Constructie Paleta

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 8

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/329612453

Hybrid Cells for Multi-Layer Prepreg Composite Sheet Layup

Conference Paper · August 2018


DOI: 10.1109/COASE.2018.8560586

CITATIONS READS
21 1,109

6 authors, including:

Rishi Malhan Ariyan Kabir


University of Southern California University of Southern California
37 PUBLICATIONS 695 CITATIONS 41 PUBLICATIONS 750 CITATIONS

SEE PROFILE SEE PROFILE

Aniruddha Shembekar Brual Shah


University of Southern California University of Southern California
14 PUBLICATIONS 386 CITATIONS 37 PUBLICATIONS 959 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Rishi Malhan on 23 April 2021.

The user has requested enhancement of the downloaded file.


Hybrid Cells for Multi-Layer Prepreg Composite Sheet Layup
Rishi K. Malhan1 and Ariyan M. Kabir1 and Aniruddha V. Shembekar1 and
Brual Shah1 and Timotei Centea2 and Satyandra K. Gupta1

Abstract— We propose a hybrid human-robot cell for reduc- or overlaps between parallel fiber or tape courses, leading to
ing labor in multi-layer prepreg composite sheet layup process. possible defects in cured parts. More importantly, AFP and
The human expert performs initial experiments to assess the ATL systems are often incompatible with small or medium-
process sequence, and the direction of the layup. The expert
knowledge is then transferred to the system in the form of high sized parts with complex (e.g. multi-curved) geometries due
level commands which are converted into low-level collision free to technical, size, or cost constraints.
trajectories for the robots in the cell. This paper demonstrates Currently, geometrically-complex parts are most often
the feasibility of the composite sheet layup using the robotic fabricated from prepreg sheets by human operators, us-
manipulators along with the design of end-effector tools that are ing hand layup. Prepreg sheets offer advantages over fiber
used during the layup process. We give a comparison between
manual and hybrid cell for execution times in different steps tows and tapes, including higher areal weights (200 - 400
involved. Finally, we give an overview of process consistency g/m2 , enabling rapid thickness build-up), multi-directional
benefits of the hybrid cell. fiber orientations (in woven fabrics), and consistent fiber
alignment (without gaps and overlaps). However, to date,
I. I NTRODUCTION layup of prepreg sheets over complex molds has not been
Robots are increasingly being used in a wide range of rigorously automated, leading to low throughput, part-to-part
industrial application areas like assembly [1], sheet lami- variability due to human factors, and high labor costs. Hybrid
nation based additive manufacturing [2], robotic finishing manufacturing cells can increase automation levels in this
[3], and composite manufacturing. Composites are advanced context, complement the use of AFP/ATL by addressing a
structural materials consisting of fiber reinforcements with current capability gap, and serve as a disruptive technology
high specific strength and stiffness embedded within an by replacing human activity with rigorous, science-based
engineered polymer matrix. The U.S. composites industry methodologies that improve efficiency, increase consistency,
is growing at a compound annual rate of 5%, and was and allow humans to focus on high-value tasks.
valued at $7.5 billion in 2015, with 62% of this value located Layup is challenging because prepreg, a flat fiber-
in manufacturing-intensive aerospace, ground transportation, reinforced material that is relatively inextensible along the
marine and clean energy sectors [4]. High-performance com- fiber direction, must be draped onto a complex (often doubly-
posites are most often produced by laying down sheets of curved) mold geometry while retaining desired fiber orien-
”prepreg,”, i.e. a sheet of carbon fiber reinforcements pre- tations and avoiding damage. Shapes with double curvature
mixed with an uncured, liquid polymer resin on a con- are typically “undevelopable,” such that a flat sheet cannot
toured tool (or mold). Multiple prepreg layers (sometimes be wrapped around the shape without wrinkling or splicing.
