9th European LS-DYNA Conference 2013

_________________________________________________________________________________

Springback Analysis and Optimization in Sheet

Metal Forming

Abdulaziz Alghtani, P.C. Brooks, D.C. Barton, V.V. Toropov

Institute of Engineering Systems and Design, School of Mechanical Engineering

University of Leeds, Leeds LS2 9JT, UK

Abstract
An accurate prediction of springback in sheet metal forming processes requires complex hardening
material models. In this research, numerical analysis of the springback in U-bending was conducted
using the well-known Yoshida model, available and known as the YU model in LS-DYNA. This model
has seven main parameters which describe the behaviour of the material as it undergoes metal
forming processes. Initially, mesh sensitivity studies were conducted to derive a suitable mesh that
represents an appropriate compromise between accuracy and computer time. Secondly design of
experiment (DoE) was employed to make 30 combinations of two design variables (die radius and
clearance) uniformly through a design space. Parametric optimisation studies were also conducted to
investigate the influence of these variables and to make recommendations to minimise the springback.
The results show that the blank element mesh density has a significant effect on the springback
prediction. Additionally the results demonstrate that certain geometrical parameters have a significant
impact in controlling the springback but that optimised values can be identified to minimise the effect.
1 Introduction
Sheet metal forming processes are widely used in the automobile industry. The most frequently used
techniques are bending, stretching, stamping and other. In these sheet metal forming processes,
defects such as rupture, wrinkling, galling and springback in formed parts might occur, whether during
or after the process. The springback phenomenon is the most complex and challenging issue in
industry. It is defined when the sheet metal is removed from the tools, the formed part tends to return
to its original shape [1].

Over the last decades, many researchers have investigated the springback phenomenon in sheet
metal forming processes. In particular Yoshida and Uemori have improved a model of large strain
plasticity to predict more precisely the springback in sheet metal forming [2].

There are several factors that can control the springback in deformed sheet such as blank material
properties, tooling configuration and/or process conditions [3]. For example, researchers have
investigated the influence of die radius and clearance on springback of a deformed steel sheet [4, 5].
Investigators have also examined the influence of lubrication and clearance in springback during the
square cup deep drawing process [6].

In the past many researchers have investigated the process of elastic-plastic deformation and they
proposed many models but within small deformation. Sheet metal forming involves large deformations
followed by an attempt to return to the blank‘s natural shape describing as springback phenomena.
Recently, some researchers mentioned that for a precise prediction of springback, the Bauschinger
effect should be taken into the account[2]. Few researchers investigated experimentally large-strain
cyclic plasticity[7] , while many papers have been published on reverse deformation after the plastic
deformation [8, 9]. Recently, Yoshida has successed doing cyclic tension-compression deformation
experiments for sheet metal at large strain[2]

The numerical model which describes the large deformation and nonlinear material behaviour is the
main challenge of using any kind of simulation package for metal forming analysis. Many researchers
have proposed models but most of them have short falls in a full description of material behaviour.

© 2013 Copyright by Arup

 9th European LS-DYNA Conference 2013
_________________________________________________________________________________

Some of these models are suitable for a small range of deformation and others do not take into the
account the Bauschinger effect which is an essential aspect to predict springback phenomenon. In the
last decade, good improvement in terms of a more precise prediction of springback has been
achieved. Researchers have examined the influence of five different hardening models on springback
prediction and they have concluded that the Yoshida model has acheived good agreement with
experiment [10]. Therefore this model will be used in the present work to investigate the influence of
certain geometrical parameters in the U-bending process which are the die radius and clearance and
to optimize these parameters to minimize the springback.

