Está en la página 1de 10

Uniaxial tension/compression tests and cyclic bending

tests for hardening parameter identification


P.-A. Eggertsen and K. Mattiasson 

Div. of Material and Computational Mechanics
Dept. of Applied Mechanics
Chalmers University of Technology, SE-412 96 Göteborg, Sweden
e-mail: eggepera@chalmers.se, kjellm@chalmers.se


Dept. 91432, PV22
Volvo Cars Safety Center
SE-405 31 Göteborg, Sweden
e-mail: kmattias@volvocars.com

Abstract: An experimental method that frequently has been used for the determination
of material hardening parameters is the three-point bending test. The advantage of this
test is that it is simple to perform, and standard test equipments can be used. The
disadvantage is that the material parameters have to be determined by some kind of
inverse approach.
An alternative method is the tensile/compression test of a sheet strip. In practice such a
test is very difficult to perform, due to the tendency of the strip to buckle in
compression. In spite of these difficulties some successful attempts to perform cyclic
tension/compression tests have been reported in the literature.
In an ideal world both these test approaches should result in the same material
parameter set-ups. However, a few writers have reported that there are substantial
differences between hardening parameters determined from bending tests and those
from tensile/compression tests. These observations have been partly verified by
investigations performed by the current authors.

Keywords: Bending test, Tension/compression test, Cyclic test, Hardening law,


Parameter identification

1. INTRODUCTION

The residual stresses in the blank after forming are the main cause for the subsequent
springback in a sheet forming operation. The accuracy of the predicted springback in a
Finite Element simulation of the forming operation is very much determined by the
quality of the material modeling. Those parts of the workpiece, which in particular
contribute to the global springback, have usually been subjected to a bending/unbending
deformation mode, when the sheet material has slipped over a tool radius. It is thus of
utmost importance that the material model can accurately describe the material
response, when it is subjected to such a deformation mode. This is considered by the so-
called "hardening law" of the material model. All hardening models involve material
parameters, which have to be determined from some kind of cyclic test.
An experimental method that frequently has been used for the determination of
material hardening parameters is the three-point bending test. The advantage of this test
is that it is simple to perform, and standard test equipments can be used. The
disadvantage is that the material parameters have to be determined by some kind of
inverse approach. The current authors have previously, successfully been utilizing this
method. The test has then been simulated by means of the Finite Element Method, and
the material parameters have been determined by finding a best fit to the experimental
results by means of a Response Surface Methodology.
An alternative method is the tensile/compression test of a sheet strip. In practice
such a test is very difficult to perform, due to the tendency of the strip to buckle in
compression. In spite of these difficulties some successful attempts to perform cyclic
tension/compression tests have been reported in the literature.
However, a few writers have reported that there are substantial differences between
hardening parameters determined from bending tests and those from
tensile/compression tests. These observations have been partly verified by
investigations performed by the current authors.
The purpose of the present study is to try to understand the background of these
differences, to find out the influence on predicted springback, and to determine an
“optimal” methodology for hardening parameter identification.

2. MATERIAL AND MATERIAL CHARACTERIZATION

Only one material is considered in this report. It is a DP600 steel with a thickness of
1.46 mm. The material characterization is governed by uniaxial tests that have been
complemented with viscous bulging tests (see [Sigvant et al., 2009]) aiming at
providing plastic hardening data for strain levels much higher than what can be
achieved in ordinary tensile tests. The resulting plastic hardening curve can be seen in
Figure 1.

1000

800
Effective Stress

600

400

200

0
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
Effective Plastic Strain

Figure 1; Plastic hardening curve for the DP600 material


3. CONSTITUTVE EQUATIONS

Two hardening laws of different complexity are used in this report. First the hardening
law by Armstrong and Frederick [Armstrong and Frederick, 1966], which stated that the
back-stress evolution is given by:
 σ α 
α  Cx sat  α  p (1)
  
where αsat and Cx are material parameters.
The second hardening law is a modified version of the Yoshida and Uemori hardening
law [Yoshida and Uemori, 2002]. The evolution of the back-stress is expressed as:
α  α *  β (2)
with
R 
α *  Cx     σ - α   α*   p
 Y 
(3)
 b 
β  k    σ - β   β   p
 Y  R 
where α* is the relative kinematic motion of the yield surface with respect to the
bounding surface, β is the centre of the bounding surface, Y is the initial size of the yield
surface, Cx and k are material parameters, and R is the isotropic hardening of the
bounding surface:
R ( p )  H ( p )   ( p ) (4)
where H ( p ) is the given, experimental plastic hardening curve.
A further explanation of these two hardening laws can be found in previous works
by the authors [Eggertsen and Mattiasson, 2009-1],[Eggertsen, 2009].

