.
• Design: These steps capture the initial design and engineering work performed on a tool.
Ascent performs both design-and-build (where Ascent designs the tools from scratch based
amount of design time on a given project. However, because Ascent is familiar with these
materials, their manufacturing process, and has necessary software, the process is able to
efficiently adapt.
• Manufacturing: Physically building these tools is the most time-consuming part of the produc-
tion process. As described in Section 2.2.1, Ascent is vertically-integrated, meaning they are
capable of transforming raw material inputs into finished tooling outputs without the need to
source and assemble parts from outside suppliers. This translates into a lengthy process and
is the primary source of lead time in this category of tooling. The initial steps are essentially
process—which involves using a brake press to form the facesheet from a flat plate of metal
to the desired contour—is a unique source of variability. This process requires a highly skilled
29
considered more art than science and an apprenticeship-type model is required to become
proficient. Machining is yet another key driver of lead time, as cutting metal to these precise
tolerances can be a lengthy process (often on the order of tens of hours). The final steps involve
assembly of additional components, hand benching, finishing, quality checks, and shipping.
From this traditional process map, it is possible to visualize the significant number of steps
required in the manufacture of traditional trim and drill fixtures and mold tooling. Ultimately, this
can drive lead times to be in excess of 12 weeks for a standard-sized tool in this category. From a
process flow standpoint, the "lumpiness" of the market dynamics and the variation in the geometries
and sizes of the tools moving through the shop make this a difficult production process to manage
and optimize.
AM stands to help address some of the key manufacturing issues identified in the preceding
section and create a significant competitive advantage for Ascent if it is implemented correctly. Now
that an understanding of the industry and company context have been described, this section will
focus on the AM technology itself. First, a basic grounding in AM will be presented to provide a
common understanding and vocabulary. Thermwood’s LSAM will then be described in detail.
of professional manufacturing contexts—have been under development since the 1970s [36]. AM
enables a user to manufacture physical parts directly from a 3D digital file by successively applying
layers of material. ASTM International defines AM as "a process of joining materials to make
objects from 3D model data, usually layer upon layer, as opposed to subtractive manufacturing
methodologies" [36]. Figure 2-4 shows a generalized overview of the AM process from a Computer
Aided Design (CAD) model all the way through to finished part. While AM was originally used
for rapid prototyping, the technology has matured sufficiently to enable direct manufacturing of
finished parts.
30
Figure 2-4: Generalized AM Process Flow [36]
Whether prototyping or manufacturing finished parts, the AM process flow generally appears
similar. First, a part is designed in a CAD software suite—Ascent utilizes CATIA, but essentially
any software that can output .STL files or similar 3D data file can be employed. This design process,
as will be explored in the LSAM’s context in Section 3.5, is driven by the constraints of the AM
technology and material being utilized. By accounting for the AM technology, material, and end
use, an additively manufactured solution can be optimized for printability and performance. Next,
software—typically called a "slicer"—uses information specific to the printer being utilized to cut
the 3D model into successive "2D" layers. These layers are translated into G-Code, which commu-
nicate to the machine the instructions necessary to execute the print program. With these print
instructions, the printer is able to print the desired object using the intended material. Following
the print process, post-processing is often required to remove support structures, relieve residual
stresses, or achieve a desired surface finish or dimensional tolerance. In the case of an aerospace
tool, the post-processing required to transform a raw print into a finished tool can be extensive and
sions. These traditional methods are termed formative and subtractive [35]. Formative methods,
like injection molding and sand and investment casting, typically involve a process that forms a
desired finished-part material around some kind of tooling to achieve its desired geometry. Sub-
tractive methods, like milling, drilling, and machining, typically involve removing material from
a block or billet of a standard shape to achieve desired geometry. AM can eliminate or reduce
the cost of the tooling required to support formative methods and eliminate the material waste
required to support subtractive methods. There are many other potential benefits that AM unlocks
for businesses: reduced supply chain risk, simplified manufacturing processes, part consolidation,
31
"free" customization and geometric complexity, light-weighting, faster speed to market, virtual
Over the course of the last 40 years, the relatively straightforward concept of AM has evolved
into myriad methods: binder jetting, directed energy deposition, material extrusion, material jetting,
powder bed fusion, sheet lamination, and vat photopolymerization [35]. For each of these methods,
there are multiple material types: filaments, pellets, liquid resins, powder, plastics, metals, etc.
For each of these method-material combinations, there exist multiple OEMs offering printers with
varying degrees of controllability, consistency, and other options. Importantly, most of these printer
options have build volumes that are < 1m3 . Using this framework, the LSAM can be said to utilize
the material extrusion method and a pelletized, carbon-fiber reinforced, thermoplastic material.
The LSAM’s crucial differentiator, as discussed in Section 2.3.2, is its massive size.
ation for employing AM in a business context. The value of AM can be difficult to quantify given
compare the cost of an AM part directly with a traditionally manufactured alternative. While the
value model for AM undoubtedly varies with the AM technology, material, and industry, there are
In this equation, Ctot represents the total cost per part via AM. Cmachine represents the
cost of the machine amortized over an individual part, which would decrease with quantity of parts
produced and also include necessary maintenance and utilities. Clabor includes all of the direct labor
costs associated with the part’s pre-processing, printing, and post-processing. Cmaterial accounts for
all of the material used to produce the part. As a result of this cost model, most comparisons in
literature suggest that AM—when not employed to achieve an AM-unique performance benefit—
32
Figure 2-5: Cost Comparison of AM and Traditional Methods [10]
Clearly, there is a quantity beyond which AM is not necessarily the cost-effective manufac-
turing method. With AM, the cost per part does not drop significantly as number of parts increases.
As a result, it can be reasonably concluded that AM is unlikely to ever fully replace traditional man-
ufacturing methods. With respect to aerospace tooling however, quantities produced per design are
typically 1:1, highlighting the potential cost advantage to leveraging AM in this context.
Another shortcoming of AM in most cases is variation in print process and its effects on the
material properties of the printed part. Because print parameters and ambient effects can influence
the inter-layer bonding of additively manufactured parts, process control is especially important.
Without controlling the process, it is difficult or impossible to control ultimate part quality. This
With an understanding of the aerospace tooling industry’s dynamics and its internal tradi-
tional tooling production process, Ascent recognized an opportunity to leverage AM to create value
in its business. However, only a printer with a build envelope large enough to accommodate its
monolithic products would be useful. Thermwood’s LSAM meets this size requirement and also
prints with the materials and precision necessary for aerospace tooling applications.
Figure 2-6 identifies some of the key components of the LSAM. The machine has a 10ft
wide, 40ft long, and 5ft tall print volume, making it amongst the largest commercially available.
Additionally, it has a dual-gantry system, which enables it to print to near net shape and machine
33
Figure 2-6: LSAM’s Functional Components [34]
to final tolerance in the same footprint. Some of the key features and their implications are captured
in Table 2.1.
It is important to note that the LSAM does not have the ability to print any "support
structures" like are common with other AM technologies. As a result, the LSAM must use its own
primary print material as support structure. This dynamic, as well as other operational issues are
discussed in more detail in Chapter 3, as they were discovered and captured as part of a deliberate
34
2.4 Chapter Summary
In Chapter 2, the full context for integrating LFAM into Ascent’s manufacturing operation
was presented. First, a summary of the aerospace tooling industry was described. Next, Ascent’s
role as a leading tooling supplier was discussed as well as the segment of their business that LFAM
is intended to impact first. Lastly, AM and Thermwood’s LSAM were introduced to provide an
understanding of their processes and capabilities. The LSAM was purchased with the intent of
reducing lead time and cost for Ascent, creating meaningful differentiation in a dynamic aerospace
tooling environment. With an awareness of this context, explorations of the production process,
35
Chapter 3
Chapter 3 describes a framework for operationally integrating the LSAM into Ascent’s production
process. Rather than incorporating the commercial and market based dynamics that would be
essential to an NPD framework or discussion of AM adoption more generally, this Chapter focuses
on factory-level implementation from an operational perspective. In some ways, this Chapter and
its findings remain a hypothesis that is continuously being updated by Ascent. Given that the
machine was installed in April of 2020, only four months of deliberate testing and validation have
occurred as of this document’s writing. However, in that time, a framework for the best practices
for integrating LFAM into a manufacturing setting have emerged. This chapter will begin by
addressing existing frameworks for the adoption and integration of AM. Focus will be placed on
organizational impacts and best practices to foster efficient adoption rather than market-based
considerations. Next, a discussion of the LSAM-specific, future-state process map will highlight the
changes from the traditional, current-state map and their operational implications. The next three
sections describe the testing, training, and continuous improvement initiatives found to be most
effective and impactful in the integration of the LSAM. This chapter provides the foundation for a
36
3.1 Existing Frameworks for Integrating Additive Manufacturing
As discussed in Chapter 2, AM was initially used as a rapid prototyping tool. This remains
true and valuable for significant portions of the AM market. However, as AM technology and ma-
terials have matured, rapid manufacturing of finished products or end-use parts has emerged as a
viable application [26]. Much of the existing body of research on the adoption of manufacturing-
specific AM considers the entire business case. This research focuses on questions like determining
whether a given firm would benefit from AM [4], making "make-versus-buy" decisions [31], under-
standing AM’s effects on supply chains [17], or even mapping and choosing which AM technologies
would be applicable for a given use-case [3]. While understanding these implications are undoubt-
edly critical for Ascent and other firms to understand, the discussion in this section focuses on
frameworks for operationally integrating a single type of AM into an existing factory or production
system. Ascent has already made the investment in a single type of LFAM. The discussion assumes
that the market-facing business case is sound and that LSAM-printed tools are, in fact, desired
by customers. Thus, the investigation here centers on optimizing this machine’s integration rather
than making higher-level make-vs-buy choices or other decisions around optimal selection of AM
process or materials.
Martens et al. use multiple case studies to propose successful approaches for implementing
AM in a manufacturing context. Three themes emerge from their research regarding firms that
have successfully adopted AM: these firms identify business opportunities, experiment with AM
technology, and embed AM technology [24]. The first step in their approach involves identifying
a compelling business case by understanding the market and how AM might create a competitive
advantage for the firm. In Ascent’s case, this analysis was summarized in Chapter 2 and is assumed
to be sound. The second step in this framework involves experimenting with AM technology to
"pilot" it, whether in-sourcing the AM technology or working through outside vendors. This fosters
the familiarity and expertise necessary to make informed decisions about the technology from a
company-specific perspective. In Ascent’s case, the decision was made to in-source the technology—
this decision was made even simpler because contract manufacturers simply do not exist for printing
at this scale. The last step, "embedding" AM, is the most applicable to the discussion around
37
operational integration.
Martens et al. suggest that in order to embed AM, a firm must conduct business model
innovation and technology adoption. Stated differently, they must change their business practices
and manufacturing practices. Zanetti et al. suggest that small manufacturing enterprises, especially,
[39]. This is because embedding AM requires new employees with different expertise, possible
establishment of new autonomous business units, and adherence to unchanging quality standards.
In other words, proper adoption requires significant, wholesale changes to existing practices to access
Cremona et al. take a deeper look into the integration of new AM-required resources into
a firm’s production process. They divide these resources into tangible and intangible categories.