In a typical optimization problem, selected result responses are optimized by determining the best
combination of design variables located in the design space. The design of experiment (DoE)
approach is used to gain an accurate approximation of response surface which is called. Simulation of
every possible set of the combination of these variables is complex. In such situation meta-modelling
is frequently used reported in [11].The response surface method evaluates an objective function at
several points in the design space to gain a good approximation [12]. Researchers have designed an
optimum blank for sheet metal forming by using the interaction of high- and low fidelity models [13].
Other researchers have designed an optimization scheme to minimize the springback in L-bending
[14]. They developed Gauss-Newton techniques by coupling the Abaqus/standard code with Python.
In the prsent study the genetic algorithm method is used to optimize parameters to give the minimum
springback in the U-bending process.
2 Numerical Analysis
2.1 Finite Element Model
The U- bending process model consists of a punch, die, blank and blank holder. The whole assembly
was modelled using 3-D quadratic shell elements as shown in Fig. 1. The punch, die and blank holder
were considered to be rigid body while the blank was considered to be a deformable body. The
baseline geometrical parameters are listed in Table 1.

Punch
Blank

Holder
F Symmetry Line
X=Y=0
ZX=ZO=0

Die
L
W

D
M

Fig. 1: U-Bending Model

Table 1: Dimensions of the U-bending model

Die Punch
Geometrical Blank
L D W M radius Radius Clearance
Parameters thickness
(Dr) (Pr)
Dimension (mm) 148 50 30 112 4 4 1 1.4

© 2013 Copyright by Arup

 9th European LS-DYNA Conference 2013
_________________________________________________________________________________

2.2 Material model

Many commercial FE codes gain constitutive model to simulate the mechanical behaviour of different
materials. One of them is mixed isotropic-kinematic hardening model. Another one is by adding a
linear component to the previous model. However, these models have limitations in predicting some
aspects of material behaviour. For example Bauschinger effect, workhardening stagnation; when the
rate of workhardening during large deformation is almost zero ,and the reduction in Young’s modulus
in the case of unloading[2]. Also some researchers have suggested constitutive models which
describes both Bauschinger effect, and workhardening stagnation [15]. However, they do not pay
much effort to stress-strain responses in the small scale re-yielding region which is essential to predict
the springback[2].

In the recent past, Yoshida and Uemori [2] have achieved a successful model that can precisely
predict springback. This model consists of two surfaces: a yield surface and a bounding surface. The
yield surface of the kinematic hardening model is surrounded by bounding surface of mixed isotropic-
kinematic hardening as illustrated in Fig. 2. In the other word the yield surface is fixed in its size but its
centre moves with the deformation as the bounding surface is allowed to be change in both size and
location [2].The yield function is expressed by:

(1)

where s and α are the Cauchy stress and the backstress, respectively. Y denotes the yield surface
radius. The bounding surface expression is as follwos:

(2)

where β is the centre of bounding surface and B+R is its initial size with R being associated with
isotropic hardening.

Fig. 2: Yoshida model of two surfaces [2]

This model contains seven material parameters (Y, C, B, R sat, b, m, and h) where can experimentally
be determined with help of some constitutive equations as follows:
1. Y is simply the radius of the yield surface which is the elastic limit.
2. B, (Rsat+b) and m can be determined by equation (3) which comes from the evolution
equation of the mixed isotropic- kinematic hardening of the bounding surface:

) (3)

© 2013 Copyright by Arup

 9th European LS-DYNA Conference 2013
_________________________________________________________________________________

Here denotes the yield stress of the bounding surface under uniaxial tension as shown in Fig.
3. R is the isotropic hardening stress and Rsat is the saturated value of the R at large plastic strain. m
denotes the material parameter that controls the rate of isotropic hardening. b is also a material
p
parameter. ε is plastic strain.

3 b is found using equation (4):

) (4)

Here, βo is the kinematic hardening of the bounding surface at the point of reverse stress, see Fig. 3.
denotes the plastic prestrain. From the experiment can be obtained as shown in Fig. 3. and the
parameter m is found in the step 2, then it is easy to find b by using equation (4)

4 Parameter C is determined from the stress strain curve of the transient Bauschinger
deformation using

(5)

Fig. 3: The motion of a) The yield surface and b) The bounding surface under uniaxial forward-revers
deformation [1]

5 h parameter is determined by simulations of the stress-strain response and it varies from 0 to

1.
Also in the Yoshida model, the effect of plastic strain on Young’s modulus is taken into account by the
following equation:

(6)

Here E is Young’s modulus and Eo and Ea are the Young’s modulus for the original and for the plastic
region, respectively. γ and p stand for a further material constant and the plastic strain respectively.

