Journal of Materials Processing Technology 228 (2016) 34–42

Contents lists available at ScienceDirect

Journal of Materials Processing Technology

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

Constitutive behaviour under hot stamping conditions

Michael Abspoel ∗ , Bas M. Neelis, Peter van Liempt
Tata Steel, The Netherlands

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

Article history: The modelling of the hardening behaviour at high temperatures and a range of strain rates is extensively
Received 10 January 2015 discussed. The hardening behaviour is characterized using tensile tests done in a Gleeble testing machine.
Received in revised form 30 March 2015 The specimen is heated by an electric current, soaked to get fully austenized, cooled down to its desired
Accepted 9 May 2015
testing temperature and then drawn to fracture with a given strain rate without a further drop of the
Available online 16 May 2015
temperature. One of the major challenges of this test is to achieve a uniform temperature distribution
over the sample, to ensure homogeneous austenization. A dedicated tensile sample geometry enables a
Keywords:
much more uniform temperature distribution than a regular sample geometry. Yielding and hardening
Hot stamping
Material properties
behaviour have been characterized with a Kocks–Mecking plot. These characterizations have been used
Hardening curves to fit parameters for a physically based hardening model that is applicable in a wide range of strain rates
FEA input and temperatures. The predicted strains in FEM simulations that use these hardening curves match well
with thickness measurements on hot stamped parts.
© 2015 Elsevier B.V. All rights reserved.

1. Introduction
Nowadays direct hot stamping is common practice in automo-
tive manufacturing to produce parts with complex shapes and high
strength. The hot stamping process starts with a heating step to
austenize the blanks in a furnace, typically at 900 ◦ C. When the
blank is released from the furnace, it is quickly transferred to a
press. During this transport the blank remains in the austenitic
phase by virtue of its slow transformation kinetics. In a quick
press stroke, the product gets its final shape, and is quenched to
a hardening structure. The final in-press transformation ensures
that no residual stresses are present in the product, and spring-
back is minimized. For FE analysis of the hot stamping process, an
accurate description of the material model is essential. Therefore,
the hardening behaviour as a function of temperature and strain
rate must be described for the austenitic phase. Åkerström (2004)
developed a method to determine the mechanical response (flow
stress) for the austenite, based on multiple overlapping continuous
cooling and compression tests in combination with inverse mod-
elling. Hein (2005) described a global approach for FE analysis of
hot stamping where all parameters needed for a good simulation
are described. Flow curves for the austenitic state between 650 and
900 ◦ C for different strain rates are recommended. It is also stated
∗ Corresponding author. Tel.: +31 251491735.
E-mail addresses: michael.abspoel@tatasteel.com (M. Abspoel),
bas.neelis@tatasteel.com (B.M. Neelis), peter.van-liempt@tatasteel.com
(P. van Liempt).
that at temperatures lower than 600 ◦ C, problems can occur due to
phase transformations. Merklein et al. (2006) described the effect
of strain rate on 22MnB5 tensile tests. Turetta (2008) showed hot
tensile tests on standard tensile specimens according to ISO 10130.
Hardening curves for different temperatures and strain rates were
measured. Lechler et al. (2008) used a phenomenological model for
the generation of hardening curves as a function of strain rate and
temperature. This model is a multiplicative model: it assumes that
the work hardening rate of the flow stress is a product of the strain
rate dependency and the temperature sensitivity.
An improved sample geometry and test set-up for use in a
Gleeble tensile testing machine is given to avoid the inhomo-
geneous heating as is observed in the standard sample. Also
a physical modelling approach for the dependence from strain
rate and temperature is proposed since a correct description
of strain-rate/temperature sensitivity is crucial for predicting
strain non-uniformities. The model recognizes two distinct strain-
rate/temperature effects on the flow stress, the first being due to
dislocation glide resistance which pertains to dislocation propa-
gation, the second due to dynamic recovery and which pertains to
dislocation multiplication (work hardening). They can be separated
experimentally by first validating the dislocation glide resistance
at the yield point, where work hardening is absent, followed by
a fit of the work hardening function to the post-yield part of the
hardening curve. The model is based on dislocation theory by the
principle of additive contributions of yield stress, glide resistance
and work hardening to the flow stress, described e.g. by Klepaczko
and Chiem (1986), who discuss the fundamentals of how to con-
struct constitutive relations on the basis that the total flow stress is

http://dx.doi.org/10.1016/j.jmatprotec.2015.05.007
0924-0136/© 2015 Elsevier B.V. All rights reserved.
 M. Abspoel et al. / Journal of Materials Processing Technology 228 (2016) 34–42 35

the sum of the effective stress (or glide resistance due to surmount- 1000
ing local obstacles by dislocations) and the internal stress (which Austenizing Quenching
pertains to dislocation multiplication). They show the validity of
this concept from rate jump experiments, and recognize that these