Tangible resources include the printers themselves, software (CAD) compatibility, and new person-
nel. The intangible resources are primarily the production "know-how" required to produce quality
products. Cremona et al. are clear to highlight this "know-how" as "one of the most valuable intan-
gible resources for a company" which is "fundamental for the future to not only be competitive, but
to survive in the market environment [11]. This division between tangible and intangible resources
is a helpful mental model when planning the integration of AM into an existing business.
production systems. They conclude that the transition from traditional manufacturing to AM
raw materials" [28]. Expanding on this idea, they further suggest that "new know-how in R&D is
required to seize the greater freedom of design enabled by AM technology adoption." From a process
flow perspective, they also observe that firms may "transition from single unit to batch production"
in the production of customized objects. In other words, AM may have "far-reaching implications
not simply buy a 3D printer and become an additive manufacturer. From the literature, it is clear
that in order to unlock all of AM’s value creation ability, firms must re-evaluate both their business
38
processes and manufacturing processes. This re-evaluation should include deliberate thought about
tangible and intangible resources, with particular attention paid to methods for capturing production
"know-how" to ensure quality and consistency. Additionally, firms must understand that operating
AM processes simultaneously with traditional manufacturing processes may create flow management
process map is developed to better visualize the necessary changes in the LSAM-specific production
process (Figure 3-1). Because the LSAM is still not generating work-product at rate-production
levels, the process map seen in Figure 3-1 represents a kind of hypothesis about the flow of LSAM
jobs through the factory. However, the map remains a helpful tool in performing a difference analysis
and determining the scope and nature of changes to Ascent’s business and manufacturing processes.
Further, by analyzing this future-state process map, it is possible to preemptively identify potential
First, it is helpful to describe the changes in the LSAM-specific process shown in Figure 3-1
compared with the traditional process shown in Figure 2-3. At a high level, the process can still be
grouped into the same three categories: inbound, design, and manufacturing. The core value add
of the LSAM—lead time reduction—can fairly quickly be visualized: the number of manufacturing
steps is reduced from 13 to eight. Through eliminating six metalworking steps in traditional tooling
and replacing them with printing at near-net shape, the LSAM can dramatically reduce the time
required to produce a tool. However, this map also highlights potential areas of concern, which are
discussed by category in the following paragraphs and are summarized in Table 3.1.
Beginning in the Inbound category, the primary bottleneck identified is inbound customer
demand. While seemingly trivial, it is critical to consider that integration of the LSAM will be
fundamentally impossible without the customer demand necessary to develop and improve processes.
Because this is a somewhat novel product offering to Ascent’s customer base, deliberate effort must
be focused on educating customers. The upstream effect of this is that Ascent must also train its
39
Figure 3-1: Process Flow Diagram, LSAM-Specific (Future State)
sales and marketing teams in the LSAM’s capabilities, constraints, and value to customers. Ascent’s
approach to addressing this bottleneck includes training of a core cadre of LSAM operators. The
Program Manager (PM) responsible for the LSAM scheduled training events with the sales and
marketing teams tailored to ensure they understood the machine well enough to prime customers
in their routine interactions. Augmenting these training events, the PM also conducted the initial
engagements with most primary target customers and hosted multiple virtual launch events. The
number of LSAM quotes received and number of LSAM-specific engagements with customers were
identified as Key Performance Indicators (KPI) and continue to be tracked closely to determine the
The LSAM-specific Design category shows two primary changes from the traditional process:
the insertion of the Manufacturing Engineering (initial slicing) step and the feedback loop from
Stress Analysis back to Mechanical Design. As discussed in Section 2.3, slicing is a critical step
in determining the process path of a additively manufactured object. The process proposed here
suggests that immediately following a draft mechanical design, that shape then be sliced to ensure
it can be printed. This gives the Manufacturing Engineer who has the best understanding of
the LSAM’s capabilities a chance to review the design for printability prior to engaging the Stress
Analysis team. Following the initial slicing, the Stress Analysis team can then ensure that the design
meets identified performance criteria (deformation under mechanical load, thermal compensation,
etc.). Output from the Stress Analysis step can then be fed back to the Design Engineering team
40
Table 3.1: Bottlenecks Identified in the LSAM-Specific Production Process with Corresponding Mitigation
to optimize or refine. This new Design cycle stands to potentially consume a significant amount (a
double-digit percentage) of overall job production time. Given that the Design Engineering team
is initially unfamiliar with the LSAM’s design space and that the Stress Analysis team does not
have a validated model for assessing LSAM-specific tools, there is significant risk in this portion
of the process map. In order to address this risk, Design for LSAM Guidelines were developed in
conjunction with Ascent’s Design Engineering team. This initiative is discussed in detail in Section
3.5. Additionally, an approach to developing an effective stress analysis modelling capability was
Lastly, the Manufacturing category changes most significantly in comparison with the tra-
ditional process. As previously discussed, the printing process replaces six metalworking steps.
However, in order to effectively replace these steps and realize the time and cost benefits of the
machine, the printing process itself must be well understood and controlled. Errors in the printing
process can lead to costly rework, breakdowns in quality, and significant impacts to the overall
41
process flow. In order to address these concerns, Ascent implemented a deliberate test plan to
explore the process and constraints. By understanding the printer’s corner and edge cases, it is pos-
sible to print correctly the first time. Additionally, lessons-learned processes and work instruction
codification are implemented to reduce variation. These initiatives are discussed in greater detail
in Section 3.4. The second critical bottleneck identified in the Manufacturing category was the
machining step. Because the LSAM can print in batches, there can be periodic "surges" of objects
requiring machining. In contrast with LSAM printing, machining is conducted on a 1:1 basis. At
rate-production level, this could create queuing at the machining step, negatively impacting overall
production time. By minimizing machining requirements for LSAM tools and implementing process
steps to reduce setup time on LSAM machining jobs, Ascent can mitigate the impacts of surging
As identified in the preceding section, in order to control the printing process and create
predictable output, it is critically important to understand the LSAM’s print parameters. One
approach would be to fully explore the design space of the LSAM using a design of experiments
approach. Such a process would involve tuning all relevant print parameters and empirically mea-
suring their effect on the printed object. While comprehensive, this approach would undoubtedly
consume a prohibitive amount of resources—in material, time, labor, speed to market, etc.
Rather than exploring the entire design space of the LSAM, an approach was developed that
began by working backwards from Ascent’s initially targeted product offerings and then investigating
the variables more relevant to those geometries. Thus, initial focus was given to the characteristics
necessary to print Trim and Drill Fixtures as well as Mold Tooling. By starting with this end state
in mind, it was possible to narrow the parameter space and increase the speed and efficiency of
testing. This constrained space was more efficient to explore and would enable the bootstrapping
The first step in developing the test plan involved determining the critical parameter to be
tested. These were grouped, generally into the following categories: organizational, production-
42
specific, and performance-specific. After assessing the initial target use cases, the following charac-
teristics were deemed critical to understand. Following this classification, a test plan was designed
to minimize cost (print time and material) to gather the data necessary to support customer-facing
production as soon as possible. This involved developing printed geometries and scheduling them
in such a way to minimize cost and maximize findings. Due to the proprietary nature of this test
schedule, it is omitted from this document, however, the summary of characterized parameters is
provided below:
• Organizational
– Machine and Team Functionality: The LSAM team, having attended Thermwood-
sponsored training, needed to validate that functionality of the machine and understand
its operation. This represents an on-going process. Through some of the mechanisms
described in the proceeding sections, critical findings were captured and processes began
to be codified. This not only reduce variability in output, but also enabled Ascent to
readily scale the team once production rates were achieved because it was easier to on-
• Production-Specific
– Printer Parameters per Material: The LSAM allows the operator to specify many
print parameters which can be tuned to ensure a print is successful. These param-
eters include drying time, nozzle temperature, layer time, and others. The optimal
settings for these parameters vary depending on material printed (ABS, Polyetherim-
ide (PEI), Polycarbonate (PC), Polyethersulfone (PESU) etc.) and the geometry of the
part. Thus, a number of prints were executed to explore the space of these variables,
simple, 1-dimensional strips of material all the way to large-scale tools with complex
curvature and overhang. These print parameters have a direct impact on the quality of
the resulting print. Once optimal parameters were identified for a given material and
43
– Layer-to-Layer Bonding: The quality and consistency of the bonding between succes-
sive layers is critical to end-part viability. As printer parameters were explored, assessing
the quality of these bonds was a central quality metric. If bonding between layer was
poor, it would compromise both the structural and vacuum integrity of the resulting tool.
Inter-laminar bonding consistency across large tools can be difficult to control because
of the thermal gradient across the large surface area. The LSAM’s fans can be used to
– Overhang Angle: Because the LSAM does not print with a separate support material,
the amount of overhang a material can withstand is a key design-space constraint. Spe-
cialized geometries were printed to explore this space and understand the limitations of
various materials. The maximum overhang angle determined remains proprietary, but
assessing these limits by material is a critical component of defining the parameter within
– Machinability: LSAM tools must be machined to final tolerances. Thus, it was crit-
ical to understand how the printed material behaved when machined and observe any
chipping, cracking, or other defects so that they could be designed around. Especially
when printing larger objects, any catastrophic error in machining that would necessitate
a re-print could be significantly costly. Because the material is bonded across layers,
some flaking and chipping can be expected. Multiple prints were used to assess the
• Performance-Specific
– Vacuum Integrity: It was important to understand the ability of printed tools to hold
vacuum, as customers typically use this method to hold a part in position for machining
part, especially after machining. The print must be generally smooth and consistent to
ensure the part will hold vacuum. Poor inter-laminar bonding will also typically result
– Fluid Reactions: Printed tools are exposed to various industrial chemicals and solvents
in the course of their use. It is important to understand the impact that these fluids
44
have on the printed tool to ensure they will function in a customer’s manufacturing
environment. Multiple difference solvents and fluids were tested on printed materials to
to maximize the LSAM’s utilization. Arranging multiple prints in the print area must
be done thoughtfully to maximize efficiency and minimize the risk that a print will fail.
Additionally, because of the layout of the LSAM’s gantry, thought must be given to
where heavier prints are placed in the print area to enable the print’s movement out of
the prints area to follow-on work centers. As the machine is capacitized, it will become
increasingly critical to thoughtfully arrange objects on the print bed during the same
printing operation.
– Thermal Performance: Printed materials and geometries must also be tested to ob-
serve how they perform at elevated temperatures. Understanding this behavior and
the material under thermal loads can induce warping in the finished part, and thus must
be properly compensated for. The first step in this process is to determine the coefficient
of thermal expansion for various material types using representatively large samples.
Deliberately enumerating these characteristics and then developing a test plan around them
allowed Ascent to optimally explore the relevant design space. It also enabled a rapid progression
toward revenue-generating jobs for customers, as these characteristics were explored sequentially.
A deliberate test plan is critical to minimizing the payback period for the LSAM and generating a
customer base as quickly as possible. Once additional research and development has been conducted
to continue exploring these key characteristics in support of more complex tooling applications,
As discussed in Section 3.1, cultivating and capturing "know-how" is critical in the successful
implementation of AM technology. This becomes particularly critical as production scales and the
45
immediate team supporting the LSAM must grow commensurately. Settings and processes should be
captured in work instructions that can be reliably replicated to reduce variability and the frequency
ensure that best practices were codified. In conjunction with exploring the LSAM’s capabilities and
processes, a Lessons Learned Log and detailed work instruction were developed.
