You are on page 1of 16

Journal of Materials Processing Technology 228 (2016) 43–58

Contents lists available at ScienceDirect

Journal of Materials Processing Technology


journal homepage: www.elsevier.com/locate/jmatprotec

Determination of strain hardening parameters of tailor hardened


boron steel up to high strains using inverse FEM optimization and
strain field matching
T.K. Eller a,b,∗ , L. Greve a , M. Andres a , M. Medricky a , V.T. Meinders b ,
A.H. van den Boogaard b
a
Volkswagen AG, Group Research, P.O. Box 1777, 38436 Wolfsburg, Germany
b
University of Twente, P.O. Box 217, 7500 AE Enschede, The Netherlands

a r t i c l e i n f o a b s t r a c t

Article history: In this article, an inverse FEM optimization strategy is proposed for identification of the strain hardening
Received 16 February 2015 parameters of boron alloyed steel 22MnB5 in five different hardness grades. In the proposed elasto-plastic
Received in revised form constitutive model, the strain hardening is represented by a nonlinear combination of the Swift hardening
16 September 2015
law and a modified version of the Voce law. Initial fits of these two classical strain hardening equations
Accepted 20 September 2015
are constructed based on experimental data from uni-axial tensile tests up to the point of diffuse necking.
Available online 25 September 2015
The strain hardening response beyond the point of diffuse necking is determined from a 3D strain field
analysis of notched tensile and equibiaxial tension tests. Both the measured force–displacement curves
Keywords:
Hot forming
and the strain fields are used as inputs for the optimization algorithm that identifies suitable material
Tailored properties model parameters by minimizing the differences between experimental and simulated results. In order to
Strain hardening show the contribution of the different parts of the elasto-plastic model for representing the real material
Hardness response, three simplified versions of the proposed model and the parameter identification procedure are
22MnB5 applied on two selected hardness grades, confirming the importance of a flexible strain hardening law,
suitable yield criteria and accurate experimental data up to high plastic strains. The calibrated model was
shown to accurately capture the elasto-plastic response of 22MnB5 in different hardness grades, with an
excellent representation of the strain fields up to the point of fracture.
© 2015 Elsevier B.V. All rights reserved.

1. Introduction state. Karbasian and Tekkaya (2010) provide a general overview


on current hot stamping technologies, a more in-depth view on
The automotive industry is continuously working on the weight the underlying microstructural transformations and on modeling
reduction of their vehicles in order to lower both fuel consump- approaches for simulation of the hot stamping process can be found
tion and CO2 emissions, while maintaining or even improving the in Naderi (2007) and Åkerström (2006), respectively.
crashworthiness in accordance with increasing safety demands. In Although hot stamped components benefit from an exception-
recent years, the hot stamping process has gained popularity as a ally high strength, car manufacturers are exploring new production
manufacturing method for crash-relevant structural components. methods that allow the introduction of regions of reduced strength
In the conventional hot stamping process, boron steel blanks are and higher ductility for local, improved energy absorption. A well-
fully austenitized in a furnace, after which they are simultaneously known application is the B-pillar: for optimal performance in a side
formed and quenched in a cooled stamping tool. Due to the high crash, the bottom part should show a high energy absorption capac-
cooling rates during the forming process, the austenitic microstruc- ity, while the upper part should ensure a high intrusion resistance
ture transforms into martensite, causing the tensile strength of the (Maikranz-Valentin et al., 2008). Common methods to produce
material to increase from an initial 600 MPa to 1500 MPa in the final parts with such ‘tailored’ mechanical properties are: local reduc-
tion of the in-die cooling rate through the use of tool materials with
varying thermal conductivities or using increased die tempera-
∗ Corresponding author at: Volkswagen AG, Group Research, P.O. Box 1777, 38436
tures (heated/tailored tooling), or by post tempering fully hardened
Wolfsburg, Germany.
parts with differential heating in the furnace (George et al.,
E-mail address: tom.karl.eller@volkswagen.de (T.K. Eller). 2012).

http://dx.doi.org/10.1016/j.jmatprotec.2015.09.036
0924-0136/© 2015 Elsevier B.V. All rights reserved.
44 T.K. Eller et al. / Journal of Materials Processing Technology 228 (2016) 43–58

To be able to fully exploit the possibilities of such tailor hardened


components, it is of paramount importance to attain accurate pre-
dictive models of their crash response. In particular the modeling of
transition zones between harder and softer regions is a challenging
task. In previous work of the authors, it was found that the strain
hardening behavior has a crucial influence on the onset and location
of fracture initiation in hardness transition zones: a small decrease
in the gradient of the strain hardening curve leads to a higher prob-
ability of localized necking, which will ultimately lead to fracture
(Eller et al., 2014). Before considering the fracture characteristics of
tailor hardened 22MnB5, it is thus required to obtain an appropri-
ate elasto-plastic constitutive model with accurately determined
parameters.
Several approaches concerning the modeling of the elasto-
plastic response of boron steels and identification of the
corresponding parameters have been presented in literature. Mohr
and Ebnoether (2009) used a planar isotropic version of the Hill
(1948) yield surface in combination with a piecewise linear hard-
ening equation k(ε̄p ) to model the plastic behavior of fully hardened
22MnB5. The strain hardening function was obtained from a pure Fig. 1. Measured cooling curves of the samples superimposed on the CCT diagram
shear experiment, which yielded experimental data up to an engi- of 22MnB5. The cooling rates indicated in the diagram are instantaneous values.

