Comparison of Numerical Methods for Geometric

Warpage Compensation
Steffen Tillmann

RWTH Aachen University: Rheinisch-Westfalische Technische Hochschule Aachen

https://orcid.org/0009-0006-5895-0406
Stefan Basermann
Stefanie Elgeti

Research Article

Keywords: Injection molding, Warpage compensation, Numerical methods, Cavity shape adaption,
Algorithm comparison

Posted Date: February 20th, 2024

DOI: https://doi.org/10.21203/rs.3.rs-3959260/v1

License:   This work is licensed under a Creative Commons Attribution 4.0 International License.
Read Full License
Comparison of Numerical Methods for Geometric
Warpage Compensation
Steffen Tillmann1*, Stefan Basermann1 and Stefanie Elgeti1,2
1* Chair for Computational Analysis of Technical Systems, RWTH
Aachen University, Schinkelstraße 2, Aachen, 52062, Germany.
2 Institute of Lightweight Design and Structural Biomechanics, TU

Wien, Gumpendorfer Straße 7, Vienna, A-1060, Austria,.

*Corresponding author(s). E-mail(s): [email protected];

Contributing authors: [email protected];
[email protected];

Abstract
In injection molding processes, shrinkage and warpage often cause deviations in
the shape of produced parts compared to the cavity shape. These deviations arise
due to uneven cooling and internal stresses within the part. One approach to mit-
igate these effects is by adjusting the cavity shape to anticipate the deformation.
This can be achieved by simulating the expected deformation using suitable mod-
els, which then inform the optimization of the cavity shape for injection molded
parts with minimal deformation.
In this study, we evaluate various numerical algorithms from existing literature
to identify the optimal cavity shape. Each method is briefly outlined regarding
how it adapts the geometry, and we discuss their respective strengths and weak-
nesses for different scenarios. We conduct comparisons using 3D geometries of
varying complexity. Our findings demonstrate that, for geometric warpage com-
pensation, the node-based reverse geometry method yields the least warpage and
is computationally cost-effective. Furthermore, it is straightforward to implement
and consistently performs well across different geometries.

Keywords: Injection molding, Warpage compensation, Numerical methods, Cavity

shape adaption, Algorithm comparison

1
1 Introduction
Injection molding is one of the most widely-used manufacturing process for plastics
products. Naturally, finding new means of increasing both process efficiency and prod-
uct quality is an ongoing field of research. One key challenge in terms of product
quality are lack of shape accuracy due to shrinkage and warpage. While shrinkage is
caused by the density-dependency on both temperature and pressure of plastics melts,
warpage results from uneven shrinkage either due to shape constraints inside the mold
or an inhomogeneous temperature distribution within the part during and after solid-
ification. [1]
How extensive the shrinkage and warpage is, is influenced by many factors; next to
the material properties, these are in particular the component design, the process
parameters–such as melt temperature, injection time, injection pressure, holding pres-
sure, and cooling time–as well as the cavity design [2]. Each aspect listed can be
optimized numerically to reduce warpage. The main means of influence with respect
to component design is wall thickness. Large, thin-walled parts are particularly prone
to warpage [3, 4]. If the wall thickness distribution can be changed from a functional
standpoint, warpage can be reduced [5]. For the optimization of process parame-
ters, various algorithms have been utilized, such as genetic algorithms [6, 7], response
surface methodology [8–10], Bayesian optimization [11, 12], and particle swarm opti-
mization [2, 13, 14]. Furthermore, there are methods that use locally adjusted cooling
to reduce warpage [15, 16].
This paper will concentrate on the third category of methods for minimizing warpage,
involving the adjustment of the mold cavity shape to offset the occurring warpage.
In this approach, the produced part from the modified cavity shape still undergoes
warpage, but it ultimately attains the desired shape after deformation [17]. While this
method can be implemented through iterative experiments, every time measuring the
newly produced part and adjusting the cavity accordingly, this proves to be both costly
and time-consuming [18, 19]. As an alternative, several simulation-based numerical
methods have been suggested to calculate the optimal geometry for the cavity shape.
We have selected four particularly common methods from literature. The purpose of
this paper is to compare these methods in terms of result quality and computational
effort. This will enable future users to make an informed choice of the appropriate
method for their application. In the following, we will briefly describe the selected
methods.
Method (1): The first approach for warpage compensation involves using an inverse
model to determine the cavity geometry, as proposed by Zwicke et al. [20–22]. The
method begins with a forward simulation to calculate the temperature distribution
before part ejection. The inverse model then uses this initial temperature distribution
to compute the adjusted cavity shape. This approach is computationally inexpensive,
requiring only one forward and one inverse run of the model. However, the initial
temperature distribution is computed only once and remains unchanged for the com-
pensated cavity shape. A similar inverse method was used in [23] for 3D printing.
Note that the first method can provide an answer to the optimal cavity shape in one
shot. The three remaining methods instead are iterative methods.
Method (2): A simple method involves comparing the positions of each mesh node

