Accepted Manuscript

Introduction of Elastic Die Deformations in Sheet Metal Forming

Simulations

Johan Pilthammar , Mats Sigvant , Sharon Kao-Walter

PII: S0020-7683(17)30212-3
DOI: 10.1016/j.ijsolstr.2017.05.009
Reference: SAS 9571

To appear in: International Journal of Solids and Structures

Received date: 29 November 2016

Revised date: 4 May 2017
Accepted date: 6 May 2017

Please cite this article as: Johan Pilthammar , Mats Sigvant , Sharon Kao-Walter , Introduction of
Elastic Die Deformations in Sheet Metal Forming Simulations, International Journal of Solids and
Structures (2017), doi: 10.1016/j.ijsolstr.2017.05.009

This is a PDF file of an unedited manuscript that has been accepted for publication. As a service
to our customers we are providing this early version of the manuscript. The manuscript will undergo
copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please
note that during the production process errors may be discovered which could affect the content, and
all legal disclaimers that apply to the journal pertain.
 ACCEPTED MANUSCRIPT

Highlights:

 Reliable prediction of stamping die deformations and pressure distributions.

 Forming simulations with elastic die and press deformations using 2D die surfaces.
 Combining two FE-models for inclusion of elastic die and press deformations.
 Techniques for transferring deformations from structural model to SMF simulation.
 Reliable forming simulations with advanced friction modeling.

T
IP
CR
US
AN
M
ED
PT
CE
AC

Abbreviations: sheet metal forming (SMF), finite element models (FE-models), AutoFormplus R6.0
(AutoForm), three-dimensional (3D), two-dimensional (2D), Volvo Cars (Volvo), Hypermesh 13.0
(Hypermesh), ABAQUS 6.14 (ABAQUS), stereolithography (stl), computer-aided design (CAD)
 ACCEPTED MANUSCRIPT

Introduction of Elastic Die Deformations in Sheet

Metal Forming Simulations

T
Johan Pilthammar1,2,3

IP
Mats Sigvant1,2
Sharon Kao-Walter2

CR
1
Volvo Cars, Dept. 81153 Stamping CAE & Die Development, Olofström, Sweden
2

3
US
Blekinge Institute of Technology, Valhallavägen, 371 41 Karlskrona, Sweden
Corresponding Author
AN
E-mail Johan Pilthammar: [email protected]
E-mail Mats Sigvant: [email protected]
M

E-mail Sharon Kao-Walter: [email protected]

ED

Abstract
PT

Simulations of sheet metal forming (SMF) with finite element models (FE-models) for

stamped parts in the car industry are useful for detecting and solving forming problems.
CE

However, there are several issues that are challenging to analyze. Virtual tryout and analyzes

of stamping dies in running production are two important cases where many of these
AC

challenging issues are present. Elastic deformations of dies and press lines and a physically

based friction model is often missing when these types of cases are analyzed. To address this,

this research aims to develop a method wherein the results of two separate FE-models are

combined to enable SMF simulations with the inclusion of elastic tool and press deformations.

The two FE-models are one SMF model with two-dimensional (2D) rigid tool surfaces and
 ACCEPTED MANUSCRIPT

one structural model of the die and press. The structural model can predict surface shapes and

pressure distributions for a loaded stamping die. It can also visualize relatively large and

unexpected deformations of the die structure. The recommended method of transferring the

deformations from the structural model to the 2D surfaces is through an FE technique called

submodeling. The subsequent SMF simulations show that the method for calculating and

using the deformed surfaces together with the TriboForm friction model yields a result that

T
matches measured draw-in and strains. It is verified that the ability to virtually deform a die

IP
and include the resulting geometry in forming simulations is of high importance. It can be

CR
used for the virtual tryout and optimization of new dies or analyses of existing dies in running

production. It is suggested that future research focus on a more efficient and automated

US
workflow. More experimental data and simulations are also needed to verify the assumptions
AN
made for the simulation models. This will enable the method to be adopted in a reliable way

for standard SMF simulations.

M

Keywords
ED

Sheet Metal Forming, Elastic Tooling, Structural Analysis, Surface Compensation, Friction

1. Introduction
PT

Simulations of SMF with FE-models for stamped parts in the car industry are useful for
detecting and solving forming problems. However, there are several issues that are
CE

challenging to analyze. Virtual tryout and analyzes of stamping dies in running production
are two important cases where many of these challenging issues are present. Elastic
AC

deformations of dies and press lines and a physically based friction model is often missing
when these types of cases are analyzed.

A correct representation of the forming surfaces in a stamping die is key for SMF simulations.
2D rigid representations of the nominal surfaces are normally used during the development
phase of a die. However, if a virtual tryout or an analysis of an existing die is performed, it is
vital to include both the shape of the real physical surfaces and the elastic behavior of the die
and the press line. Scanning tool surfaces is one way to obtain data for SMF simulations,
 ACCEPTED MANUSCRIPT

which was the one used in this research. The work in this paper was done in cooperation with
a project by the commercial software TriboForm, which was used to study and create friction
models for the simulations in this research [1-4]. This friction study was performed through
experiments and simulations of the forming operation of the existing die for the XC90 rear
door inner, shown in Fig. 1.1. It is common knowledge in the field of SMF simulations that
die and press deformations influence the stamping process. It is crucial to represent these
deformations in SMF simulations, both for accurate shape representation and correct
prediction of contact conditions, which influence the friction model. These deformations are

T
difficult to predict in standard SMF simulations where the forming surfaces are represented

IP
as 2D rigid surfaces and the SMF software is often not capable of performing 3D structural

CR
modeling.