neering shear strain of 0.28. Although the model provided excellent


results for specimens with the same geometry under different
loading angles, the stress level of the uni-axial tensile test was Yld2000-2D yield function (Barlat et al., 2003) with an extension for
overestimated by 10%. Bardelcik et al. (2012) performed minia- general three-dimensional stress states by Dunand et al. (2012)
ture tensile tests at four different strain rates from 0.003 s−1 to with ˛k = 1 is used and compared to results obtained with the Von
1075 s−1 with 22MnB5 in five different hardness grades ranging Mises yield criterion. For fitting of the initial stress–strain response,
from 268 HV to 466 HV. For the quasi-static tests, experimental true force–displacement curves of uni-axial tensile tests are used up
stress–strain data reached up to strains of 0.035 for the hardest to the point of necking initiation. Because the model is planned
grade and 0.07 for the softest grade. A modified Voce (1955) law to be used for calibration of a strain based fracture criterion, it
was fitted to the data to model the strain hardening behavior in should be able to accurately represent the strain fields under dif-
dependence of strain rate and Vickers hardness. Ten Kortenaar et al. ferent loading conditions up to fracture initiation, which for the
(2013) quenched 22MnB5 samples to fully martensitic and fully uni-axial case is far beyond the point of uniform elongation. For
bainitic microstructures in order to determine the fracture strains this purpose, full field strain measurements are performed using
under different stress triaxialities using tensile specimens with dif- a 3D digital image correlation (DIC) system. These measurements
ferent notch geometries. Their piecewise linear hardening model are used as input for the optimization routine that optimizes the
was calibrated using flow stress curves from uni-axial tensile tests strain hardening parameters in the post-necking regime. The uni-
up to diffuse necking, after which extrapolation was applied based axial tensile test is not suited for this purpose, because localization
on the apparent hardening rate at this point. When using the cali- may occur anywhere in the relatively long gauge section and often
brated model to simulate the notched tensile tests, it was found that initiates outside the DIC measuring window. Therefore, strain fields
the stresses were overestimated and that the extrapolation of post- of notched tensile tests are used, which have a very well defined
uniform hardening behavior is critical to the accuracy of the model. strain distribution. For calibration up to even higher strains than
Östlund et al. (2014) calibrated the localization and failure behav- obtained from the notched tensile tests, an equibiaxial tension test
ior of three grades of 22MnB5 with yield strengths of 400, 550 and is used.
800 MPa. They used a digital image correlation system to perform
full field measurements of tensile tests and calibrated piecewise
linear flow and localization curves with these data. The full field 2. Material description
measurements allowed for evaluation of mechanical properties at
different analysis lengths, providing parameters for a model which The material used for this study is 22MnB5. Known by the
accounts for shell element size. The objective of their work was commercial name Usibor® 1500 P, ArcelorMittal has provided the
not to model the phenomenon of localized necking, but rather its 22MnB5 steel grade with an aluminum-silicon coating that pro-
effect on load response and subsequent rupture, which is why no tects the metal against oxidation and decarburization during the
direct comparisons of measured and simulated strain fields are press hardening process (ArcelorMittal, 2012). In the as-delivered
shown. state 22MnB5 has a ferritic/pearlitic microstructure, an ultimate
In the present study, an inverse FEM optimization routine tensile strength of 600 MPa, and a uniform elongation of 0.22.
for determination of the strain hardening parameters is applied After quenching in cooled stamping tools, a fully martensitic
to five different hardness grades of quench-hardenable boron microstructure is obtained resulting in an ultimate tensile strength
steel 22MnB5. In the proposed model, the strain hardening is of 1500 MPa and a uniform elongation of 0.06. The sheets used in
represented by a nonlinear combination of the Swift hardening this work have a thickness of 1.5 mm; the chemical composition is
law and a modified version of the Voce law. The non-quadratic given in Table 1.

Table 1
Chemical composition of the 22MnB5 used in this work (wt.%) (ArcelorMittal, 2011).

C Mn P S Si Al Ti B

0.2–0.25 1.1–1.4 ≤0.025 ≤0.008 0.15–0.35 ≥0.015 0.02–0.05 0.002–0.005


T.K. Eller et al. / Journal of Materials Processing Technology 228 (2016) 43–58 45

Fig. 2. Optical micrographs of the five material hardness grades of 22MnB5.

3. Experimental work The press, which applies a holding pressure of 12.5 MPa, is solely
used to ensure good contact between tool surface and sheet, no
3.1. Material preparation plastic deformation is applied. Sheet temperatures during the hold-
ing phase are measured using surface probes.
In order to create material samples with different hardnesses, The first set of sheets is transferred to a stamping tool that is
the as-delivered sheets are first fully austenitized in a furnace at cooled to a constant temperature of 25 ◦ C. After a holding time of
950 ◦ C for approximately 6 min, after which they are subjected to 15 s, the sheets are removed and left to cool down to room tem-
carefully controlled cooling processes in cooled and heated stamp- perature in air. A measured cooling curve is superimposed on the
ing tools. After opening the furnace and taking the sheets out, CCT diagram of 22MnB5 in Fig. 1 (‘Cooled tool’), from which it can
transfer from furnace to tool takes on average 3.8 s, after which be seen that the achieved cooling rates clearly exceed the critical
another 5.8 s pass until full closure of the tool. During the total cooling rate for martensite formation. At the martensite start tem-
transfer time of 9.6 s, the sheets cool down to approximately 700 ◦ C. perature of approximately 400 ◦ C (Åkerström et al., 2005), a small
46 T.K. Eller et al. / Journal of Materials Processing Technology 228 (2016) 43–58

Fig. 3. SEM micrographs of material hardness grades 3, 4 and 5.

change in gradient can be observed in the cooling curve, which is are presented in Table 2. The sheets that were quenched in the
caused by the latent heat released during the martensite transfor- water-cooled stamping tool have a hardness of 497 HV0.1, which
mation. corresponds to the manufacturer specifications of fully hardened
For the second, third and fourth sets of sheets, a heated tool 22MnB5. The heated tool sheets have intermediate hardnesses of
is used that is held at constant temperatures of 350 ◦ C, 425 ◦ C and 404 HV0.1, 320 HV0.1 and 252 HV0.1, respectively. The sheets that
500 ◦ C, respectively. For the 350 ◦ C and 425 ◦ C treatments, a holding cooled down in the opened furnace have similar hardness values
time of 30 s is used, for the 500 ◦ C treatment the holding time is 20 s. as the material in the as-delivered state (181 HV0.1 and 183 HV0.1,
Again, no plastic deformation is applied. Looking at the measured respectively). The found standard deviations are common for HV0.1
temperature curves in Fig. 1, it can be seen that the cooling rates measurements and indicate that material properties are relatively
increase quickly after tool closure at approximately 10 s. The latent constant throughout the sheets.
heat of the bainitic transformation causes the sheets to approach Because of the relatively unstable conditions during the cooling
a constant temperature at approximately 50 ◦ C above the respec- process in the open furnace and the accompanying warping of the
tive tool temperatures and even leads to a slight reheating. The sheets, the material in the as-delivered state will be used for further
sheets are removed and left to cool down to room temperature mechanical characterization and will be denoted as hardness grade
in air before they have reached the respective tool temperatures. 1 in the remainder of this work. This is analogous to the approach
These three material grades are expected to have a mainly bainitic used in previous work of the authors (Eller et al., 2014). The sheets
microstructure. from the heated and cooled tools are denoted hardness grade 2–5,
The fifth set of sheets is left inside the furnace to cool down, with see Table 2.
the furnace turned off and the door slightly opened. The resulting
cooling rate of approximately 2 ◦ C/s is significantly lower than the
cooling rate of normal air cooling (approximately 19 ◦ C/s at 750 ◦ C, 3.3. Metallographic analysis
see Fig. 1).
In Fig. 2, the optical micrographs of the five material hardness
grades are presented. After several grinding and polishing steps,
3.2. Hardness measurements hardness grade 1 was etched with a 2% nital solution and hardness
grades 2–5 with a solution of 2 g picric acid and 1 ml HNO3 in 200 ml
Hardness measurements are carried out on a Duramin micro- ethanol. Of each hardness grade, several micrographs are taken to
hardness tester. A Vickers pyramid is used in combination with a ensure that the graphs presented here are representative for the
0.1 kg load and 15 s of loading time. All specimens used for hard- material set.
ness measurements are first mounted in epoxy resin, after which The micrograph of hardness grade 1 shows a fully fer-
they are subjected to several grinding and polishing steps. A total ritic/pearlitic microstructure. The pearlite was found to be located
of 50 measurements are taken on the cross section of each grade, both at the grain boundaries and in the form of granular pearlite
the average of these measurements will be considered as represen- in the ferrite matrix. Hardness grade 2 has an upper bainitic
tative hardness value. The results of the hardness measurements microstructure, which can be recognized by the characteristic
T.K. Eller et al. / Journal of Materials Processing Technology 228 (2016) 43–58 47

Table 2
Overview of the five material hardness grades used in this work.