> 100 layers) can be required to achieve target stiffness and However, prepregs based on woven fabric reinforcements can
strength. After layup, the resulting laminate is cured using a exhibit large in-plane deformation via trellis shear, allowing
prescribed heat and pressure cycle in an autoclave or oven forming over doubly curved shapes. During hand layup,
to convert the resin from a fluid into a vitrified solid and human laminators manually apply shear and compression
maximize its properties. loads to the woven prepreg in localized regions, typically
Prepreg can be supplied in narrow forms (fiber tows or 1-10 cm2 at a time [6] [7].
tapes measuring 5 - 50 mm) or roll sheet stock (up to In this paper, we present an approach for automating
1200 mm wide). Fiber tows and tapes are unidirectional, prepreg sheet layup on three complex-shaped molds using
lightweight (145 g/m2 ), compliant, and can be placed using a hybrid cell. A planning module generates collision-free
automated fiber placement (AFP), automated tape layup trajectories and instructions for a multi-robot system based
(ATL) or automated dry fiber placement (ADFP) methods, on expert user input. The robotic system performs repeated
which consist of a large, complex material-dispensing head low-level tasks associated with layup. We will be referring to
attached to a gantry or robotic arm programmed to move the geometry on which the sheet is being conformed as the
in prescribed patterns relative to the mold. Such automation “mold” and to the end-effectors being used as “tools” hereon.
enables rapid material laydown rates but also presents disad- Fig. 1 shows the CAD model of the three parts which are to
vantages [5]. Material placement often results in either gaps be made by our cell. We chose a mock-up of a GE-90-115B
fan blade and a jet engine half-nacelle as commercially-used
1 Center for Advanced Manufacturing, University of Southern California,
parts, as well as a custom-designed part incorporating com-
Los Angeles, CA, USA [guptask]@usc.edu
2 M.C. Gill Composites Center, University of Southern California, Los plex features typically considered challenging for prepreg
Angeles, CA, USA [centea]@usc.edu layup.
multitude of different grasps and actions during layup to form
prepreg over complex parts [6]. Automating this process
requires the robots to change holding position and trajectory
during the process because prepreg behaves non-linearly and
its state and shape can differ between runs [7].
III. OVERVIEW OF A PPROACH
The primary goal of our approach is to allow automated
placement of prepreg sheets onto a complex-shaped mold
using a hybrid cell. The placement process should be robust
and generally-applicable, as well as capable of conforming
Fig. 1: CAD model of three parts that are targeted for this work. A. prepreg to the mold surface without formation of layup-
GE90-115B Fan Blade mock-up. B. Jet Engine Nacelle half mock- induced defects (e.g. wrinkling, bridging over concave cor-
up. C. Custom Designed Mold. CAD model of three Molds used
ners, fiber misalignment, or local damage caused by exces-
for this work. P. GE90-115B Fan Blade. Q. Jet Engine Nacelle half.
R. Custom Designed sive application of force). It should also minimize the time
required to perform layup. The major steps in our approach
II. R ELATED W ORK
are described below.
A. Automation in Composites Manufacturing
Robots are used primarily for AFP, ATL and similar 1) Mold Design: Mold design is primarily dictated by the
variants. However, these methods have disadvantages that desired geometry and strength of the part being manu-
include high setup costs, defects caused by fiber steering, factured. Complex geometries often impose challenges
and technical, economic, and physical limitations that render for manufacturing due to the presence of features such
such processes unsuitable for small, highly-complex parts as double curvatures and sharp corner radii, for which
[5]. Here, we demonstrate that advances in collaborative design-for-manufacturing techniques are not prevalent.
robotics can be used in a hybrid cell to perform placement For our work, we have chosen three molds. Figure
of prepreg sheets, increasing the level of automation in the 1 shows a simple concave shape replicating the jet