The Lessons Learned Log is a living document meant to capture practices and findings—both
good and bad—immediately after they occur. Following the execution of print or maintenance oper-
ations, the team conducts a thorough debrief, discusses the intended plan, examines how execution
diverged from the plan, and ensures these notes are captured in the document.
By logging these findings, Ascent minimizes the risk that failures are repeated. Additionally,
Ascent captures best practices. This document provides the raw material from which formalized
Work instructions are the formalized, step-by-step guides to executing a given process on the
LSAM. The target audience for these work instructions are machine operators so that no critical
steps are omitted. As the investigation of various processes was explored as described in Section
After work instructions reached a functional state, they were printed, bound, and placed at
46
Figure 3-3: Sample from an Operator Checklist Used to Minimize Errors and Variation
the operators work station next to the printer. These instructions are critical to minimizing print
As the capabilities and operation of the machine were understood and systematized, the
focus shifted farther upstream in the production process to the tool design process itself. Design
for Additive Manufacturing (DfAM) is a well-studied problem in literature meant to optimize the
design of a 3D-printed object for the specific printer and material being utilized [20] [1] [30]. Relying
on traditional design approaches can make the printing process inefficient and prevent the design
from fully leveraging the unique capabilities of AM versus traditional manufacturing processes. The
process of exploring the LSAM’s design space had to precede the creation of new, LSAM-specific
design guidelines. However, with the design space more completely understood via the approach
described in the above subsections, detailed guidelines could be created for various tool types.
These guidelines were codified in "Design for LSAM" (DfLSAM) documentation that were
created with the primary input of both the LSAM team and the Design Engineering team. The
constraints, and overall design space. Major components of the documentation include LSAM
Description and Capabilities, Key Processing Parameters, Design Considerations, Design Method-
ology, and Stress Analysis. Importantly, the documentation includes CATIA "go-by" files that
design engineers can use to reference things like recommended widths, maximum overhang angles,
backup-structure design, and more. The inclusion of these CATIA files greatly helps to minimize
47
error in communication and enables the rapid on-boarding of new design engineers.
Figure 3-4: Design for LSAM Guidelines Made to Increase Design Efficiency
The DfLSAM documentation provides a common vocabulary for LSAM-centric tool produc-
tion, which facilitates faster development of tooling (translating to reduced lead time) with less
re-design and re-work. The primary target audience for the guidelines are Ascent’s Design Engi-
neers and Stress Analysis team. Secondary audiences include the LSAM operating team, Product
The details of the DfLSAM guidelines are considered by Ascent to be proprietary informa-
tion, and thus the guidelines themselves are beyond the scope of this document. However, their
structure and the act of creating them represent best practices for codifying LFAM design proce-
dures and accelerating the organization’s adoption of the technology. The exercise of creating the
DfLSAM documentation itself helped to facilitate the inclusion of various stakeholders across As-
cent’s organization. Other organizations looking to rapidly adopt LFAM should consider pursuing
a similar path.
Chapter 3 uses the LSAM’s integration at Ascent as a case study in the integration of LFAM
into manufacturing operations more generally. From this example, it is clear that unlocking the
process. From customers, to sales, to project managers, engineers, and technicians, existing pro-
that must be planned for. Stakeholders across the value chain must be educated in the technol-
ogy’s capabilities and limitations. By thoroughly understanding and controlling the print process,
48
streamlining the design process, and simultaneously educating stakeholders, LFAM can be more
easily integrated into existing manufacturing operations. By mapping the future-state LSAM pro-
duction process, a number of potential bottlenecks are identified. These bottlenecks are then used
49
Chapter 4
Chapter 4 explores a methodology for determining the physical and mechanical properties of LSAM-
printed materials for use in FEA simulation. As there are no currently accepted ASTM standards
or industry norms for characterizing these types of large-scale, additively manufactured materials,
the values quoted on material-supplier data sheets and in literature can be confusing to interpret,
resulting in "apples-to-oranges" comparisons across different material types, printers, and print pa-
rameters. It is critical to think deliberately about how to conduct these tests and how to best
interpret and compare their results. A proper methodology accounts for the unique scale, compos-
ite quality, and anisotropic nature of the material and manufacturing process and also accounts for
differences in print parameters. This chapter begins by reviewing current approaches for charac-
terizing 3D-printed materials and CFRPs. A methodology for testing the engineering properties
of these materials is then suggested. Following this, the results of experimental testing done on a
First, the structure of the printed material is important to understand. As previously dis-
cussed, the LSAM uses thermoplastics impregnated with chopped carbon fiber reinforcement. This
50
reinforcement substantially increases the strength of the printed material [25] and is necessary,
given the side of the LSAM’s bead, to support printed objects at this scale. The thermoplastic
used—whether ABS, PC, PEI, PESU, etc.—is referred to as the matrix material. The carbon fiber
reinforcement is referred to simply as the fiber. As the material is extruded out of the nozzle and
onto its desired location in the print, the fiber tends to align longitudinally in the printed direction
[7]. This alignment occurs more on the periphery of the bead than in its center, where the flow of
the matrix material is less effected by shearing forces on the edges of the bead. The fiber alignment
can be visualized in the right side of Figure 4-1. Considering only a single bead in isolation, this
directional alignment results in anisotropic effects. Even ignoring the impact of the fiber align-
ment on the material properties, the additive manufacturing process itself imparts anisotropy on
the printed material. As seen in the left side of Figure 4-1, which can be assumed to consist of
a just matrix material, the printed object is not homogeneous, and different factors dominate in
different directions such as bead-to-bead bond strength. The presence of aligned fibers and the
bead-to-bead interactions make characterizing FDM-printed CFRP challenging. One must account
Figure 4-1: (L) Orientation of Extruded Beads within an LSAM-Printed Object and Relation to Engineering
Properties [13]; (R) Orientation of Carbon-Fiber Reinforcements in an Extruded Bead [7]
printing and additive manufacturing. However, given the diversity of AM methodologies and mate-
rials and their relatively recent development, no definitive, generalized methodology for character-
izing 3D-printed objects exists. The unique size of the LSAM and its relative rarity make published
standards less likely to be developed in the near term. Thus, Ascent requires a method of charac-
51
terizing this material internally, which could be used to quantify and compare the performance of
to begin with non-standard testing methodologies that have been explored for other 3D-printing
technologies [12].
CFR-ABS, extrusion-printed structures for numerical simulations [13]. They note that the presence
of reinforcing fibers further exaggerates the already anisotropic nature of FDM-printed ABS, making
direction" (also called direction "1"), the material is well-aligned along the long axis of the extruded
material, resulting in increase tensile strength. In either the "stacking direction" (direction "3") or
the "transverse in-plane" (direction "2") directions, the bead-to-bead bond strength becomes the
dominant determinant of the material’s tensile strength, reducing it compared to the 1 direction.
In the case of short fiber reinforcement, Kissling, et al., show that it is possible to approximate
this anisotropic material as orthotropic. This orthotropic approximation is helpful in reducing the
52
While described in the preceding paragraph, it is important to be explicit about the notation used
to describe the printed material’s orientation. This nomenclature is central to understanding both
the analytical equations described below as well as the discussion of testing results in subsequent
sections. Rather than using X, Y, and Z to describe the material’s orientation, the numbers 1, 2,
and 3 will be utilized. X, Y, and Z will be used to refer to "global" parameters describing the entire
printed object with respect to the LSAM’s print area. 1, 2, and 3 directions are defined as follows,
• 1-Direction: The "1-direction" refers to the direction that the print head is moving—the
printing direction. It can be thought of as the "long axis" of a printed bead. While this
1-direction is likely to change during the course of a print with respect to the global frame
(i.e. the print head will change directions while printing), this is convenient notation for
• 2-Direction: The "2-direction" refers to the transverse in-plane direction with respect to the
• 3-Direction: The "3-direction" refers to the "stacking" direction above and below the print-
ing direction. This direction is parallel to the height of a printed bead. While on the print bed,
the 3-direction is identical to the global Z direction. However, even once a part is removed
from the print bed and reorientation, the stacking direction will still be referred to as the
3-direction.
and strain for these orthotropically-modelled materials. In the linear elastic case, the stress-strain
relationship is simply:
{︂ }︂ {︂ }︂
−1
𝜖 = [𝐸] · 𝜎 (4.1)
Using the directional numbering system described in the preceding paragraph, the constitu-
53
tive equation can be written compactly as follows:
⎧ ⎫ ⎡ ⎤ ⎧ ⎫
⎪
⎪
⎪
⎪ 𝜖1 ⎪
⎪
⎪
⎪ ⎢
1
𝐸1 − 𝜈𝐸122 − 𝜈𝐸133 0 0 0 ⎥ ⎪
⎪ 𝜎1 ⎪
⎪
⎪ ⎪
⎪
⎪
⎪
⎪ ⎪ ⎥ ⎪ ⎪
⎪ ⎪ ⎢
⎢− 𝜈21
⎪ ⎪
1
− 𝜈𝐸233
⎪ ⎥ ⎪ ⎪
⎪
⎪
⎪ 𝜖 ⎪
2 ⎪
⎪ ⎢ 𝐸1 𝐸2 0 0 0 ⎥ ⎪
⎪
⎪ 𝜎 2
⎪
⎪
⎪
⎪
⎪ ⎪
⎪ ⎢ ⎥ ⎪⎪ ⎪
⎪ ⎪ ⎢ 𝜈31 ⎪ ⎪ ⎪
⎨ 𝜖3 ⎪ ⎢− 𝐸1 − 𝜈𝐸322 1
0 0 0 𝜎3 ⎬
⎪ ⎬ ⎥ ⎪⎨ ⎪
𝐸3
(4.2)
⎥
=⎢ ⎥·
1
⎢ ⎥ ⎪ ⎪
𝛾 ⎢ 0 0 0 0 0 ⎪𝜏23 ⎪
⎪ ⎪
⎪
⎪
⎪ 23 ⎪
⎪
⎪ ⎢ 𝐺23
⎥ ⎪
⎥ ⎪
⎪
⎪
⎪
⎪ ⎪
⎪ ⎢ ⎥ ⎪⎪ ⎪
⎪
⎪ ⎪ 1 ⎥ ⎪ ⎪
⎪
⎪
⎪
⎪𝛾 13
⎪
⎪ ⎢
⎪
⎪ ⎢ 0 0 0 0 𝐺13 0 ⎥ ⎪⎪𝜏13 ⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪ ⎪
⎪ ⎣ ⎦ ⎪⎪ ⎪
⎪
⎩𝛾12 ⎪ 1
0 0 0 0 0 ⎩𝜏12 ⎪
⎪ ⎭ ⎪ ⎭
𝐺12
The Young’s Modulus, Poisson’s Ratio and Shear Modulus can be defined, respectively, as
follows:
∆𝜎𝑖
𝐸𝑖 = (4.3)
∆𝜖𝑖
𝜖𝑖
𝜈𝑖𝑗 = − (4.4)
𝜖𝑗
𝜏𝑖𝑗
𝐺𝑖𝑗 = (4.5)
𝛾𝑖𝑗
Thus, there are twelve parameters which must be experimentally determined in order to
describe the orthotropic material’s mechanical behavior. It should be noted that although there
are six Poisson’s ratios in total, only three need to be determined experimentally because the other
𝜈𝑖𝑗 𝜈𝑗𝑖
= (4.6)
𝐸𝑗 𝐸𝑖
54
This reduces the overall number of required, experimentally determined mechanical param-
eters to nine.