US
AN
M
ED

Fig. 1.1. Formed blank from AutoForm simulation of the XC90 rear door inner.
PT

Previous work has been done regarding elastic tool deformations and their influence on the
forming process and simulations. One example, which is one of the most extensive works in
CE

the area of combining structural behavior and SMF simulations, by R.A. Lingbeek, describes
different methods for analysis and virtual rework of tool structures and surfaces [5]. Another
AC

example [6] outlines a method for compensating against elastic tool deformations and
demonstrates it through simulations. Most previous work in the area of combining structural
analysis and SMF simulations has been performed on small industrial or experimental dies.
Several studies assert that more work is required, especially simulations and validations for
dies on an industrial scale. Many suggested methods combine the structural analysis and
forming simulation into one FE-model. This leads to large models, which are time-consuming
to solve when scaled to dies on an industrial scale. A couple of suggested methods to reduce
the degrees of freedom in a combined FE-model is coupling of coarse and fine meshes or an
 ACCEPTED MANUSCRIPT

approach called static condensation [7]. Methods of reduction are not possible in most
widely-used SMF software, such as AutoForm, which currently offer no possibility to include
solid structures. All methods for reduction of structural models have advantages and
disadvantages, since some information and accuracy is always lost. Software for structural
analysis are also generally not developed for advanced SMF-simulations. AutoForm, which is
used for the SMF simulations in this paper, has some simplified capabilities for modeling the
blankholder deformations through its different binder models. These binder models are not
dependent on any real die geometry and the surfaces in the models are not really deforming.

T
It is instead the contact conditions that are controlled to simulate the effect of the deforming

IP
structure. The user of AutoForm gives input corresponding to die dimensions such as width,

CR
length and height together with stiffness values for the die and contact algorithm. The input
can also include the positions of cushion pins and gas springs. All these parameters will
impact the contact conditions for the die surfaces.

US
To address the presented issues, the aim and focus of this research is to develop and discuss a
method wherein two separate FE-models are combined to enable SMF simulations that
AN
include the elastic deformations of stamping dies and press lines. One model, solved in the
software ABAQUS, is for structural analysis of the die and press deformations. The other
M

model is for the SMF simulations with AutoForm. Information will be transferred from the
structural model to the SMF model in order to achieve reliable simulation results. This
ED

method has two benefits. The first is the potential for die and press deformations to be
incorporated into SMF software without the possibility to model 3D solids. The second is its
PT

ability to reduce solver time when SMF and structural behavior must be simultaneously
analyzed, see [8].
CE

Previously, Volvo has also conducted research on structural analysis and optimization of
stamping dies. Two master theses were performed at Volvo Cars: one in 2007, summarized in
AC

a conference paper [9], and one in 2016 [10]. Both theses indicated a high potential for
improving die designs by using structural analysis and other numerical tools.

The research question in this paper is to investigate and develop a method for combining a
structural FE-model and a SMF simulation model. This is achieved by transferring
information about the structural behavior of the die and press line from the structural FE-
model into the SMF model. When used together, the two models yield a total solving time
that is significantly shorter than that of a model which combines structural analysis and
 ACCEPTED MANUSCRIPT

forming simulation. An accurate and fast method is vital if used for virtual tryout or for
analyses of real dies in running production. This method is valid under certain assumptions,
presented in Chapter 2.

This article is organized into five chapters. Chapter 1 is an introduction to the research
question and its context. Chapter 2 presents the experimental and simulation methods
together with a more thorough background on the theory and assumptions that this research is
based upon. Chapter 3 presents the results from the models used in the different simulations.

T
Chapter 4 is a discussion about the methods and the results. Chapter 5 presents the

IP
conclusions of the research and suggestions for follow-up research.

CR
2. Experimental and Simulation Methods

US
This chapter presents experimental data from a production run of the XC90 rear door inner. It
also describes the theoretical background, assumptions and setup for the different numerical
AN
methods and FE-models used for analyzing the forming procedure of the stamping die.

2.1 Full-scale production run

M

A production run with the XC90 rear door inner was enacted in order to gather data for the
following simulations. The press was a mechanical transfer press line at Volvo Cars in
ED

Olofström, Sweden. The velocity profile was recorded for use in the TriboForm friction
model during forming simulations.
PT

The die material was GGG70L, with a die and punch that was chrome plated in selected areas.
The forming surfaces of the blankholder were polished and laser hardened. The geometries of
CE

the forming surfaces were scanned. The blank material was VDA239-CR4 GI, with a
thickness of 0.7 mm and Fuchs 4107 as pre-lube. Measurement of the lubrication amount in
the production indicated a variance between 0.7 g/m2 and 2.2 g/m2 across the blank.
AC

Some collected data, from the production run, is presented in Fig. 2.1 – Fig. 2.4. Fig. 2.1.
shows a spotting image. Spotting is a technique wherein the metal sheet is painted before the
die is closed – in this case, the blankholder and matrix are painted in the areas where contact
is expected when the blankholder is closed. When the tool is opened, the paint is removed
where there is pressure caused by contact between the sheet and the tool surfaces. Fig. 2.2
and Fig. 2.3. are showing the scanned outline and strains measured on a real formed sheet.
 ACCEPTED MANUSCRIPT

Fig. 2.4. demonstrates that the distance between the blankholder and the matrix exceeds two
sheet thicknesses in some areas when the die is unloaded. This distance is expected to be
close to 0.7 mm which is the blank thickness and the distance in the original CAD-model of
the die. The surfaces are positioned for this measurement such that the closest distance
between them is 0.7 mm, corresponding to the blank thickness. This measurement is based on
scanned data of the die surfaces obtained with Atos/GOM compact scan together with the
software Atos Professional V7.5 SR1. The alignment of the scanned surfaces was made with
a Canon EOS 1D with the software Tritop Professional V7.5 SR1.