2
between the deformed and ideal geometries. In each iteration, the cavity geometry is
adjusted in the opposite direction of the disparity between the current deformed geom-
etry and the ideal geometry. This method is inspired by experimental cavity shape
compensation, where discrete measurements of produced parts inform adjustments
[18, 19, 24]. However, this so-called reverse geometry method [25] requires identical
meshes with corresponding nodes for both the deformed and ideal shapes.
Method (3): Kastelic et al. [25] introduced an alternative method for updating the cav-
ity shape. The method determines the intersection point of the surface normal vector
with the ideal geometry and adjusts the surface mesh nodes along the direction of the
normal vector to compute the new cavity shape. This method is advantageous when
the deformed geometry and the ideal geometry do not share the same mesh. Both the
reverse geometry method and the normal vector method by Kastelic et al. [25] require
very few iterations to compute the cavity shape.
Method (4): Another approach is to formulate the problem as a shape optimization
task [26]. This can be performed in many different ways, e.g., the shape parameteriza-
tion can be parametric [27] or nonparametric [28–30]. In this paper, we have selected
a spline-based parametrization of the geometry, known as free-form deformation [31].
Here, the control points of the spline serve as the optimization parameters. Subse-
quently, Bayesian optimization is used as the optimization algorithm [32, 33].
This paper evaluates different warpage compensation methods on 3D geometries
of varying complexity. We will begin by presenting the governing equations of our
warpage simulation model. Since the primary focus is on the algorithms, we will use
a simple warpage model with a linear elastic material model. It is important to note
that the warpage algorithms can be adapted to more advanced warpage models if
they are available. The following section offers brief explanations of each algorithm,
followed by a comprehensive comparison of the methods. Our goal is to determine the
most efficient method for each geometry based on the results obtained.

2 Simulation Model
This section describes the simulation model used in this paper, including the gov-
erning equations and boundary conditions. Since the main focus of this paper is the
assessment of the optimization procedures, the simulation model is kept as simple as
possible and comprises two components: a heat conduction simulation and a warpage
simulation. The heat conduction simulation is used to determine the initial temper-
ature distribution within the mold cavity prior to part ejection. Subsequently, the
warpage simulation is used to calculate the deformation that occurs as the part cools
down to ambient temperature after ejection. The simulation model is implemented
using the open-source finite element library Nutils [34].

2.1 Heat Conduction Model

To establish a realistic initial condition for the warpage simulation without conducting
a full filling simulation, we solve the heat equation within the cavity domain. The

3
standard heat equation is given by

∂T
= κ∆T in Ω , (1)
∂t
where T denotes the temperature and t the time. Ω indicates the computational
domain and κ the thermal diffusivity. The latter is generally dependent on the thermal
conductivity, the isobaric heat capacity and the density, which itself are temperature
dependent. Here, we use a simplified material model with constant κ. As initial con-
dition we impose a constant temperature TI across the part. In addition, we impose
a Dirichlet boundary condition TD on the entire boundary ∂Ω. TD is constant over
time and has a lower value than the initial condition. In particular,

TD − TI = ∆TD = −240K on ∂Ω . (2)

The heat conduction simulation is run only for a time period τ , during which the
exterior of the part cools while a hot core remains. The resulting temperature field is
called T0 and is used in the subsequent warpage simulation. The utilized parameters
of the heat equation simulation are summarized in Table 1.