© 2013 Copyright by Arup

 9th European LS-DYNA Conference 2013
_________________________________________________________________________________

The material are used in this study is high strength steel DP600 and the parameters of the Yoshida
model for this material are shown in Table 2 as reported in [16].
Table 2: The Yoshida model parameters for DP600 [16]
Y B Rsat b C m h Eo Ea γ
(MPa) (MPa) (MPa) (MPa) (GPa) GPa)

360 435 255 66 200 26 0.4 206 152 61

2.3 Methods of analysis

The U-bending process consists of two sequential operations; loading and unloading. The loading
process is initiated as the blank sheet is clamped by the blank holder. Then, the punch is moved down
to draw the blank sheet into the die cavity making a U-shape. Subsequentaly, the unloading process
can be defined when the work piece is freed from any constrains. Two types of analysis are typically
used to solve this kind of process: explicit analysis for the large deformation forming operation and
implicit analysis for the subsequent low deformation unloading operation leading to springback [17]

The boundary conditions assumed for this process were as follows:

1. The punch is constrained in all rotations and translations in both direction X and Y.
2. The blank holder is constrained in all rotation and displacements in both direction X and Y.
3. The die is fixed in all degrees of freedom.
4. The blank fixed along symmetry line in both x and y-direction for translation and ZX and ZO for
rotation as shown in Fig. 1.

The punch is moved at a rate of 2 mm/ms in the Z-direction until it reaches the end of the operation
which is at 50.5 mm of punch displacement. 50 kN load is applied on the middle of the blank holder in
the Z-direction. A one way surface to surface contact was used to define the interaction between
punch, die, and blank and blank holder components. In this contact definition, the blank holder, punch
and die were considered as the master and the blank was considered to be the slave surface. Static
and dynamic coefficients of friction were assumed to be 0.1.

In this study, the explicit method is firstly used to analyse the U-bending process. Then the implicit
mode is utilized to calculate the springback that occurs in the blank during unloading process. In this
implicit analysis, all constrains are removed from the workpiece so it is completly free to take up to its
final deformation shape.

3 Mesh Sensitivity Study

3.1 Mesh definition
In finite element analysis, the mesh density is an important parameter to obtain accurate results. A
small element size for discretisation of the blank provides precise results. On the other hand, the finer
mesh leads to increased computation time. During the U- bending process, a certain area of the blank
experiences much more severe stress than the rest. Moreover as the blank width is much bigger than
its thickness, the effect of changing the mesh in the Y direction is negligible. Based on the above
consideration, the blank has been divided into three areas designated A, B and C as shown in Fig. 4.
The mesh size in region B is constant and minimum. Beyond zone B, the element size in zone A and
C is increased slowly up to each free end. The element aspect ratio in zone B is set as shown in Fig.
4.

© 2013 Copyright by Arup

 9th European LS-DYNA Conference 2013
_________________________________________________________________________________

w (A)

(B)

(C)

Fig. 4: The blank model for U-bending Simulation

The sensitivity of the results to the mesh density of the blank has been investigated to achieve a
balance between the accuracy of the result and computation time. Table 3 shows the six different
mesh sizes of the blank for the U-bending model.

Table 3: Different element size of the blank in U-bending and springback results

Elapse CPU time (hr : min : sec) Springback

w Number of Forming Springback

Mesh l (mm) Θ1 Θ2
(mm) Elements (explicit) (implicit)

1 1.50 3.0 680 00:07:41 00:00:17 88.72 91.94

2 1.25 2.5 972 00:12:17 00:00:24 88.41 92.90
3 1.00 2.0 1530 00:31:25 00:00:35 87.83 94.12
4 0.75 1.5 2700 01:02:08 00:01:05 87.08 96.67
5 0.50 1.0 6090 03:31:29 00:02:54 86.57 97.54
6 0.25 0.5 24360 26:14:36 01:05:02 86.55 97.58