Thermal history Average hardness Standard deviation Average tensile Microstructure


[HV0.1] [HV0.1] strength [MPa]

Open furnace 181 7.9


Hardness grade 1 As-delivered 183 8.2 587 Granular pearlite, ferrite
Hardness grade 2 Heated tool (500 ◦ C, 20 s) 252 7.0 742 Upper bainite, pearlite traces
Hardness grade 3 Heated tool (425 ◦ C, 30 s) 320 8.0 956 Lower bainite
Hardness grade 4 Heated tool (350 ◦ C, 30 s) 404 7.0 1231 Lower bainite
Hardness grade 5 Cooled tool (25 ◦ C, 15 s) 497 11.3 1504 Martensite

1.8
0° to rolling direction
45° to rolling direction
1.6
90° to rolling direction
Hardness grade 5

1.4
Fig. 4. Geometry of the uni-axial tensile test specimen (dimensions in mm).
Hardness grade 4

True stress [GPa]


1.2
carbides that are dispersed throughout the ferrite matrix and
Hardness grade 3
oriented along a common direction (Bhadeshia, 2001). The opti- 1
cal micrographs of hardness grades 3 and 4 show very similar Hardness grade 2
microstructures that may both be described as lower bainite, char- 0.8
Hardness grade 1
acterized by the typical plate-like morphology. The bainite in
hardness grades 2, 3 and 4 has been formed almost isothermally 0.6
during the holding phase in the heated tool (see Fig. 1), resulting
in clearly distinguishable upper and lower bainitic microstructures 0.4
(Naderi, 2007). For hardness grade 4 a tool temperature of 350 ◦ C
was used, which might result in some traces of martensite when 0.2
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14
considering the martensite start temperature of 400 ◦ C. However, True plastic strain [−]
during a SEM-analysis of the samples of both hardness grades 3
and 4, see Fig. 3, no martensite was found. The optical micro- Fig. 5. True stress–true plastic strain curves from the uni-axial tensile tests.
graph of hardness grade 5 reveals a typical needle-like martensitic
microstructure without clearly recognizable grain boundaries. A
SEM micrograph of hardness grade 5 is shown in Fig. 3.

3.4. Mechanical testing

3.4.1. Uni-axial tensile tests Fig. 6. Geometry of the notched tensile test specimen (dimensions in mm).
Standard uni-axial tensile tests according to DIN 50125 are per-
formed with specimens from all five hardness grades extracted
at 0◦ , 45◦ and 90◦ relative to the rolling direction. The specimen standard tensile tests described in Section 3.4.1. Fig. 7 presents
geometry is shown in Fig. 4. The tests are carried out on a Zwick the force–displacement curves of all five hardness grades. It can
Roell universal material testing machine which is able to apply be seen that there is a slight reduction in displacement to fracture
a maximum force of 250 kN (force measurement inaccuracy of for hardness grade 4 in comparison to hardness grade 5, which is
0.5% starting from 0.5 kN). The extensions are measured using a not what would be expected from the typical strength–ductility
macro extensometer, which is a high accuracy (1 ␮m) extensome-
ter operating on the contact principle, and a measuring length of 30
L0 = 50 mm. A small preforce of 60 N is applied to minimize the Hardness grade 5
Hardness grade 4
effects of specimen curvature and initial grip misalignments on the Hardness grade 3
extensometer output. All tensile tests are performed at a quasi- 25 Hardness grade 2
static strain rate of 0.002 s−1 . Hardness grade 1

The measured force–displacement curves of the uni-axial ten-


20
sile tests have been converted to true stress–true plastic strain
curves, see Fig. 5. Note that this conversion is only valid up to the
Force [kN]

point of maximum uniform elongation, which is the strain at which 15


diffuse necking begins. For hardness grades 1 and 5, the curves
could be evaluated up to strains of 0.135 and 0.039, respectively.
A DIC system was used to measure the axial and width strains 10
during the experiments, from which, under the assumption of plas-
tic incompressibility, the thickness strains and the r-values can be
5
calculated. The resulting r-values and the yield stresses, extracted
from Fig. 5, are summarized in Table 3.
0
3.4.2. Notched tensile tests 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
Displacement [mm]
The geometry of the notched tensile specimen is shown in
Fig. 6; testing equipment and procedures are the same as for the Fig. 7. Force–displacement results from the notched tensile tests.
48 T.K. Eller et al. / Journal of Materials Processing Technology 228 (2016) 43–58

Table 3
R-values and yield stresses of the 5 hardness grades measured from uni-axial tensile tests, mean values and standard deviations (SD) from 5 repeat experiments.

Hardness grade r0 r45 r90  y,0 [MPa]  y,45 [MPa]  y,90 [MPa]

Mean 0.980 0.890 0.749 439 427 400


1
SD 0.034 0.011 0.006 0.9 1.4 2.0

Mean 0.763 0.934 0.766 537 517 538


2
SD 0.018 0.024 0.006 6.8 2.8 3.6

Mean 0.834 0.906 0.840 722 725 712


3
SD 0.018 0.005 0.025 9.0 4.2 7.1

Mean 0.826 0.818 0.820 966 965 971


4
SD 0.013 0.031 0.032 5.8 4.1 5.5

Mean 0.839 0.834 0.785 964 990 953


5
SD 0.038 0.035 0.013 17.3 9.6 12.6

Fig. 8. Total strains εxx of the notched tensile tests of hardness grade 5 measured Fig. 9. Total strains εxx of the notched tensile tests of all hardness grades mea-
along the longitudinal axis of symmetry at displacements of 0.5 mm and 0.7 mm. sured along the longitudinal axis of symmetry at approximately 85% of the fracture
displacement.