Table 1: Parameters used in the

heat conduction simulation
Parameter Value
Thermal diffusivity κ 0.0025 m2 /s
Simulation time τ 0.1 s

2.2 Warpage Model

The warpage simulation approximates the deformation of the material due to shrink-
age and warpage after ejection. We assume a fully solidified, isotropic and stress-free
material with linear elasticity. The problem is described by the linear thermoelastic
constitutive equation, also known as the Duhamel-Neumann law [35], which is given
by:
σ = λ tr(ε)1 + 2µε − α(3λ + 2µ(Ta − T0 )1 . (3)
Here, σ denotes the stress, ε the strain, and α the thermal expansion coefficient. The
temperatures Ta and T0 represent the ambient and initial temperature respectively, λ
and µ are the first and second Lamé constants. The utilized values are summarized in
Table 2.

3 Arrangement for Method Comparison

This section provides a detailed description of the setup utilized for comparing the
methods. We commence by presenting a summary of each method under consideration.

4
 Table 2: Parameters used in the
warpage simulation
Parameter Value
Ambient Temperature Ta 0◦ C
Thermal expansion coeff. α 10−4 m ◦ C
First Lamé const. λ 4 GPa
Second Lamé const. µ 1 GPa

Additionally, we elaborate on the test geometries and the quality criteria employed to
assess the methods.

3.1 Numerical Methods

3.1.1 Inverse Shape Design
In [21], the authors introduce the inverse shape design method, which aims to calcu-
late the optimal shape of the cavity directly. The concept entails creating an inverse
warpage simulation. The desired shape is used as input, and the unknown cavity shape
is the output. Under certain assumptions, it is indeed possible to create an inverse
model. The requirements for this task are to restrict the simulation to warpage and
allow enough time for cool-down to reach a steady state solution. Note that linear
material behavior is explicitly not a requirement. Additionally, an approximation of
the initial temperature field is required as the final cavity shape is unknown prior to
the simulation. To estimate the temperature field in the final cavity shape, we use
the temperature field computed for a cavity that matches the desired geometry. For a
more accurate temperature approximation, please refer to [36].

3.1.2 Reverse Geometry Method

The reverse geometry method is an iterative technique that corrects the cavity shape
by adjusting the mesh coordinates in the opposite direction of the deviation from
the desired shape. That is, starting from an initial estimate of the cavity, a forward
warpage simulation is performed. The deviation between the warped geometry and
the ideal geometry is the basis for the cavity shape correction. Consider the following:
xi represents the mesh coordinates of the cavity shape in iteration i. The deformed
domain resulting from the warpage simulation is denoted by x̃i and the ideal shape is
denoted by xideal . The update scheme is then given by:

x0 = xideal , (4)
xi+1 = xideal − (x̃i − xi ) . (5)

The advantages of the method are that the computational cost does not scale with
the complexity of the geometry and the implementation is simple. The disadvantage
is that the method does not work directly if the geometry is re-meshed, which means
that the same mesh is not available for the ideal and the deformed geometry.

5
3.1.3 Normal-Vector Method
The normal-vector method, developed by Kastelic et al. [25], involves an iterative
adjustment of the cavity shape. This adjustment is carried out by computing the
normal vector at each surface node of the warped geometry mesh and identifying its
intersection point with the corresponding desired geometry surface. Subsequently, the
mesh coordinates of the cavity shape are modified based on the displacement vector
between the warped mesh node and the intersection point.
It is important to note that there is no guarantee that the normal vector at a mesh
node of the deformed geometry will intersect the desired geometry surface. To address
this, surfaces may need to be extended to ensure identification of an intersection point.
In addition, since moving mesh nodes only in the direction of the surface normal
can result in an irregular mesh, mesh regularization techniques must be incorporated.
The method offers advantages such as localized computation of changes and non-
intrusiveness. However, it has the disadvantage of requiring additional computations,
and mesh regularization requires an appropriate choice of hyperparameters to be set
for each geometry.
Remark. The normal vector method, when applied exactly as described in [25], faces
challenges when it comes to geometries with intricate details such as small holes or
ribs. To overcome this problem, additional steps are required to accurately locate the
intersection points of the associated normal vectors. As these supplementary imple-
mentations performed by the authors were not available to us, for a fair comparison,
we use the normal vector method only on a simplified representation of geometry C
of Section 3.3.