3.2 Punch Force prediction

The punch force versus its displacement has been plotted for the six different blank meshes as
defined in Table 3. The trends are very similar for all meshes with slight differences in their behaviour.
The punch force rises swiftly from the first contact point until it reaches almost 13 kN at 10 mm of
punch travel. After that the punch force increases very slightly to the end of the operation.
Fig. 5 shows fluctuated forces for meshes 1-3 and Fig. 6 shows consistent trends with reasonable
fluctuation for the last three blank meshes.
16

14

12
Punch Force (kN)

10 Mesh 1

8
Mesh 2
6
Mesh 3
4

0
0 5 10 15 20 25 30 35 40 45 50 55
Punch displacement (mm)

Fig. 5: Punch force versus displacement in U-bending process for meshes 1-3

© 2013 Copyright by Arup

 9th European LS-DYNA Conference 2013
_________________________________________________________________________________

16

14

12

Punch Force (kN)

10 Mesh 4

8
Mesh 5
6

4
Mesh 6
2

0
0 5 10 15 20 25 30 35 40 45 50 55
Punch displacement (mm)

Fig. 6: Punch force versus displacement in U-bending process for meshes 4-6
3.3 Springback prediction
The springback in U-bending first is identified by measuring the maximum z-displacement of the blank
flange after the forming operation [10, 18]. In some cases; the calculation of the springback by
measuring the maximum-z displacement is not sufficient. Fig. 7 (a) shows almost the same maximum
z-displacement along the top side of the U-shape. However, Fig. 7 (b) displays different z-
displacement along the flange side of U-shape. Therefore the U-shape in both figures is different
though the maximum z-displacement along the flange side is very similar.

Z-
displacement
along the top
side

(a) (b)

Fig. 7: Z-displacement for U-bending after the springback

Another approach for measuring the springback is by calculating the exact angle that follows the
springback action [14, 19] as illustrated in Fig 8. Two angles are considered which are θ1 and θ2. The
angle of θ1 is between line AB and CD, whilst θ2 is identified as the angle between line CD and EF as
can be seen in Fig 8. This study has utilized this second technique since it seems to be a more
reliable methodology to measure the springback.
B Θ1
A

20 mm

D
15 mm

E F
Θ2

Fig 8: Springback angles (θ1 and θ2)

Table 3 shows a remarkable increase for θ2 and slight decline for θ1 from mesh 1 to mesh 4. However
there is only a slight difference in the angles between mesh 5 and mesh 6 with a significant increase in
elapse time for both the explicit (forming0 and implicit (springback) analyses. Therefore mesh 5 will be
used for all further investigations.

© 2013 Copyright by Arup

 9th European LS-DYNA Conference 2013
_________________________________________________________________________________

4 Design of Experiments and Optimization

In a typical optimization problem, selected result responses are optimized by determining the best
combination of design variables located within the design space. The location of the selecting points is
important to gain an accurate approximation of the response surface. In this study, the Design of
Experiments (DoEs) approach has beeen used to locate there points.

This study will investigate the influence of two design variables which are the die radius and
clearance. Simulation of every possible set of the combination of these variables is complex. In such
situations meta-modelling is frequently used [11]. This involves carrying out numerical analysis for
certain combinations of design variables within a design space and fitting an approximate response
surface to the actual results. In other words, the meta-modelling is used to provide an accurate
response surface from a minimum number of simulations or experiments.

In this study, an Optimal Latin Hypercube (OLH) Design of Experiments (DoE) is used to make 30
combination of the two design variables uniformly through a design space that is determined by the
upper and lower limit for each variable. The DoE is divided into two parts; one is a build model and the
second is a validation model. The purpose of each model is to maximize their uniformity taking
account of the space-filling properties of the designs. As a result two non-overlapping DoEs are
obtained which are the union of build points and validation points. There are several approaches to
generate the OLH using criteria such as integrated mean squared error, maximizing entropy[20,
21]and maxi-min distance[22].