production of parts unsuitable to traditional layup techniques engine nacelle half, a mold for a mock-up of General
like AFP and ATL. Electric GE90-115B fan blade, and a custom-designed
mold with hemisphere, ramp, and corner regions. We
B. Automated Sheet Layup machined the custom-designed mold with Aluminum
Researchers at the University of Bristol performed the 6061, while the other two molds were additively-
first rigorous analyses and automation studies for prepreg manufactured from ABS plastic to limit costs.
sheet placement. A press was first used to shear the prepreg 2) Material Selection: Material selection is mainly dic-
material, circumventing the need for extensive local shearing tated by structural performance requirements, but can
during layup. Then, the sheets were placed using a 6-axis also be affected by costs and availability. The fiber
industrial robot (ABB IRB 140). Use of specific end effector bed and resin affect the layup process because they
tools based on geometric mold features was also investigated influence the prepreg sheet compliance, drapeability,
[7]. Pre-shearing simplifies the layup step, but requires an and tack (or level of adhesiveness before cure). The
additional step and separate press tooling for every part prepreg used in this paper consisted of an aerospace-
geometry, reducing efficiency and generality, and adding to grade woven carbon fiber fabric (Hexcel F3C, five
equipment costs. Furthermore, since the substrate geometry harness satin, 285 g/m2 areal weight) pre-impregnated
changes as multiple prepreg plies are laid down, pre-shearing with an epoxy (Hexcel HexPly 913).
may become progressively less effective for thick laminates. 3) Drape Simulation: Draping simulators (e.g. Siemens
Here, we focus on automated prepreg sheet layup without FiberSIM [16]) are used to determine the sheet geome-
pre-shearing, showing that a heating system can be incorpo- tries required for draping on a given CAD model [17],
rated into end effector and adequate shear can be achieved [18]. Figure 2 shows the simulations of a prepreg sheet
using impedance motion control and collaboration between on three molds used in our experiments. Simulations
multiple robots. predict whether two-dimensional sheets with known
C. Manipulation of Flexible Material using Multiple Robots material properties can be draped onto a contour, or
Manipulation of flexible materials requires coordinated if fiber locking angles are reached, represented by
motion and control of multiple different robots [8]–[11]. red regions in Fig. 2. Fiber locking angles define the
Learning from demonstration methods have been adapted to minimum relative orientation of two tows in a weave,
learn tasks specific to deformable parts [12], [13]. Previous below which wrinkles can arise and cutting/splicing
attempts at automating sheet layup have had limited success. might be required. Simulation results can include a
For instance, systems created by Newell [14] and Molfino fiber bed shearing map (Fig. 2), which can predict fiber
[15] used four robotic arms to grasp corners of the plies direction along the surface of the mold. Such infor-
using only the initial grasping locations without changing mation allows human experts to assess start locations
or readjusting the sheet. In contrast, human laminators use a and layup sequences, and provide insights into how
wrinkles and voids during layup can be avoided. and collision avoidance. We used a hybrid of sampling, opti-
mization, and search-based algorithm [19] to find the optimal
placement location for each agent in the hybrid cell. During
each iteration, the robot trajectories were simulated and Flex-
ible Collision Library [20] was used to detect collision. Fig.
3 illustrates an on-going layup using the multi-robot system.
A video for this cell working on the mock-up of GE90 fan
blade can be found online at https://youtu.be/3XT9nzCwlOs
We also conducted experiments on custom-made aluminum
mold we mentioned before, using a cell with 3 DOF grasping
robots in [21]. We concluded that manipulation of sheet was
not possible with 3 DOF system and 6 DOF grasping robots
were needed. A video for our previous work can be found
online at https://youtu.be/yUAUaATqINw