Thermal properties are also critical to Ascent’s use case. Thermal loads play a significant
role in the additive manufacturing process itself because of the heat required to make the pelletized
feedstock extrudable. The dynamic heating and cooling involved can induce residual stresses leading
to warping and cracking. One of the additional benefits of adding carbon-fiber reinforcement to the
thermoplastic matrix is to minimize the effects of these induced stresses on the quality of the final
print [23]. Nascent software simulations of the printing process exist that model the formation and
effect of these residual stresses for a given print. They will be discussed briefly in Chapter 5, but
In this characterization methodology, it is assumed that a print is successful and does not
have catastrophic warping or cracking. Given this as a starting point, Ascent is primarily concerned
with the performance of the tool when subjected to operational thermal loads in autoclave and
oven cure cycles. Thus, the coefficient of thermal expansion (CTE) and heat deflection temperature
(HDT) are of interest. Zaldivar investigated the effect of different orientations and fill methodologies
on CTE [38]. It is clear from their work that CTE changes based on the orientation of the material.
Similar to the mechanical properties discussed above, these thermal properties can also be assumed
to behave orthotropically.
Kissling, et al, also argue that a given set of experimentally-determined properties are only
applicable for the corresponding print parameters used for the specimens’ printing. Christiyan pro-
vides an experimental investigation into the impact of print parameters on mechanical properties for
3D-printed ABS, supporting this assertion [9]. Thus, it is important to standardize print parameters
if consistent material properties are desired. Because the products that Ascent is initially targeting
are well-known, it is possible to settle on a narrow range of typical parameters. This enables the
material characterization methodology to be useful, because the test specimens can be printed using
Lastly, a testing methodology must account for the LSAM’s unique size. Due to its uniquely
55
large bead dimensions, directly applying ASTM standards used for injection molding or composites
can generate misleading conclusions. A test specimen must be sufficiently large to capture the
bead-to-bead interactions that dominate in certain orientations. Otherwise, one risks omitting those
critical factors in the characterization of the material. For example, the dimensions of the bounding
box surround the specimen for an ASTM D638 Standard Test Method for Tensile Properties of
Plastics test are 9.7"L x 1.13"W x 0.55"T. As discussed in Section 2.3.2, a single extruded bead
from the LSAM is 0.83"W and 0.2"T. The standard D638 specimen is only incrementally wider than
a sinlge LSAM bead. Thus, the standard D638 sample is poorly scaled to capture the true, macro-
level nature of the material’s behavior. As a result, each test must be considered for modification
the unique challenge of the LSAM’s size, a tailored testing methodology is required. The intent of
this approach is to collect the material properties necessary to understand a printed tool’s behavior
under operational mechanical and thermal loads. The list of properties suggested here is not ex-
haustive, but rather focuses on those material properties necessary for modelling the performance
characteristics relevant to Ascent’s LSAM-produced tooling. Table 4.1 shows the tests performed,
grouped by category.
Table 4.1: Engineering Properties Determined Minimally Sufficient to Provide Input to Numerical Simulation
This batch of testing would need to be repeated for each material type and printer parameter
set of interest. Additionally, this batch of testing could be periodically re-executed on a time-based or
event-based trigger. For example, testing on a particular material-parameter set could be completed
56
every six months in order to determine if there is any drift in material properties. Such drift could
be due to wear on the machine itself or deviations in the feedstock. It may also be advisable to
re-execute testing after a new lot of feedstock is introduced, and certainly if a new material supplier
is utilized. By periodically re-testing according to this methodology, an LSAM user can be more
The mechanical tests proposed include determining the Tensile or Young’s Modulus, Poisson’s
Ratio, and Shear Modulus. Both the Young’s Modulus and Poisson’s Ratio can be determined using
ASTM D638. However, the specimen dimensions must be scaled up to properly capture the bead-
to-bead interactions described in the previous section. The modified specimens can be seen in
Figure 4-2. They maintain the same shape and proportions as the D638 standard, however, they
are large enough to produce multiple bead-to-bead interfaces in each direction. D638 is the only
ASTM standard that required modification to accommodate the LSAM’s unique scale. All of the
remaining mechanical, physical, and thermal tests were able to utilize prescribed specimen sizes
Figure 4-2: (L) Modified Tensile Test Specimen Model; (R) Representative Finished Tensile Specimens for the
1-Direction and 3-Direction [27]
In order to conveniently produce the number of samples suggested for this testing methodol-
ogy, three-dimensional, hexagonal structures are utilized. By utilizing these hexagonal structures,
it is possible to replicate the test specimens accurately and efficiently as required rather than ma-
chining the specimens out of different printed objects. The structure of these hexagons can be
seen in Figures 4-3 and 4-4. Two hexagons—one for the 1-direction specimens, and one for the
3-direction specimens—are used to produce the needed specimens for each orientation. The two
57
hexagonal shapes are of different dimension to accommodate the orientation of the specimens (i.e.
the face of a 1-direction hexagon is wider than it is tall, and the face of a 2-direction hexagon is
taller than it is wide). With two tensile specimens produced per face of each hexagon, only 2.5
faces of the hexagons are required to produce the necessary number of over-sized tensile specimens.
Thus, 2.5 usable faces remain to produce the specimens required for other material tests for a given
orientation. The sixth face of each hexagon possesses a "zipper" where each layer starts and stops.
Because of this discontinuity in the layer, this face is discarded for the purposes of testing.
It should be noted that a 2-direction specimen is omitted from this testing methodology.
In order to reduce the number of tests required, it was assumed that the "Z" properties would
be sufficiently similar to the "Y" properties to consider them equal. This assumption is based on
the bead-to-bead interface being similar in both the 2 and 3 directions. Because this bead-to-bead
interface is the determinant of the material properties in these orientations, cost can be saved by
testing only one of these directions. A more exhaustive characterization would include testing in
the 2-direction.
Figure 4-3: (L) Methodology for the Printing and Machining of 1-Direction Tensile Coupons
Figure 4-4: (L) Methodology for the Printing and Machining of 3-Direction Tensile Coupons
58
The required number of shear specimens are also machined out of the hexagons for each
orientation. These specimens use the dimensions specified in ASTM D5379. The specimens can be
The remaining physical and thermal test specimens are machined from the remaining portions
of the hexagons corresponding to their orientations. These specimens are all smaller than the
modified tensile specimens and shear specimens, and are sized according to their respective ASTM
standards.
The testing conducted here is restricted only to CFR-ABS and would need to be repeared
for any additional material types. While trivially easy following this procedure, this obviously
incurs additional cost. Additional samples per material types could certainly be tested for each test
specified in this methodology. However, testing was completed by a third party lab, and there is
substantial cost associated with each test. Thus, the number of tests performed in this methodology
were deemed to be minimally sufficient to produce meaningful results. The test process itself, even
when printing with known print parameters can produce variation even with this simple hexagonal
geometry. Thus, it is important to use as many samples as possible to reduce the impact of any
59
4.3 Laboratory Testing Results and Discussion of Findings
Testing was conducted according to the suggested test methodology via a third-party lab.
The results of this testing are presented in Table 4.2. In the following sub-sections, these values will
Table 4.2: Material Properties from Experimental Characterization of ABS Material (numbers are scaled by factors
to preserve proprietary information, but relative relationships are preserved) [27]
The experimental setup of the tensile testing can be seen on the left side of Figure 4-6.
Additionally, a representative example of the failure modes for both the 1 and 3 direction can be
seen on the right side of Figure 4-6. Clearly, in the case of both orientations, the specimens break at
the center of the specimen, where the cross-sectional area is minimized, as intended. Additionally,
the bead-to-bead overlap zone is also visible in both the 1 and 3 orientation cross-sectional views.
It is also immediately evident when inspecting these specimens that there is a qualitative difference
between the 1 and 3 failure modes. The specimens pulled in the 1 direction—longitudinally along
the length of an extruded bead and in the direction of fiber alignment—present with a jagged
pattern. Conversely, the specimens pulled in the 3 direction—pulling the stacked layers away from
60
one another—have a cleaner break. This clean break is because the specimen broke at a bead-to-
bead interface, where the inter-laminar bond was the weak point in the specimen.
Figure 4-6: (L) Tensile Test Setup [27]; (R) Representative Failure Modes for X and Z Specimens [27]
The material possesses much better tensile strength performance in its 1 direction than its 3 direc-
tion. In the 1 direction, the material’s strength is dominated by the micro-level interactions between
matrix and aligned-fibers. However, in the 3 direction, the inter-laminar bond becomes the lower
bound on the material’s tensile performance. It is logical that the bonds between layers, even if
the print parameters are set optimally, will be less strong compared to the homogeneous, extruded
material itself.
Poisson’s Ratio can also be calculated from the results of D638 testing by using strain gauges
on the relevant sides of the specimens. Differences in Poisson’s Ratio can also be observed between
the three orientations. This also makes intuitive sense given the material’s composition.
Shear testing was conducted according to ASTM D5379. No alterations were made to the
specimen dimensions in this case. Specimens in all three orientations were machined out of the
hexagonal shape previously discussed. Variation in the shear modulus can be seen based on the
specimen’s orientation. Figure 4-7 shows the test setup as well as the failure modes of each specimen
tested.
The plots of the stress strain curves for each orientation can be seen in Figure 4-8. It can
61
Figure 4-7: (L) Shear Test Setup [27]; (R) Representative Failure Modes for all Three Orientations [27]
be seen that when stress is applied in the 1 and 2 directions (in the left and center positions of the
figure), there is a significant amount of variation across samples. This can be contrasted with the
relatively consistent curves seen when stress is applied in the 3 direction. This effect can be visually
observed in Figure 4-7. The left-most column of specimens (the Z -labelled specimens are in the
3-direction) all exhibit similar, qualitative failure modes at the central, narrowest point. However,
both the Y and X -labelled specimens exhibit a wider degree of variation in failure mode. Some
specimens break to the left or the right of the narrowest point. This difference is attributable to
the density of bead-to-bead interfaces present in the X and Y faces. Thus, the shear strength of
the material in the X and Y directions is mostly dominated by the location of the shear force: if
shear force is applied at or near a bead-to-bead interface, the ultimate shear strength tends to vary
more significantly. If, however, shear force is applied in the Z direction, where each bead-to-bead
interface has a prescribed overlap zone, this variation is muted, resulting in more consistent results.
This property is useful when designing the orientation of tool components that will see higher shear
loads.
Figure 4-8: Stress-Strain Plots for Shear Testing in the 1 (L) 2 (C) and 3 (R) Orientations (axes redacted to preserve
proprietary information) [27]
62
4.3.3 Physical Properties
With respect to the physical properties of this CFR-ABS, both density and hardness were
examined. Neither test required modification from its ASTM standard. The density value of 0.04
𝑙𝑏𝑠/𝑖𝑛3 roughly coincides with material supplier data sheet values. However, accurate measure of
this density value is critical for calculating tool weight and tool deformation under its own load.