3.1.4 Bayesian Optimization Approach

Bayesian optimization (BO) is a powerful technique used in robust optimization to
find the optimal solutions for complex problems with uncertainty or variability in the
system. BO combines probabilistic modeling with optimization to efficiently find the
best solution in a noisy or uncertain environment. It uses an acquisition function, such
as expected improvement (EI) or probability of improvement (PI), to balance explo-
ration (sampling uncertain regions) and exploitation (sampling promising regions) of
the parameter space. The acquisition function guides the search for the best solution
while considering uncertainty. In the frame of robust optimization, BO is strongly
interlinked with Gaussian process regression.
How BO can be used in cavity design is described in detail in [26]. Here, we give a brief
recap. When approaching the warpage compensation problem as a shape optimization
problem, the first ingredient that needs to be determined is the objective function J.
Here, we use a mesh-based measure that sums the squared distance between the ideal
and deformed shapes at each mesh node. This yields:
v
u N
u1 X
J =t ||xideal,j − x̃j ||22 , (6)
N j=0

6
where xideal and x̃ are the mesh node coordinates of the ideal and deformed shape.
N is the total number of mesh nodes, and j indicates the number of the mesh node.
For the deformation and parametrization of the geometry, the free-form deformation
method [31] is employed. In this method, the finite element mesh is embedded in
a box-spline. The positions of the spline control points then serve as optimization
parameters. As these control point positions are adjusted, also the mesh deforms.
Bayesian optimization begins by running the forward simulation model with a small
initial set of random inputs. Using this data, the Gaussian process regression surrogate
model is trained. The optimal next values for balancing exploration and exploitation
lie at the global minimum of the acquisition function. An optimizer seeks the global
optimum of the acquisition function, which is then chosen as the next training point.

3.2 Evaluation Metric

To compare the effectiveness of the different methods we need an evaluation mea-
surement of the current warpage. For the reverse geometry method, the Bayesian
optimization approach and the inverse method, we use Equation 6. This is the objec-
tive function used in the Bayesian optimization approach which calculates the mesh
node based root mean square (RMS) difference. Since the normal vector method moves
the mesh nodes in the surface for regularization purposes, we cannot use the node-to-
node difference. Instead, we calculate the distance d between the mesh nodes of the
deformed surface and the ideal surface:
v
u N
u1 X
J =t ||dj ||22 . (7)
N j=0

In the results section these metrics are normalized to the initial value of one. This is
done to have a meaningful comparison between the different geometries.

3.3 Test Geometries

In order to perform the comparison, a wide range of geometries is used. Figure 1 gives
an overview over these five 3D geometries with different level of complexity. The edge
lengths for these parts are between 30 and 200 mm. Geometry A (Figure 1a) is a box
with a half cylindrical opening and geometry B (Figure 1b) is a parallelogram-shaped
part. Geometry C (Figure 1c) is a symmetrical part with boxes and fins, geometry D
(Figure 1d) is a box with cross shaped fins and geometry E (Figure 1e) is a bicycle
saddle.

7
 (a) Box-shaped geometry A with a half cylindrical (b) Geometry B with a parallelogram-shaped
opening. base.

(c) Geometry C with fins and two upwards opened (d) Geometry D with cross shaped installation
boxes. mounts and U-shaped holes.

(e) Bicycle saddle geometry E.

Fig. 1: The five geometries A-E which are used as test cases for the methods.

4 Numerical Results
This section presents the outcomes of the comparison among the various methods.
Initially, the results obtained from the Bayesian optimization method, the inverse
method, and the reverse geometry method are showcased for all geometries. Subse-
quently, a comparison is made between the reverse geometry method and the normal
vector method, specifically on the simplified geometry.

4.1 Comparison of Bayesian Optimization Method, Inverse

Method and Reverse Geometry Method
Now, the Bayesian optimization method, inverse method and reverse geometry method
are applied to all geometries shown in Section 3.3.

4.1.1 Results of the Bayesian optimization method

In this section, the results of the Bayesian optimization are presented and discussed.
The convergence of the RMS measure (Equation 6) is plotted over 300 iterations
for each of the five geometries is displayed in Figure 2. The Bayesian optimization
algorithm decreased the warpage in the best case of geometry E (1e) to 31 % and in
the worst case of geometry D (1d) to only 70 %. Throughout the iterations of each
geometry, there are sections where the RMS does not decrease. This happens when
the computed training point of the Bayesian optimization is used for exploring the
domain.

8
Fig. 2: Convergence of the normalized RMS difference on a logarithmic scale for the
geometries A-E with the Bayesian optimization method. The RMS difference measures
the difference between the warped part and the ideal geometry. The x-axis shows the
number of iterations.