T
IP
CR
US
AN
M

Fig. 2.1. Spotting of the blank.

ED
PT
CE
AC

Fig. 2.2. Scanned outline of the drawn blank.

 ACCEPTED MANUSCRIPT

T
IP
Fig. 2.3. Strain measurements with Argus. Major strain is plotted in the figure.

CR
US
AN
M
ED

Fig. 2.4. Distance between scanned blankholder and scanned matrix in unloaded condition.
PT

2.2 Structural analysis of the stamping die

CE

Since the gap between the surfaces of the matrix and the blankholder differs significantly, it
AC

is assumed that this distance will vary when the die is loaded. Therefore, it is necessary to
virtually deform the surfaces before they can be used in a forming simulation. The aim of the
structural analysis is to replicate the shapes of the blankholder and matrix surfaces at a single
point in time, when the blankholder is closed before the forming has started. This is the same
position as in the spotting image in Fig. 2.1. The calculated shapes of the surfaces are then
used in forming simulations. The reason for choosing this point in time is that it represents
the most common position used at Volvo for spotting of the blankholder.
 ACCEPTED MANUSCRIPT

Using the deformed surfaces from the structural model in a forming simulation is valid under
the assumption that the change in tool deformations is relatively small after the blankholder
has been closed. Additional deformation should be manageable using functions as the various
support types in AutoForm, which are approximations or metamodels of how a die deforms.

In order to virtually deform the tool surfaces, a structural FE-model of the die was created in
Hypermesh and solved with ABAQUS. The die was modeled as elastic bodies and solved
with a dynamic quasi-static method. The ram was assumed to be rigid and an elastic model of

T
the cushion was included in the model. The cushion force, measured in the real press line,

IP
was applied at the points where the cylinders connected to the cushion. The lower shoe and
the punch was excluded since it has no influence on the deformations of the die in the

CR
calculated position.

No metal sheet is present in the structural model; instead, the sheet is represented by a contact

US
surface which was given the same thickness as the sheet: 0.7 mm. No forming is done in this
model because the desired result is the shape and behavior of the die. There is an additional
AN
assumption that the thickness deformation of the blank is negligible in the position where the
blankholder has just closed. These assumptions simplifies the modelling and reduces the
solver time.
M

The structural FE-model was created from the original CAD-model with nominal surfaces.
ED

The forming surfaces in the real die deviates from the CAD-model since many changes are
typically made during the tryout of the physical die. The scanned surfaces from the real die
were imported into the pre-processor Hypermesh, and the nodes on the surface of the meshed
PT

blankholder and matrix surfaces were moved to the scanned surfaces with the morphing tools
in Hypermesh. Only the surfaces that should be in contact with one another were morphed.
CE

This operation modified the surfaces from the nominal shape to the shape of the real die
surfaces. This allows for solving the structural model with the scanned tool surfaces. Fig. 2.5.
AC

and Fig. 2.6. are showing the structural FE-model.

Material parameters used for the die in the simulations are a Young´s modulus of 176 GPa
and a Poisson ratio of 0.275. Cushion pins and the Press cushion are modelled as steel with
Young´s modulus 210 GPa and a Poisson ratio of 0.3.
 ACCEPTED MANUSCRIPT

T
IP
CR
Fig. 2.5. Structural FE-model.

US
AN
M
ED
PT

Fig. 2.6. Solid FE-model of the matrix (blue) with the scanned surface (red). Nodes on the
surfaces of the blue solid mesh are morphed with the red surface as destination.
CE
AC

2.3 Modifying the shape of the scanned die surfaces

The structural FE-model of the die is quickly solved and gives the shape of the matrix and
blankholder in the loaded condition at the moment when the blankholder is closed. Although
the shape is correct, it is not possible to create a surface based on the solid mesh and use it in
AutoForm. This is because the structural mesh is too coarse for use in AutoForm for SMF. It
has to be coarse for a reasonable solving time of the solid model. A solution is to use the
scanned surfaces, which have meshes that are sufficiently fine, and deform those surfaces
 ACCEPTED MANUSCRIPT

based on the deformations from the structural model. Two different strategies have been used:
morphing in AutoForm and submodeling in ABAQUS.

2.3.1 Deforming the scanned surfaces by morphing in AutoForm

In AutoForm, morphing is done by using guide curves. These guide curves describe the
parameters of where the surfaces should be morphed. Catia V5 was used to create guide
curves on the original scan surfaces and on the deformed surfaces from the structural FE-
simulation. The curves are identically positioned in the xy-plane; the only difference is the

T
curvature in the z-direction. The position of the curves on the blankholder are depicted in Fig.

IP
2.7. Because the distance between the curves is minute, it is difficult to identify that there are
two sets of curves in Fig. 2.7. These curves are created to cover the deformation of the entire

CR
blankholder surface.

US
AN
M
ED

Fig. 2.7. Scanned blankholder surface with guide curves for morphing in AutoForm.
PT

2.3.2 Deforming the scanned surfaces by submodeling in ABAQUS

Submodeling is a method that applies a variable, such as the deformations, from one
CE

structural model as a load or boundary condition to a second model [11]. It is often used to
apply deformations from a large global model as boundary conditions to a smaller part of the
AC

model that is of high interest and modeled in greater detail in the submodel. In this work, the
method is instead used for applying the deformations from the 3D model of the stamping die
to the scanned 2D surface of the die surfaces. The scanned 2D surfaces are triangulated
surfaces representations in stl-format and are therefore possible to import into most FE-
software for use as a mesh in an analysis. Fig. 2.8. shows the deformations transferred from
the structural model to the scanned stl-surface.
 ACCEPTED MANUSCRIPT

T
IP
CR
Fig. 2.8. Deformation of the matrix in the solid model (upper) and deformation of the
scanned matrix in the submodel (lower) (mm).