Hardness is also a critical value of interest because these tools will frequently be used around metal
Density was not tested at multiple orientations because, intuitively, this would not produce
any differing results. Hardness, however, was tested in both the 1-3 and 1-2 planes. There was very
Thermal testing—both CTE and HDT—was conducted on samples from all three orienta-
tions. The experimental setup for each of these tests can be seen in Figure 4-9. Both tests were
Figure 4-9: (L) Coefficient of Thermal Expansion Apparatus [27]; (R) Heat Deflection Temperature Apparatus [27]
Very little difference in HDT was observed between orientations. This is because HDT
measures the temperature at which the specimen deforms a specified distance under a specific load.
While the printed material is not perfectly thermally isotropic, given the size of the tested samples,
63
they are close enough in each orientation so as to not produce a significant change in HDT. If larger
samples were used, it is possible that more significant differences in HDT would begin to emerge.
𝜇𝑚 𝜇𝑚
Conversely, CTE varies significantly between orientations: 80 𝑚𝐶 𝑜 in the 1-direction, 80 𝑚𝐶 𝑜
𝜇𝑚
in the 2-direction, and 120 𝑚𝐶 𝑜 in the 3-direction. It is particularly important to note the 50%
increase in CTE in the 3 direction. This is the "stacking" direction, with the highest density of
bead-to-bead interfaces. The presence of more interfaces reduces the material’s ability to resist
thermal expansion in this direction and is consistent with other investigations into the CTE of 3D-
printed material. This difference confirms the need to think deliberately about the build orientation
Chapter 4 provides the methodology and results necessary for characterizing LSAM-printed
objects and numerically simulating their behavior. The chapter began with a review of current
empirical testing is conducted on a representative CFR-ABS feedstock, and the results are sub-
sequently discussed. While these results are specific to CFR-ABS, the same testing methodology
could be readily utilized on any LFAM feedstock. Such a testing methodology could be repeated on
time-based or event-based triggers, such as each time a new lot of material is introduced, to confirm
64
Chapter 5
Chapter 5 develops an approach to the Finite Element Analysis (FEA) of LSAM-printed objects
using the material properties determined in Chapter 4 as inputs. The goal is to produce a simulation
solution which is sufficient to inform design decisions and confirm that a tool will perform to
customer specification. The chapter begins by exploring existing efforts to model 3D-printed objects,
specifically those using composite feedstocks where print orientation is especially important. Next,
the overall approach of the modelling effort is described, moving from engineering properties to
simulating geometries of increasing complexity. Two applications of the model are then discussed:
the tensile tests described in Chapter 4 and the deflection testing of a more complex sub-assembly.
From these explorations, the model’s applications and limitations are described.
FEA is a well-understood and powerful tool that can be used to simulate the behavior of a
digital model subject to various conditions [2]. These various conditions might include structural
loads, thermal gradients, fluid flow, and others. FEA simulations, which rely on the Finite Element
Methods (FEM) to solve, can provide critical insight into a structure’s performance at far cheaper
rate than physical testing would require. FEM is a numerical method for solving partial differential
65
equations that involves discretizing a model (of one, two, or three dimensions) into smaller pieces.
These smaller pieces are referred to as finite elements. The process of discretizing the model into
finite elements is called "meshing" [22]. With the model subdivided into elements and loading
conditions specified, a large system of equations can be generated and iteratively solved to produce
solutions for each element. This resulting set of solutions can then be analyzed to provide insight
Many software solutions, both proprietary and open source, exist that can perform FEA sim-
ulations. Such FEA software packages include: Abaqus, Autodesk Simulation, ANSYS, LS-DYNA,
HyperWorks, Patran, and Nastran. Each software suite comes with its own features, advantages,
and disadvantages, which can either inhibit or enable its use in this LFAM-specific modelling appli-
cation. For example, as discussed in previous sections, the orientation of the material is critical to
its mechanical and thermal properties. Any FEA software package intended for use in this LSAM-
specific modelling application must be capable of specifying material orientation with relatively high
fidelity. Ascent’s stress analysis team already uses the Altair’s HyperWorks software package, and
thus, this approach also leverages this software. However, other software solutions are likely capable
There have been previous attempts at modelling the performance of additively manufactured
objects using FEA. These attempts generally fall into one of two categories: those that simulate
the entire print process and those that more naively model the finished part’s performance. Those
that model the entire print process—which are referred to here as "process simulation"—are more
computationally intensive and focus predicting the final shape and residual stresses of a print by
capturing its full thermal history during printing [8] [14] [7] [19] [29] [33]. These process simulations
use the sliced geometry of a part and previously known material properties and print parameters
as inputs. The print process is then simulated in software, replicating the heat transfer, crystal-
lization kinetics, and mechanical properties as new simulated material is successively extruded.
This inclusive process has the benefit of being able to accurately predict warping, cracking, and
residual stresses in a printed object. However, most of these software approaches are still under
development in university laboratories and private companies. Once released, the LSAM-specific
print parameters and material properties would still need to be input into the software to generate
66
meaningful results. Such a software solution would certainly be a powerful design tool that would
enable Ascent to simulate a wide variety (virtually any) slicable geometry and confirm that it would
print properly. However, such a powerful tool may not be necessary for Ascent’s applications. As
previously discussed, because the range of tool types Ascent aims to print is known and relatively
well-defined, this full process simulation approach may be excessive given its likely cost.
FEA has also been employed to model the performance of finished parts [32] [21] [37] [16] [2] [6]
[40]. These investigations typically follow the following procedure: a 3D geometry is designed; this
3D geometry is printed with material whose material properties are known at various orientations;
the printed objects are then subjected to a battery of physical tests; the 3D geometry is processed
through an FEA modelling approach; the results of the physical test are then compared with the
results of the FEA output. This approach has the advantage of "closing the loop" and enabling
one to compare the simulated expectations with the physical reality. These approaches generally
assume that the printed object maintains the geometry specified in the digitally-designed model.
More directly, they ignore the effects of residual stresses and warping in the finished parts which
are subjected to loading conditions. In spite of this assumption, many investigations are able to
achieve relatively closely matching results. For example, Bahandari, et al., using a PEI blend with
carbon-fiber reinforcement, was able to obtain tension and shear moduli with <10% error.
The relatively desirable results of these investigations motivated the framework discussed in
the following sections. However, few of these previous investigations have explored LFAM specifi-
cally, with all of its unique constraints. The investigative framework proposed in the next section
provides a methodology for modelling LFAM-generated objects that are printed within a well-defined
The current state of Ascent’s LSAM design approach is reliant on operator experience and
intuition. The ability of a design and eventual printed tool to meet customer specifications is the
result of some degree of trial-and-error. However, this does not scale well and incurs excessive cost.
If effective simulation tools can be developed to provide some quantitative backing to this intuition,
67
design iteration can happen digitally and new LSAM operators do not have to be the only check on
a design’s viability. The current model can be seen in Figure 5-1. Stated simply: a tool is designed
by a design engineer, the tool design is vetted by an LSAM-specific expert, the tool is then printed,
and it is subsequently successful or fails. Failure is defined as an overall structure failure (collapse
or failure to print properly), or an inability of the printed tool to meet customer specifications. If
a tool warps excessively under mechanical or thermal loads, it has failed. Customers are able to
Figure 5-1: Current Physical Approach to the Performance Modelling of LSAM-Printed Designs
A more desirable future state would allow Ascent to test designs in FEA software before
expending time and materials on a physical print. A designer would create a tool which could be
subsequently analyzed in FEA software under operational loading conditions. Pending successful
completion of this simulation step, the design could be released to the LSAM team for final validation
and printing. This future-state process can be seen in 5-2. This sort of process flow is already in
use in Ascent’s traditional, metallic tooling manufacturing process. It would enable digital iteration
on a design. In the best case, it could not only minimize failed prints and re-work, but enable some
degree of design optimization by reducing the amount of material required to meet operational
specifications. By vetting a design through simulation, Ascent can prevent over-engineering a tool
design and save print time and material cost. This FEA process would ideally include the ability
to model thermal expansion as well, enabling a designer to confirm that a tool expands as expected
and does not fall outside of acceptable ranges for specified cure cycle parameters.
However, there is a capability gap in the modelling of LSAM-printed tools. A naive approach
would be to run FEA on a designer’s 3D model, treating the material as a homogeneous solid with
consistent, globally-defined material properties. However, this approach ignored the anisotropy
described in Chapter 4. On the other end of the spectrum, current commercial process simulation
modelling approaches are expensive, untested on the LSAM, and perhaps unnecessarily detailed
68
Figure 5-2: Future Approach to the Performance Modelling of LSAM-Printed Designs
for Ascent’s application. Due to Ascent’s relatively well-defined parameter set, this prompts the
following question: is it possible to generate an FEA simulation solution using known material
properties from Chapter 4, well-defined geometries from Ascent’s tooling, and known near-optimal
print parameters using in-house simulation software? Would such a simulation approach be sufficient
To answer this question, the approach represented in Figure 5-3 was developed. Generally,
it proceeds from simulating simple geometries through moderately complex sub-assemblies, to full
tools. For each geometry type, both a simulation and a physical test are performed. This enables
the results of the simulation to be compared to a physical result, enabling the model to be adjusted
and validated.
subjected to tensile loads identical to the tensile tests described in Chapter 4. These simple geome-
tries are an ideal starting point because they are large enough to capture macro-level behavior, but
69
do not have any printed curvature. All of the curvature in the specimens is the result of machining
operations. Thus, the material orientation is consistent throughout the specimen. The orientation
This enables the simulation to focus on whether the material properties generated in Chapter 4 are
After confirming the material representation in simulation, the approach moves onto arc-
shaped sub-assemblies. These sub-assemblies represent a cross-section of a typical trim tool or layup
mold, ignoring any backup structure. These geometries possess printed curvature, meaning that the
local orientation of the material changes throughout the length of the sub-assembly. Simulations
can be run on these geometries wherein the material is treated as homogeneous throughout and
wherein the material orientation is assigned properly. Comparing the solutions of these simulation
approaches reveals the benefit of properly representing the material orientation in simulation. This
is an important step, as assigning material orientation properly throughout the tool is a non-trivial
Finally, after confirming the material representation and the effect of proper material orien-
tation, a final validation step on a full tool should be performed. This full tool would include the
facesheet and any supporting backup structure. This final check would ensure that the modelling
approach does, in fact, capture the behavior of a fully printed tool. Unfortunately, testing on a full
tool was not possible during the timeline of this research. However, testing on tensile specimens
and sub-assemblies was conducted and is discussed in Sections 5.4 and 5.5, respectively.
Using this framework, there are generally three sources of error that can arise when comparing
the results of the simulation with that of the physical experiment. These three sources of error are
• Physical Experimental Error: As in any physical experiment, there are errors that arise from
the nature of the experimental setup and in the methods used to measure relevant values.
For example, if a specimen slips when placed under load, errors in displacement may occur.
Unintended warping or flaws in a particular specimen could also influence the results gener-
ated. Additionally, the method of measurement may be error-prone as well. For example,
70
when measuring the displacement of the sub-assemblies under load, a laser tracker was em-
ployed. Any calibration issues with the laser tracker could lead to downstream error in the
displacement values. This error was minimized to the greatest extent possible, but still exists.
• Test Modelling Error: This is the simulation analog on the Physical Experiment Error de-
scribed above. Each physical test must be modelled in software. Some assumptions must be
made as to where a load is applied, whether and how surfaces should be allowed to move in
each axis, and how to model slippage between two faces or contact points. Deviations between
the modelled test and the way the physical test was actually executed will lead to differences
in the results.