Temperature [°C]
contributions cause, respectively an instantaneous and a strain
dependent strain rate sensitivity. van Liempt et al. (2002) published
a flow stress model based on this principle of additive flow stress 750
contributions. Also Sarkar and Militzer (2009) propose a similar
Tensile test
approach of flow stress modelling at elevated temperatures. The
individual flow stress contributions are described in several papers.
Bailey and Hirsch (1960) experimentally validated the correlation
between dislocation density and flow stress that was first proposed
by Taylor (1934). Krabiell and Dahl (1981) described a function for 500
the dislocation glide resistance which is the stress to move mobile Time [s]
dislocations at the required velocity at a specific temperature.
The work hardening theory used is based on theories using Fig. 1. Temperature cycle.
dislocation density and dynamic recovery and annihilation or alter-
natively remobilization of dislocations. Kocks (1976) characterized
recovery as dislocations getting “annihilated or becomes ineffec-
tive in some other way at each potential recovery site”, where
Bergström (1969–1970), interpreted it as remobilization of stored
dislocations. The Bergström model was later developed further by
Vetter and van den Beukel (1977) by incorporating the effect of dis-
location density in the storage term, yielding the Bergström model
mathematically identical to the Kocks–Mecking theory. Kocks and
Mecking (2003) later revisited Kocks’ model, discussing further
implications of the theory.
With the parameters found for the equations, a strain rate and
temperature dependent hardening description can be made. The
parameters were derived from a limited dataset within the hot Fig. 2. Tensile sample with equal current density.
stamping regime. To verify the accuracy of the predicted curves,
they were compared with the measurements and simulations with
these curves were compared with thickness measurements on a cycle is shown in Fig. 1. During tensile testing, the temperature is
hot formed part. measured and controlled. Generally a slight heat increase is seen
due to the deformation of the sample. The test is performed in
vacuum to avoid decarburisation of the steel.
2. Experimental The temperature is measured with a thermocouple welded in
the centre of the top surface of the sample. The thermocouple is
2.1. Basic principle also an input for the temperature control loop. The strain is mea-
sured with either a retractable strain gauge or by the crosshead
The goal for this testing programme is to have accurate mea- displacement corrected for machine stiffness. An online strain mea-
surements, which requires tensile tests with a homogeneous surement with Aramis is also possible but not used in this case.
temperature along the deformation area of the sample, including
the shoulders, for which special sample shape and set-up have been
2.3. Tensile sample shape
designed. Another goal is to cover a wide range of temperatures
and strain rates with a limited amount of tests. Therefore the tests
Since the tensile sample is heated internally and the grips are
are parameterized with a model based on physical parameters. The
at room temperature, undesirable temperature gradients develop
use of a physical model increases the understanding of the test
along the gauge section. A special sample shape was designed to
itself and relations between parameters are investigated and can
ensure a homogeneous temperature distribution. The temperature
be combined to a predictive model.
homogeneity of this sample is ±10 ◦ C over the gauge length and a
large part of the shoulders. The sample shape and its dimensions is
2.2. Tensile test equipment and test set up shown in Fig. 2. The sample has a gauge length of 30 mm × 10 mm
which is the minimum allowable for a uniaxial tensile condition.
For the tensile tests a Gleeble testing machine at Delft University The four “legs” are connected with thick wires to create a shunt
of Technology is used. The Gleeble system was chosen for its excel- for the current as shown in Fig. 2. This ensures equal current den-
lent temperature control. According to the DSI datasheet Gleeble sity and consequently heating rate in the gauge and shoulder, so
3800, the Gleeble is able to create heat rates up to 10.000 ◦ C/s (for that the temperature gradient in the latter is minimized. Never-
welding simulations, depending on sample size), achieve steady theless there will be a local maximum in the current density when
state equilibrium temperatures within ±1 ◦ C and cooling rates: up the tensile sample is necking. This will locally increase the tem-
to 250 ◦ C/s (depending on material and sample size, quench media). perature just before fracture. Since this happens after incipient
The tested material is a regular 22MnB5 with a thickness of 1.5 mm. necking this is not important for the further evaluation of the mea-
The zinc coating was removed to avoid zinc deposition in the Glee- sured stress strain curve. Only the stresses and strains before diffuse
ble. The sample is heated by electrical current and is subjected to a necking are considered. In the centre of the sample a thermocou-
programmed cycle consisting of 60 s of heating to 900 ◦ C, followed ple is attached on the surface with a spot weld to measure the
by an industrial soaking time for 240 s, quenching to the desired temperature. The temperature distribution can be analyzed with
