Anda di halaman 1dari 7

Journal of Materials Processing Technology 125–126 (2002) 785–791

Finite element simulation of springback in sheet metal forming


Luc Papeleux, Jean-Philippe Ponthot*
LTAS-Milieux Continus et Thermomécanique, Université de Liège, 1, Chemin des Chevreuils, B-4000 Liège-1, Belgium
Received 20 December 2001; received in revised form 24 February 2002; accepted 24 February 2002

Abstract

Although finite element analysis (FEA) is successful in simulating complex industrial sheet forming operations, the accurate and reliable
application of this technique to springback has not been widely demonstrated. Several physical parameters, as well as numerical, influence this
phenomenon and its numerical prediction. In this paper, we investigate the impact of these parameters on the springback appearing in a 2D U-
draw bending.
# 2002 Elsevier Science B.V. All rights reserved.

Keywords: Springback; Sheet metal forming; Kinematic hardening; Finite element

1. Introduction in Sections 5–8, numerical experimentation will be carried


out under various combinations of numerical and physical
Springback describes the change in shape of a formed factors in order to determine how they influence the process.
sheet upon removal from the tooling. It is more severe for
materials with higher strength-to-modulus ratios (e.g., alu-
minum and high strength steel (HSS)). The finite element 2. Benchmark description
analysis (FEA) of springback is shown to be very sensitive to
many numerical parameters, including the number of Let us start with a description of the benchmark we will
through-thickness integration points, type of element, mesh study throughout this paper, namely the U-draw bending
size, angle of contact per element on die shoulder, possible presented at the NUMISHEET’93 Conference [3]. It is very
inertia effects and contact algorithm. Moreover, springback useful for our study since, on the one hand, it is very sensitive
is also sensitive to many physical parameters including to many parameters from both the experimental and the
material properties, hardening laws, coefficient of friction, numerical point of view and, on the other hand, springback
blankholder force (BHF) and possible unloading procedure. can be quantified by three simple geometric measurements.
All that makes springback simulation very cumbersome. A The schematic drawing of the U-draw bending is shown in
challenging benchmark, named the U-draw bending was Fig. 1. The three geometric quantities that quantify spring-
presented at the NUMISHEET’93 Conference [3] where a back are shown in Fig. 2. These consist of two angles y1 and
collection of numerical, as well as experimental results can y2 and the radius r.
be found. This benchmark will be analyzed here in details. The main hypothesis used throughout this simulation are
The structure of the paper is the following. In Section 2, the following:
we will describe the benchmark. In Section 3, we will
 Element type: 2D linear shell element.
introduce how we actually perform the simulation. Then,
 Time integration scheme: Chung–Hulbert dynamic impli-
in Section 4, we will show that, under the benchmark
cit [1].
conditions (given material parameters, friction coefficient,
 Springback simulation: complete tool removal including
BHF, etc.), our numerical results are bracketed by the
contact interactions.
experimental ones and that they are consistent with numer-
 Contact algorithm: penalty method on analytically
ical results obtained by the commercial code OPTRIS. Then,
defined rigid bodies.
*
Corresponding author. Tel.: þ32-4366-9310; fax: þ32-4366-9141.
Further details about the shell element used can be found
E-mail addresses: l.papeleux@ulg.ac.be (L. Papeleux), in [8,7] and further details about METAFOR, the software
jp.ponthot@ulg.ac.be (J.-P. Ponthot). used here can be found in [6]. Result of simulations will be

0924-0136/02/$ – see front matter # 2002 Elsevier Science B.V. All rights reserved.
PII: S 0 9 2 4 - 0 1 3 6 ( 0 2 ) 0 0 3 9 3 - X
786 L. Papeleux, J.-P. Ponthot / Journal of Materials Processing Technology 125–126 (2002) 785–791

Fig. 1. Schematic U-draw bending process.

Fig. 3. Punching phase.

Fig. 2. Description of the three quantities y1, y2 and r to quantify


springback. r is the radius of the circle defined by A, B and C.

Table 1
Elastic, thickness and frictional properties for U-draw bending test

Mild steel HSS Aluminum

Young’s modulus 206000 206000 71000


Poisson’s ratio 0.3 0.3 0.33
Thickness (mm) 0.78 0.74 0.81
Frictional coefficient 0.144 0.129 0.162

Table 2
Isotropic hardening properties

Material Hardening law

Mild steel 157:1281ð1 þ 140:5086epl Þ0:2589 Fig. 4. End of punching phase.


pl 0:2182
HSS 265:45ð1 þ 80e Þ
Aluminum 132:23ð1 þ 60:31epl Þ0:3593

shown for the three following materials: mild steel, HSS and
aluminum. Materials properties are given in Tables 1 and 2.

3. Simulation process

In the simulations performed, special attention is given in


order to remain as close as possible to the physical process.
Using an implicit dynamic time integration scheme, we are
able to simulate the whole process including both the
loading and the unloading phases with the same time Fig. 5. Punch withdrawal.
L. Papeleux, J.-P. Ponthot / Journal of Materials Processing Technology 125–126 (2002) 785–791 787

Fig. 6. After complete punch withdrawal. Fig. 10. Final geometry—note the contact forces.