• Material Representation Error: FEA simulations require a user to specify material properties.
These will be discussed in greater detail in the following section. Any errors in the constants
used for these material properties (Young’s Modulus, Poisson’s Ratio, Shear Modulus, etc.)
will influence the solution the simulation produces. Additionally, errors in material orientation
or how the printed materials anisotropy is modelled will influence results. Minimizing this
HyperWorks was employed as the FEA simulation software in this modelling approach. Hy-
perWorks is a powerful tool that enables a high degree of pre-processing capability and provides
many different solvers from which to choose. This freedom enabled significant experimentation in
choosing optimal simulation parameters. Several decisions were made to inform the ultimate mod-
elling approach including: element type, material card, mesh density and quality, order, linearity,
and solver type. The choices of these parameters were primarily driven by balancing the function-
ality that each provided in terms of accurately representing the material and the computational
requirements to run the simulation. These two factors were typically in conflict with one another,
and striking a feasible balance between fidelity and run time was the ultimate driver of the modelling
Multiple element types are available within HyperWorks. They include zero, one, two, and
71
three-dimensional elements. Multiple element types can be included in the same simulation for
complex models. However, the benefit of these LSAM-printed objects is that they are comprised of
a single material type, requiring one element type. The primary decision here was between using
have the benefit of solving more quickly and are frequently used when homogeneous materials are
being considered. However, because the printed material is thick and the internal interactions
between elements is critical to ultimate part performance, solid elements were chosen. Additionally,
the script developed and employed to dynamically assign local material orientation worked only
with solid elements. Due to their ability to provide a qualitatively cleaner mesh, CHEXA elements
were chosen. Additionally, CHEXA elements are compatible with the MAT9 material card. The
MAT9 material card is used to define the properties of anisotropic elastic materials. This material
card selection enables a user to specify all of the parameters necessary to model the material as
an orthotropic solid. The MAT1, isotropic, material card was also used in this investigation to
compare the effect of modelling isotropically versus orthotropically. Because there is a non-trivial
amount of labor required to properly align the material in software, it was important to perform
this comparison to determine if the additional labor significantly improved simulation results.
Mesh density and quality were also critical parameters in developing the modelling approach.
Too dense of a mesh will result in unreasonable long run times. However, too crude of a mesh will
result in a low-fidelity solution that does not provide the sort of meaningful insight required to
model performance. Because the bead width of the LSAM is 0.83"W by 0.2"T, each side length of a
CHEXA element was specified as 0.1". This enabled sufficient fidelity to produce meaningful results
while not creating an unreasonably dense mesh. Mesh quality was also a critical consideration.
Care should be taken to ensure the mesh is relatively symmetrical and does not have regions with
significantly misshapen elements. Such regions will result in solver failures or unreasonable solutions.
HyperWorks provides the ability to specify 1st or 2nd order elements. Solving using 2nd order
elements provides additional nodes at the midpoint between two connected nodes on an element. For
a 8-element CHEXA element, this doubles the number of nodes per element to 16. While increasing
the fidelity of the solution, running a simulation using 2nd order elements also significantly increases
the run time. Multiple iterations were run to determine whether there was any material benefit to
72
running simulations in 2nd order. However, it was determined that 1st order was generally sufficient
Because these simulations investigated the transient effect of different loads on the specified
geometries and large deformations were expected, a nonlinear solver was utilized. Multiple solvers
were investigated, however, Optistruct was used for all of the analyses presented in this document.
The general modelling approach eventually arrived at through some iterative parameter tuning is
Lastly, as previously referenced, a script was developed and implemented that enabled the
dynamic assignment of local material orientation. This script allows a user to specify the orientation
of a particular element in the meshed model. This is a critical step necessary to properly orient
the modelled material such that the orthotropic properties are accurately represented as printed.
The script enables a stress analysis engineer to quickly select blocks of elements and specify their
orientation without having to do so individually. This significantly reduces the amount of time
The first simulation tested involved a tensile test of the tensile specimens used during the
material characterization approach described in Chapter 4. Because this test was executed by a
professional, third-party lab according to a modified ASTM-standard tensile test, it had the lowest
physical experimental error. Additionally, because the tensile specimens have no curvature, there
was no requirement to dynamically assign local material orientation. Material orientation could be
specified at the global level, making these iterations relatively quick to execute. Modelling these
73
tensile tests allowed a focus on whether the material properties were generally correct before moving
The overall setup of these tensile tests can be visualized in Figure 5-4. The green elements
represent the LSAM-printed material itself. As previously discussed, each element has a side length
of 0.1", resulting in 32,396 elements to represent the full geometry. The nodes represented in
yellow are fixed in three dimensions. They represent the clamped faces of one end of the specimen
in the test apparatus as seen in Figure 4-6. The nodes represented in red are where the tensile
force is applied. They represent the opposite clamped faces in Figure 4-6. A number of alternative
formulations of this simulation setup were also attempted, where different faces were held in place
and the force was applied to different element selections. However, this setup, which best captures
the nature of the physical test, produced the most desirable results.
After the tensile specimen was meshed and the clamped faces were assigned, the material ori-
entation of each element was assigned. As previously discussed, in these samples, all of the material
is oriented globally in the same direction, making this assignment straightforward in HyperWorks.
Because the physical tests to which these simulation results are compared correspond to 1-direction
and 3-direction samples, this is how the material was oriented in simulation as well. Figure 5-5
Material properties were assigned using two different material cards, to determine the best
74
Figure 5-5: 1-Direction Specimen (top) and 3-Direction Specimen (bottom) with Red Arrow Denoting the 1-
Direction and Blue Arrow Denoting the 3-Direction
way to represent the material: MAT1 (isotropic) and MAT9 (orthotropic). First, the average
material property values for the respective orientation from Chapter 4 were applied homogeneously
using a MAT1 material card. In this case, the simulation naively assumes an isotropic material
and ignores local material orientation. Second, all of the relevant material properties for respective
orientations were input into simulation using a MAT9 material card. In this case, the simulation
incorporates all of the experimentally determined material properties and takes material orientation
into account. Ten load steps ranging from 1000 to 10,000 lbf were then successively applied to each
specimen. The outputs of these two modelling approaches can then be compared with experimental
results.
Run times for these simulations, both for isotropic and orthotropic runs, were less than 1
minute. Representative output of the simulation can be seen in Figure 5-6. The stress and strain
values are maximized at middle portion of the specimen, where the cross-sectional area is minimized.
This behavior is consistent with intuition. Both the isotropic and orthotropic runs produced similar
qualitative results.
The results of both the isotropic and orthotropic simulation approaches can be quantitatively
compared with the results of the physical tests. At each load step, the error between the simulated
and actual stress and strain is computed and averaged. The results of this calculation are seen in
75
Figure 5-6: Representative Output of Tensile Test Modelling Approach Showing Von Mises Stress (L), Strain (C),
and Displacement (R) for a 1-Direction Orthotropic Model at 2000lbf
Table 5.2. These relatively low error values suggest that both modelling approaches may be suitable
for such a simple geometry with no printed curvature. However, the orthotropic approach can be
seen to minimize error when considering both stress and strain values. The stress-strain diagrams
seen in Figure 5-7 show how the orthotropic simulation’s output corresponds with the actual stress-
strain curves generated for each of the experimental runs for both X and Z orientations.
Table 5.2: Average Percent Error between Simulation Solution and Physical Experiment at each Load Step
Figure 5-7: Comparison of Orthotropic Modelling Output and Physical Tensile Tests for 1-Direction Specimens (L)
and 3-Direction Specimens (axes removed to preserve proprietary information) (R)
The orthotropic simulation’s output, using the material properties determined in Chapter 4
and the modelling approach developed above appear to correlate well with physical experiments.
76
Some error can be seen at the lowest stress values. This is largely due to specimen slippage in
the test apparatus’s clamps in the physical experiment. The findings of the tensile tests suggested
that an orthotropic model was likely the preferred simulation choice. However, for such a simple
geometry, the gains for modelling orthotropically are not dramatic compared to the isotropic model.
Using the insight gained from the tensile test simulations, it makes sense to next investigate
the simulation’s efficacy on a geometry with more printed curvature. Additional curvature should
increase the influence of the material’s anisotropy on its behavior. Because most trim tools and
layup molds have some amount of curvature, this is particularly relevant to Ascent’s use case.
This test involved the printing of two arc-shaped geometries that resemble a cross-section
of a tool’s facesheet, one six feet in length and the other approximately nine feet in length. The
shape of these specimens and the overall experimental setup can be seen in Figure 5-8. The arcs
themselves are two beads in width. In the reference frame depicted on the right of Figure 5-8, the
arcs have been rotated 90 degrees about the X axis from their printed orientation to arrive at their
tested orientation. Thus, the Y axis depicted in Figure 5-8 represents the 3-direction.
For both lengths, the flat portions of the specimen are placed on immobile support structures.
A pallet is then suspended underneath the specimen where weight can be incrementally added,
placing the specimen under load in the negative Z direction in the manner of a three-point-bending
test. Load steps progressed as follows for each sub-assembly: free-hanging under its own weight,
addition of the rigging required, and then the successive addition of seven 50lbs steel plates. The
maximum load applied to each sub-assembly was self weight plus an additional 350lbs. The flat
portions of the specimen are not secured to the support structure, allowing the ends to slide freely
as weight is added. The flat portions of the specimen are, however, abutted against immobile
guides secured to the support structure. These guides minimize the specimen’s movement in the Y
Checkpoints (CPs) are mounted on the side of each specimen at fixed locations. A laser
tracker is then calibrated against these checkpoints. As additional load is placed on the suspended
77
Figure 5-8: Experimental Setup for Sub-Assembly Deflection Testing Showing Arc with Loading Applied (L), Laser
Tracking of CP Positions (C), and the Common Coordinate System and CP Locations (R)
pallet, the laser tracker is able to locate these checkpoints in three dimensions. This enables an
accurate measure of the specimen’s deflection as load is applied. These laser-tracker generated 3-
dimensional positions are used as the physical truth against which simulation results are compared.
The setup of the sub-assembly deflection test is similar to that of the tensile test described
in the previous section. The simulation setup can be seen in Figure 5-9. The maroon colored
elements represent the printed material. The purple colored elements represent the immobile support
structures. The load is applied over the same surface area in simulation as in the physical experiment
and can be visualized by the pink rays extending from the pink arrow. The contact surface between
the printed material and the support structures were modelled as frictionless, allowing the sub-
assembly to freely slide against the support structure in the manner of the physical test. Load steps
were successively applied using the same progression as in the physical experiment, allowing a direct
comparison of results.