2.4 SMF simulations

US
The forming simulations are performed in AutoForm with different configurations for seven
AN
models labeled A-G. The different configurations includes deformed surfaces and the original
scanned surfaces, the velocity of the press, the TriboForm friction model and constant friction.
M

By using this input, the shape of the physical die and lubrication system is represented in
different ways in the SMF simulations. The configuration for each model is presented in
ED

sections 2.4.1-2.4.7. Simulation model A is chosen as reference because it is using the

TriboForm friction model, the surfaces are compensated in a reliable way with submodeling,
and it yields the best results. This is further discussed in section 3.3.
PT
CE
AC

The BBC2005 material model is used for all SMF simulations [12]. The material parameters
are presented in Table 2.1., Fig. 2.9, and Fig. 2.10.

σ0 σ45 σ90 σb R0 R45 R90 Rb M

(MPa) (MPa) (MPa) (MPa)
156.6 160.0 156.0 187.0 1.81 1.34 1.88 0.98 4.5
Table 2.1. Material parameters used in the BBC2005 material model
 ACCEPTED MANUSCRIPT

T
IP
Fig. 2.9. Hardening curve used in the BBC2005 material model

CR
US
AN
M
ED
PT

Fig. 2.10. Forming limit curve at onset of localization used in the SMF simulations

Most of the simulations are run with a binder model in AutoForm with support type Force
CE

Controlled and with loading condition Uniform. These settings are appropriate because the
shape of the surfaces will be kept constant and the blankholder force is a constant force of the
AC

same magnitude as the real force in the press. The modified variable in the simulations is the
tool stiffness. Modifying this parameter has the same effect as adjusting the contact stiffness
between two bodies in an FE-analysis, an example of this is visualized in Fig. 2.11.
 ACCEPTED MANUSCRIPT

T
IP
CR
Fig. 2.11. Predicted pressure distribution on the blank, at the end of the clamping stage.
Equal force magnitude with different tool stiffness: 1000 MPa/mm (left), 5000 MPa/mm

US
(right).

2.4.1 Model A: Simulation with compensated surfaces (submodeling) and TriboForm

AN
friction model (reference simulation)

The following simulation parameters are used in AutoForm:

M

Tool surfaces: Tool surfaces are compensated for the elastic deformation by
submodeling in ABAQUS
ED

Friction: TriboForm friction model

PT

Support type for blank holder: Force controlled with uniform loading condition

Tool stiffness: 5000 MPa/mm

CE

2.4.2 Model B: Simulation with uncompensated scanned surfaces

Difference from reference simulation model A in section 2.4.1:

AC

Original scanned surfaces without any type of surface compensation.

2.4.3 Model C: Simulation with compensated surfaces (morphing)

Difference from reference simulation model A in section 2.4.1:

Tool surfaces are compensated for by morphing in AutoForm.

2.4.4 Model D: Simulation with lower tool stiffness: 100 MPa/mm

 ACCEPTED MANUSCRIPT

Difference from reference simulation model A in section 2.4.1:

Tool stiffness: 100 MPa/mm (instead of 5000 MPa/mm)

2.4.5 Model E: Simulation with higher tool stiffness: 15000 MPa/mm

Difference from reference simulation model A in section 2.4.1:

Tool stiffness: 15000 MPa/mm (instead of 5000 MPa/mm)

2.4.6 Model F: Simulation with constant friction of 0.12

T
IP
Difference from reference simulation model A in section 2.4.1:

Friction coefficient: 0.12 (instead of TriboForm model)

CR
2.4.7 Model G: Simulation with constant friction of 0.15

US
Difference from reference simulation model A in section 2.4.1:

Friction coefficient: 0.15 (instead of TriboForm model)

AN
2.5 TriboForm friction model

The TriboForm software is used together with experimental data to create a physically-based
M

friction model that is used in most of the following forming simulations. In this case, the
friction model is 3D and depends on contact pressure, relative sliding velocity, and plastic
ED

strain in the sheet material [2].

A constant friction can often produce reliable results. However, constant friction must be
PT

inversely modeled by studying the output from a real die or applied with extensive experience
of which friction coefficient to use for many different lubrication systems. Therefore, one of
CE

the main reasons for a physically-based friction model is to predict the friction conditions in a
simulation before a die is manufactured.
AC

The TriboForm software is based on the research presented in [4]. The creation of each
friction model starts with defining a tribological system, i.e. a sheet material with or without
coating, a lubricant and a tool material with or without coating and hardening. The input to
the model are the mechanical properties for the sheet material, the surface roughness of both
the sheet and the die surfaces, together with the viscosity and amount of lubricant. Each
model is then calibrated with rotational friction tests and a library is created for the
tribological system. The software TriboForm Analyser reads the library and the user can
 ACCEPTED MANUSCRIPT

easily change the surface roughness for the sheet and the die as well as the amount of
lubrication. It is possible to visualize both the selected surfaces and the resulting friction
model, see Fig 2.12-13. When the user is satisfied with the settings, a file containing
information about the defined friction model is exported to AutoForm for use in SMF
simulation. This process and a more in-depth description of the creation of the specific model
used in these simulations is described in [1-4].