4.1.2 Results of the inverse method

In Table 3 the results of the inverse method are shown for each geometry. For every
geometry the warpage is reduced to roughly 1 %. The inverse method worked best
on geometry C (1c) with a reduction to 0.95 % and worst on geometry A (1a) with
1.73 %. This shows, that the inverse method is an effective tool to reduce the warpage
regardless of the complexity of the geometry. Additionally, the inverse method is com-
putationally much cheaper than the Bayesian optimization method, as it only requires
one iteration instead of 300 iterations.

Table 3: Inverse method results:

Remaining final warpage as percent-
age of initial warpage of the RMS
difference.
Geometry Normalized RMS Difference

A 1.73%
B 0.97%
C 0.95%
D 1.12%
E 1.71%

9
Fig. 3: Convergence of the normalized RMS difference for the geometries A-E with
the reverse geometry method. The RMS difference measures the difference between
the warped part and the ideal geometry. The x-axis shows the number of iterations.

4.1.3 Results of the reverse geometry method

Now we are showing the results of the reverse geometry method applied to all geome-
tries (Figure 3). Note that the y-axis of the figure is a logarithmic scale and only five
iterations are shown. The warpage is reduced to a factor on 10−4 to 10−8 depending
on the geometry. This method needs way less iterations than the Bayesian optimiza-
tion method and reduces the warpage to a higher degree. As mentioned before, in the
case of a linear elastic material law, the first iteration of the reverse geometry method
and the inverse method lead to the same result. But the reverse geometry method
further decreases the warpage in the subsequent iterations.

4.1.4 Summary
Overall, the reverse geometry performed the best in reducing the objective function.
The Bayesian optimization method required more iterations and lead to a worse result.
The inverse method is the computationally cheapest method as it only needs on iter-
ation, but the warpage reduction is lower than for the reverse geometry method with
five iterations.

4.2 Comparison of reverse geometry and normal vector method

Now we want to compare the normal vector method to the reverse geometry method
on a simplified version of geometry C (1c). The simplification is that the small round
elevations with holes in the middle of each box are removed. In Figure 4 the warpage
measurement of each method is plotted in a logarithmic scale over five iterations. Here,

10
Fig. 4: Comparison of the normalized warpage measurement in a logarithmic scale
over the number of iterations of the normal vector method and the reverse geometry
method. The comparison is done on a simplified version of geometry C (Figure 1c).

the normal vector method was able to reduce the warpage to a factor of 10−2 from
the initial warpage in five iterations. This is a good reduction in warpage but it is
still lower than the reverse geometry method, which reduced the warpage to a factor
below 10−4 .

5 Discussion
The numerical findings indicate that the Bayesian optimization method requires a sig-
nificantly higher number of iterations compared to the other methods. In fact, the
method is simply too powerful for the performed design test and its practical util-
ity only becomes evident when the geometric compensation of the cavity shape is
integrated with additional optimization objectives. Instead, the inverse method yields
promising results, but being an intrusive approach, it is impractical for industrial
applications using commercial simulation tools.
Overall, the normal vector method and the reverse geometry method yielded the
most favorable results, with the reverse geometry method demonstrating superior
performance in the tested geometry. While the practical significance of this better
performance may be debated, it’s crucial to acknowledge that the normal vector
method encounters difficulties in accommodating small geometric features. Further-
more, achieving satisfactory results in shape regularization with the normal vector
method requires hyperparameter tuning.

11
The reverse geometry works on any geometry without additional effort or hyperpa-
rameter tuning and it is very easy to implement. It has to be noted that the reverse
geometry method does not work directly when the desired geometry and the deformed
geometry are not available in the same mesh. This can happen when the simulation
model forces a remeshing of the geometry in the subsequent iterations of the reverse
geometry method. Further research will be done to overcome this problem by applying
interpolation methods to the remeshed geometry.
Acknowledgments. The presented investigations were carried out at RWTH
Aachen University within the framework of the Collaborative Research Centre
SFB1120-236616214 “Bauteilpräzision durch Beherrschung von Schmelze und Erstar-
rung in Produktionsprozessen” and funded by the Deutsche Forschungsgemeinschaft
e.V. (DFG, German Research Foundation). The sponsorship and support is gratefully
acknowledged.