The printed material was modelled using the same parameters as the tensile tests. Using the
same 0.1in element sizes resulted in substantially more elements for each sub-assembly due to their
larger size: 507,780 CHEXA elements for the 6ft arc and 734,580 CHEXA elements for the 9ft arc.
the local material orientation was thought to be more important in this application. Thus, the
material orientation script previously described was implemented to assign proper orientations to
all CHEXA elements in each sub-assembly model. This can be visualized in Figure 5-10. The thick
red arrow on the right side of the Figure describes the path of the LSAM’s nozzle as it deposited
78
Figure 5-9: Modelling Sub-Assembly Deflecting Test Setup in Hypermesh
the printed material. On the left side of the Figure, the material orientation of selected elements
is shown. The red arrows point in the local 1-direction and the green arrows point in the local
2-direction. The 3-direction follows the right-hand rule. Thus, using this script, it was possible to
Figure 5-10: Material Orientation of Sub-Assemblies in Hypermesh Showing the 1-Direction with Red Arrows
For each sub-assembly, four different simulations were run: isotropic using the 1-direction ten-
sile modulus from the material supplier’s data sheet, isotropic using the 3-direction tensile modulus
from the material supplier’s data sheet, isotropic using the 1-direction tensile modulus from the ex-
perimental results, and orthotropic using the material properties from the experimental results. The
material supplier data sheet values were used simply as a point of comparison. The material-supplier
provided tensile values were substantially higher than the experimentally-determined values, and,
The general approach taken was to use the 6ft sub-assembly as a "training set" to optimize
the simulation setup, and then use the 9ft sub-assembly as a "test set." This enabled a better
validation of the modelling approach, as additional simulation parameters were not modified for
79
the 9ft sub-assembly simulation. The only change that was made during the 9ft simulation was the
physical geometry of the sub-assembly itself. Using this approach, it was possible to determine the
The increase in the number of elements in these simulations lead to longer run times. Run
times were generally less than 5 minutes. However, when simulating larger geometries, this may
time.
Figure 5-11: Representative Output of Sub-Assembly Deflection Testing Showing X Displacement (top) and Z
Displacement(bottom) at Max Load Using Orthotropic Approach
After tuning the simulation parameters using the 6ft arc, representative output from the 9ft
arc can be seen in Figure 5-11. This figure shows the sub-assembly’s displacement profile when
maximum load is applied using the orthotropic modelling approach. The arc can be seen to deflect
in accordance with intuition, qualitatively following the same trends seen in the physical experiment.
For the purpose of the following discussion, the coordinate system seen in Figure 5-8. Along the
the X axis, the arc can be seen to "collapse" inward at the highest load. Along the Y=Z axis, the
flat portions of the sub-assembly flex upwards, while the middle of the arc sinks in the negative
80
Figure 5-12: Representative Levering Effect Captured by Modelling Approach Shows CP1 Deflecting Upwards and
Inwards Under Load
Z direction. Looking more closely at the flat portions of the sub-assembly in Figure 5-12, the arc
can be seen to slide and rotate around the edge of the immobile support structure. Qualitatively,
this behavior corresponds with the physical experiment. The isotropic simulations also had similar
qualitative behavior.
With the simulation providing reasonable output, the next step was to compare the simula-
tion’s output with that of the physical experiment. Figures 5-13 and 5-14 show this comparison.
Figure 5-13 depicts the deflection in both Z (L) and X (R) for the 9ft arc. The sign flip in the
X direction represents the arc’s "collapse" inward as increasing load is applied. No values from
the Y direction are reported, as they are essentially negligible. Again, qualitatively, the simulation
Figure 5-13: Experimental Deflection of 9ft Sub-Assembly in Z (L) and X (R) at Each Load Step via Laser Tracker
In order to quantitatively compare the results of the simulation and physical experiment,
the simulated position of each checkpoint was compared with its actual movement in the physical
experiment. Any deviation in X and Z between simulation and experiment was considered error.
81
Figure 5-14: Modelled Deflection of 9ft Sub-Assembly in Z (L) and X (R) at Each Load Step via Orthotropic
Approach
This absolute error of each CP in both X and Z for each simulation approach can be visualized in
Figure 5-15 with the 6ft sub-assembly on the left and 9ft sub-assembly on the right. The magnitudes
of these errors are less than 0.5" for the 6ft arc and less than 2.0" for the 9ft arc, centered around
the origin.
Figure 5-15: Absolute Error in X and Z Directions for the 6ft Sub-Assembly (L) and 9ft Sub-Assembly (R) using
Various Modelling Approaches
A portion of this error is attributable to the physical experiment itself. In contrast with
the tensile tests described in the previous section, this deflection test was not an ASTM standard
conducted in a certified laboratory. As a result, the weight was not perfectly applied or evenly
distributed. Imperfect slippage against the support structure due to friction also was not modelled
perfectly. Additionally, some degree of "rocking" about the X -axis occurred because the flat portions
of the test specimens were not, in fact, perfectly flat. These sources of physical experimental
error contribute to the difference between the physical and simulation results. It is likely that the
82
simulation overestimates the amount of slippage in the X -direction. As a result, the simulation is
Figure 5-16: Orthotropic Modelling Approach is Shown to Reduce RMSE for both the 6ft and 9ft Sub-Assemblies
Using the absolute error values it was possible to compute a room mean squared error (RMSE)
value to quantitatively measure the accuracy of a simulation approach. These RMSE values can
be seen in Figure 5-16. From these RMSE values, it is evident that the orthotropic modelling
approach does outperform the isotropic approach for these sub-assembly geometries. This suggests
that modelling the material’s orientation does provide a meaningful increase in simulation accuracy.
It is likely that a larger, more dramatically shaped geometry would further exacerbate the difference
This orthotropic modelling approach appears to closely follow the physical performance of
LSAM-printed objects. However, there are a number of limitations to this approach which must be
acknowledged.
First, this simulation approach does not explicitly model layer-to-layer bonding. Because
only macro-level material properties are represented (i.e. 1, 2, and 3 direction tensile and shear
moduli), it is not possible to simulate cracking at layer interfaces. As this is the most likely failure
mode, it is important to use this simulation approach for load values substantially less than the
ultimate tensile strength of the simulated material. Additionally, the number of elements required
to model layer-to-layer bonds would likely be computationally prohibitive. Consider again that the
bead dimensions are 0.83"W and 0.2"T. Ascent frequently produces tools with facesheet surface
83
areas exceeding 100 square feet in surface area. Thus, the number of elements required to model
such a tool would be on the order of millions or billions. Without substantial computation resources,
Second, this modelling approach does not include the results of a process simulation. It
assumes that a print was executed successfully without substantial warping or cracking. Because of
this, it ignores residual stresses induced by the manufacturing process. Residual stresses in a printed
object could influence the object’s performance under load. A more comprehensive approach would
simulate the print process itself, capture these residual stresses and strains, and then conduct
simulation on this object. Because this approach ignores the effect of residual stresses in the printed
object, it is important not to operate near the ultimate stress limits of the simulated material.
Chapter 5 described the motivation, methodology, and initial results justifying this modelling
approach as sufficient for Ascent’s applications. First, a review of FEA simulation approaches in
literature was reviewed. Next, the investigative framework for determining Ascent’s approach was
described. Using tensile testing and sub-assembly deflection testing, the FEA simulation approach
was then investigated. Assuming an object was successfully printed, the orthotropic model most ac-
curately correlated with physical tests. With this foundation, the model could be validated against
a full printed tool’s behavior, rather than sub-assemblies. By physically testing and simulating a
fully printed, representative tool geometry, Ascent can be increasingly confident in the validity of
this simulation approach. Additionally, given that thermal properties were also collected during the
material testing phase of this investigation, the model could be subjected to thermal loading. Fur-
ther, this modelling approach should be validated on other material types. With a simulation model
validated across multiple material types, Ascent would have a powerful tool for design improvement
84
Chapter 6
The previous chapters have explored the operational integration of the LSAM, a characterization
methodology for its printed material, and a simulation approach for modelling the behavior of
LSAM-printed objects. In Chapter 6, a summary of key findings and conclusions are presented.
Selected next steps are then described, providing a road map for follow-on work.
6.1 Conclusions
This project set out to effectively implement LFAM capability at Ascent Aerospace, an
aerospace tooling manufacturer, to maximize value impact. Due to the low-volume/high-mix As-
cent’s manufacturing environment, AM stands to substantially reduce the production cost of tooling.
Additionally, because of the bespoke nature of each tooling solution, AM stands to substantially
reduce lead time from on the order of months to on the order of weeks. This project focused on two
establish the context of the project. Then the focus shifts to a discussion of the operational integra-
tion of the LSAM. A comparison of current and future state process maps is conducted to inform
production planning and identify critical bottlenecks. Next, a test plan is described to establish the
85
LSAM’s functionality and reduce variability in its operation. Multiple continuous improvement ini-
tiatives are then described, culminating in the Design for LSAM Guidelines. Taken together, these
strategies of process description, printer testing, and continuous improvement initiatives comprise
Next, a process for characterizing LFAM-printed materials is described using CFR-ABS. This
material is shown to behave anisotropically due both to the chopped carbon fiber reinforcement and
the layered nature of its manufacture. An orthotropic assumption is described to simplify the
number of material properties required to represent the materials behavior. Finally, the results of
physical tests on LSAM-printed samples are presented and discussed. The material is shown to vary
tion process would work for other pelletized feedstocks used in LFAM. Additionally, a hexagonal
geometry is described to quickly and consistently produce the test specimens necessary for this
characterization process.
gested. Effective simulation allows production to scale more rapidly by decreasing reliance on op-
erator expertise and enables more optimal designs. Rather than conducting full process simulation,
the simulation approach aims to use Ascent’s existing software to achieve a useful solution assuming
a successful print. Results of a quality simulation could be used to inform the design process as well
as make performance claims regarding a printed tool’s performance under operational conditions.
An orthotropic model using three-dimensional solid elements is shown to closely correlate with the
results of physical experiments for the tensile tests and sub-assembly deflection tests.
well. First, executive-level emphasis was critical to providing the necessary resources. Some degree
who can become subject matter experts on this new technology and its associated processes. These
associated processes extend all the way from customer education through all steps of manufacturing
to quality assurance and quality control. Establishing a culture of continuous improvement is critical
to effectively capturing the "know-how" central to effective integration. Only with these mechanisms
86
in place can production scale to meaningful quantities.
Abstracted away from the Ascent-specific context, many of the processes and approaches
described in this document could be used by other manufacturing companies to efficiently integrate
LFAM. Far more than a rapid prototyping tool, LFAM has matured to a point where it can be
used to manufacture tooling for an industry as exacting and demanding as aerospace. Projected
to generate over $10M in annual revenue, Ascent’s embrace of this emergent technology stands to
This research primarily focused on process integration and performance modelling. Along
both of the dimensions, there exist substantial additional work that could be performed to better
extract value from LFAM in an industrial manufacturing environment. Below, three directed areas
of further research are suggested, which would build naturally upon the work captured in this thesis.
how pelletized raw material should be transformed into finished aerospace tooling. Much of the work
the LSAM from its installation. However, because this project occurred as the LSAM was being
installed, there was limited data available as to the actual flow of LSAM-generated work product
through the factory. As the number of customer-paid jobs continue to increase, a greater variety
and quantity of data will be available to quantitatively analyze this future-state process flow. Using
this information, it will be possible to identify process bottlenecks and better inform manufacturing
process decisions.
As part of this quantitative approach to LFAM integration, a nascent initiative was begun
to measure the utilization of the LSAM by unit time and by job. The LSAM does not have a native
capacity tracking capability that enables this functionality directly. Thus, a script was developed
to take the raw logs from the LSAM and transform them into understandable output. The process
87
is depicted in Figure 6-1. Refining this script and integrating it into other aspects of the LSAM-
specific workflow could generate useful insights as the LSAM continues to be capacitized. A number
of commercial alternatives to this script were investigated during the course of this research, however,
an internally developed solution integrated with Ascent’s existing SAP system might be optimal.