4. DETERMINATION OF KINEMATIC HARDENING PARAMETERS

Two types of experimental set-ups were used in this study in order to determine
unknown material parameters. A three-point bending test and a tension/compression
test.

4.1. Three point bending test


The unknown material parameters in the different kinematic hardening laws described
above, were identified by an inverse approach from a three-point bending test. The
experimental set-up of the bending test is shown in Figure 2. The bearings are free to
rotate around the bearing centre. Furthermore, the sheet is free to slip between two
rollers at the bearings. The punch in the middle is moved with a prescribed
displacement and the applied force is measured. The resulting force-displacement curve
is then used as a target curve in an optimization procedure, where the three-point
bending test is simulated by the finite element code LS-DYNA and the optimization
procedure is performed by the optimization tool LS-OPT.

(a) (b)
Figure 2; Experimental set-up used in the three point cyclic bending tests: (a) a picture
of the experimental equipment; (b) a sketch of the test arrangement

4.2. Tension/compression test


The experimental test equipment for the tension-compression test is shown in Figure 3
below. The test specimen is cut out according to Figure 3a. The specimen is painted
with a randomized dot pattern that is used for the measurements of strains in the
specimen. The specimen is clamped in a holder, with the purpose to prevent buckling
during the test. The holder has a peek hole that is used for measurements of the strains.
A reinforced Teflon film is put between the specimen and the holder in order to
eliminate the influence of friction forces. The silicone pieces are used to prevent the
specimen to buckle between the specimen holding texture and the chucks. The strain
distribution in the specimen was measured by an ARAMIS optical measuring system
during the whole process.

(a) (b)
Figure 3; Experimental equipment used in the tension-compression test: (a) a picture of
the experimental equipment; (b) a picture of the assembled test equipment
Two sets of experiments are performed in this study in which the specimen is loaded up
to 2% and 4%, respectively, in tension and then reloaded to 2% and 4% in compression.

4.3. Optimization procedure


The resulting force-displacement curves from the bending tests and the resulting stress-
strain relationships from the tension/compression tests were used as target curves in an
optimization procedure. For the three-point bending test the experimental set-up was
simulated by the FE-method, while the predicted stress-strain relationship for the
tension/compression test could be determined directly. The parameters in the various
hardening laws served as design variables in this optimization procedure, in which the
normalized error between the predicted and experimental curves was minimized. The
mean squared error (MSE) is defined as
2 2
1 P  f  x   Gp  1 P  ep  x   (5)
MSE    Wp  p   Wp  

P p1  sp  P p1  s p 

where f p  x , p=1, …, P are the values on the computed curve, Gp , p=1, …, P are the
values on the target curve, the Sp, p=1, …, P are residual scale factors, and the Wp,
p=1, …, P are weights applied to the square of the scaled residual (fp - Gp) / sp at point
p. That is, the smaller error, the better fit to the experimental data. In this work all parts
of the target curve are considered to be of equal importance and the weight and residual
scale factors are therefore set to 1.
The identification of the hardening parameters was performed by means of the
optimization code LS-OPT [Stander et al., 2007] and a Response Surface Methodology
(RSM). The RSM is especially advantageous for problems, in which gradients to the
object function are difficult to calculate, such as in this highly nonlinear problem. A
further explanation of the identification procedure can be found in: [Eggertsen and
Mattiasson, 2009-2].

5. SPRINGBACK EXPERIMENT

The problem chosen in this study is the well known U-bend benchmark from the
NUMISHEET’93 conference. The experimental set-up is described in Figure 4 below.
Just as for the three point bending test, the FE-code LS-DYNA was used to solve the
problem. The forming step was solved by means of an explicit, dynamic solver, and the
springback step with an implicit one. The symmetry of the problem was utilized, and
only one quarter of the model was analyzed. The geometry of the tools is shown in
Figure 4a, and the total size of the blank is 300x50mm. The blank holder force was
chosen to be 100kN. Fully integrated quadrilateral shell elements were utilized in the
FE-model. It should be mentioned that for accurate springback predictions it is of great
importance to include the reduction of elastic properties with increased plastic work. A
detailed explanation of this phenomenon can be found in: [Eggertsen and Mattiasson,
2010].

(a) (b)

Figure 4; The NUMISHEET’93 benchmark problem. (a) Experimental set-up; (b)


Definition of the angles θ1 and θ2, used for the evaluation of the springback.