temperature, and tensile testing. An example of the cooling-testing standard video camera, because the light intensity correlates well
36 M. Abspoel et al. / Journal of Materials Processing Technology 228 (2016) 34–42
Fig. 3. Temperature distributions in the improved shape.
Fig. 4. Temperature and phase distribution in the standard shape.
Fig. 5. Double neck in ferrite shoulders.
to the temperature for this temperature range. Ranges from red
towards yellow/white correspond to austenite temperature range.
Black areas correspond to temperatures around 700 ◦ C or lower. In
Fig. 3, the improved sample and the standard sample are shown at
the end of the austenization cycle. At this stage, a homogeneous
temperature distribution is required. The standard sample how-
ever does not show a homogeneous temperature distribution over
the gauge length. The black areas in the standard sample are below
700 ◦ C, and are still in the ferrite phase (Fig. 4). The different areas
within the gauge length will each have different yield strengths
because the ferrite phase at this temperature can be softer than the
austenite as shown by Barnett and Jonas (1999). The ferrite will
deform while the austenite does not or hardly deforms until the
ferrite has reached the yielding condition of the austenite. Due to
this, the measured yield point and flow stress are those for fer-
rite rather than austenite. In the improved sample, all the material
between the shoulders is homogeneous austenitic. The measure-
ment of the austenitic yield point for the given temperature is much
more accurate when the improved sample shape is used.
An impression of what could happen with inhomogeneous heat-
ing is shown in Fig. 5 where a standard shaped sample with shorter
gauge length is necked in the ferrite shoulders.
Cooling/quenching is performed by blowing Ar gas on the
sample. Next to homogeneous heating and austenizing, also homo-
geneous cooling before performing the tensile test is essential. The
standard Gleeble quench nozzle system in combination with the
low initial temperature gradient of the austenization phase of the
test proved to be satisfactory in this respect.
3. Hot tensile fitting and modelling
3.1. Workflow and boundary conditions
For the input of the finite element code, a dataset of hardening
curves spanning a broad range of forming temperatures and strain
rates is necessary. To limit the amount of tests, and also find a solu-
tion for some difficult combinations of strain rate and temperature
where phase transformation occurs, a parameterized model is a
useful method to determine the hardening curves. For hot stamping
simulations the hardening description must be valid in a range of
temperatures and strain rates. The boundaries of the domain for the
temperature when hot stamping austenite are 500 ◦ C and 900 ◦ C,
and for the strain rate the test boundaries are chosen 0.1/s and 10/s.
These ranges are representative for the hot-stamping process.
The workflow for determining a hardening law is listed below:
• A set of 20 different combinations of temperature and strain rate
in this domain is tested. Ten conditions have two repetitions.
Combinations of test conditions of 500 ◦ C, 600 ◦ C, 700 ◦ C, 800 ◦ C
and 900 ◦ C at strain rates of 0.1/s, 1/s and 10/s were made.
• A hardening law is chosen to describe each tested curve depen-
dent on strain rate and temperature.
• To describe the hardening behaviour, each measured curve is
parameterized and correlations of these parameters with test
temperature and strain rate are evaluated.
• The strain rate and temperature dependence of the yield point is
derived separately.
• A hardening law describing curves dependent on strain rate and
temperature is found.
• The predicted hardening curves are compared with the measured
data and with hardening curves found in the literature.
• The curves derived from the found hardening law are used in a
FEM simulation to verify the validity of the curves.
3.2. Hardening law
The flow stress can be described as a summation of a yield
stress, a stress due to thermally activated dislocation glide which
is directly dependent on strain rate and temperature and a strain
dependent work hardening component which is indirectly depen-
dent on strain rate and temperature only, since it is the hardening
rate  = d/dε that is dependent on ε̇ and T, making this effect
strain dependent. van Liempt and Sietsma (2011) showed that the
indirect strain rate sensitivity can be responsible for the onset of
plastic instabilities, in that case intrinsic instabilities due to neg-
ative strain rate sensitivity of the hardening rate associated with
dynamic strain ageing. The relevance of that result, which pertains
to a form of work softening, to the present work is that a correct
description of strain rate sensitivity is extremely important to the
prediction of the sensitivity of a deformation process to geometri-
cal softening as explained by Backofen (1972). Softening can lead to
local thinning in sheet metal forming. Backofen stated: “The soft-
ening may be the result of geometrical changes, or of changes in
the material or both”.
The structure of the model is now:
f = 0 +  ∗ (ε̇, T ) + w (ε, ε̇, T ) (1)
where:
 f = flow stress [MPa]
 0 = athermal limit of the yield stress [MPa]
 * = thermally activated stress component [MPa]
 w = work hardening component to the flow stress [MPa]
ε = plastic strain [–]
 M. Abspoel et al. / Journal of Materials Processing Technology 228 (2016) 34–42 37