Fig. 2: Drape simulation in Siemens Fibersim.

4) Prototyping by Human Expert: Experts carry out a


limited set of manual trials with different placement
strategies. Previous studies have shown that the angles
and fiber behavior depends upon the start point on the
mold [18]. The experts record the sequence of start
and end point pairs for each successful layup strategy.
This dataset forms the basis of a generic sequence of
paths on the mold, though it does not take into account
the non-linear behavior of the sheet during the process
which poses the main challenge for automation.
5) High-Level Planning by Human Expert: Through our
interface, the experts can indicate the layup sequence
to the automated system (Section V-B, V-C). For Fig. 3: An ongoing composite sheet layup in the hybrid cell on the
mock-up of GE90-115B mold. Kuka iiwa 7 equipped with a custom
example, they can click on the CAD model of the
designed roller is used as conforming robot. Epson S5 (left) and
mold to denote the start and end points for each layup Epson C3(right) are used for grasping.
trajectory the robot needs to carry out, or for robot
grasping points.
6) Automated Generation of Robot Instructions: Motion B. Mold Geometry
planning algorithms take the expert user’s input and The GE90 Fan blade mold measures 292×160×93 mm,
generate instructions for the robots. We have used jet engine nacelle is 225×200×90 mm, and custom designed
three robots for experimentation: one conforming robot mold measures 254×254×88.9 mm. The fan blade mold
(KUKA iiwa 7) and two holding (or grasping) robots consists of a complex saddle-shaped surface. The jet engine
(Epson S5 and C3). The conforming robot executes nacelle is formed from a steep concave double curvature with
draping and dibbing (local compression) motions to radius decreasing from 90 mm to 67.5 mm along its length.
conform the prepreg sheet to the mold surface. This The custom-designed mold features slopes (ramps) angled at
robot also shears the sheet while in motion. The 55◦ and 50◦ , intersecting in a corner of radius 4.5 mm. The
holding robots grasp and move the sheet to avoid molds were designed according to standard guidelines. The
undesired contact between the sheet and the mold. Our mold designs are shown in Fig. 1.
motion planning algorithms (Section V-D) generate the C. End Effector Design
instructions for the robots to carry out the layup in KUKA iiwa 7’s end-effector was equipped with a custom
collaboration. roller and dibber tools (Fig. 4 A, B). During layup, draping
IV. C ELL D ESIGN tools are mostly used for convex and flat regions. The roller
A. Robot Layout was chosen as a draping tool because it can apply pressure
KUKA iiwa 7 was chosen based on the requirement of without slipping and orientation can be changed without
impedance and force control during conformation, and the affecting contact and shear forces due to its circular geom-
two holding robots (Epson S5 and C3 for this cell) were etry. The dibber is a sharply curved tool approximating a
required based on the motion mapping discussed in Section human finger, useful for tight concave corners. End-effectors
V-B. The optimum locations of the mold and robots were were additively manufactured from ABS plastic. Silicone
selected based on the kinematic constraints of the robots (Smooth-On) with a Shore hardness of 30a was used for the
{(g11 , g12 ), (g21 , g22 ), . . . , (gm
1 2
, gm )}.
3) Sequence of start and end point pairs for the draping
paths S = {{ps1 , pe1 }, {ps2 , pe2 }, . . . , {psm , pem }}
The system generates the following robot instructions
using this input.
1) The draping trajectories (τdrape ) and dibbing trajecto-
ries (τdib ) as an ordered sequence of waypoints.
2) The process parameters draping force (Fdrape ), dibbing
force (Fdib ), tool velocity (v), temperature of airflow
(T ), Stiffness of the conforming robot (K))
3) The repositioning trajectories (τreposition ).
Fig. 4: Design of the draping and dibbing tools used for the prepreg 4) The sequence of grasping and moving motions for the
sheet layup. (a) The roller for convex surfaces, (b) The dibber for holding robots. (τgrasp )
concave surfaces, (c) The draping and dibbing tools attached to the
consolidating robot while the robot executes dibbing trajectories on B. Identification of Motion Primitives
the concave feature of custom designed mold. Human motions need to be mapped to the robots to
determine the grasping techniques. The research presented
roller and dibber to achieve the desired stiffness and avoid
in [6] performs an extensive study of human motions during
adhesion with prepreg.
the layup process and categorizes all the motions into eight
The roller radius is limited by the part geometry and
generic techniques. Below are the techniques that can be
minimum curvature being draped. The roller material is
mapped to the motions performed by the robot.
limited by the desired stiffness K (11 N mm−1 for our
case). Stiffer rollers can fail to conform to tight curvatures, 1) Sheet Guiding: Holding robots guide the sheet, align-
requiring more force to achieve the desired contact patch. ing and manipulating it while the conformation robot