6. RESULTS AND DISCUSSION

The U-bend springback example used in this work is commonly used in the literature on
springback prediction. It is a very simple example and the forming procedure consists of
bending and subsequent re-bending of the material. That is, the material is subjected to
one half of a full loading/unloading stress-strain cycle. Therefore, the optimization of
the unknown material hardening parameters is performed on a half cycle both in the
three-point bending test and the tension/compression tests. However, in a real industrial
application the part geometry is often more complicated, and more than just a half
loading cycle can ocurr. For this reason the optimization procedure is also performed on
two whole loading cycles, both in the three-point bending test and in the
tension/compression test. The choice of two cycles is based on the assumption that this
is an upper limit for what a typical automotive part can be subjected to during forming.
Since the behaviour of the various hardening laws varies from one model to
another, the results are presented individually for each hardening law. For each model,
the results are presented in two tables. The first table describes the results in terms of
the obtained material parameters, the obtained MSE-value, and the resulting tip
deflection in the U-bend springback example. The second table describes how well a set
of hardening parameters, obtained from a certain experiment, can fit the results from
other experiments. The accuracy is measued in terms of MSE-values. A further
explanation of how this table should be read follows in the first sub-chapter below.
6.1. Armstrong-Frederick hardening
As can be seen in Table I, quite good springback results are obtained for the Armstrong-
Frederick hardening law, especially for the cases where only half loading cycles are
considered. In those cases, also the prediction of the tip deflection becomes very good.
For the cases where two whole loading cycles are considered, the MSE-values are larger
and the springback predictions are less accurate. The reason for the larger errors in the
cases with two whole loading cycles is due to the fact that the Armstrong-Frederick
hardening law cannot account for the permanent softening effect (see [Eggertsen and
Mattiasson, 2009]). The results from fitting to a half and two full loading cycles,
respectively, are illustrated in Figure 5b. One of the characteristics of the Armstrong-
Frederick hardening law is that the stress at reverse loading always saturates towards the
isotropic hardening curve. However, as can be seen in Figure 5b, the stress level in the
experimental data saturates towards a fixed value of about 700 MPa. This is the reason
why the MSE-values and the predicted springback become worse for two whole loading
cycles than for a half cycle. In the case shown in Figure 5b, the stress after a half
loading cycle is under-estimated, which leads to that even the springback becomes
under-estimated. Figure 5a shows the predicted and experimental stress-strain
relationships based on the 4% tension-compression test for a half loading cycle. In this
case the stress level after a half cycle is slightly over-estimated, and, thus, also the
springback becomes slightly over-estimated.

Table I; Material parameters, MSE-Values and resulting springback for the Armstrong-
Frederick hardening law.
Springback
Cx sat MSE
27.3 ±0.2
2% 75.1 195.7 0.001321 27.5
½ cycle

4% 77.13 191.7 0.001857 27.5


bending 144.7 199.3 0.001498 27.2
2% 66.69 230.8 0.005793 27.0
2 cycles

4% 23.89 247.7 0.009250 26.1


bending 49.109 248.3 0.002630 29.1
2 97
Table II shows the errors obtained, when the material parameters from a certain
test are used to simulate the other tests. The numbers in the grey-shaded cells represent
MSE-values obtained when fitting the material model to that particular test. The
percentage-values in the other cells in the same row represent the additional errors
obtained, when the material parameters from the test corresponding to the shaded cell
are used to simulate the other tests. It is worth mentioning that the “best” MSE-values in
Table II differs a lot for a half and two whole loading cycles, respectively. Thus, care
should be taken when comparing the errors in terms of percent.
6.2. Modified Yoshida-Uemori
The Yoshida-Uemori model is a more advanced hardening law and accounts also for the
permanent softening effect, besides the same characteristics as the Armstrong-Frederick
law. Therefore, quite good results could be expected, both for the fit to experimental
data and for the springback prediction. As can be seen in Table III, those expectations
are more or less fulfilled. The springback predictions are good in five of six cases, and
the MSE-values are quite low in all cases. An explanation of the bad tip deflection
prediction of the half cycle, 2% tension-compression test is that a strain level of 2% is

(a) (b)

Figure 5; Predicted and experimental stress-strain relationships for the Armstrong-


Frederick hardening: (a) a half cycle based on the 4% tension compression test. (b)
two whole cycles based on the 4% tension compression test.

Table II; MSE-values for various experiments based on different optimizations for the
Armstrong-Frederick hardening law.
½ cycle 2 cycles
Optimized
from 2% 4% bending 2% 4% bending
2% 0.00132 +60% +22% +134% +285% +134%
½ cycle

4% 1
+50% 0.00185 +2.3% +35% +165% +54%
bending +120% 7
+114% 0.001498 +150% +29% +59%
2% +324% +55% +68% 0.0057 +91% +8.1%
2 cycles

4% +504% +339% +23% 93


+99% 0.0092 +57%
bending +552% +187% +89% +27% 5
+36% 0.002630

simply too far from the real experimental conditions, where the strains can reach levels
that are more than five times higher. Table IV reveals that the half cycle, 2% tension-
compression test yields quite bad results when its material parameters are applied to the
five other experiments.
When comparing the results for the Armstrong-Frederick model in Table II with
the results for the modified Yoshida-Uemori model in Table IV, the most striking
observation is that, for the latter model, even the results for the two-cycle tests are quite
accurate. Also for this material model it can be emphasized that, even a large deviation
in terms of percent, can still signify a relatively small MSE-value, since the “best” fits
result in low MSE-values for all experiments.