ε̇ = strain rate [1/s] 30000

T = absolute temperature [K] E modulus

At the moment of yielding, the work hardening contribution is Pre yield Linear stage III work hardening
20000
still zero which allows isolating the glide resistance contribution.

dσ/dε
As stated by Krabiell and Dahl (1981), the yield stress  y can be
described as a function of a static and a dynamic part, where the a=Ω/2
dynamic part is a function of strain rate and temperature only: 10000

 kb · T
 ε̇ m y = -ax + b
y = 0 + 0∗ · 1 + · ln (2) U'/2
G0 ε̇0
0
where: 0 150 300

σy σ [Mpa]
 y = yield stress [MPa]
0∗ = dynamic stress at zero thermal activation [MPa] Fig. 6. Kocks–Mecking plot.
kb = Boltzmann-constant = 8.617 × 10−5 [eV/K]
G0 = maximum activation energy [eV]
Now Eqs. (3), (4) and (6) are substituted in Eq. (5) and the equa-
ε̇0 = limit strain rate for thermally activated movement [1/s]
tion is reduced to:
m = strain rate exponent [–]
d 1
= · (U  −  · ( − y )) (7)
The parameters ε̇0 , the limit strain rate for thermally activated dε 2
movement and G0 , the maximum activation energy are mate- Integrating Eq. (5) gives:
rial specific constants. How the parameters for this equation are  U  
established from measured data will be explained in Section 3.4. 
f = y + · 1 − e−((ε̇,T )/2)·ε (8)
For the description of the work hardening, a physical model 
using dislocation theory is used. According to Bailey and Hirsch
Eq. (8) now consists of a yield stress which we describe with Eq.
(1960) the flow stress can be written as a function of the shear
(2) and a work hardening part dependent on strain rate, temper-
modulus G and the dislocation density :
ature and strain. Eq. (8) with a constant  is identical to the well
√ known Voce hardening law. For the equation, first the unknown
f = y + ˛ · G · b ·  (3)
parameters U ,  and  y for each individual curve must be found.
where: Then these parameters can be correlated with strain rate and tem-
perature. For the U and  this is done directly, for  y this is done
˛ = Schmidt factor [–] with the help of Eq. (2).
G = shear modulus [eV]
b = Burgers vector [nm]
3.3. Parameters from individual curves
 = dislocation density [m−2 ]
Eq. (8) is only valid for constant strain rate and temperature
where the term on the right-hand side is the relation between dis- and is used to describe the flow stress of the individual curve. The
location density and work hardening as described by Taylor (1934). parameters  y ,  and U are unknown but they can be derived from
The evolution equation for the dislocation density as a function of a Kocks–Mecking plot. The steps to create a Kocks–Mecking plot are
plastic strain according to Vetter and van den Beukel (1977) reads: as follows. Before using the stress–strain data, the measured sig-
d √ nal is smoothed using a Fast Fourier transformation. The smoothed
=U· −· (4) engineering stress signal and the engineering strain are converted
dε
into true stress true strain value. The Kocks–Mecking plot (Fig. 6)
where:
shows d/dε versus  and can be easily derived from the true stress
true strain curve. For stresses between zero and the yield stress,
U = immobilization parameter [MPa−1 ]
the elastic modulus is degraded relative to the theoretical modu-
 = parameter describing annihilation and remobilization of dis-
lus and is a non-linear decreasing function of stress. This is due to
locations [–]
dislocation anelasticity (reversible dislocation glide) as suggested
by Mott (1952) who explained the effect by the bowing out of ini-
The immobilization parameter U is conventionally taken as
tially present dislocation segments under stress. They considered
athermal as posed by Mecking and Kocks (1981). According
only the degradation of the initial modulus (at zero stress). Ghosh
to Bergström and Hallén (1982), at elevated temperatures, the
(1980) later pointed out that the pre-yield decreases nonlinearly
dynamic recovery parameter  = (ε̇, T ) is a function of strain rate
as a function of stress. The flat plateau in Fig. 8 is a good esti-
and temperature at temperatures far above room temperature.
mate for the apparent initial E-modulus. The pre yield stiffness then
Eq. (3) is differentiated:
degrades increasingly steeply to zero. This behaviour is caused by
√
d d  1 d the decrease of the resistance of the segments at increasing stress
=˛·G·b· =˛·G·b· √ (5)
dε dε 2  dε and is the subject of a current theoretical study at Tata Steel and
TU Delft. It is followed by a quite abrupt transition to linear stage
With:
III work hardening (Fig. 8). This transition indicates the yield stress,
U = U · ˛ · G · b (6) which is calculated by fitting two lines: one in the transition zone
and one in the fully plastic zone. The intersection of the two lines
where: is the yield point  y . This is a physically based definition of the
yield point. It is defined by the start of dislocation multiplication
U = normalized hardening parameter [nm−1 ] and therefore it is not arbitrary, in contraction to a yield definition
38 M. Abspoel et al. / Journal of Materials Processing Technology 228 (2016) 34–42

Fit Measured curve 25

400

U*α*b=U'/G_g(Ghosh) [Mpa-1]
20

300
True stress [MPa]

15

200
10

100 5

0 0
0 0.1 0.2 0.3 0.4 0.5 0.6 400 500 600 700 800 900 1000
True strain [-] T [°C]

Fig. 7. Measured and fitted curve. Fig. 9. U * ˛ * b = U /G Ghosh versus temperature.