The roller width is restricted by the geometry of the mold, performs draping.
with narrower rollers preferable for tight corners. 2) Tension Shearing: Fabrics can be sheared using tensile
force. Holding robots can grip the sheet and stretch
D. Heating System Design it in opposite directions, or the roller can dynamically
Prepreg sheets can be heated during layup to increase tack shear the sheet due to tension between the fixed and/or
(adhesiveness), reduce resin viscosity, and facilitate deforma- already conformed area behind the roller and the slack
tion during placement. In our setup, a warm air flow (45◦ C at regions in front of it.
0.5 Mpa) was maintained through internal channels created 3) Tool Interaction: Humans use a technique of mold
in the draping and dibbing tools. The temperature was interaction wherein the sheet is stretched using a corner
selected based on experiments. The pressure was determined or other suitable profile of the mold [6]. Robots use
according to heat convection in the prepreg at the given tool interaction wherein the sheet is held under friction
velocity. Nozzles allowed air flow vectoring and, hence, as by the holding robot till there is a lack of sheet at
shear was applied during conformation, the prepreg sheet the roller end. Force generated by roller forces more
deformed in desired regions rather than globally, allowing it sheet to be released from the grippers overcoming the
to conform to local contoured surfaces. Fig. 4 shows a nozzle friction and thus shearing it.
attached to the end effector.
C. Identification of Process Parameters
V. AUTOMATED I NSTRUCTION G ENERATION The process parameters affect the performance of
A. Problem Statement layup. Initial process parameter values lie in the bounds:
Let, C be the CAD model of the mold/part and M Fdrape , Fdib ∈ [0.5, 6] N, vdrape ∈ [10, 100] mm-s−1 ,
be the surface where the composite prepreg sheet will be T ∈ [35, 70] ◦ C, and Kdrape , Kdib ∈ [0, 5000] N-m−1 which
consolidated. M can be represented as a collection of paths, needs to be constrained for any layup setup. Each of these
M = {p1 , p2 , . . . , pm }. Each path is defined with a start parameters are limited by the maximum physical constraints
point and an end point, i.e. pi = (psi , pei ) provided by expert of the KUKA iiwa 7 robot, and the heating system. We
using our interface. From the ordered set, S, of start and identify the right values of these parameters through carefully
end point pairs, our motion planning module can generate designed experiments conducted on double curvature, single
the draping trajectories. For each path a geodesic curve [22] curvature, flat (zero Gaussian), concave, and convex surfaces.
can be generated on M with (psi , pei ) as input. Each geodesic The experiments were designed over three levels
curve (i.e. each path) can be represented as a collection of X1 , X2 , X3 . In each level, three values (low, medium, high)
waypoints, i.e. pi = {w1 , w2 , . . . , wni } and pi ⊂ SE(3). of each parameter were considered for both draping and
The expert user provides the following inputs to the dibbing. The parameters which caused a violation of any
system- draping or dibbing characteristics at X1 were discarded. X2
1) The CAD model of the mold/part, C. and X3 comprised of a refined range of the parameters based
2) Sequence of grasping locations for the on the experiments of the previous level. The results of the
holding robots as a set of tuples, G = experiments at level X3 were then used for the sheet layup.
We needed feedback on layup or surface quality during surface normals around the face represents concave regions.
process parameter identification. Inspection of the surface If normals do not intersect, it is convex region.
after conformation was carried out using a digital micro- 2) Generating Draping Trajectories
scope (DinoLite AM7915MZT). Reference standards were Once we have S, we generate a geodesic curve on the
established for comparison as shown in Fig. 5. Carbon fiber target surface (M) for each start-end point pair ({psi , pei }).
is visible as a woven fabric, with wet shining resin areas The geodesic curves are shortest paths on free-form para-
representing an undamaged surface. Damaged regions and metric surfaces [22] [23]. We computed geodesics on the
a comparison of roller and hand layup are also shown. triangular mesh of the mold that the system takes as an input
The following quality metrics were determined for each [24]. Each geodesic constitutes a sequence of way-points. We
experiment: need to define collision-free configurations of the robot along
1) Resin Quality: At high pressure and temperature, any these points. Collision-free trajectories between successive
applied contact displacement, parallel to the surface geodesic paths is generated by an algorithm similar to
can disturb the resin distribution by displacing resin CODES3 described in [25]. As stated earlier in Section IV-C,
which should be avoided. roller was chosen as its circular geometry constrains only 5
2) Conformity: Layup should produce excellent confor- degrees of freedom in the pose of the robot. The sixth degree