T
IP
CR
US
Fig. 2.12. Start and midpoint of the simulation for the friction model. Blank, tool, and
lubricant can be visualized in the friction simulations.
AN
M
ED
PT
CE
AC

Fig. 2.13. Surfaces representing the friction model for the tribology system used in the
simulations. The various surfaces represent different levels of strain in the sheet material.
The highest surface represents 0% strain and the lowest represents 20% strain. The step
between surfaces is 5% strain.
 ACCEPTED MANUSCRIPT

3. Results

This chapter contains the results from the different methods and models used in the research.
It describes the results from the structural FE-model of the stamping die, how the results were
transferred to the SMF model and the final results from the SMF model itself.

3.1 Structural analysis

Section 3.1.1. – 3.1.3 presents the results from the structural analysis of the stamping die. The

T
loadstep for the deformation of the die is solved in 15 hours on 64 cores. The convergence of
the contact problem consumes the vast majority of the solver time. This is because the mesh

IP
is coarse and based on scanned data. Reduction of the solver time is discussed in section 5.2.

CR
3.1.1 Deformations

The deformations of the matrix and the blankholder are shown in Fig. 3.1. and Fig. 3.2. The

US
legend for the deformation of the blankholder does not start at zero due to some rigid body
motion in the model prior to contact with the matrix. In Fig. 3.3, the deformation is magnified
AN
250X and shows that the blankholder bends in an arc shape. The deformation of the
blankholder is significantly larger than the deformation of the matrix, even when the rigid
body motion is neglected for the blankholder.
M
ED
PT
CE
AC

Fig. 3.1. Deformation of the matrix (mm).

 ACCEPTED MANUSCRIPT

T
IP
Fig. 3.2. Deformation of the blankholder.

CR
US
AN
M

Fig. 3.3. Deformation of the blankholder magnified 250X.

ED

3.1.2 Contact between cushion pins and cushion

PT

Since the blankholder bends upwards, many of the cushion pins lose contact with the cushion.
This is visible in Fig. 3.4.
CE
AC
 ACCEPTED MANUSCRIPT

T
IP
CR
Fig. 3.4. Contact pressure between pins and cushion (MPa).

3.1.3 Virtual visualization of the spotting

US
Visualization of the spotting is facilitated by setting the values to appropriate levels for the
variable COPEN in ABAQUS. COPEN represents the distance between the slave and master
AN
contact surfaces in the structural analysis. The variable COPEN or contact pressure are both
variables that are possible to use for a virtual spotting image. The values in the color legend
are identified simply by attempting different combinations. This reveals the potential to
M

predict and compare the contact surface with real photos. However, the full contact area in
the structural model will not have a perfect match with the real spotting image, because areas
ED

in contact with a low pressure will leave the paint intact. Further research is therefore
necessary to determine how to set the values in the legend for a robust and reliable
PT

comparison. The values should be based on experimental data for a number of dies compared
with simulations. A comparison of the virtual and real spotting is visualized in Fig. 3.5.
CE
AC
 ACCEPTED MANUSCRIPT

T
IP
CR
Fig. 3.5. Comparison between the spotting predicted in the Abaqus model (left) and physical
spotting (right), with the aid of the variable COPEN.

3.2 Modification of the scanned surfaces US

AN
Fig. 3.6.-3.8. visualizes the resulting geometries with the morphing and submodeling methods,
used for transferring the structural deformations from the structural FE-model in ABAQUS to
the SMF model in AutoForm. The morphing and submodeling are based on the deformations
M

predicted by the structural model.

ED
PT
CE
AC

Fig. 3.6. Distance between compensated and uncompensated blankholder surfaces.

 ACCEPTED MANUSCRIPT

T
IP
CR
Fig. 3.7. Distance between compensated and uncompensated matrix surfaces.

US
AN
M
ED
PT

Fig. 3.8. Distance between compensated and uncompensated blankholder and matrix.
CE

3.3 Results of SMF simulations

AC

This chapter presents the results from seven SMF simulations conducted in this research.
Results are presented for the models, labeled A-G, in sections 3.3.1-3.3.9. The SMF
simulation models is solved in 3 hours on 8 cores.

Figure 3.9-3.15 in subsection 3.3.8 are visualizing the difference between the simulated part
and the scanned outline of a real part for each model. The figures depicts the geometries from
each SMF simulation together with a scanned draw-in curve from a real part.
 ACCEPTED MANUSCRIPT

Figure 3.16-3.22 in subsection 3.3.9 shows the difference between the SMF simulations
predicted major strain and the experimental one, for model A-G. Red and black indicates
over-prediction while blue and white indicates under-prediction.

The results for model A in section 3.3.1 are the most reliable of all the forming simulations
and includes both compensated tool surfaces and the TriboForm friction model. Therefore,
these results are used as a reference point for comparison with the rest of the simulations
presented in this article. Model C in section 3.3.3 yields very similar results. However, model

T
C is not chosen as reference since its surfaces were compensated by morphing in AutoForm

IP
instead of submodeling in ABAQUS. There are potential issues with the morphing method
that are visualized in section 3.2 and discussed in section 4.2.

CR
The parameters changed with respect to the reference simulation model A are stated in the
title of each subsection 3.3.1-3.3.7 and more in detail in subsections 2.4.1-2.4.7.

US
The importance of being able to virtually deform and use surfaces with a correct shape in
SMF simulations is also necessitated by the inability to obtain satisfying results when using
AN
the original scanned surfaces in AutoForm. The best achievable results with these surfaces
are depicted in Fig. 3.10 and Fig. 3.17, regardless of simulation settings.
M

3.3.1 Model A: Simulation with compensated surfaces (submodeling) and TriboForm

friction model (reference simulation)
ED

There is a reliable match with the measured draw-in curve. The strain prediction is generally
accurate, except around the five holes and in some areas on the edge of the blank.
PT

3.3.2 Model B: Simulation with uncompensated scanned surfaces

CE

The draw-in is severely over-predicted and the strains are both under- and over-predicted in
several areas.
AC

3.3.3 Model C: Simulation with compensated surfaces (morphing)

The results for model C are similar to the reference simulation model A. This is expected
since all settings are the same. The only difference is the method for compensating the die
surfaces for elastic deformations.
 ACCEPTED MANUSCRIPT

3.3.4 Model D: Simulation with lower tool stiffness: 100 MPa/mm

The predicted draw-in is under-predicted in several areas when compared to the reference
model A. There is also significant over-prediction of the strains.