0.1 1 3 10 U' trendline 0.1 1 3 10

U' [nm-1]

Ω [-]

700 800 900 1000 1100 1200

T [K] 700 800 900 1000 1100 1200
T [K]
Fig. 8. Parameter U versus temperature for different groups of strain rates.
Fig. 10. Parameter  versus temperature for different groups of strain rates.

with Rp0 . The slope of the right fitted line is equal to −/2 and the
intersection of the two lines is equal to U /2.
Now all parameters are known for Eq. (8) and the fitted curve can
be plotted. For the example used in Fig. 6, the fitted curve is shown
in Fig. 7. The experimental curve diverges from the extrapolated
Ω average [-]

curve for strains beyond the uniform strain due to necking of the
specimen.

3.4. Correlations with strain rate and temperature

For four strain rates (0.1, 1, 3 and 10/s) and a range of forming
temperatures between 500 ◦ C and 900 ◦ C tensile tests were done.
The parameters  y ,  and U derived from the Kocks–Mecking
plots for all individual tensile curves are shown in Figs. 8–10. The
parameters are plotted versus the temperature and divided in three
groups of strain rates. For the parameter U the found values for the
strain rates 1, 3 and 10/s follow the same trend line (Fig. 8).
Interesting is that U is depending on temperature. Kocks and
Mecking however posed that the parameter U was athermal. As
defined in Eq. (6), the difference between U and U is the shear
modulus G. According to Ghosh and Olson (2002) the shear modulus
of austenitic steels is dependent on temperature:
G Ghosh (T ) = 9.2648 · (1 − 7.9921 × 10−7 · (T )2
+ 3.3171 × 10−10 · (T )3 ) × 104
Now taking into account the temperature dependence for the
shear modulus, the temperature dependence of U is shown in Fig. 9
after rewriting Eq. (6).
0 .1 1 10
Strain rate [1/s]
Fig. 11. Average value of parameter  versus strain rate.
This remaining U(T) dependence may be caused by dynamic
strain ageing (DSA) as described by Bergström and Roberts (1971).
Because the amount of measurements is too low, the dynamic strain
ageing effect is difficult to determine and therefore it will not be
taken into account.
For the parameter  there is no correlation identified between 
(9) and the temperature (Fig. 10). Apparently the temperature does not
have a large role in the chosen test regime. The strain rate groups
however show a slight trend and therefore the average value of 
for each strain rate group was taken and plotted on a logarithmic
scale versus the strain rate (Fig. 11). There is a light trend found
depending on the strain rate.
 M. Abspoel et al. / Journal of Materials Processing Technology 228 (2016) 34–42 39

0.1 1 3 10 0.1 1 3 10 Linear (all)

200 300
σ y = σ 0 + σ 0*

200
Slope = σ 0*

σy [Mpa]
σy [Mpa]

100

100

0 0
700 800 900 1000 1100 1200 0 0.2 0.4 0.6 0.8 1
T [K] ΔG / ΔG0

Fig. 12. Parameter  y versus temperature for different groups of strain rates. Fig. 13. Plot to determine ε̇0 and G0 .

In Fig. 12, the yield strengths are shown for different strain on temperature, strain rate and strain, the equation for the flow
rate groups versus temperature. The dependency from strain rate stress is:
and temperature of the yield strength requires a more advanced       
U  (T ) kb · T ε̇
approach. The yield strength is described as a function of a static  = 0 + · 1 − e−((ε̇,T )/2)·ε + 0∗ · 1 + · ln (12)
(ε̇) G0 ε̇0
and a dynamic part as shown in Eq. (2) where  0 is the athermal
limit of the yield stress for the material. There are several parame-
4. Discussion
ters and constants in the equation of Krabiell and Dahl of which the
strain rate and the temperature for each test, and the Boltzmann
With all parameters and constants known, curves can be cal-
constant kb are known. The remaining parameters and constants
culated for all temperature and strain rate combinations for the
are: the limit strain rate ε̇0 , the maximum activation energy G0 ,
austenitic condition. Note that the dynamic strain rate influence
the athermal limit of the yield stress  0 , the maximal thermal yield
cannot be negative and needs to be cut off at zero for negative val-
stress contribution 0∗ and the strain rate exponent m . The strain
ues. The hardening curves from the model were compared with the
rate exponent m is chosen to be 1, which means, as is described by
direct fits on the measured data. In Fig. 14, the comparison for the
Kocks et al. (1975) that the interaction between dislocations and an
measurements at different strain rates and temperatures is shown.
obstacle follows a square force–distance relationship and is an ade-
To quantify the accuracy of the model, the variance of the resid-
quate assumption when the range of experimental strain rates and
ual of the whole curve is determined by calculating the sum of
temperatures is limited. When the exponent is chosen different,
squares of the datapoints and dividing them by the degrees of
this interaction has a more Gaussian shape. The activation energy
freedom which is total amount of points.
for a certain temperature and strain rate is given by:

 ε̇  [ym − yF ]2
G = −kb · T · ln (10) Var = (13)
ε̇0 DOF

where
G = activation energy (eV)
With m = 1 and Eq. (10), the Krabiell and Dahl equation is then
reduced to:
 G
   G
y = 0 + 0∗ · 1 − → y = 0 + 0∗ − 0∗ ·
G0 G0
The limit strain rate and maximum activation energy are by def-
inition constants for all tensile tests in the complete domain. Eq.
(11) is then reduced to a simple linear equation. Fig. 13 shows the
measured  y values, plotted versus G/G0 . From Eq. (11) it fol-
lows that the yield stress is a single valued function of G. This
is only possible for a correct estimated combination of limit strain
rate and the maximum activation enthalpy. Since G/G0 ≤ 1 the
yield stress is never lower than  0 .
Due to the linear behaviour of Eq. (10), 0∗ can be derived from
the trendline. When G/G0 is 0,  y is equal to the intersection
value of the trendline with the y-axis. From this  0 can be cal-
culated. The linear correlation shows also that the chosen m is
sufficient for this range of temperatures and strain rates. Now all the
parameters in the Krabiell and Dahl equation are known, describing
the yield strength dependent from strain rate and temperature, and
both the work hardening and the yielding are described depending
where ym is the stress from the model and yF is the stress in the
fit of the measured curve. More datapoints are taken at the start of
the curve and then gradually the sampling frequency is lowered.
The average accuracy of the model is 15 MPa. Because the hard-
ening curves from the model show a good agreement with the
measured data. This also implicates that m = 1 is an adequate esti-
mate. To verify the sensitivity for m , values of 1.2 and 1.5 are
considered (Fig. 15). With an m of 1.5 the curve already starts to
(11) deviate from the measured points but an m of 1.2 is possible. In
both cases the values of  0 and 0∗ will rise. The impact for the
hardening curve is also shown in Fig. 15. Combining this impact
and the comparisons with the measured data confirms that m of
1 is a value that is valid. Even though the range for the tempera-
tures is from 500 ◦ C to 900 ◦ C and the strain rates range from 0.1 to
10/s, this results in no points for the lower range of G/G0 . Tests
for much lower temperatures and other strain rates to create more
points on the left side are not possible due to limitations on the
speed range of the Gleeble and due to transformation issues with
lower temperatures and lower strain rates.
For further verification of the model an additional set of nine
in between strain rates and temperatures is tested. The additional
set had temperatures of 750 ◦ C, 800 ◦ C and 900 ◦ C at a strain rate of
3/s and 450 ◦ C, 550 ◦ C, 700 ◦ C, 750 ◦ C at a strain rate of 1/s. All these
combinations were not taken in the set used to calibrate the model.
This additional set also had an average deviation of the modelled
curve of 15 MPa.
40 M. Abspoel et al. / Journal of Materials Processing Technology 228 (2016) 34–42

500°C & 0.1 model 600°C & 0.1 model 900°C & 0.1 model 600°C & 1 model 700°C & 1 model 800°C & 1 model
500°C & 0.1 Fit 600°C & 0.1 Fit 900°C & 0.1 Fit 900°C & 1 model 600°C & 1 Fit 700°C & 1 Fit
600°C & 0.1 Fit 800°C & 1 Fit 900°C & 1 Fit
400 500

400
True stress [MPa]

True stress [MPa]

300

300
200
200
100
100

0 0
0 0.1 0.2 0.3 0.4 0.5 0.6 0 0.1 0.2 0.3 0.4 0.5 0.6
True strain [-] True strain [-]

750°C & 3 model 800°C & 3 model 900°C & 3 model 800°C & 10 model 900°C & 10 model
750°C & 3 Fit 800°C & 3 Fit 900°C & 3 Fit 800°C & 10 Fit 900°C & 10 Fit
900°C & 10 Fit
400 400
True stress [MPa]

True stress [MPa]

300 300

200 200

100 100

0 0
0 0.1 0.2 0.3 0.4 0.5 0.6 0 0.1 0.2 0.3 0.4 0.5 0.6
True strain [-] True strain [-]

Fig. 14. Hardening curves from the model versus fitted curves on measured data.