mity of sheet to the mold surface without voids (air of freedom or angle made by the surface normals of roller
entrapped) and wrinkles. and the mold can be set as a variable. For planning in a space
3) Fiber Alignment: Draping can lead to fiber misalign- with lesser degrees of freedom, we use search to determine
ment relative to neighboring tows and the global mold a trajectory which offers continuous motion over each of
orientation. Fiber angle deviations are permissible up the paths. Each depth in the graph computed corresponds to
to a certain extent (since they can be unavoidable all the valid collision-free configurations of the robot at that
during shearing). High deviations cause wrinkling. way-point. We search through this tree for different initial
configurations of the robot at the start point. We used the
norm of angle change in each joint as the cost to go to a
new state. Manipulator is programmed to keep the cartesian
velocity of roller constant. This velocity is selected through
identification of process parameters (Section V-C).
3) Motion Planning for Holding Robots
For each patch (Pi ), a grasp location is determined for
holding robots. Holding robots need to manipulate the sheet
such that there is no contact between the sheet and the mold
except the point at which the conforming robot is applying
the force. Based on the geometry of the mold and sheet, our
system determines where and how the grasping robots need
to move the sheet. We consider end-points of all trajectories
in Pi as a frontier. We translate this frontier away from the Pi
Fig. 5: Different surfaces as captured by digital microscope (a)
Original surface. Few resin areas marked in red. (b) Resin devoid which is being currently draped by the conforming robot by
surface showing dry and damaged fibers marked in red. (c) Surface a distance greater than or equal to some positive value such
after layup by a Roller. (d) Surface after a Hand Layup. that there is no collision between conforming and holding
robots. We constrain the holding robots to grasp from the
edges of the sheet. We project rays from initial frontier to
D. Generation of Joint Space Trajectories
the translated frontier and iteratively increase the angle made
The system goes through the following steps to generate by the rays with the mold until there is no collision between
the joint space trajectories for the robots. the rays and the mold. Grippers of the holding robots are
1) Identify concave and convex regions on the surface and aligned with the normals of the surface at the grasping point.
generate dibbing and draping paths respectively. Trajectories from one grasp point to the other are calculated
2) Generate draping (τdrape ), dibbing (τdib ), and repo- by an algorithm similar to CODES3 [25].
sitioning (τreposition ) trajectories for consolidating
E. Process Parameter Experiments
robot.
Under impedance control, the minimum Fdrape necessary
3) Generate corresponding moving and shearing trajecto-
for layup was 10 N, below which conformity degraded.
ries (τgrasp ) for the holding robots.
Higher force of Fdrape =20 N was required for double
4) Start executing planned motions after taking approval
curvature compared to flat and single curvature. Corners
from human expert.
required Fdrape =30 N. These results were obtained at v=20
1) Identifying Concave and Convex Regions mm-s−1 and T =45 ◦ C. For v=40 mm-s−1 , the conformity
Neighbors of a specific face on the surface were computed deteriorated at Fdrape =10 N but remained unchanged at
for regions having a change in gradient. Intersection of Fdrape =20 and 30 N. Resin quality was satisfactory for
all the above cases. Draping characteristics were unaffected maximum deviation tolerance of 15◦ was kept. 75% of the
by v as high as 60 mm-s−1 beyond which the motion region was found to have fibers aligned within 5◦ of average
was too haphazard causing low conformity and high fiber deviation. Layup on the custom-made mold was cured under
misalignment. Trajectories were executed without slippage manufacturer recommended temperature and pressure cycles
of the roller. Fdib =30 N was essential to prevent poor to produce a final component. Visual inspection showed no
conformity in concave regions. Fdrape =35 N caused damage signs of defects caused by resin dislocation showing that the
to the prepreg resin surface but allowed high conformity parameters chosen were appropriate for the process. Fig. 7
by shearing the sheet to a great extent. A temperature of shows the cured part.
45◦ C was suitable in all cases. Lower temperature reduces
tack, lowers conformity, and increases required shear forces.
Higher temperatures cause excessive tack and resin dislo-
cation (or adhesion to the roller). Velocity and pressure are
dependent on dimensions of the roller (in our case, φ40 mm,
width=28 mm & K=11 N-mm−1 ), which define a contact
patch area. The pressure vs force relation helps to generalize
these results and implement process parameters over different
designs. Pressure at Fdrape =10, 20, 30, 40 N was determined
to be 0.07, 0.1, 0.2, 0.25 MPa, respectively.
F. Manufacturing Quality Analysis