Fig. 11. Comparison of the final geometries for the different materials.

Fig. 7. Blankholder withdrawal.


force. In these figures, it is clearly seen that contact forces do
exist during the unloading phase. Fig. 11 shows the final
geometries, after complete unloading for the three different
materials used.
The simulation usually cannot be performed in the same
way using commercial codes where the tendency is to use two
different algorithms. Indeed, an explicit algorithm is generally
used for the loading phase, including the contact interactions
while an implicit algorithm, without any contact capabilities
is used to simulate the unloading phase, see e.g. [2,4].
Fig. 8. After complete blankholder withdrawal.

integration algorithm and with a frictional contact algorithm 4. Results according to the material
active all along the process.
The whole process is described in Figs. 3–10. In these Results of these simulations with elastoplastic isotropic
figures, the arrows represent the contact forces. The length material and isotropic hardening are described in Figs. 12–14
of the arrow is proportional to the magnitude of the contact

Fig. 9. Specimen ejection—note the contact forces. Fig. 12. Parameter Y1 vs material.
788 L. Papeleux, J.-P. Ponthot / Journal of Materials Processing Technology 125–126 (2002) 785–791

Fig. 13. Parameter Y2 vs material. Fig. 16. Parameter Y2 vs BHF.

Fig. 14. Parameter r vs material.


Fig. 17. Parameter r vs BHF.

for the three materials. Result obtained with our code META- increases with small forces, but decreases as the force
FOR [6] are compared to experimental data (Min, Max and increases for large force values. The same phenomenon is
mean value) [3] and simulation data obtained using the found in the FE simulation using mild steel as shown in
commercial code OPTRIS [5]. All these results show that Figs. 15–17. The curves exhibit an extremum for some
our numerical results are consistent with the results from values of the BHF and one can also find a BHF value where
OPTRIS and are bracketed by the Min and Max values from the considered springback parameter almost vanishes.
the experimental data. This phenomenon can be explained by the fact that with
low BHF, the punch induces mostly bending stresses in the
material, but as the blankholder holds the blank more
5. Sensitivity to the BHF severely, the stresses induced by the punching phase become
mostly tensile stresses.
We will now study the influence of several physical and
numerical parameters on springback. The first of those is the
BHF. It is experimentally observed that the springback 6. Sensitivity to the coefficient of friction

Friction is an important but not very well known factor


influencing springback. Its modeling is difficult because this
coefficient is probably different on the curved and flat parts of
both the die and punch. Moreover, it is very difficult to
measure those coefficients experimentally. So, we have cho-
sen to use the same coefficient on all parts of all the tools and
study the results obtained for different values. Figs. 18–20
show that the same evolution as for BHF is observed: spring-
back curves also exhibit an extremum (we used mild steel for
this case). It is also important to note that the values of the
parameters for low coefficient of friction are significantly
different than the values for mean (i.e. m ¼ 0:15) friction
Fig. 15. Parameter Y1 vs BHF. coefficients. That makes friction a very sensitive parameter.
L. Papeleux, J.-P. Ponthot / Journal of Materials Processing Technology 125–126 (2002) 785–791 789

physically correct nor enables us to measure the springback


obtained since the blank would have no static final shape. We
could have used a scheme containing artificial damping but
this does not represents what physically happens. Further-
more, people using the vanishing tools method generally use
those kind of mixed schemes. We did not use a quasi-static
scheme for the blankholder removal phase because in such a
case we need to constrain a degree of freedom vertically in
order to prevent a rigid body movement from making the
iterative process diverge and this would have prevented us
from modeling all the springback effects correctly. We have
used three kinds of explicit schemes:
Fig. 18. Parameter Y1 vs friction coefficient.
 the classical central difference scheme;
 a scheme where we have scaled up the mass in order to
artificially increase the maximum time step allowed by a
factor 10 (density has been multiplied by 100);
 a scheme where the punch speed is 10 times larger so that
the total forming time has been divided by 10.
Figs. 21–23 show the values obtained for the springback
parameters while using different schemes. In these figures,
the minimum and maximum values of the experimental
results are also represented.
As can be seen from the figures, results are slightly
different whether we use implicit or explicit scheme, but
they are still in the range of the experimentally obtained
values. The main difference is in the CPU cost, as shown in
Fig. 19. Parameter Y2 vs friction coefficient.

Fig. 21. Parameter Y1 obtained with various integration schemes.


Fig. 20. Parameter r vs friction coefficient.

7. Influence of the time integration algorithm

Until now, we have used the dynamic implicit Chung–


Hulbert [1] scheme in every case we have studied. We are
now going to show why we did so and why being able to deal
with contact using implicit schemes is so important. For this
purpose, we used an explicit scheme for the punching phase,
an implicit dynamic one for the punch removal and for the
blankholder removal and a quasi-static implicit scheme for
the ejection phase. We did not use an explicit scheme for the
whole process because we would have obtained undamped
vibrations of the upper part of the blank after ejection and
those vibrations would never be damped out, which is not Fig. 22. Parameter Y2 obtained with various integration schemes.
790 L. Papeleux, J.-P. Ponthot / Journal of Materials Processing Technology 125–126 (2002) 785–791

Fig. 25. Parameter y1 vs b.


