Journal of Materials Processing Technology 227 (2016) 227–243

Contents lists available at ScienceDirect

Journal of Materials Processing Technology

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

Variation and consistency of Young’s modulus in steel

Zhong Chen a , Umesh Gandhi b , Jinwoo Lee c , R.H. Wagoner a,∗
a
Department of Materials Science and Engineering, 2041 College Road Ohio State University, Columbus, OH 43210, USA
b
Toyota Research Institute North America, Ann Arbor, MI, 48105, USA
c
Materials Deformation Department, Process Commercialization Center, Korea Institute of Materials Science (KIMS), Changwon, South Korea

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

Article history: The mechanically-measured Young’s modulus of metals is consistently lower than the physically mea-
Received 22 May 2015 sured one, particularly after plastic straining. Furthermore, the nominally elastic loading and unloading
Received in revised form 20 August 2015 behavior is not linear; it shows significant curvature and hysteresis. While many reports of this so-called
Accepted 21 August 2015
“modulus effect” have appeared, the consistency of the behavior among grades of steel, or within a single
Available online 28 August 2015
grade produced by alternate methods and suppliers, is unknown. That is, there is little information on
whether it is necessary for manufacturers to measure and control the mechanical modulus for every coil
Keywords:
of steel in order to guarantee accurate simulations, consistent forming, and reliable in-service behav-
Steel
Unloading
ior. In order to address these issues, 12 steels (4 diverse grades: IF, HSLA, DP600, DP980; 3 producers
Reloading per grade) were subjected to high-precision modulus measurements using mechanical testing, resonant
Modulus frequency damping analysis, and ultrasonic pulse-echo techniques. All of these measurements show
Nonlinearity remarkable consistency among not only suppliers but also among grades. The primary determinant of
Resonant frequency hysteresis/curvature of the stress–strain response was found to be the nominal flow stress of the alloy.
Ultrasonic Other variations of overall mechanical modulus are minor compared with hysteresis/curvature. The fol-
Hysteresis lowing conclusions were reached: 1) there is no significant difference among suppliers of a single steel
grade, 2) there is very little difference between grades of steel, except for that attributable to differ-
ing strengths, 3) mechanical unloading and reloading after pre-strain are similar, 4) cyclic loading and
unloading cycles have no accumulated effect except through a minor change of flow stress, and 5) the ini-
tial loading or unloading modulus is very similar to the physical modulus, but the mechanically measured
slope degrades very rapidly as loading or unloading proceeds, and plateaus at even small strain (<2%). The
measured unloading and reloading behavior is more consistent and reproducible than that during initial
loading, and unloading behavior is more consistent and reproducible than reloading behavior. Therefore,
it is recommended that unloading after pre-strain is used to represent all of nominally elastic nonlinear
behavior most accurately.
© 2015 Elsevier B.V. All rights reserved.

1. Introduction becomes a significant problem for the adoption of AHSS, particu-

larly in certain formed part areas such as corner radii (Opbroek,
Advanced high strength steels (“AHSS”) are increasingly being 2009) and sidewalls (Lee et al., 2011).
used for strength-constrained automotive applications in order to It has been long known that the mechanically-measured Young’s
meet the demand of vehicle mass reduction and fuel economy. The modulus is reduced by plastic straining (Lems, 1963). However,
higher strengths of the AHSS correspond to larger elastic recovery this complexity has been almost universally ignored in industrial
(spring back) after forming, consistent with theory and experience forming applications; instead, handbook values of Young’s mod-
(Wagoner et al., 2006). In addition to this well-known consequence, ulus (i.e. physically-measured ones, or arbitrarily adjusted ones)
it has also proven more difficult to model such springback consis- are used in virtually all simulations. Morestin and Boivin (1996)
tently and reliably with these alloys as compared with traditional showed that using the mechanically-measured Young’s modulus
ones. The accuracy and reliability of springback simulation thus value, which for their materials degraded up to approximately
20% during the first 5% of plastic strain, produced more accurate
springback predictions. Subsequent research has shown similar
degradation magnitudes, depending on material: up to 30% for mild
∗ Corresponding author. steels (Yoshida et al., 2002), 20% for high-strength steels (Cleveland
E-mail address: [email protected] (R.H. Wagoner).