Figure 6-1: LSAM Utilization Data Processing Script to Transform Raw LSAM Logs into Interpretable Insights
Lastly, significant further work could be done on optimal batching of LSAM prints. As
Ascent increases the number of printed jobs, optimally grouping prints into the same jobs to fill the
print area will become critical. However, because virtually all LSAM tools must be subsequently
machined, this impact must also be investigated to reduce idle work in progress as it flows into the
machining work center. Developing useful heuristics and tools to optimize the flow of LSAM-printed
The material properties data presented in this document only addresses carbon-fiber re-
inforced ABS. However, there are many other material types that are of interest to Ascent and
other manufacturers of LFAM objects. These alternative materials include PC, PESU, PEI, and
others. ABS was chosen initially because it is the cheapest material and is generally used for room-
temperature applications—the first tool types targeted. However, as Ascent begins to shift focus
to tools that will be exposed to higher temperatures, understanding the behavior of these other
The relatively good performance of the FEA simulations with ABS suggest that the char-
acterization methodology developed in Chapter 4 would also be applicable to these other material
types. It generates sufficient printed material in a now-known geometry to create all specimens
88
required for characterization. As time or event based triggers for re-characterization are developed,
it may make sense to purchase testing equipment to conduct particular tests. This will not only
reduce the cost compared to using a third-party lab, but will also enable faster execution time.
Ascent and other LFAM adopters should continue to monitor ASTM Committee F42 as it
publishes standardized testing processes for AM. While they are unlikely to generate LFAM-specific
standards in the near future, it may be possible to adapt best practices for other AM technologies,
particularly those surrounding FDM and carbon-fiber reinforced materials. Research around these
two areas was particularly useful in developing the characterization and simulation approaches
As discussed at the end of Chapter 5, significant further work remains to continue developing
and refining the simulation approach to LFAM-printed materials. The first step is to conduct
physical testing and simulation on a full tool. Such an investigation would provide the best insight
on the efficacy of the simulation approach. As the timeline of this research concluded, steps were
being taken to conduct this work. Figure 6-2 shows an example tool, with significant curvature both
in the facesheet and due to the fork-lift tubes along the base surface. This tool has been properly
meshed and its material correctly oriented using the processes previously described. By subjecting
this tool to loading and tracking its deflection, it would be possible to gain further insight into
whether this simulation approach provides meaningful feedback. Ultimately, the simulation is only
useful if it can meaningfully influence design decisions and enable Ascent to make meaningful claims
Figure 6-2: LSAM-Printed Full Tool Model in HyperWorks, Meshed with Material Orientation Properly Assigned
Beyond full-tool testing, it will also be critical to simulate the thermal performance of LFAM-
89
printed tools. The characterization required to conduct this simulation has already been performed
for ABS. However, the physical experiment must be developed and executed such that the results
can be compared. Such a test—measuring the tool’s facesheet’s deflection during a standard cure
cycle in an oven or autoclave—would be non-trivial to conduct. However, the insights gained from
such a test and the refinement of a simulation approach would enable Ascent to more rapidly and
As new materials are characterized, they can also be implemented in simulation using the
approach described in Chapter 5. Developing a comprehensive simulation approach for all common
material types will be critical to having a comprehensive simulation solution. Further, as commer-
cial simulation software developers and research laboratories continue to improve their products,
these process simulation approaches are likely to become increasingly relevant. If tools like Purdue
University’s Composites Manufacturing and Simulation Center’s Additive3D become effective and
cheap enough, these may eclipse the internally developed simulation approach described in this doc-
ument [8]. Dassault System’s Simulia software suite also has a process simulation capability that
may prove useful as well. However, the transition to any future commercial alternative will require
testing and validation by Ascent, and thus represents a rich area of future research.
90
Bibliography
[1] How do you design for manufacturing excellence in the world of Additive Manufacturing?,
March 2017. Library Catalog: www.technia.com Section: DASSAULT SYSTEMES.
[2] D. W. Abbot, D. V. V. Kallon, C. Anghel, and P. Dube. Finite Element Analysis of 3D Printed
Model via Compression Tests. Procedia Manufacturing, 35:164–173, January 2019.
[4] Charisios Achillas, Dimitrios Tzetzis, and Maria Olga Raimondo. Alternative production strate-
gies based on the comparison of additive and traditional manufacturing technologies. Inter-
national Journal of Production Research, 55(12):3497–3509, June 2017. Publisher: Taylor &
Francis Ltd.
[6] Sunil Bhandari and Roberto Lopez-Anido. Finite element modeling of 3D-printed part with
cellular internal structure using homogenized properties. Progress in Additive Manufacturing,
4(2):143–154, June 2019.
[7] Bastian Brenken. Extrusion Deposition Additive Manufacturing of Fiber Reinforced Semi-
Crystalline Polymers. Ph.D., Purdue University, United States – Indiana, 2017. ISBN:
9780355612837.
[8] Bastian Brenken, Eduardo Barocio, Anthony Favaloro, Vlastimil Kunc, and R. Byron Pipes.
Development and validation of extrusion deposition additive manufacturing process simula-
tions. Additive Manufacturing, 25:218–226, January 2019.
[9] K.G. Jaya Christiyan, U. Chandrasekhar, and K. Venkateswarlu. A study on the influence of
process parameters on the Mechanical Properties of 3D printed ABS composite. IOP Conference
Series: Materials Science and Engineering, 114:012109, February 2016.
[10] G. Costabile, M. Fera, F. Fruggiero, A. Lambiase, and D. Pham. Cost models of additive man-
ufacturing: A literature review. International Journal of Industrial Engineering Computations,
pages 263–283, 2017.
[11] Luca Cremona, Mauro Mezzenzana, Aurelio Ravarini, and Giacomo Buonanno. How Additive
Manufacturing Adoption Would Influence a Company Strategy and Business Model. page 14,
2016.
91
[12] John Ryan C. Dizon, Alejandro H. Espera, Qiyi Chen, and Rigoberto C. Advincula. Mechanical
characterization of 3D-printed polymers. Additive Manufacturing, 20:44–67, March 2018.
[14] Anthony J Favaloro, Bastian Brenken, Eduardo Barocio, and R Byron Pipes. Simulation of
Polymeric Composites Additive Manufacturing using Abaqus. page 13, 2017.
[16] Filip Górski, Wiesław Kuczko, Radosław Wichniarek, and Adam Hamrol. Computation of
Mechanical Properties of Parts Manufactured by Fused Deposition Modeling Using Finite Ele-
ment Method. In Álvaro Herrero, Javier Sedano, Bruno Baruque, Héctor Quintián, and Emilio
Corchado, editors, 10th International Conference on Soft Computing Models in Industrial and
Environmental Applications, Advances in Intelligent Systems and Computing, pages 403–413,
Cham, 2015. Springer International Publishing.
[17] Raed Handal. An implementation framework for additive manufacturing in supply chains.
Journal of Operations and Supply Chain Management, 10(2):18, December 2017.
[18] John Hart. MIT xPRO: AMx, Additive Manufacturing for Innovative Design and Production,
May 2020.
[19] Mathias L Jensen, Mads R Sonne, Jesper H Hattel, and Hans N Hansen. Thermal modelling
of extrusion based additive manufacturing of composite materials. page 13.
[20] Martin Kumke, Hagen Watschke, and Thomas Vietor. A new methodological framework for
design for additive manufacturing. Virtual and Physical Prototyping, 11(1):3–19, January 2016.
[21] Ningrong Lei, Xiling Yao, Seung Ki Moon, and Guijun Bi. An additive manufacturing process
model for product family design. Journal of Engineering Design, 27(11):751–767, November
2016. Publisher: Taylor & Francis Ltd.
[22] Daryl L. Logan. A First Course in the Finite Element Method. Cengage Learning, January
2011. Google-Books-ID: KptPymzHa_gC.
[23] Lonnie J. Love, Vlastamil Kunc, Orlando Rios, Chad E. Duty, Amelia M. Elliott, Brian K.
Post, Rachel J. Smith, and Craig A. Blue. The importance of carbon fiber to polymer additive
manufacturing. Journal of Materials Research, 29(17):1893–1898, September 2014. Publisher:
Cambridge University Press.
[24] Robert Martens, Susan K. Fan, and Rocky J. Dwyer. Successful approaches for implementing
additive manufacturing. World Journal of Entrepreneurship, Management and Sustainable
Development, 16(2):131–148, April 2020.
[25] Nagendra Kumar Maurya, Vikas Rastogi, and Pushpendra Singh. Experimental and Compu-
tational Investigation on Mechanical Properties of Reinforced Additive Manufactured Compo-
nent. Evergreen, 6(3):207–214, September 2019.
[26] Stephen Mellor, Liang Hao, and David Zhang. Additive manufacturing: A framework for
implementation. International Journal of Production Economics, 149:194–201, March 2014.
92
[27] National Technical Systems. Additive Manufacturing Carbon Fiber Mechanical Testing, August
2020.
[28] Lauri Ojala, Juuso Töyli, Tomi Solakivi, Harri Lorentz, Sini Laari, and Ninni Lehtinen. NO-
FOMA 2016 - PROCEEDINGS OF THE 28TH ANNUAL NORDIC LOGISTICS RESEARCH
NETWORK CONFERENCE. page 761.
[29] Deepankar Pal, Nachiket Patil, Kai Zeng, and Brent Stucker. An Integrated Approach to
Additive Manufacturing Simulations Using Physics Based, Coupled Multiscale Process Mod-
eling. Journal of Manufacturing Science and Engineering, 136(6), December 2014. Publisher:
American Society of Mechanical Engineers Digital Collection.
[30] David W. Rosen. Research supporting principles for design for additive manufacturing: This
paper provides a comprehensive review on current design principles and strategies for AM.
Virtual and Physical Prototyping, 9(4):225–232, October 2014.
[31] Massimilliano Ruffo. Make or buy analysis for rapid manufacturing | Request PDF.
[35] Syed A. M. Tofail, Elias P. Koumoulos, Amit Bandyopadhyay, Susmita Bose, Lisa O’Donoghue,
and Costas Charitidis. Additive manufacturing: scientific and technological challenges, market
uptake and opportunities. Materials Today, 21(1):22–37, January 2018.
[36] Mark Vitale, Mark Cotteleer, and Jonathan Holdowsky. An Overview of Additive Manufac-
turing. page 8, 2016.
[37] Brice Matthew Willis. Applying Finite Element Analysis with a Focus on Tensile Damage Mod-
eling of Carbon Fiber Reinforced Polymer Laminates. PhD thesis, The Ohio State University,
2013.
[39] Vittorio Zanetti, Sergio Cavalieri, and Giuditta Pezzotta. Additive Manufacturing and PSS: a
Solution Life-Cycle Perspective. IFAC-PapersOnLine, 49(12):1573–1578, January 2016.
[40] Javad Zarbakhsh, Armin Iravani, and Zeinab Amin-Akhlaghi. Sub-modeling Finite Element
Analysis of 3D printed structures. In 2015 16th International Conference on Thermal, Me-
chanical and Multi-Physics Simulation and Experiments in Microelectronics and Microsystems,
pages 1–4, April 2015.
93