In this study, the PermGA algorithm reported in [11] is used to generate the OLH build and validation
DoEs as uniformly as possible. This has the following physical meaning, the design space has points
of mass unit and these points apply force on each other leading to a system that has a minimum
potential energy. Using this method 30 combinations of the two design variables were uniformly
distributed in a design space that is determined by upper and lower limit for each variable. The upper
and lower limits for the two design variables are listed in Table 4. 30 points repersenting the
combination set of die radius and clearance are uniformly distributed in the design space as shown in
Fig. 9.

Table 4 Upper and lower limits for the two design variables
Variables Upper) Lower
Die Radius (mm) 10.7 2
Clearance (mm) 1.07 0.2

1.1

0.9
Clearance (mm)

0.8

0.7

0.6

0.5

0.4

0.3

0.2
2 3 4 5 6 7 8 9 10 11
Die Radius (mm)
Bulid Points Validation Points

Fig. 9: Build and validation points

U-bending operations for the 30 selected points were simulated as described in section 2.3 using Ls
Dyna software and the springback results are obtained. From this information, an approximation
response surface of the springback within the design space was created using the moving least
squares approximation method.

© 2013 Copyright by Arup

 9th European LS-DYNA Conference 2013
_________________________________________________________________________________

The response surface for the springback angles θ1 and θ2 is shown in Fig. 10 (a) and (b) respectively.
Inspection of the thirty solutions showed that the die radius influences the springback angles much
more than the clearance. The springback that represented by θ 2 shows that the springback increases
dramatically with increase of the die radius. Interestingly the springback angle θ1 has different
behaviour with variation of the die radius. Fig. 10 (a) illustrated that the θ1 is highest at the lower limit
of the die radius (2 mm) followed by a sharp decrease up to 6 mm of die radius then the angle
increases again to the upper die radius limit.
Finallty the genetic algorithm GA optimization technique is employed using Hyperstudy v11 of Altiar
packge. The main objective is to find optimal design variables that provides angles closet to 90
degrees or near. Table 5 shows the optimum design variables that leads to minimum springback
angles. Also this table compares the predicted springback with the simulated one with mild error.

Table 5: Comparsion of the predicted springback angles with the simulated one for the optimum
design variables
Design variables optimization Simulation Error %
Die Radius (mm) Clearance (mm) θ1 θ2 θ1 θ2 θ1 θ2
2 0.245 91.755 91.707 92.162 92.141 0.43 0.54

θ1
Θ2

Fig. 10: Response surface of springbacjk for a) θ1 and b) θ2

Fig. 11: The springback angles θ1 and θ2 for the optimum design variables

© 2013 Copyright by Arup

 9th European LS-DYNA Conference 2013
_________________________________________________________________________________

5 Conclusion
The numerical modelling of the U bending processes was developed using the Yoshida material
model availble in LS-DYNA. The mesh size of the deformable sheet is an important factor that
influences the accuracy of the numerical results. Therefore this study conducted a mesh sensitivity
study for the blank and selected an appropriate mesh size for the rest of the study. Important
parameters such as die profile radius and clearance that influences the springback in U bending
processes were investigated. A meta-modelling was used to generate suitable points of combination
of die radius and clearance within the design space. These points were then simulated using LS-
DYNA. It was found that the die radius influences the springback angles much more than the
clearance. The GA optimization technique was employed using Hyperstudy version 11 to search for
the optimum springback within the design space. It is demonstrated that the optimum springback is
when the die radius and clearance equal to 2 mm and 0.245 mm respectively for the particular
operation considered. These values of the design variables wre then employed in LS-DYNA for
validation purposes. The result showed that the springback is indeed minimized using this
methodology.