3.3.5 Model E: Simulation with higher tool stiffness: 15000 MPa/mm

The predicted draw-in is similar to the reference simulation model A. The strains are also
similar, except for some over-prediction in minor areas.

T
3.3.6 Model F: Simulation with constant friction of 0.12

IP
The draw-in is similar to the reference simulation model A. The strains are also similar,

CR
excluding some minor areas that are over-predicted compared to the reference simulation.

3.3.7 Model G: Simulation with constant friction of 0.15

US
The draw-in is under-predicted in several areas compared to the reference model A. There is
also significant over-prediction of the strains.
AN
3.3.8 Draw-in comparison for model A-G
M
ED
PT
CE
AC

Fig. 3.9. Model A: SMF simulation compared to scanned outline of real part.
 ACCEPTED MANUSCRIPT

T
IP
CR
US
AN

Fig. 3.10. Model B: SMF simulation compared to scanned outline of real part.
M
ED
PT
CE
AC

Fig. 3.11. Model C: SMF simulation compared to scanned outline of real part.
 ACCEPTED MANUSCRIPT

T
IP
CR
US
AN
Fig. 3.12. Model D: SMF simulation compared to scanned outline of real part.
M
ED
PT
CE
AC

Fig. 3.13. Model E: SMF simulation compared to scanned outline of real part.
 ACCEPTED MANUSCRIPT

T
IP
CR
US
AN
Fig. 3.14. Model F: SMF simulation compared to scanned outline of real part.
M
ED
PT
CE
AC

Fig. 3.15. Model G: SMF simulation compared to scanned outline of real part.
 ACCEPTED MANUSCRIPT

3.3.9 Over- and under-prediction of major strain for model A-G

T
IP
CR
.
US
Fig. 3.16. Model A: Over- and under-prediction of first major strain.
AN
M
ED
PT
CE
AC

Fig. 3.17. Model B: Over- and under-prediction of first major strain.

 ACCEPTED MANUSCRIPT

T
IP
CR
US
Fig. 3.18. Model C: Over- and under-prediction of first major strain.
AN
M
ED
PT
CE
AC

Fig. 3.19. Model D: Over- and under-prediction of first major strain.

 ACCEPTED MANUSCRIPT

T
IP
CR
US
Fig. 3.20. Model E: Over- and under-prediction of first major strain.
AN
M
ED
PT
CE
AC

Fig. 3.21. Model F: Over- and under-prediction of first major strain.

 ACCEPTED MANUSCRIPT

T
IP
CR
4. Discussion
US
Fig. 3.22. Model G: Over- and under-prediction of first major strain.
AN
This chapter is a discussion of the methods and results presented in this paper.

4.1 Discussion of Structural analysis

M

The structural model is a relatively simple linear elastic FE-model. This model can accurately
predict the surface shapes and pressure distributions. This is indicated by the fact that the
ED

initial gap between the die surfaces is closed and that the use of the resulting surfaces in SMF
simulations yields reliable results close to real measurements. However, the results are
PT

somewhat unexpected. The large deformations of the blankholder and the loss of contact
between the pins and the cushion are not the desired behavior for a stamping die. According
CE

to the results, it will be difficult to obtain a satisfactory blankholder pressure on some areas of
the blankholder surface. This has been confirmed by individuals working with the real die at
AC

Volvo. One element missing in the model is the elastic behavior of the ram. No full CAD
model or drawing is available, so including it in the model requires the deformations of the
ram to be characterized and represented in the model through other means. Likely, the best
approach is to use certain deformation measurements. In this specific case, it is expected that
the influence of the extra deformation of the ram will be minor, since the deformation of the
blankholder is large and will therefore govern the resulting pressure distribution. The model
also indicates a possibility to visualize the spotting in a virtual environment. More experience
is needed for the values used in the color legend in order to represent it correctly.
 ACCEPTED MANUSCRIPT

It is likely possible to avoid the problems visualized for this die. Numerically, this can occur
by making a structural optimization of the solid structure and performing a virtual spotting of
the die surfaces before the die is manufactured.

4.2 Modification of the scanned surfaces

Both methods presented in this paper perform effectively in moving the deformations from
the structural model to the SMF model. The recommended method is submodeling because it
transfers the exact deformation field from the global model to the submodel. Furthermore, it

T
can be scripted and automatized to run directly after the global structural model is solved.

IP
This requires no extra manual work beyond the import into AutoForm.