banana shape for metals. An explanation for this apparent drop of to exhibit stronger strain localization than grade 5, resulting in a
ductility is given in Section 5. Grade 4 also seems to exhibit more smaller fracture displacement at a higher fracture strain.
experimental scatter than the other grades.
In the notch area, full field strain measurements were performed 3.4.3. Equibiaxial tension tests
using a 3D DIC system. The DIC system, equipped with two cam- An Erichsen-like cupping test with punch and die diameters
eras with each a resolution of 2448 × 2050 pixels, was connected of 60 mm and 90 mm, respectively, is used to obtain strain field
to the material testing machine and able to receive both force and measurements up to even higher strains than the ones obtained
displacement signals, such that every measured strain field can be from the notched tensile tests in Section 3.4.2. An overview of the
linked to a point on the corresponding force–displacement curve. test setup is shown in Fig. 10, square 127 mm × 127 mm specimens
Fig. 8 shows the measured surface strains in x-direction (εxx ) along are used. The experiments are carried out with a custom designed
the longitudinal axis of symmetry of the specimens of hardness specimen clamp and punch which are fitted in a standard material
grade 5 at displacements of 0.5 mm and 0.7 mm. At a displace- testing machine. Quasi-static testing conditions are created with a
ment of 0.5 mm, which is approximately at the force maximum, testing velocity of 0.1 mm s−1 . To minimize friction forces between
a bell-shaped strain distribution is found with maximum strains of punch and specimen, a tribological system is applied, consisting of
0.05. At 0.7 mm displacement, two strain peaks are found on the two layers of 0.17 mm thick PTFE film and lubricant. The resulting
specimen surface, corresponding to the two tips of an X-shaped force–displacement curves are presented in Fig. 11. Displacements
through-thickness necking zone, see the illustration in Fig. 8. are measured locally with a 2D DIC system to avoid influences from
Fig. 9 shows the measured strains of all five hardness grades at elastic deformations of the clamping tool and the material test-
approximately 85% of the fracture displacement during through- ing machine. The displacement is defined as the relative vertical
thickness necking; the exact displacements are listed in the legend displacement between punch and specimen clamp, see Fig. 10.
on the left. In the strain fields of grades 3, 4 and 5, two strain A second DIC system is used to obtain full field 3D strain field
peaks are found, whereas grades 1 and 2 show a bell-shaped strain measurements of the top surface of the specimens, see Fig. 12.
distribution. It should be noted that due to the smaller fracture Fig. 13 shows the measured strains of all hardness grades extracted
displacement of grade 4 relative to grade 5, the strain field is also along the x-axis at approximately 90% of the fracture displacement.
taken at a smaller displacement (0.6 mm for grade 4 and 0.7 mm The strain fields of hardness grades 3, 4 and 5 are taken at similar
for grade 5). Remarkably, the maximum strains measured in the punch displacements (19, 18 and 19 mm, respectively) and show
necking area of grade 4 at 0.6 mm displacement are higher than similar strain distributions along the x-axis. Although the strain
the strains of grade 5 at 0.7 mm displacement. Grade 4 thus seems field of hardness grade 4 is taken at a smaller displacement than
T.K. Eller et al. / Journal of Materials Processing Technology 228 (2016) 43–58 49

Fig. 10. Test setup of the equibiaxial tension test.

Fig. 12. 3D-strain field measurement of the equibiaxial tension test of hardness
grade 5 at a punch displacement of 18.8 mm, left camera view (top) and measured
3D strain field (bottom).

Fig. 11. Force–displacement results from the equibiaxial tension tests.

Fig. 13. Total equivalent Von Mises strains ε̄eq of the equibiaxial tension tests of all
hardness grades at approximately 90% of the fracture displacement.

the strain field of grade 5, the average maximum strain of grade 4 is 4. Constitutive modeling
slightly higher (0.57 for grade 4 and 0.55 for grade 5). This confirms
the findings of Section 3.4.2, where it was concluded that grade 4.1. Model assumptions
4 exhibits stronger strain localization than grade 5, resulting in a
lower fracture displacement at a higher fracture strain. The strain In the early design phase of a car, the rolling direction is gen-
fields of hardness grades 1 and 2 are taken at significantly larger erally not known. For full vehicle crash simulations, it is thus
punch displacements and show much wider strain distributions, preferred to use isotropic constitutive models. As the experimental
which is an indication for a higher strain hardening rate. Remark- results presented in Section 3.4.1 revealed that all five considered
ably, the measured strain fields do not feature a dip in strain at the hardness grades show only minor anisotropy of the stress–strain
punch center, which is normally seen in punch tests subjected to response, and that all r-values are in the range 0.75 < r < 1, an
friction. This was found to be caused by a slight eccentricity of the isotropic model will be used. The optimization routine for determi-
punch: 1 mm in x-direction and 2.5 mm in y-direction. To account nation of the strain hardening parameters presented in this section
for this eccentricity, the coordinate system for strain evaluation thus aims at finding the best possible results with an isotropic
in Fig. 12 has been shifted with respect to the die and placed in model. The isotropic assumption will be validated by comparing
the punch center. The simulations in Sections 4 and 5 will also be simulated and measured full strain fields of the notched tensile
performed with an eccentric punch. and equibiaxial tension tests.
50 T.K. Eller et al. / Journal of Materials Processing Technology 228 (2016) 43–58

Fig. 14. Experimental true stress–strain curves and fitted lower (modified Voce)
and upper (Swift) bounds for small (a) and large (b) strains (hardness grade 5). Fig. 15. Force–displacement curves and strain fields of the notched tensile test of
hardness grade 5: experimental data, optimization steps, lower (modified Voce) and
upper bounds (Swift). (For interpretation of the references to color in the text, the
First, the Yld2000-2D yield function (Barlat et al., 2003) with an reader is referred to the web version of the article.)

extension for general three-dimensional stress states by Dunand


et al. (2012) will be used. The eight anisotropy coefficients ˛k are
set to 1 to obtain an isotropic model. Material parameter a is set to 6, are fitted to the experimental data from the uni-axial tensile tests,
which is a common assumption for BCC materials. The optimization see Fig. 14a and b. Instead of using the classical Voce saturation law,
results obtained with the Yld2000-2D model will then be compared a modified Voce function with an asymptotic hardening rate ˛ for
to results obtained with the Von Mises yield criterion. large strains, similar to the one introduced by Tome et al. (1984)
is used. When Eq. (2) is fitted to experimental data without using
4.2. Lower and upper bounds the additional linear term ˛ε̄p , the strain hardening rate will vanish
at strains not far outside the experimental range. It has, however,
Average ( − ε̄p )-curves are extracted from the measured strain often been reported that strain hardening persists up to very high
hardening curves in the 0◦ -direction in Fig. 5; the results of hard- strains (e.g. Sevillano et al., 1980; Van Liempt, 1994). With this in
ness grade 5 are plotted in Fig. 14a. Because the conversion from mind, and to ensure numerical stability of the material model, ˛ is
engineering stresses and strains to true stresses and true plastic set to 0.05 GPa.
strains is only valid up to the strain at which necking begins, exper- From Fig. 14a and b, it can be seen that both the modified
imental data only reaches up to a strain of approximately 0.035. Voce and Swift law provide acceptable fits for small strains, but
The strain field measurements of the equibiaxial tension tests in behave very differently in the extrapolated area. Assuming that
Fig. 13, however, show local strains of almost 0.60, which means the true strain hardening curve lies somewhere in between these
that extrapolation of the ( − ε̄p )-curve is required. For this pur- lower (modified Voce) and upper (Swift) bounds, a strain hardening
pose, both the classical Swift (1952) power law: model is introduced that can capture both extreme cases as well as
intermediate kinds of strain hardening. Similar to the approaches
S (ε̄p ) = ks (ε̄p + ε0,s )ns (1) used by Sung et al. (2010) and Roth and Mohr (2014), the modified
Voce and Swift hardening laws are combined:
and a modified Voce (1955) law:

V (ε̄p ) = kv + Qv (1 − e−ˇv ε̄p ) + ˛ε̄p (2) y (ε̄p ) = (1 − )S (ε̄p ) + V (ε̄p ) (3)
T.K. Eller et al. / Journal of Materials Processing Technology 228 (2016) 43–58 51

0.16 where initial represents the Voce-influence for small strains, final
Mesh size 0.5 mm
Mesh size 0.2 mm represents the Voce-influence for large strains and shape is a shape
0.14
Mesh size 0.1 mm parameter. The parameters of the Swift ( S ) and modified Voce ( V )
Mesh size 0.05 mm
laws have been determined from fitting to uni-axial tensile test
[−]

0.12
data; the remaining model parameters (initial , final , shape ) will be
xx
Total strain in x−direction ε

0.1 determined from inverse FEM optimization. In some cases, when


initial is close to 0 and final close to 1 (Swift-dominated at small
0.08 strains and Voce-dominated at large strains), the overall strain
hardening curve y (ε̄p ) (Eq. (3)) might have a negative gradient.
0.06 To avoid this, and to ensure continuous differentiability of y (ε̄p ),
the gradient of the overall strain hardening curve should always be
0.04
at least equal to the smallest gradient of its two components  S and
V:
 
0.02

dy dS dV


0 ≥ min , (5)
0 0.5 1 1.5 2 2.5 3 dε̄p dε̄p dε̄p
x−coordinate [mm]

Fig. 16. Predicted strains of the notched tensile test of hardness grade 5 at a dis-
placement of 0.7 mm for different mesh sizes.
4.3. Parameter optimization

Instead of using a linear mixing approach with constant  as To identify the three free model parameters of Eq. (4) in
proposed by Sung et al. and Roth and Mohr, a new exponential the frame of large plastic deformations, an inverse FEM opti-
mixing law is introduced in which mixing parameter  becomes a mization scheme is set up that takes into account the measured
function of the equivalent plastic strain (ε̄p ): force–displacement curves of the uni-axial and notched tensile
tests and the strain fields of the notched tensile tests and equibiax-
 = (ε̄p ) = final + (initial − final )e−shape ε̄p (4) ial tension tests. The force–displacement curves of the equibiaxial

Fig. 17. Comparison of measured and simulated strain fields of hardness grade 5: axial and lateral strains of the notched tensile test at 0.70 mm displacement (a) and
equivalent strains of the equibiaxial tension test at 19 mm displacement (b).
52 T.K. Eller et al. / Journal of Materials Processing Technology 228 (2016) 43–58

Fig. 19. Comparison of simulation results from full optimization routine and three
simplified variants: strain fields of notched tensile test (a) and equibiaxial tension
test (b).
Fig. 18. Comparison of simulation results from full optimization routine and three
simplified variants: force–displacement curves of uni-axial (a) and notched tensile
tests (b). function containing the force–displacement curves of the notched
tensile test YF,notched is defined likewise. The strain field cost func-
tension tests are not included in the optimization, because it was tion Yε contains the strain fields of the notched tensile test and the
found that they are relatively insensitive to changes in the strain equibiaxial tension test:
hardening curve. Similar to the approach used by Haddadi and
Belhabib (2012), the overall cost function Y to be minimized is Yε (X) = ˇε Yε,notched (X) + (1 − ˇε )Yε,biax (X) (9)
defined by a weighted least-squares deviation between simulated
and measured forces and strains:
Y(X) = ˇYF (X) + (1 − ˇ)Yε (X) (6)
in which Y(X), YF (X) and Yε (X) are the overall, the force and
the strain field cost functions, respectively, and X = (initial , final ,
shape ) is the hardening parameter vector. Weighting factor ˇ can
be used to adjust the influence of the force–displacement curves
and strain fields on the cost function, and is set to 0.5. YF contains
the force–displacement curves of both the uni-axial and notched
tensile tests:
YF (X) = ˇF YF,uni (X) + (1 − ˇF )YF,notched (X) (7)
in which ˇF is a weighting factor which is set to 0.5 and YF,uni is
defined as:
 2
1 
nF
Fsim (di , X) − Fexp (di )
YF,uni (X) = nF (8)
nF 1
F (d )
i=1 nF j=1 exp j

where nF is the number of sample points on the force–displacement Fig. 20. Predicted strain fields of the equibiaxial tension test of hardness grade 1 for
curve and F(di ) is the force at sample displacement di . The cost different punch diameters, simulated with linear mixing approach ( = constant).
T.K. Eller et al. / Journal of Materials Processing Technology 228 (2016) 43–58 53

tension test, which is taken at approximately 70% and 90% of the


fracture displacement. On all strain fields and force–displacement
curves, 40 sample points are defined at which the devia-
tion between experimental and simulated value is evaluated
(nF = nmp = 40).
For the optimization procedure, Eqs. (1)–(5) are implemented in
the modular material model (MMM) framework (Greve, 2012) for
the explicit FEM code VPS (Virtual Performance Solution). As stated
in Section 4.1, the Yld2000-2D yield function with an extension
for general three-dimensional stress states is used with ˛k = 1 and
material parameter a = 6 for BCC materials. To ensure an accurate
representation of the strain fields, under-integrated 8-node solid
elements with one integration point are used. A mesh size study
in previous work of the authors showed that the predicted strain
field of the notched tensile test is very sensitive to mesh size. For
the optimization routine in this work, a mesh size of 0.1 mm will
be used, which was found to provide a converged solution (Eller
et al., 2014). To validate this for the current material model, an
additional mesh size study with the optimized parameters of hard-
ness grade 5 is presented at the end of this section. The strain field
prediction of the equibiaxial tension test was found to be virtu-
ally mesh size independent due to the large punch diameter and
the resulting small strain gradients. A mesh size of 0.5 mm will
be used for this test, which can be considered as converged (Eller
et al., 2014). The artificial stiffness method given in Flanagan and
Belytschko (1981) is used to control the hourglass modes of the ele-
ments, using the default hourglass parameter of 0.01 (VPS, 2013).
Because of the tribological system that was applied in the equibiax-
ial tension experiments, see Section 3.4.3, friction between punch
and sheet is neglected in the simulations. A trust-region-reflective
algorithm is used to minimize the cost function with [initial , final ,
shape ] = [0.5, 0.5, 1] as initial values.
Figs. 15a and b show the experimental force–displacement
results and the measured strain fields of the notched tensile tests
of hardness grade 5, together with several simulation results. The
used mesh size of 0.1 mm ensures that the geometry of the necking
area and the strain localization can be accurately represented by
the model. The blue and green curves are simulation results with
the Modified Voce and Swift lower and upper bounds, as they have
Fig. 21. Optimized true stress–strain curves of all five hardness grades including
fitted lower and upper bounds (a) and simulation results of the uni-axial tensile
been presented in Figs. 14a and b. The Swift law clearly overesti-
tests (b). mates the strain hardening rate, leading to overestimation of the
forces. Furthermore, there is virtually no strain localization, lead-
in which ˇε is a weighting factor which is set to 0.5 and Yε,notched is ing to an underestimation of the strains of almost 50% at d = 0.7 mm.
defined as: The modified Voce law underestimates the strain hardening rate,
and thus also the forces, and clearly overestimates the strains. It
nimg should be emphasized that both laws were fitted to the available
1  1 
nmp
experimental data from the tensile test, which is still a commonly
Yε,notched (X) =
nimg nmp used method in many applications.
i=1 j=1
The red curves are results of the optimization procedure, visu-
 2 alizing the convergence behavior of the algorithm. Looking at
εxx,sim (di , Pj , X) − εxx,exp (di , Pj ) the simulation results in Fig. 15a, it can be seen that up to the
1
nmp (10)
nmp ε (d , Pk )
k=1 xx,exp i
force-maximum the exponential mixing law does not have much
influence on the simulated force–displacement curve, which is con-
where nimg and nmp are the numbers of strain fields and measur-
 firmed when looking at the simulated strain field at d = 0.5 mm,
ing points in each strain field, respectively, and εxx di , Pj is the see Fig. 15b. This can be explained when looking at Figs. 14a
strain at global displacement di on position Pj . The cost function and b: up to a strain of 0.04, the Swift and modified Voce laws
containing the strain fields of the equibiaxial tension test Yε,biax is are intentionally very similar, meaning that mixing them will not
defined likewise. The displacement di can be related to the corre- change much in the final result. At higher strains, e.g. after ini-
sponding strain field because of the data connection between DIC tiation of through-thickness necking (d = 0.7 mm in Fig. 15b), the
system and material testing machine. The position on the spec- lower and upper bounds diverge, leading to an increased influence
imen surface at which the strain is evaluated is defined by Pj , of the mixing law. After a total of 16 iterations the algorithm has
which represents the initial coordinates vector of the jth measur- converged for a predefined threshold. When comparing the sim-
ing point. The strain field of the notched tensile test is taken at two ulation results with initial parameters to the simulation results
instances of time during the experiment: at the force-maximum with optimized parameters (‘Step 16’), it is clear that both the pre-
and during through-thickness necking (see Fig. 8), such that nimg diction of force–displacement curves and strain fields is greatly
equals 2. The same is done for the strain field of the equibiaxial improved.
54 T.K. Eller et al. / Journal of Materials Processing Technology 228 (2016) 43–58