© 2013 Copyright by Arup

 9th European LS-DYNA Conference 2013
_________________________________________________________________________________

References

1. Gau, J.-T. and G.L. Kinzel, A new model for springback prediction in which the Bauschinger
effect is considered. International Journal of Mechanical Sciences, 2001. 43(8): p. 1813-1832.
2. Yoshida, F. and T. Uemori, A model of large-strain cyclic plasticity describing the Bauschinger
effect and workhardening stagnation. International Journal of Plasticity, 2002. 18(5-6): p. 661-
686.
3. Z. Marciniak, J.L.D.a.S.J.H., Mechanics of Sheet Metal Forming. 2002, Butterworth-
Heinemann.
4. LIU, R.G.D.a.Y.C., Control of Springback in a Flanging Operation. APPLIED
METALWORKING, 2 January 1984. 3.
5. Chen, F.-K. and S.-F. Ko, Deformation Analysis of Springback in L-Bending of Sheet Metal.
Advanced Science Letters, 2011. 4(6-7): p. 1928-1932.
6. Lei, L.P., S.M. Hwang, and B.S. Kang, Finite element analysis and design in stainless steel
sheet forming and its experimental comparison. Journal of Materials Processing Technology,
2001. 110(1): p. 70-77.
7. Wilson, D.V. and P.S. Bate, Reversibility in the work hardening of spheroidised steels. Acta
Metallurgica, 1986. 34(6): p. 1107-1120.
8. Hasegawa, T., T. Yakou, and S. Karashima, Deformation behaviour and dislocation structures
upon stress reversal in polycrystalline aluminium. Materials Science and Engineering, 1975.
20(0): p. 267-276.
9. Christodoulou, N., O.T. Woo, and S.R. MacEwen, Effect of stress reversals on the work
hardening behaviour of polycrystalline copper. Acta Metallurgica, 1986. 34(8): p. 1553-1562.
10. Eggertsen, P.A. and K. Mattiasson, On the modelling of the bending–unbending behaviour for
accurate springback predictions. International Journal of Mechanical Sciences, 2009. 51(7): p.
547-563.
11. Narayanan, A., et al., Simultaneous model building and validation with uniform designs of
experiments. Engineering Optimization, 2007. 39(5): p. 497-512.
12. Mkaddem, A. and R. Bahloul, Experimental and numerical optimisation of the sheet products
geometry using response surface methodology. Journal of Materials Processing Technology,
2007. 189(1-3): p. 441-449.
13. Hino, R., F. Yoshida, and V.V. Toropov, Optimum blank design for sheet metal forming based
on the interaction of high- and low-fidelity FE models. Archive of Applied Mechanics, 2006.
75(10-12): p. 679-91.
14. Gassara, F., et al., Optimization of springback in L-bending process using a coupled
Abaqus/Python algorithm. International Journal of Advanced Manufacturing Technology, 2009.
44(1-2): p. 61-7.
15. Hu, Z., E.F. Rauch, and C. Teodosiu, Work-hardening behavior of mild steel under stress
reversal at large strains. International Journal of Plasticity, 1992. 8(7): p. 839-856.
16. Ninshu Ma, Y.U., Yuko Watanabe, Takaki Ogawa, Springback Prediction by Yoshida-Uemori
Model and Compenstion of Tool Surface Using JSTAMP. Numisheet 2008, 2008.
17. Prior, A.M., Applications of implicit and explicit finite element techniques to metal forming.
Journal of Materials Processing Technology, 1994. 45(1–4): p. 649-656.
18. Eggertsen, P.A. and K. Mattiasson, On constitutive modeling for springback analysis.
International Journal of Mechanical Sciences, 2010. 52(6): p. 804-818.
19. Chen, P. and M. Koç, Simulation of springback variation in forming of advanced high strength
steels. Journal of Materials Processing Technology, 2007. 190(1-3): p. 189-198.
20. Park, J.-S., Optimal Latin-hypercube designs for computer experiments. Journal of Statistical
Planning and Inference, 1994. 39(1): p. 95-111.
21. Ye, K.Q., W. Li, and A. Sudjianto, Algorithmic construction of optimal symmetric Latin
hypercube designs. Journal of Statistical Planning and Inference, 2000. 90(1): p. 145-159.
22. Johnson, M.E., L.M. Moore, and D. Ylvisaker, Minimax and maximin distance designs. Journal
of Statistical Planning and Inference, 1990. 26(2): p. 131-148.

© 2013 Copyright by Arup