Fig. 8: Deviation color map obtained by laser scanning. The color


map shows deviations of the parts from tool dimensions.

Fig. 6: Automated Layups A. GE90-115B Fan Blade replica. B. Jet


Engine Nacelle half. C. Custom Designed Mold.

Fig. 6 illustrates the three


layups performed using our cell.
A laser coordinate measurement
machine (Hexagon AB, Romer
Absolute Arm with integrated
laser scanner) was used to scan
the parts for conformity (Fig.
Fig. 7: Cured part made 8). Three layers of prepreg were
from the layup on custom deposited building up 2.25 mm
Fig. 9: Color map shows the fiber alignment on layup of mold
designed mold. thickness. Results showed the replica of GE90 fan blade. 75% of region was observed within
surface was within a tolerance deviation of 5◦ was observed, and some of remaining regions went
of [−0.4, +0.4] mm indicating satisfactory conformation up to 15◦ .
with the mold contour. Fiber alignment for automated layup
was closer to drape simulation output and consistent for all
layers. Fig. 9 shows the results for layup on mold replica of G. Execution Time and consistency Comparison
GE90 fan blade. Apodius Vision System 3D integrated with Table I gives a comparison of execution times for both
Hexagon AB absolute Romer arm was used to determine manual and hybrid cell at different steps. Hybrid process is
the fiber alignment of the uncured layups. It determines the more consistent as identical trajectories are executed from
average fiber distribution in a region and detects the areas one layer to the other layer. Hand layup can cause fiber
where this alignment changes with respect to the regional misalignment due to human variability. A higher number of
average. Consistent direction of fibers was observed when a layers and safety factors are needed to avoid this problem in
TABLE I: Time taken by different operations in Human Operated
[5] D. H. J. A. Lukaszewicz, C. Ward, and K. D. Potter, “The engineering
Cell and Hybrid Cell aspects of automated prepreg layup: History, present and future,”
Item Time (minute) Applicable for Composites Part B: Engineering, vol. 43, no. 3, pp. 997 – 1009, 2012.
Drape Simulation 5.00 Both Cells [6] M. Elkington, D. Bloom, C. Ward, A. Chatzimichali, and K. D.
Human Trials 28.50 Both Cells Potter, “Hand layup: understanding the manual process,” Advanced
Contingency Handling Manufacturing: Polymer & Composites Science, vol. 1, no. 3, pp. 138–
0.75 Both Cells
by Human 151, 2015.
Single sheet layup Only Human [7] M. Elkington, C. Ward, and K. D. Potter, “Automated layup of sheet
9.13
by human (Average Time) Operated Cell prepregs on complex moulds,” in SAMPE Long Beach Conference,
Instruction Generation 2016.
4.00 Hybrid Cell
by Human [8] Z. Peng and L. Yuanchun, “Position/force control of two manipula-
Automated Plan Generation 0.10 Hybrid Cell tors handling a flexible payload based on finite-element model,” in
Plan Verification 2007 IEEE International Conference on Robotics and Biomimetics
5.00 Hybrid Cell
and Adjustment by Human (ROBIO), Dec 2007, pp. 2178–2182.
Reposition of Consolidating Robot 3.11 Hybrid Cell [9] J. K. Mills and J. G. L. Ing, “Dynamic modeling and control of a
Reposition of Holding Robots 0.07 Hybrid Cell multi-robot system for assembly of flexible payloads with applications
Tool Change 0.18 Hybrid Cell to automotive body assembly,” Journal of Robotic Systems, vol. 13,
Grasping and Releasing 0.12 Hybrid Cell no. 12, pp. 817–836, 1996.
Single sheet layup by robots 5.39 Hybrid Cell [10] K. Kosuge, H. Yoshida, T. Fukuda, M. Sakai, and K. Kanitani, “Ma-
nipulation of a flexible object by dual manipulators,” in Proceedings
hand layup. This can lead to higher part weight and higher of 1995 IEEE International Conference on Robotics and Automation,
cost. vol. 1, May 1995, pp. 318–323 vol.1.
[11] H. Dominik and W. Heinz, Robot Manipulation of Deformable Ob-
VI. C ONCLUSIONS jects. Springer Science & Business Media, 2012.
[12] A. X. Lee, H. Lu, A. Gupta, S. Levine, and P. Abbeel, “Learning
In this paper, we have presented an automated composite force-based manipulation of deformable objects from multiple demon-
layup hybrid cell that automates nearly 60 % of the layup strations,” in 2015 IEEE International Conference on Robotics and
Automation (ICRA), May 2015, pp. 177–184.
process (considering the worst-case custom designed mold). [13] H. Lu, “Learning manipulation of deformable objects from multiple
The instructions and trajectories for conforming and grasping demonstrations,” 2015.