http://dx.doi.org/10.1016/j.jmatprotec.2015.08.024
0924-0136/© 2015 Elsevier B.V. All rights reserved.
228 Z. Chen et al. / Journal of Materials Processing Technology 227 (2016) 227–243

and Ghosh, 2002; Yoshida et al., 2002), 12% for copper (Agnew
and Weertman, 1998), or 10% for aluminum (Cleveland and Ghosh,
2002).
Yoshida et al. (2002) introduced an exponential representation
of the change of chord modulus (i.e. Echord , the slope of a straight
line drawn through the two stress–strain endpoints for a nominally
elastic loading or unloading excursion) with plastic strain, p :
Echord = E0 − (E0 − Ea )[1 − exp(− p )] (1)
where E0 and Ea represent the chord modulus for unstrained and
highly strained materials, respectively, and  is a material constant.
A chord modulus can be thought of as a kind of average mod-
ulus over the nominally elastic range, i.e. ignoring the change of
slope or the curvature of the stress–strain relationship. In general,
the authors subsequent to Morestin and Boivin have recommended
using the chord modulus to represent nominally elastic loading and
unloading in simulations in order to improve springback predic-
tion in finite element programs. That is, to ignore the curvature
and instead focus on the change of overall slope. In addition to
the role of plastic straining, and of course also the choice of mate-
rial, other secondary sources of chord modulus variation have been
reported: chemical composition (Fei and Hodgson, 2006), time
between pre-strain and loading or unloading (Morestin et al., 1996),
heat treatment (Yang et al., 2004), and strain path (Pavlina et al.,
2008).
In spite of such simplifications, the stress–strain relationship
of unloading or reloading when chord modulus degradation is
observed is not linear. The unloading and reloading paths form a
hysteresis loop that exhibits strain recovery/reversal but not energy
preservation (Sun and Wagoner, 2011). In addition to the reversed
concavity, reports of other differences in loading and unloading
chord moduli have varied: Zang et al. (2006) reported a lower
value in unloading for aluminum alloys while Andar et al. (2010)
found the opposite for high strength steel. Some researchers iden-
tified three stages describing the change of slope during loading or
unloading within the nominally elastic strain range (Cleveland and
Ghosh, 2002) or stress range (Luo and Ghosh, 2003).
Fig. 1. Schematics of automotive pillar cross-sections: (a) flat design (b) jogged
Sun and Wagoner (2011) developed the QPE design.
(Quasi–Plastic–Elastic) model for the governing material behavior
and showed that its use reduced the error of springback simu-
lations by factors of 2–6 as compared with linear elastic chord
models. They also clarified many aspects of the nonlinear response
of steels, including to first order the following: strain-rate inde- on tensile tests conducted on the sheet before forming and weld-
pendence, single-cycle vs. multi-cycle independence, and even ing. Young’s modulus was taken to be the standard handbook
the same hysteresis behavior of two markedly different steels (a value: 208 GPa. As is readily seen, the results are not satisfac-
DP980 and a DP780) with the same strength before unloading. tory, particularly in view of the fact that the consistency of the
The motivation for the current work lies not only with the well- load-displacement response of such a member, and the ability to
known need for accurate springback simulation: as noted earlier, simulate it, is critical for vehicle structural performance, particu-
the unloading nonlinearity, or even the use of a chord modulus, larly for the location and calibration of accelerometers and other
is not widely used in such simulations today. The current motiva- such sensors that may be mounted on it.
tion derives from experiments on deformation characteristics of After extensive simulations (See Appendix A for more detail)
two automotive structural pillars from experiments conducted in in an unsuccessful task to determine a model-based source of the
2009 and with associated simulations performed between 2009 discrepancy, a modified version of Sun and Wagoner’s (2011) QPE
and 2013. Those results are not available in public publications; model was implemented and used in a new simulation with an
they have only been presented to a limited audience (Gandhi and otherwise identical FE model. The QPE parameters were based on a
Wagoner, 2013). For that reason, the details of the experiments single tensile test of the material used to construct the pillars, and
and simulations are presented in the Appendix A to this paper. The followed the procedures in Sun and Wagoner (2011). The result,
principal results appear in Figs. 1 and 2; additional presentation shown as red long-dashed lines in Fig. 2, shows greatly improved
and discussion of these results will appear toward the end of this accuracy with respect to the experimental results; the standard
paper using the results of the current research. deviations between the simulations and experiments were reduced
Fig. 1 shows the cross-sections of two types of 600 mm-long pro- by 55–56%.
totype automotive structural pillars, designated (a) the flat design The Toyota pillar research suggests that it is necessary to know
and (b) the jogged design. The points in Fig. 2 represent low-speed the details of the nonlinear stress–strain response of a material in
compression experiments performed using a mechanical testing the nominally elastic region in order to simulate the performance
machine. The blue/solid lines represent standard industrial simu- of structural members made of that material accurately. With that
lation practice, i.e. using an elastic-plastic material model based in mind, corollary questions arise:
 Z. Chen et al. / Journal of Materials Processing Technology 227 (2016) 227–243 229

Fig. 3. Tensile test results for 12 steels in 4 grades, standard deviations shown
(<S.D.>).