m' = 1.0 m' = 1.2 m' = 1.5 Measured points 700°C & 0.001 model 700°C & 0 numisheet
400 700°C & 0.1 model 700°C & 0.1 numisheet
700°C & 1 model 700°C & 1 numisheet
700°C & 10 model 700°C & 10 numisheet
300
400
σy [MPa]

200
True stress [MPa]

100
200

0
0 0.2 0.4 0.6 0.8 1
ΔG/ΔG0

750°C & 3 model m'= 1.0 750°C & 3 model m' = 1.2 0
750°C & 3 model m' = 1.5 750°C & 3 Fit 0 0.1 0.2 0.3 0.4 0.5
True strain [-]
400
Fig. 16. Comparison with 22MnB5 dataset Numisheet 2008.
True stress [MPa]

300

200
100
0
0 0.1
Fig. 15. Impact of m value.
For a comparison with other available data, the hardening curves
from the derived model are compared (Fig. 16) with data used
in the Numisheet 2008 benchmark (BM3) (2008) for the contin-
uous press hardening of a B-Pillar, which is also standard data in
some FEA packages. The main difference between the curves is the
yielding behaviour for low strain values. The curves from the
Numisheet 2008 benchmark show higher yield values, and less
hardening. Possible explanations can be: the more accurate deter-
mination of the yield stress from the Kocks Mecking plots, a possible
inhomogeneous temperature distribution in the Numisheet sam-
0.2 0.3 0.4 0.5 0.6
ples, or the use of different substrate.
True strain [-]
To validate the hardening curves, a simulation in AutoForm ver-
sion R5.2 was made (Fig. 17). The part is a simplified B-pillar and
made of zinc-coated boron steel with a thickness of 1.5 mm, iden-
tical to the tensile specimens. The simulation mimics the process
in the press. It starts with heating of the blank to 900 ◦ C, a transfer
time of 6 s to the press, and closure of the tools with 350 mm/s. The
speed of closure is important for the strain rate effect found in the
measured tensile curves. A quenching time of 10 s is applied after
which cooling in air remains. Due to the low transfer time, the part
 M. Abspoel et al. / Journal of Materials Processing Technology 228 (2016) 34–42
Fig. 17. Simulation and thickness measurement of a lab type B-pillar. (Left: Numisheet, right: Tata Steel.)
remains austenitic and the stamping occurs while the temperature
is still above 500 ◦ C. Both strain rate and temperature are within
the test domain of the tensile specimens. In the corner the high-
est thinning occurs. Three hot formed parts were measured along
a cross section. Two of the parts were only measured in the flange
and on the thinnest part just after the die radius. One part was
measured at multiple spots along the cross section. The simulation
with the Numisheet hardening curves predicts a lower thickness
in the flange and the side wall. The simulation with the Tata Steel
constitutive model is quite successful in predicting the measured
thickness distribution. In both simulations all conditions, settings
and yield locus parameters, were kept identical, only the applied
hardening models are different. To check whether the friction could
play a role in the predictability, the simulation with the Numisheet
hardening curves was repeated with a low friction value of 0.3.
Even with this unrealistic value, the thickness in the flange was not
predicted correctly.
5. Conclusions
The heat distribution in the tensile bar for tensile tests under
hot conditions is very important. It is possible to create a
41
homogeneous temperature distribution in the tensile bar by
using shunts. These shunts allow for an equal current density
in the complete sample. Samples without these shunts have the
risk of an inhomogeneous temperature distribution resulting in
non-austenized areas of that sample. Considering that for high tem-
peratures ferrite behaves softer than austenite, necking can occur in
this softer area. An inhomogeneously heated sample may also result
in invalid measurements. Furthermore, from the tensile test curve a
Kocks–Mecking plot can be derived, which provides the parameters
needed for hardening description. By correlating these parameters
with strain rate and temperature, a parametric model was made
to describe hardening curves within the hot stamping domain for
austenitic forming conditions. The measurements and the calcu-
lated curves are in good agreement. The proposed approach for
determining the strain rate and temperature model can be used for
different geometries of the tensile sample. It is however important
that the temperature distribution in the tensile bar is homoge-
nous and a wide enough range of temperatures and strain rates
is recommended.
Simulation with the found curves on a hot formed part and