Declarations
• a. Funding: This research was funded by the Deutsche Forschungsgemeinschaft e.V.
(DFG, German Research Foundation) within the framework of the Collaborative
Research Centre SFB1120-236616214.
• b. Conflicts of interest/Competing interests: All authors declare that they have no
conflicts of interest.
• c. Availability of data and material: The data and materials are available upon
reasonable request.
• d. Code availability: The code is available upon reasonable request.
• e. Ethics approval: Not applicable
• f. Consent to participate: Not applicable
• g. Consent for publication: Not applicable

References
[1] Fischer, J.M.: 2 - shrinkage and warpage. In: Fischer, J.M. (ed.) Handbook
of Molded Part Shrinkage and Warpage (Second Edition), Second edition edn.
Plastics Design Library, pp. 9–17. William Andrew Publishing, Boston (2013).
https://doi.org/10.1016/B978-1-4557-2597-7.00002-1

[2] Zhao, N.-y., Lian, J.-y., Wang, P.-f., Xu, Z.-b.: Recent progress in minimizing the
warpage and shrinkage deformations by the optimization of process parameters
in plastic injection molding: A review. The International Journal of Advanced
Manufacturing Technology 120(1-2), 85–101 (2022)

[3] Azad, R., Shahrajabian, H.: Experimental study of warpage and shrinkage in
injection molding of hdpe/rpet/wood composites with multiobjective optimiza-
tion. Materials and Manufacturing Processes 34(3), 274–282 (2019)

[4] Chen, Y., Zhu, J.: Warpage analysis and optimization of thin-walled injection
molding parts based on numerical simulation and orthogonal experiment. In: IOP

12
 Conference Series: Materials Science and Engineering, vol. 688, p. 033027 (2019).
IOP Publishing

[5] Lee, B., Kim, B.: Variation of part wall thicknesses to reduce warpage of injection-
molded part: robust design against process variability. Polymer—Plastics Tech-
nology and Engineering 36(5), 791–807 (1997)

[6] Zhao, J., Cheng, G., Ruan, S., Li, Z.: Multi-objective optimization design of
injection molding process parameters based on the improved efficient global
optimization algorithm and non-dominated sorting-based genetic algorithm. The
International Journal of Advanced Manufacturing Technology 78, 1813–1826
(2015)

[7] Li, K., Yan, S., Zhong, Y., Pan, W., Zhao, G.: Multi-objective optimization of the
fiber-reinforced composite injection molding process using taguchi method, rsm,
and nsga-ii. Simulation modelling practice and theory 91, 69–82 (2019)

[8] Heidari, B.S., Oliaei, E., Shayesteh, H., Davachi, S.M., Hejazi, I., Seyfi, J.,
Bahrami, M., Rashedi, H.: Simulation of mechanical behavior and optimiza-
tion of simulated injection molding process for pla based antibacterial composite
and nanocomposite bone screws using central composite design. Journal of the
mechanical behavior of biomedical materials 65, 160–176 (2017)

[9] Rosli, M., Termizi, S.A., Khor, C., Nawi, M., Omar, A.A., Ishak, M.I.: Simulation
based optimization of thin wall injection molding parameter using response sur-
face methodology. In: IOP Conference Series: Materials Science and Engineering,
vol. 864, p. 012193 (2020). IOP Publishing

[10] Rizvi, S., Singh, A.K., Bhadu, G.R.: Optimization of tensile properties of injection
molded α-nucleated polypropylene using response surface methodology. Polymer
Testing 60, 198–210 (2017)

[11] Wang, X., Gu, J., Shen, C., Wang, X.: Warpage optimization with dynamic injec-
tion molding technology and sequential optimization method. The International
Journal of Advanced Manufacturing Technology 78, 177–187 (2015)

[12] Li, H., Liu, K., Zhao, D., Wang, M., Li, Q., Hou, J.: Multi-objective optimizations
for microinjection molding process parameters of biodegradable polymer stent.
Materials 11(11), 2322 (2018)

[13] Xu, Y., Zhang, Q., Zhang, W., Zhang, P.: Optimization of injection mold-
ing process parameters to improve the mechanical performance of polymer
product against impact. The International Journal of Advanced Manufacturing
Technology 76, 2199–2208 (2015)

[14] Zhang, J., Wang, J., Lin, J., Guo, Q., Chen, K., Ma, L.: Multiobjective optimiza-
tion of injection molding process parameters based on opt lhd, ebfnn, and mopso.