CR
The morphing method needs manual post-processing and includes an extra step in CAD
software before the guide curves can be imported into AutoForm. In AutoForm, the surfaces

US
must be morphed by the guide curves in an additional operation. The morphing is also
problematic with advanced surfaces, as evidenced in the results from the morphing of the
matrix in Fig. 3.7. The resulting deformations look abnormal, while the morphing of the
AN
simpler blankholder effectively resembles the surface modified by submodeling. The
influence of the unsuccessful morphing of the matrix on the final results is likely of minimal
M

significance since the absolute deformation of the matrix was so small compared to the
blankholder.
ED

4.3 Discussion of SMF simulations

The simulations presented in this paper show that the method for calculating and using the
PT

deformed surfaces together with the TriboForm friction model yields reliable results,
provided that friction and contact conditions are represented effectively in AutoForm.
CE

One issue is the over-prediction of strains around the holes; this is likely due to strain rate
effects. Generally, the plastic strain rates in the SMF simulations in this paper are much
AC

higher than standard simulations at Volvo Cars. The reason for this is that the real press
velocity is used instead of a standard value of 1 mm/s, since the TriboForm friction model is
dependent on the sliding velocity of the blank. There are areas in the part that reaches strain
rates of more than 1 1/s which leads to a stronger material locally. That would lead to reduced
strains in those areas if strain rate effects were to be introduced in the SMF model.

It is interesting to note that a constant friction of 0.12 produces a result equally reliable as that
of the TriboForm friction model. However, the benefit of the TriboForm model is the ability
 ACCEPTED MANUSCRIPT

to predict the friction without using inverse modeling or depending on years of experience to
achieve accuracy.

For some or all dies, there will be elastic deformations of the die structure during the forming
operation. It could potentially be handled by applying an average deformation based on
different points in time during the drawing of the blank. Another option could be to use the
presented method to correct the surface shapes when the blankholder is closed, and then
investigate whether some of the support types and load conditions for the blankholder in

T
AutoForm can account for the remaining deformations.

IP
5. Conclusions

CR
This chapter concludes the research presented in this paper and suggests topics for follow up
research.

US
5.1. Conclusions about suggested methods and results

The ability to virtually deform a die and include the resulting geometry in forming
AN
simulations is of high importance. The aim of this research was to combine two simulation
models to create a new strategy for including die and press deformation into SMF simulations.
One example of the method has been demonstrated in this paper with successful results. The
M

main benefit of the method is that it makes it possible to include die and press deformations
in SMF simulations where the die surfaces are represented as 2D surfaces. SMF simulations
ED

with 2D surfaces is the most common simulation method, often in software without the
possibility to model 3D solid structures such as AutoForm, which is used for SMF
PT

simulations in this paper. The simplified binder models implemented in AutoForm are found
insufficient to represent the deformations of the forming surfaces in these simulations.
CE

Another benefit of the method is the relatively short total simulation time of both models
when compared to some other methods for incorporating die deformations, such as
AC

combining both structural and forming simulations in the same model without any type of
reduction method for the degree of freedoms.

The suggested method can be used to offer support to running production or analyses of an
existing die. This link between structural analysis and SMF will also enable additional
methods for virtual design and rework of stamping dies. Structural optimization and methods
for virtual spotting should be used together for controlling the die deformations. This would
provide an optimum die design.
 ACCEPTED MANUSCRIPT

5.2. Follow-up research

Follow-up research is suggested in the following areas:

A more efficient workflow is needed. Currently, the total solver time of both the models is
relatively short. But setting up the models has the potential to be much faster and more
automated. This is crucial if the method shall be adopted in standard forming simulations.
The increase in computational time when an elastic model of the die is used instead of rigid
surfaces is studied in [13], where it is stated that the computational time is increased by a

T
factor of 9-14 times for the specific model in that paper. If it was possible to use the elastic

IP
die presented in this paper in AutoForm there is no doubt that the computational time would
increase by a large factor for this model as well. It is probably possible to reduce the

CR
computational time substantially for the structural model by a coarsening of the elements in
the main body of the die and a smoothing of the contact surfaces [14] for an increased

US
convergence rate of the contact algorithm. Even if the computational time is reduced it is
probably not enough to achieve a combined structural and SMF model that is solved in a
AN
reasonable time by just combining them. Some reduction of the models degree of freedoms
would have to be used. Reduction methods are studied in [5], where it is concluded that static
condensation is not a reliable method to use for stamping dies, since it is accurate but it
M

increases the computational time by a factor of 10 because many nodes have to remain on the
detailed die surfaces. A promising reduction method is Deformable Rigid Bodies (DRB),
ED

which is a modal method. A reliable simulation is achieved by the DRB method with an
increase of only 8% in computational time. Since the method in this paper is suggested for
PT

software that offer no possibility to model solid structures, such as AutoForm, the different
methods for reduction are not a way forward until it is possible to include solid structures in
CE

that type of software. The structural model presented in this paper could be reduced but the
effort will probably not repay itself in a reduced computational time, since the structural
AC

model is only solved once in this case.

More research is needed that verifies the method on more and different types of stamping
dies. The assumption that the die deformations are small after the closing of the blankholder
also needs to be investigated more thorough. A suggestion is that these deformations can be
handled by the different support types and blankholder models available in AutoForm. The
structural analysis to obtain the correct surface shapes can also be made for other points in
time during the forming operation, e.g. end of the forming operation or any other point. It is
 ACCEPTED MANUSCRIPT

possible that any other point in time is the most crucial for good SMF simulation results, or
an average of several different points.

One drawback of the friction model is that it creates one single friction model for all the
forming surfaces. The simulation model has the potential to be even more accurate if separate
friction models could be created for different areas and parts of the forming surfaces. This is
implemented in AutoForm from version 7.1. and can be used in future work

Acknowledgements

T
The authors are grateful for the financial support from Volvo Cars and wishes to thank Tata

IP
Steel, AutoForm Engineering, and TriboForm Engineering for valuable cooperation and

CR
support. The authors also wish to express their appreciation of their management team at
Volvo Cars and Mats Walter, Head of Mechanical Engineering Department at Blekinge

US
Institute of Technology, for enabling an eminent cooperation.

This research did not receive any specific grant from funding agencies in the public,
AN
commercial, or not-for-profit sectors.