2. Experimental and analytical procedures

2.1. Materials

Twelve steels, 3 each from distinct suppliers in 4 widely-varied

grades were obtained from ArcelorMittal, Nucor, ThyssenKrupp, or
U.S. Steel. Suppliers asked not to be identified with individual alloys,
which are therefore designated here as being from Suppliers 1–4,
not, respectively. That is, a dual-phase steel with 980 MPa nominal
tensile strength provided by Supplier 1 will be designated DP980-1.
Typical automotive grades of steel were used. Interstitial-free
steel (IF) contains very low amounts of interstitial elements includ-
Fig. 2. Test and simulation results for automotive pillar crush: (a) flat design, (b) ing carbon (<0.005%), thus it is ductile and soft, and it will not form
jogged design. Luder’s lines during forming. High-strength low-alloy steel (HSLA)
is stronger and has a higher carbon content (0.05–0.25%), with
micro-alloying elements (such as manganese, nickel, chromium)
to improve its strength, primarily by promoting grain refinement.
Dual-phase steels (DP980 and DP600 in the current work) are com-
• Does the nonlinear stress–strain response vary from alloy to alloy,
posed of a matrix of soft ferrite and islands of hard martensite
supplier to supplier, coil to coil, and/or part to part? phase, thus exhibiting low yield stress, high work hardening rate,
• Will the structural performance vary from part to part (and
and good ductility through the stress concentration effect produced
thus vehicle to vehicle) if the normally-unmeasured nonlinear by the interaction of the two coarsely-distributed phases.
response varies? All of the tested alloys were zinc coated. By acid stripping of the
• Is it necessary to destructively test all alloys, suppliers, coils
coatings, the average coating thickness was found to be 1.3 ± 0.2%
and/or parts and discard non-conforming examples in order to of the overall sheet thickness (full range: 0.7–1.8%). Accounting for
guarantee uniform structural performance vehicle to vehicle? the coating thickness in density measurements has an estimated
Conversely, can suppliers guarantee a certain behavior although absolute effect of less than 0.2% in view of the similar coating and
they do know what controls it nor how to adjust processes accord- substrate densities. The absolute effect on stress measurement is
ingly? estimated to be less 1% in view of typical coating strengths. These
values are similar to other sources of absolute uncertainties and
scatter in raw measurements, therefore no attempt was made to
In other words, how much new testing or standard-setting compensate for, or to remove the coatings. It should be noted
would be required to assure the accuracy and reproducibility of that nearly all of the stress-based quantities used in the current
structural simulations and real-world behavior in order to conform work are differential, either for measurements on a single alloy
to the design intentions? (with coating have virtually no effect); or among alloys, for which
These questions were of interest and concern to the Steel the expected variation by ignoring coating thicknesses would be
Marketing Development Institute, [SMDI]. In order to answer them, related only to difference of fraction coating thicknesses among the
SMDI commissioned the study and results upon which the current alloys. For stress measurements among the alloys, for example, the
paper is based. The mechanical stress–strain responses of 12 sheet error would be approximately ±0.2%, less than typical test-to-test
steels were measured before and after plastic tensile deformation scatter.
and compared with physical measurements of Young’s modulus. Table 1 and Fig. 3 present standard mechanical properties
The steels represent three alternately-sourced examples in each of obtained from tensile testing. These show a wide variation of prop-
four grades used for automotive body structures: traditional grades erties (by a factor of ∼4–5 times): yield stresses from 137 to
IF (interstitial-free) and HSLA (high-strength, low alloy), and AHSS’s 790 MPa, ultimate tensile strengths from 278 to 1036 MPa, total
DP (dual-phase) 600 and DP 980. elongations from 12 to 50%. The average test-to-test scatter for all
230 Z. Chen et al. / Journal of Materials Processing Technology 227 (2016) 227–243

Table 1
The mechanical properties of tested steel samples, rolling direction.