To check whether the used mesh size of 0.1 mm provides a con- D/t = 40), a small dip is found in the strain field at the punch cen-
verged solution, Fig. 16 shows a comparison of the predicted strains ter. With an even smaller punch diameter (D = 20 mm, D/t = 13.3),
obtained from simulations with mesh sizes of 0.05 mm, 0.1 mm, this dip is even more pronounced, and with an increased value of
0.2 mm and 0.5 mm. It can be seen that the 0.2 mm and 0.5 mm D/t it disappears. The dip in the strain field is thus caused by bend-
meshes are not able to represent the details found in the measured ing effects when the punch radius is too small. In combination with
strain fields from Fig. 15b. From the simulations with 0.05 mm and an underestimation of the strain hardening rate at high strains, this
0.1 mm meshes, it can be concluded that the 0.1 mm mesh pro- causes a through-thickness necking zone to emerge in a ring around
vides a virtually converged solution. Fig. 17a shows a comparison the punch center. From Fig. 19b, it can be concluded that a suitable
of the measured and simulated full strain fields of the notched set of hardening parameters (‘Full optimization’) can prevent this.
tensile test. It can be seen that the model is able to provide an accu- When leaving the strain fields out of the cost function, the
rate representation of both the axial and lateral strains, including prediction of the force–displacement curves is slightly improved
the characteristic double strain peak that is found during through- compared to the full optimization variant (except for the notched
thickness necking. The strain field of the equibiaxal tension test, tensile test of hardness grade 1). The strains in the specimen center
see Fig. 17b, is also represented with excellent accuracy. of the notched tensile test are overestimated, whereas the strains
of the equibiaxial tension test are too low. Just as with the linear
4.4. Simplifications of the plasticity model and the optimization mixing approach, the optimized set of parameters, not taking into
procedure account the strain fields, is not able to represent the double peak in
the strain field of the notched tensile test of hardness grade 5.
The proposed plasticity model and parameter identification When using a Von Mises yield function (variant 3, ‘Von Mises’),
procedure are quite comprehensive and require advanced exper- necking of the uni-axial tensile test initiates too early for both hard-
imental techniques for determination of the strain fields. The ness grades. The force–displacement curves of the notched tensile
computational cost depends on the number of iterations (niter ) test are slightly overestimated, both at Fmax and after necking ini-
and can best be expressed in the total number of FE-simulations tiation for hardness grade 1. When the weighting factor of the
required (nsim ): force–displacement curve of the uni-axial tensile test is increased,
such that necking initiates later, the maximum force of the notched
nsim = niter · (ntest · (npar + 1)) tensile test is overestimated even further. This effect was found
for all hardness grades: with the Von Mises yield function there
in which ntest is the number of different tests involved (e.g. uni- is a competition between the uni-axial and notched tensile tests,
axial and notched tensile test) and npar is the number of free whereas the isotropic Yld2000-2D yield function is able to repre-
model parameters. For each iteration, all tests have to be simu- sent both tests with good accuracy. This can be explained when
lated (npar + 1) times: calculation of the numerical derivatives with considering the stress states of both samples: the standard tensile
respect to the free model parameters and evaluation of the cost test is in a state of uni-axial tension, whereas the notch area of the
function with the new set of parameters. To check whether it is notched tensile test is in a state of plane strain tension ( 2 = 1/2 1 ).
possible to reduce the complexity of the model and of the optimiza- Apparently, the Von Mises yield function overestimates the yield
tion procedure without losing accuracy, three simplified variants stress under plane strain tension for 22MnB5. The Yld2000-2D
are applied on hardness grades 1 and 5: yield function, which lies in between Von Mises and Tresca, pro-
vides better results. Furthermore, the Von Mises yield function
1. 1-Parameter optimization with linear mix between modified underestimates the maximum strains in the specimen center of
Voce and Swift ( = constant instead of  = (ε̄p )). the equibiaxial tension test of hardness grade 5.
2. Optimization without strain fields in the cost function (ˇ = 1 From the comparison in Figs. 18 and 19, it can be concluded that
instead of ˇ = 0.5 in Eq. (6)). the simplified variants do not provide adequate results. Further-
3. Simple Von Mises yield function instead of extended Yld2000- more, it can be concluded that the availability of experimental data
2D. at different loading conditions, including accurate information of
the corresponding strain fields, is of most importance.
A summary of the results is presented in Figs. 18 and 19. Note
that to enhance the readability of the plots, separate x- and y-
axes have been introduced for the two hardness grades. It should 5. Application of the method to other material hardness
be emphasized that all simulation results have been obtained by grades
inverse FEM-optimization, they thus represent the best possible
results obtainable with the corresponding settings. To check the performance of the proposed parameter iden-
When comparing the full optimization procedure with the lin- tification procedure, it is applied to all five available hardness
ear mixing variant, it can be seen that a good prediction of the grades presented in Section 3. With hardnesses between 183 and
force–displacement curves of both the uni-axial and notched ten- 497 HV0.1, these grades cover the full range of possible hardness
sile tests is obtained. However, the characteristic double peak in values of 22MnB5, from a soft ferritic/pearlitic microstructure to
the strain field of the notched tensile test of hardness grade 5 can- fully hardened martensite. The results presented in this section
not be represented with this simplified approach, even when using have been obtained with the standard settings presented in Sec-
an increased weighting factor for the strain fields (see Eq. (6)). The tions 4.2 and 4.3: a 3-parameter optimization with exponential
prediction of the strain field of the equibiaxial tension test is also mixing law ( = (ε̄p )), strain fields included in the cost function
less accurate, especially in the specimen center. For hardness grade and the isotropic Yld2000-2D yield function.
1, the model now predicts a dip in strain at the punch center, which Fig. 21a shows the fitted lower and upper bounds together with
is not what would be expected when using a friction coefficient of 0. the optimized strain hardening curves, the corresponding param-
This dip was found to be related to the ratio between punch diam- eters are presented in Table 4. The value of final , which represents
eter D and sheet thickness t. Fig. 20 shows simulation results of the the Voce-influence on the gradient of the strain hardening curve at
equibiaxial tension test of hardness grade 1 with different values large strains, increases with material hardness. The softer material
of D/t. The linear mixing approach ( = constant) has been used for grades thus approach their upper bound at large strains, whereas
this study. For the setup used in this work (punch diameter 60 mm, the harder grades approach their lower bound, see Fig. 21a. The
T.K. Eller et al. / Journal of Materials Processing Technology 228 (2016) 43–58 55