References
M

[1] Sigvant M, Hol J, Chezan T and van den Boogard T 2015 Friction modeling in sheet
metal forming simulations Proc. FTF 2015 (Zürich, Switzerland)
ED

[2] Sigvant M, Pilthammar J, Hol J, Wiebenga J H, Chezan T, Carleer B and van den
Boogard T 2016 Friction and lubrication modeling in sheet metal forming simulations of a
PT

Volvo XC90 inner door, IOP Conf. Series: Materials Science and Engineering 159 (2016)
012021 doi:10.1088/1757-899X/159/1/012021, IDDRG conference 2016 (Linz, Austria)
CE

[3] Sigvant M, Pilthammar J, Hol J, Wiebenga J H, Chezan T, Carleer B and van den
Boogard T 2016 Friction and lubrication modeling in sheet metal forming simulations of the
AC

Volvo XC90 inner door, Journal of Physics: Conference Series 734 (2016) 032090
doi:10.1088/1742-6596/734/3/032090, Numisheet conference 2016 (Bristol, UK)

[4] Hol, Johan. Multi-scale friction modeling for sheet metal forming. University of Twente,
2013.

[5] Lingbeek R Virtual Tool Reworking New Strategies in Die Design using Finite Element
Forming Simulations 2008 (Netherlands Enschede: Print Partners Ip-skamp)
 ACCEPTED MANUSCRIPT

[6] Groβmann K, Wiemer H, Hardtmann A, Penter L and Kriechenbauer S 2009 Adjusting

the contact surface of forming tools in order to compensate for elastic deformations during
the process Proc. 7th European LS-DYNA Conference

[7] Haufe A, Roll K and Bogon P 2008 Sheet metal forming simulation with elastic tools in
LS-Dyna, Proc. Numisheet 2008 (Interlaken, Switzerland)

[8] Pilthammar J, Sigvant M and Kao-Walter S 2016 Including Die And Press Deformations
in Sheet Metal Forming Simulations, Journal of Physics: Conference Series 734 (2016)

T
032036 doi:10.1088/1742-6596/734/3/032036, Numisheet conference 2016 (Bristol, UK)

IP
[9] Nilsson A and Birath F "Topology optimization of a stamping die." AIP Conference

CR
Proceedings. Eds. Jose MA Cesar de Sa, and Abel D. Santos. Vol. 908. No. 1. AIP, 2007.

[10] Reddy N and Tatipala S 2016 Integrated Blankholder Plate for Double Action Stamping

US
Die, Master Thesis, Blekinge Institute of Technology, (Karlskrona, Sweden). http://bth.diva-
portal.org
AN
[11] ABAQUS, User´s Manual, Version 6.14, Dassault Systems.

[12] Banabic D, Carleer B, Comsa D S, Kam E, Krasivskyv A, Mattiasson K, Sester M,

M

Sigvant M and Zhang X 2010 Sheet Metal Forming Processes, Constitutive Modelling and
Numerical Simulation (Berlin Heidelberg: Springer Verlag)
ED

[13] Neto D M, Coër J, Oliveira M C, Alves J L, Manach P Y and Menezes L F, Numerical

analysis on the elastic deformation of the tools in sheet metal forming processes, International
PT

Journal of Solids and Structures, 100-101, 270-285, doi: 10.1016/j.ijsolstr.2016.08.023, 2016

[14] Neto D M, Oliveira M C and Menezes L F, Surface Smoothing Procedures in

CE

Computational Contact Mechanics, Arch Computat Methods Eng (2017) 24:37–87 DOI
10.1007/s11831-015-9159-7
AC
 ACCEPTED MANUSCRIPT

Vitae

Johan Pilthammar

Johan Pilthammar is an Industrial Ph.D. Candidate at Blekinge Institute of Technology (BTH)

and Volvo Cars since September 2015. His research focus is elastic tooling in sheet metal
forming simulations and structural optimization of stamping dies. In 2012, Johan received
M.Sc. degrees in both mechanical engineering and mathematics from BTH. Prior to the start
of his PhD studies, Johan worked as a CAE engineer with TitanX Engine Cooling and Volvo

T
Cars. He was also an exchange student in 2011 at Kunming University of Science and

IP
Technology (KUST) in China.

CR
US
AN
M

Johan Pilthammar.

Mats Sigvant
ED

Mats Sigvant is the Technical Expert on SMF simulations at Volvo Cars in Sweden. He also
serves as an Adjunct Lecturer in Mechanical Engineering at Blekinge Institute of Technology,
PT

Sweden. Mats has a PhD in Computational Mechanics from Chalmers University of

Technology and completed his thesis on the simulation of hemming in the automotive
CE

industry. His publication list consists of two journal papers, twenty-eight conference papers,
and a contribution to a book on sheet metal forming simulations. His research areas are
AC

material modeling, fracture prediction, friction modeling, and inclusion of elastic die
deformations in forming simulations.
 ACCEPTED MANUSCRIPT

Mats Sigvant.

T
IP
Sharon Kao-Walter

CR
Professor Sharon Kao-Walter specializes in solid mechanics at Blekinge Institute of
Technology (BTH). Sharon received a BSc from Shanghai Jiao-Tong University in 1982 and
a Licentiate degree from Lund Institute of Technology in Sweden in 1991. After eight years

US
as a development engineer at Tetra Pak, Sharon completed her PhD at BTH in 2005 and
became an associate professor. She also serves as a visiting professor at the Kunming
AN
University of Science and Technology (KUST) and as an overseas scholar in Shanghai, China.
Sharon is member of the editorial board of The International Journal of Materials and
Mechanical Engineering.
M
ED
PT
CE

Sharon Kao-Walter.
AC