Fig. 23. Parameter r obtained with various integration schemes.

Fig. 26. Parameter y2 vs b.


Fig. 24. Relative CPU costs for different integration schemes.

Fig. 24. The explicit scheme without mass scaling requires represent the minimum, maximum and average value of the
more than 60 times more CPU than its implicit counterpart. experimental results.
With a 10 times larger time step, the implicit scheme is still It is obvious that complete kinematic hardening ðb ¼ 0Þ
six times faster. That is why we chose this scheme to make does not provide accurate results. A minimum of b ¼ 0:3 is
all the other tests. required for the springback parameters to be in the range of
experimental data. Furthermore, the angle y1 is less sensitive
to kinematic hardening than the other springback para-
8. Sensitivity to the constitutive parameters meters, because the region defining angle y1 is only bent
once during the process and Bauschinger effect only appears
In the previous section simulations were performed using on successive plastic loading in tension-compression. The
isotropic plastic criterion and isotropic hardening. sensitivity of y2 and r to b is higher since it corresponds to a
As the sheet is bent and unbent in the U-draw bending, we structural region where bending and unbending occur.
assume that Bauschinger effect must have some influence on Finally, it is worth noting that the last two results fit the
springback. To emphasize this, the mild steel simulation has mean experimental value for a value close to b ¼ 0:8.
been performed once again but this time with a mixed
hardening law. As the correct parameters for kinematic
hardening are not known, plastic limit stress curve (given
by a tension test) is split between Von Mises yield stress
evolution ðsv ðepl ÞÞ and an equivalent back stress evolution
ðaðepl ÞÞ through a user defined parameter b

sv ¼ s0v þ bðspl ðepl Þ  s0v Þ (1)


pl
a ¼ ð1  bÞðspl ðe Þ  s0v Þ (2)

Simulations have been performed from the complete iso-


tropic hardening model ðb ¼ 1Þ to the complete kinematic
model ðb ¼ 0Þ. The results are displayed in Figs. 25–27. In
these figures, three horizontal lines have been drawn. They Fig. 27. Parameter r vs b.
L. Papeleux, J.-P. Ponthot / Journal of Materials Processing Technology 125–126 (2002) 785–791 791

A last simulation has been performed using this time an References


anisotropic plastic criterion. Using the Hill 48 plastic criter-
ion with a mean R value (in plane isotropy) given by [1] J. Chung, J.M. Hulbert, A time integration algorithms for structural
dynamics with improved numerical dissipations: the generalized-a
R0 þ 2R45 þ R90
R¼ ¼ 1:77 (3) method, J. Appl. Mech. 60 (1993) 371–375.
4 [2] S.W. Lee, D.Y. Yang, An assessment of numerical parameters
Once again, numerical results fall in the range of experi- influencing springback in explicit finite element analysis of sheet
mental data (Figs. 25–27). metal forming process, J. Mater. Process. Technol. 80–81 (1998)
60–67.
[3] M. Makinouchi, E. Nakamichi, E. Onate, R. Wagoner, In: Proceedings
of the NUMISHEET’93 Second International Conference on Numer-
9. Conclusions ical Simulation of 3D Sheet Metal Forming Processes, Tokyo, Japan,
1993.
In this paper, a classical benchmark in springback pre- [4] K. Mattiasson, P. Thilderkvist, A. Strange, A. Samuelson, Simulation
diction was described and the results obtained were com- of springback in sheet metal forming, In: Shen, Dawson (Eds.),
pared with experimental data. A simulation method Proceedings of the NUMIFORM’95 Simulation of Materials Processing:
Theory, Methods and Applications, 1995, pp. 115–124.
describing the tools action during the whole process was [5] Technical Report, Dynamic software, OPTRIS S.A., Springback in
introduced and compared with the classical vanishing tools optris 6.0, 1996.
approximation. The influence on springback of several [6] J.-P. Ponthot, Traitement unifié de la Mécanique des Milieux Continus
parameters such as BHF, friction, spatial integration, time solides en grandes transformations par la méthode des éléments finis,
integration scheme and constitutive laws describing the Ph.D. Thesis, University of Liège, Liège, Belgium, 1995.
[7] J.-P. Ponthot, L. Papeleux, S. Gohy, X. Collard, Numerical simulation
material used were studied in details. of springback in sheet metal forming. In: Proceedings of the
ECCOMAS 2000/COMPLAS VI European Congress on Computa-
tional Methods in Applied Sciences and Engineering, Barcelona,
Acknowledgements Spain, 2000, CD-ROM.
[8] D. Quoirin, Modélisation des grandes déformations de corps minces,
The support of the Walloon Region, under grant number Application à la mise à forme, Ph.D. Thesis, University of Liège,
9713609 is gratefully acknowledged. Liège, Belgium, 1995–1996.

Anda mungkin juga menyukai