13
 The International Journal of Advanced Manufacturing Technology 85, 2857–2872
(2016)

[15] Kitayama, S., Yamazaki, Y., Takano, M., Aiba, S.: Numerical and experimental
investigation of process parameters optimization in plastic injection molding using
multi-criteria decision making. Simulation Modelling Practice and Theory 85,
95–105 (2018)

[16] Hopmann, C., Nikoleizig, P.: Inverse thermal mold design for injection molds:
Adressing the local cooling demand as quality function for an inverse heat transfer
problem. International Journal of Material Forming 11, 113–124 (2018)

[17] Keuerleber, M., Eyerer, P.: 6 gestalten, fügen, berechnungsansätze und simulation
edv-unterstützter konstruktionen und auslegung von kunststoffbauteilen. Polymer
Engineering, 466–485 (2008)

[18] Schlutter, R.: Einstieg in die Spritzgießsimulation. Carl Hanser Verlag GmbH Co
KG, Munich (2023)

[19] Zhai, H., Chang, Y., Li, X., Xiong, X., Zhu, W., Li, C., Wang, Y.: A research
method to improve the quality of box-type thin-walled parts by combining
parameter optimization and i nverse-deformation design (2022)

[20] Zwicke, F., Behr, M., Elgeti, S.: Predicting shrinkage and warpage in injection
molding: Towards automatized mold design. In: AIP Conference Proceedings, vol.
1896 (2017). AIP Publishing

[21] Zwicke, F., Elgeti, S.: Inverse design based on nonlinear thermoelastic material
models applied to injection molding. Finite Elements in Analysis and Design 165,
65–76 (2019)

[22] Zwicke, F., Hohlweck, T., Hopmann, C., Elgeti, S.: Inverse design based on
nonlinear thermoelastic material models. PAMM 20(1), 202000130 (2021)

[23] Hanuhov, T., Cohen, N.: Design principles for 3d-printed thermally activated
shape-morphing structures. International Journal of Mechanical Sciences 262,
108716 (2024)

[24] Zhai, H., Li, X., Xiong, X., Zhu, W., Li, C., Wang, Y., Chang, Y.: A method
combining optimization algorithm and inverse-deformation design for improving
the injection quality of box-shaped parts. The International Journal of Advanced
Manufacturing Technology, 1–24 (2023)

[25] Kastelic, T., Starman, B., Cafuta, G., Halilovic, M., Mole, N.: Correction of
mould cavity geometry for warpage compensation. The International Journal of
Advanced Manufacturing Technology 123(5-6), 1957–1971 (2022)

14
[26] Tillmann, S., Behr, M., Elgeti, S.: Using bayesian optimization for warpage com-
pensation in injection molding. Materialwissenschaft und Werkstofftechnik 55(1),
13–20 (2024) https://doi.org/10.1002/mawe.202300157

[27] Azegami, H., et al.: Shape Optimization Problems. Springer, Singapore (2020)

[28] Hojjat, M., Stavropoulou, E., Bletzinger, K.-U.: The vertex morphing method for
node-based shape optimization. Computer Methods in Applied Mechanics and
Engineering 268, 494–513 (2014)

[29] Giannakoglou, K.C., Papadimitriou, D.I.: Adjoint methods for shape optimiza-
tion. Optimization and computational fluid dynamics, 79–108 (2008)

[30] Jameson, A.: Aerodynamic shape optimization using the adjoint method. Lectures
at the Von Karman Institute, Brussels (2003)

[31] Sederberg, T.W., Parry, S.R.: Free-form deformation of solid geometric mod-
els. In: Proceedings of the 13th Annual Conference on Computer Graphics and
Interactive Techniques, pp. 151–160 (1986)

[32] Shahriari, B., Swersky, K., Wang, Z., Adams, R.P., De Freitas, N.: Taking the
human out of the loop: A review of bayesian optimization. Proceedings of the
IEEE 104(1), 148–175 (2015)

[33] Frazier, P.I.: A tutorial on bayesian optimization. arXiv preprint arXiv:1807.02811

(2018)

[34] Zwieten, J.S.B., Zwieten, G.J., Hoitinga, W.: Nutils 7.0. Zenodo (2022). https:
//doi.org/10.5281/zenodo.6006701

[35] Sadd, M.H.: Elasticity: Theory, Applications, and Numerics. Academic Press,
Burlington (2009)

[36] Zwicke, F., Behr, M., Veroy, K.: Inverse shape design in injection molding based
on the finite element method. PhD thesis (2020)

15