comparison with practice showed that the simulation predicts
the thickness distribution very well. This is due to the physically
correct formulation of the proposed constitutive model and the
42 M. Abspoel et al. / Journal of Materials Processing Technology 228 (2016) 34–42
improved methodology of executing and interpreting the tensile
experiments. The curves from literature were different and could
not predict the right thickness, even with different friction coeffi-
cients.
Acknowledgements
We would like to thank Delft University of Technology for facil-
itating the Gleeble test machine, Nick den Uijl, Roy Frinking and
Hans Hofman for their contribution to the testing work on the
Gleeble test machine.
References
Åkerström, P., (Licentiate thesis) 2004. Material Characterisation for Simulation of
Press Hardening. Luleå University of Technology, ISSN 1402-1757/ISRN LTU-
LIC–04/28–SE/NR 2004:28.
Backofen, W.A., 1972. Deformation Processing. Addison-Wesley, Reading, MA.
Bailey, J.E., Hirsch, P.B., 1960. The dislocation distribution, flow stress, and stored
energy in cold-worked polycrystalline silver. Philos. Mag. 5, 485–497.
Barnett, M.R., Jonas, J.J., 1999. Distinctive aspects of the physical metallurgy of warm
rolling. ISIJ Int. 39 (9), 856–873.
Bergström, Y., 1969–1970. A dislocation model for the stress–strain behaviour of
polycrystalline ␣-Fe with special emphasis on the variation of the densities of
mobile and immobile dislocations. Mater. Sci. Eng. 5, 193–200.
Bergström, Y., Roberts, W., 1971. The application of a dislocation model to dynamical
strain ageing in ␣-iron containing interstitial atoms. Acta Metall. 19, 815–823.
Bergström, Y., Hallén, H., 1982. An improved dislocation model for the stress–strain
behaviour of polycrystalline ␣-Fe. Mater. Sci. Eng. 55, 49–61.
Ghosh, A.K., 1980. A physically-based constitutive model for metal deformation.
Acta Metall. 28 (11), 1443–1465.
Ghosh, G., Olson, G.B., 2002. The isotropic shear modulus of multicomponent Fe-base
solid solutions. Acta Mater. 50, 2655–2675.
Hein, P., 2005. A global approach of the finite element simulation of hot stamping.
Adv. Mater. Res. 6, 763–770.
Klepaczko, J.R., Chiem, C.Y., 1986. On rate sensitivity of fcc metals, instantaneous
rate sensitivity and rate sensitivity of strain hardening. J. Mech. Phys. Solids 34
(1), 29–54.
Krabiell, A., Dahl, W., 1981. Zum Einfluss von Temperatur und
Dehngeschwindichkeit auf die Streckgrenze von Baustählen unterschiedlicher
Festigkeit. Arch. Eisenhüttenwes. 52, 429–436.
Kocks, U.F., 1976. Laws for work-hardening and low-temperature creep. J. Eng.
Mater. Techn. 98 (1), 76–85.
Kocks, U.F., Argon, A.S., Ashby, M.F., 1975. Thermodynamics and kinetics of slip. In:
Chalmers, B., Christian, J.W., Massalski, J.B. (Eds.), Progress in Materials Science,
vol. 19, p. 26.
Kocks, U.F., Mecking, H., 2003. Physics and phenomenology of strain hardening: the
FCC case. Prog. Mater. Sci. 48 (3), 171–273.
Merklein, M., Lechler, J., Geiger, M., 2006. Characterisation of the flow properties of
the quenchenable ultra high strength steel 22MnB5. CIRP Ann. Manuf. Technol.
55 (1), 229–232.
Lechler, J., Merklein, M., Geiger, M., 2008. Mechanical material properties of ultra-
high strength steels for hot stamping. Steel Res. Int. 79 (2), 98–104.
Mott, N.F., 1952. CXVII. A theory of work-hardening of metal crystals. Philos. Mag.
43 (346), 1151–1178.
Mecking, H., Kocks, U.F., 1981. Kinetics of flow and strain-hardening. Acta Metall. 29
(1), 1865–1875.
2008. Numisheet, www.numisheet2008.ethz.ch/benchmarks
Sarkar, S., Militzer, M., 2009. Microstructure evolution model for hot strip rolling
of Nb–Mo microalloyed complex phase steel. Mater. Sci. Technol. 25 (9),
1134–1146.
Taylor, G.I., 1934. The mechanism of plastic deformation of crystals. Part I. Theoret-
ical. Proc. R. Soc. Lond. A 145 (855), 362–387.
Turetta, A., (Thesis) 2008. Investigation of Thermal, Mechanical and Microstructural
Properties of Quenchenable High Strength Steel in Hot Stamping Operations.
University of Padova, pp. 45–62.
van Liempt, P., Onink, M., Bodin, A., 2002. Modelling the influence of dynamic
strain ageing on deformation behaviour. Adv. Eng. Mater. 4 (4), 225–
232.
van Liempt, P., Sietsma, J., 2011. A revised criterion for the Portevin–Le Châtelier
effect based on the strain-rate sensitivity of the work-hardening rate. Metall.
Mater. Trans. A 42 (13), 4008–4014.
Vetter, R., van den Beukel, A., 1977. Dislocation production in cold worked copper.
Scr. Metall. 11 (2), 143–146.