Table III; Material parameters, MSE-Values and resulting springback for the modified
Yoshida-Uemori hardening law.
Springback
Cx b k h MSE
27.3 ±0.2
2% 93.85 400 51.0 0.634 0.001363 25.7
½ cycle

4% 5 6
141.5 12.0 19.5 0.518 0.001144 27.2
bending 97.5 2 4 4
54.9 94.3 0.340 0.001098 27.1
2% 225.8 99.8 73.3 0.549 00 0.001621 27.4
2 cycles

4% 8 1 6
119.2 34.3 57.3 0.561 0.001730 27.3
bending 95.5 6 1 6
19.3 84.6 0.433 0.001217 27.1
9 5 9

Table IV; MSE-values for various experiments based on different optimizations for the
modified Yoshida-Uemori hardening law.
½ cycle 2 cycles
Optimized bendin
2% 4% bending 2% 4%
from g
2% 0.00136 +132% +309% +278% +300% +131%
½ cycle

4% 3
+156% 0.00114 +42% +51% +183% +26%
bending +175% 4
+69% 0.00109 +161% +310% +13%
2% +107% +32% 8
+25% 0.00162 +143% +21%
2 cycles

4% +172% +67% +11% 1


+32% 0.00173 +3.3%
bending +192% +72% +18% +49% +158% 0.00174
8

7. CONCLUDING REMARKS

As can be seen in the previous chapter, it is possible to find material parameter set-ups
that result in good fits to the cyclic, experimental values, and in good springback
predictions. However, for the simpler Armstrong-Frederick law, this requires that the
identification procedure is made for only a half loading cycle, and that a corresponding
deformation path is present in the springback experiment. For the more advanced
Yoshida-Uemori model good results are generated even for two full cycles.
The results of this report shows, that in order to get good springback predictions, it
is advantageous to determine the material parameters from experiments that are as
similar to the springback problem of interest, as possible. This means that one should
aim to have the same number of cycles and the same strain levels in the cyclic test as in
the springback problem. This is of particular importance for simpler material models.

8. ACKNOWLEDGEMENT

The characterization of the materials used in this study was conducted by Per
Thilderkvist and Jörgen Hertzman at the Industrial Development Center in Olofström,
Sweden. The three-point bending tests were performed by Bertil Enquist at Växjö
University. Their contribution to this work is gratefully acknowledged.
The work has been performed within the Swedish national research program MERA
(Manufacturing Engineering Research Area). Financial support has been provided by
Vinnova.

REFERENCES

[Eggertsen and Mattiasson, 2009-1] Eggertsen, P.A.; Mattiasson, K.; "On the
modelling of the bending-unbending behavior for accurate springback predictions";
International Journal of Mechanical Sciences 51, pp. 547-563
[Eggertsen and Mattiasson, 2009-2] Eggertsen, P.A.; Mattiasson, K.; "An efficient
inverse approach for material hardening parameter identification from a three-point
bending test"; Engineering with computers; DOI: 10.1007/s00366-009-0149-y
[Eggertsen, 2009] Eggertsen, P.A.; "Prediction of springback in sheet metal forming,
with emphasis on material modeling"; Licentiate thesis, Chalmers University of
Technology.
[Eggertsen and Mattiasson, 2010] Eggertsen, P.A.; Mattiasson, K.; "On constitutive
modeling for springback analysis"; International Journal of Mechanical Sciences;
DOI: 10.1016/j.ijmecsci.2010.01.008
[Sigvant et al., 2009] Sigvant, M.; Mattiasson, K.; Vegter, H.; Thilderkvist, P.; "A
viscous pressure bulge test for determination of a plastic hardening curve and
equibiaxial material data"; International Journal for Material Forming 4, pp. 235-
242
[Armstrong and Frederick, 1966] Armstrong, P.J.; Frederick, C.O.; "A mathematical
representation of the multiaxial Bauschinger effect"; G.E.G.B. report RD/B/N 731
[Yoshida and Uemori, 2002] Yoshida, F.; Uemori, T.; "A model of large-strain cyclic
plasticity describing the Bauschinger effect and work hardening stagnation";
International Journal of Plasticity 18, pp. 661-686
[Stander et al., 2007] Stander, N.; Roux, W.; Eggleston, T.; Craig, K.; "LS-OPT
user’s manual, Version 3.2"; Livermore Software Technology Corporation (LSTC)

También podría gustarte