Table 4
Optimized parameters of the exponential mixing law.

Hardness grade initial final shape

1 0.29 0.26 1.16


2 0.54 0.70 0.98
3 0.18 0.86 4.25
4 0.53 0.87 8.03
5 0.56 0.97 3.63

Fig. 23. Simulation results with optimized parameters of all five hardness grades:
force–displacement curves (a) and strain fields (b) of the equibiaxial tension tests.

is able to represent and indicate this behavior by the high value of


shape , leading to a strong curvature of the strain hardening curve
towards the lower bound at a relatively small strain, see Fig. 21a.
The strain hardening curves of grades 1 and 2 have a relatively
high gradient at high strains, causing the strain distributions of the
equibiaxial tension tests to be wider. Just as in the experiments, the
Fig. 22. Simulation results with optimized parameters of all five hardness grades: predicted local strains at the punch center of the equibiaxial ten-
force–displacement curves (a) and strain fields (b) of the notched tensile tests.
sion test of hardness grade 4 are higher than the strains of hardness
grade 5, even though the strain field of hardness grade 4 is taken at
value of shape is a measure for the curvature of the final strain hard- a smaller punch displacement (see also Fig. 13).
ening curve at small strains (ε̄p < 0.2): hardness grades 3, 4 and 5 The force–displacement curves of the uni-axial tensile tests
have a relatively high value and thus a higher curvature than grades revealed that hardness grade 4 has a lower ductility than hardness
1 and 2. The simulation results of the uni-axial tensile tests are pre- grade 5 when considering the global displacements to fracture. On
sented in Fig. 21b. It can be seen that the calibrated model predicts a local scale, however, strain field measurements showed that the
the experimental force–displacement curves with good accuracy. local strains just before fracture initiation of hardness grade 4 are
Even the early necking initiation of hardness grade 4 compared to higher than those of hardness grade 5. The results presented in this
hardness grade 5 is represented by the model. section show that the calibrated material model is able to represent
The predicted force–displacement curves and strain fields of this behavior. To further illustrate this effect, the Considère crite-
the notched tensile and equibiaxial tension tests are presented in rion (Considère, 1885) is applied to the optimized hardening curves
Figs. 22 and 23. A comparison of the measured and simulated full from Fig. 21a. According to the Considère criterion, necking initiates
strain fields can be found in Appendix A. Again, it is seen that the when the strain hardening rate equals the stress: d/dεp = . Fig. 24
model provides an excellent prediction of the experimental results shows plots of the calibrated strain hardening curves (εp ) and
for all hardness grades. In Section 3.4.2 it was found that hardness their corresponding derivatives d/dεp . According to the Considère
grade 4 exhibits stronger strain localization than hardness grade criterion, the strains at which these curves intersect are the strains
5, leading to higher strains at a smaller displacement. The model at which necking initiates, which is clearly at a smaller strain for
56 T.K. Eller et al. / Journal of Materials Processing Technology 228 (2016) 43–58

tensile tests reached up to 0.15 and of the equibiaxial tension tests


up to 0.55, thus providing a profound experimental basis for the
parameter identification routine.
From a comparison between the results obtained from the pro-
posed model and parameter identification procedure and results
obtained from three simplified variants, it can be concluded that
a flexible strain hardening equation is required to be able to accu-
rately represent the behavior of the full scope of possible 22MnB5
hardness grades. Simple curve fitting of classical hardening laws
such as Swift or Voce yields a poor prediction of strain fields and
force–displacement curves. With a linear combination between
Swift and Voce the results could be improved, but still it was not
possible to represent the measured strain fields accurately. Further-
more, it was found that strain fields have to be considered when
calibrating the strain hardening behavior of a material. The example
of hardness grade 5 showed that it is possible to obtain an excel-
lent representation of the force–displacement curve of the notched
Fig. 24. Visualization of the Considère criterion for all 5 hardness grades: plastic tensile test, with at the same time a poor prediction of the general
instability starts when d/dεp = . shape of the corresponding strain field. This could not only lead to
over or underestimation of local strains, but also to misinterpreta-
hardness grade 4 than for hardness grade 5. It can thus be concluded
tion of the stress state in the notch area. Finally, it was found that
that necking and strain localization occur earlier for hardness grade
when using the Von Mises yield function to model the constitutive
4, leading to higher local strains at the same global displacement
response of 22MnB5, there is a competition between the uni-axial
compared to grade 5.
and notched tensile tests: either necking of the uni-axial tensile
test initiates too early, or the forces of the notched tensile tests
6. Conclusions
are clearly overestimated. The isotropic version of the Yld2000-2D
yield function, in combination with the flexible strain hardening
An elasto-plastic constitutive model for five different hardness
model, is able to predict both tests with good accuracy.
grades of boron steel 22MnB5 has been proposed and calibrated
The proposed model was shown to provide excellent results for
that provides an accurate prediction of the strain fields up to the
all five available hardness grades of 22MnB5, ranging from a soft
point of fracture initiation. In the proposed model, a nonlinear
ferritic/pearlitic grade to fully hardened martensite. Besides a good
combination of the Swift hardening law and a modified Voce law
representation of the force–displacement curves, the model was
are used to represent the strain hardening behavior in combina-
able to accurately capture both the magnitude and the shape of
tion with an isotropic version of the Yld2000-2D yield function.
the measured strain fields, making it particularly suited for use in
Initial fits of the two classical strain hardening equations were con-
combination with strain-based fracture criteria.
structed based on experimental data from uni-axial tensile tests
up to the point of diffuse necking. For the hardest material grade,
which was found to have a fully martensitic microstructure and a
hardness of 497 HV0.1, this was at a strain of 0.035. The strain hard- Appendix A. Comparison of measured and simulated strain
ening response beyond the point of diffuse necking was determined fields for hardness grades 1, 2, 3 and 4
from a 3D strain field analysis of notched tensile and equibiax-
ial tension tests in combination with an inverse FE optimization In this Appendix, a comparison between the measured and sim-
routine. For the hardest material grade, strains of the notched ulated strain fields of the notched tensile and equibiaxial tension
tests of hardness grades 1, 2, 3 and 4 can be found, see Figs. A.1–A.4,
respectively.