robot motions required for layup, are automatically generated [14] R. O. Buckingham and G. C. Newell, “Automating the manufacture
of composite broadgoods,” Composites Part A: Applied Science and
from high-level expert input. Finally, we presented an exe- Manufacturing, vol. 27, no. 3, pp. 191 – 200, 1996.
cution time comparison and overview of process consistency [15] R. Molfino, M. Zoppi, F. Cepolina, J. Yousef, and E. E. Cepolina,
benefits of the hybrid cell. Future advances can include “Design of a hyper-flexible cell for handling 3d carbon fiber fabric,”
Recent advances in mechanical engineering and mechanics, vol. 165,
integrating online inspection systems to perform contingency 2014.
handling and correction of trajectories. An algorithm will be [16] E. Bernardon, “Fibersim, cad integrated software tools for composite
developed in the future to auto-generate drape trajectories manufacturing and design,” in Society for the Advancement of Material
and Process Engineering. International SAMPE Europe conference,
and grasping trajectories for robots by evaluating the drape 1998, pp. 163–173.
model which in our work, were mainly governed by expert [17] S. B. Sharma and M. P. F. Sutcliffe, “Draping of woven fabrics:
user inputs. Progressive drape model,” Plastics, Rubber and Composites, vol. 32,
no. 2, pp. 57–64, 2003.
VII. ACKNOWLEDGMENTS [18] S. G. Hancock and K. D. Potter, “The use of kinematic drape mod-
elling to inform the hand lay-up of complex composite components
This work is supported in part by National Science using woven reinforcements,” Composites Part A: Applied Science and
Foundation Grant #1634431. Opinions expressed are those Manufacturing, vol. 37, no. 3, pp. 413 – 422, 2006.
of the authors and do not necessarily reflect opinions of [19] A. M. Kabir, J. D. Langsfeld, C. Zhuang, K. N. Kaipa, and S. K.
Gupta, “A systematic approach for minimizing physical experiments
the sponsors. The authors acknowledge Siemens PLM for to identify optimal trajectory parameters for robots,” in 2017 IEEE
supplying a FiberSIM license, Hexagon AB for facilitating International Conference on Robotics and Automation (ICRA), Singa-
the work with Apodius Vision System 3D, The Boeing pore, May 2017, pp. 351–357.
[20] J. Pan, S. Chitta, and D. Manocha, “Fcl: A general purpose library for
Company for donating prepreg, and Airtech International for collision and proximity queries,” in Robotics and Automation (ICRA),
donating consumables. 2012 IEEE International Conference on. IEEE, 2012, pp. 3859–3866.
[21] R. K. Malhan, A. M. Kabir, B. Shah, T. Centea, and S. K. Gupta,
R EFERENCES “Automated prepreg sheet placement using collaborative robotics,” in
Society for the Advancement of Material and Process Engineering
[1] R. K. Malhan, Y. Shahapurkar, A. K. Kabir, B. Shah, and S. K. Gupta, (SAMPE) Technical Conference and Exhibition, Long beach, CA,
“Integrating impedance control and learning based search scheme USA, May 2018.
for robotic assemblies under uncertainty,” in ASME Manufacturing [22] T. Maekawa, “Computation of shortest paths on free-form parametric
Science and Engineering Conference, College Station, TX, USA, June surfaces,” Transactions of the ASME-R-Journal of Mechanical Design,
2018. vol. 118, no. 4, pp. 499–508, 1996.
[2] P. M. Bhatt, M. Peralta, H. A. Bruck, and S. K. Gupta, “Robot assisted [23] K. REDFIELD, “Finding geodesics on surfaces,” 2007.
additive manufacturing of thin multifunctional structures,” in ASME [24] D. Martı́nez, L. Velho, and P. C. Carvalho, “Computing geodesics on
Manufacturing Science and Engineering Conference, College Station, triangular meshes,” Computers & Graphics, vol. 29, no. 5, pp. 667–
TX, USA, June 2018. 675, 2005.
[3] A. M. Kabir, A. V. Shembekar, R. K. Malhan, R. S. Aggarwal, [25] A. M. Kabir, B. C. Shah, and S. K. Gupta, “Trajectory planning
B. C. Langsfeld, J. D.and Shah, and S. K. Gupta, “Robotic finishing for manipulators operating in confined workspaces,” in 2018 IEEE
of interior regions of geometrically complex parts,” in ASMEs 13th International Conference on Automation Science and Engineering
Manufacturing Science and Engineering Conference, College Station, (CASE), Munich, Germany, Aug 2018.
Texas, USA, June 2018.
[4] Composites Manufacturing. (2016) State of the composites industry.
[Online]. Available: http://compositesmanufacturingmagazine.com/
2016/01/state-of-the-composites-industry-lucintel-2016

View publication stats

You might also like