Fig. A.1. Comparison of measured and simulated strain fields of hardness grade 1, notched tensile test at 1.70 mm displacement (a) and equibiaxial tension test at 27 mm
displacement (b).
T.K. Eller et al. / Journal of Materials Processing Technology 228 (2016) 43–58 57

Fig. A.2. Comparison of measured and simulated strain fields of hardness grade 2, notched tensile test at 1.05 mm displacement (a) and equibiaxial tension test at 24 mm
displacement (b).

Fig. A.3. Comparison of measured and simulated strain fields of hardness grade 3, notched tensile test at 0.75 mm displacement (a) and equibiaxial tension test at 19 mm
displacement (b).

Fig. A.4. Comparison of measured and simulated strain fields of hardness grade 4, notched tensile test at 0.60 mm displacement (a) and equibiaxial tension test at 18 mm
displacement (b).
58 T.K. Eller et al. / Journal of Materials Processing Technology 228 (2016) 43–58

References Hill, R., 1948. A theory of the yielding and plastic flow of anistropic metals. Proc. R.
Soc. A 193, 281–329.
Åkerström, P., 2006. Modelling and Simulation of Hot Stamping. LuleåUniversity of Karbasian, H., Tekkaya, A.E., 2010. A review on hot stamping. J. Mater. Process.
Technology, Luleå, Sweden (Ph.D. thesis). Technol. 210, 2103–2118.
Åkerström, P., Wikman, B., Oldenburg, M., 2005. Material parameter estimation for Maikranz-Valentin, M., Weidig, U., Schoof, U., Becker, H.H., Steinhoff, K., 2008. Com-
boron steel from simultaneous cooling and compression experiments. Model. ponents with optimised properties due to advanced thermo-mechanical process
Simul. Mater. Sci. Eng. 13, 1291–1308. strategies in hot sheet metal forming. Steel Res. Int. 79, 92–97.
ArcelorMittal, 2011. A54 – quenchable boron steels, product catalog. Mohr, D., Ebnoether, F., 2009. Plasticity and fracture of martensitic boron steel under
ArcelorMittal, 2012. Steels for hot stamping, product catalog. plane stress conditions. Int. J. Solids Struct. 46, 3535–3547.
Bardelcik, A., Worswick, M.J., Winkler, S., Wells, M.A., 2012. A strain rate sensi- Naderi, M., 2007. Hot Stamping of Ultra High Strength Steels. RWTH, Aachen,
tive constitutive model for quenched boron steel with tailored properties. Int. J. Germany (Ph.D. thesis).
Impact Eng. 50, 49–62. Östlund, R., Oldenburg, M., Häggblad, H.Å, Berglund, D., 2014. Evaluation of
Barlat, F., Brem, J., Yoon, J., Chung, K., Dick, R., Lege, D., Pourboghrat, F., Choi, S.H., localization and failure of boron alloyed steels with different microstructure
Chu, E., 2003. Plane stress yield function for aluminum alloy sheets – part 1: compositions. J. Mater. Process. Technol. 214, 592–598.
theory. Int. J. Plast. 19, 1297–1319. Roth, C.C., Mohr, D., 2014. Effect of strain rate on ductile fracture initiation in
Bhadeshia, H.K.D.H., 2001. Bainite in Steels, 2nd ed. Institute of Materials. advanced high strength steel sheets: experiments and modeling. Int. J. Plast.
Considère, A., 1885. Memoire sur l’emploi du fer et de l’acier dans les constructions: 56, 19–44.
ii. Ann. Ponts Chaussées (ser. 6) 9, 574–605. Sevillano, J.G., van Houtte, P., Aernoudt, E., 1980. Large strain work hardening and
Dunand, M., Maertens, A.P., Luo, M., Mohr, D., 2012. Experiments and modeling of textures. Prog. Mater. Sci. 25, 69–134.
anisotropic aluminum extrusions under multi-axial loading – part I: plasticity. Sung, J.H., Kim, J.H., Wagoner, R., 2010. A plastic constitutive equation incorporating
Int. J. Plast. 36, 34–49. strain, strain-rate, and temperature. Int. J. Plast. 26, 1746–1771.
Eller, T.K., Greve, L., Andres, M.T., Medricky, M., Hatscher, A., Meinders, V.T., van den Swift, H., 1952. Plastic instability under plane stress. J. Mech. Phys. Solids 1,
Boogaard, A.H., 2014. Plasticity and fracture modeling of quench-hardenable 1–18.
boron steel with tailored properties. J. Mater. Process. Technol. 214, 1211–1227. Ten Kortenaar, L., Bardelcik, A., Worswick, M., Detwiler, D., 2013. The effects of stress
Flanagan, D.P., Belytschko, T., 1981. A uniform strain hexahedron and quadrilateral triaxiality on the failure response of boron steel quenched to a martensitic and
with orthogonal hourglass control. Int. J. Numer. Methods Eng. 17, 679–706. bainitic material condition. In: Proceedings of the 4th Hot Sheet Metal Forming
George, R., Bardelcik, A., Worswick, M.J., 2012. Hot forming of boron steels using of High-Performance Steel, pp. 565–572.
heated and cooled tooling for tailored properties. J. Mater. Process. Technol. Tome, C., Canova, G., Kocks, U., Christodoulou, N., Jonas, J., 1984. The relation between
212, 2386–2399. macroscopic and microscopic strain hardening in f.c.c. polycrystals. Acta Metall.
Greve, L., 2012. Modulare materialmodellierung für die simulation von 32, 1637–1653.
deformations- und bruchvorgängen. In: crashMAT 2012 - 6. Freiburg Van Liempt, P., 1994. Workhardening and substructural geometry of metals. J. Mater.
Workshop zum Werkstoff- und Strukturverhalten bei Crashvorgängen. Process. Technol. 45, 459–464.
Haddadi, H., Belhabib, S., 2012. Improving the characterization of a hardening law Voce, E., 1955. A practical strain-hardening function. Metallurgica 51, 219–226.
using digital image correlation over an enhanced heterogeneous tensile test. Int. VPS, 2013. Virtual Performance Solution 2013 – Solver Reference Manual. ESI Group.
J. Mech. Sci. 62, 47–56.

You might also like