Material t (mm)  YS  UTS (MPa) eu (%) et (%)*  (kg/m3 )

DP980-1 1.65 660 962 10.2 16.8 7821

DP980-2 1.00 576 960 10.2 17.8 7727
DP980-3 1.38 790 1036 7.6 12.4 7709
DP600-1 1.36 390 650 15.7 25.3 7767
DP600-2 1.63 417 693 15.7 25.7 7750
DP600-3 1.22 394 636 17.6 27.3 7894
HSLA-1 1.42 415 453 20.4 31.1 7724
HSLA-2 1.23 362 485 17.3 26.7 7681
HSLA-3 1.50 382 449 18.7 31 7734
IF-1 1.2 147 278 28.6 50.2 7764
IF-2 1.18 134 300 28.7 47.2 7745
IF-3 1.50 150 278 26.4 50.5 7742
DP980 Avg. 675 ± 54 986 ± 22 9±1 16 ± 1 7752 ± 30
DP600 Avg. 400 ± 7 660 ± 30 16 ± 1 26 ± 1 7804 ± 39
HSLA Avg. 386 ± 13 462 ± 10 19 ± 1 19 ± 1 7713 ± 14
IF Avg. 144 ± 4 286 ± 6 28 ± 1 49 ± ± ± 1 7750 ± 6

Key:
t – sheet thickness;
 y –0.2% offset yield strength;
 UTS – ultimate tensile strength;
eu – uniform elongation (engineering strain at maximum tensile load);
et – total elongation (total engineering strain to fracture);
 – density.
Tensile properties were determined at the strain rate of 0.0005/s, with a stroke rate of 0.04 mm/s.

steels was 5 MPa. The scatter of properties varies across suppliers Table 2
Olympus 38DL ultrasonic measurement parameters, as recommended by Olympus.
for a single steel grade varies by grade, ranging from 110 MPa for
DP980 to 20 MPa for IF steel. M208 V222-BB-RM

Detection mode Mode 3 Mode 3

2.2. Ultrasonic pulse-echo technique (US) Measurement type Standard Standard
Probe type M208 M208
For standard ultrasonic pulse –echo (US) measurement of elastic Velocity (mm/␮s) 5.99 3.16
Zero value 600 600
moduli, the ultrasonic transducer sends a pulse of ultrasonic exci- Pulser power (Volts) 200 110
tation into the tested sample and detects the reflected return from Max gain (dB) 66.1 54.5
the other side of the specimen. For sheet materials, the direction of Init gain (dB) 50 46.2
sound transmission is limited to the sheet normal direction (“ND” TDG slope (dB/␮s) 31.16 31.16
MB blank (␮s) 3.8 5.7
or “N”) by accuracy constraints, although the excitation direction
Echo window (␮s) 4.5 4.5
can be co-linear (for a longitudinal wave corresponding to a Young’s IF blank (ns) 170 0.31
Modulus through the thickness of the sheet, i.e. “N-N”) or perpen- M3 blank (ns) 110 110
dicular to the sheet normal, e.g. in rolling direction (“RD” or “R”),
transverse direction (“TD” or “T”), and 45◦ directions, correspond-
ing to shear moduli in R-N, T-N, and 45-N directions, respectively. The various elastic moduli: Poisson’s ratio (), Young’s modulus
An Olympus 38DL Plus ultrasonic thickness gage and accessories (EUS , which is ENN , i.e. in the through-thickness direction) and
were used (Olympus, 2011). Longitudinal waves were excited and shear modulus (G, which can be for any in-plane direction and
detected using an M208-RM transducer acting through a high- the through-thickness direction, depending on excitation orienta-
viscosity “SWC” couplant, while shear waves were excited and tion, e.g. GNR , GNT , GN45 ), are then computed in terms of calculated
detected using a V222-BB-RM transducer acting through glycerin. velocities as shown below (Olympus, 2011):
A set-up profile, DEFM3-20-M208, was used for both for M208
1 − 2(VT − VL )2
and V222 tests, with the parameter settings as recommended by v= (3)
Olympus technicians as shown in Table 2. 2 − 2(VT − VL )2
The primary US measurements were made on undeformed ten-
VL2 (1 + v)(1 − 2v)
sile specimens in the shoulder regions. Secondary readings were E= (4)
1−v
made after multi-cycle LUL tests along the longitudinal center line
of the specimen, 12 mm away from the fracture. G = VT2  (5)
The sheet thickness at the site of each ultrasonic modulus mea-
surement was measured using a Mitutoyo Quantumike micrometer where VT is the shear velocity, VL is the longitudinal velocity,  is
(Model: No. 293-185) (Mitutoyo, 2008), which has a measurement the density. The density was measured to an estimated accuracy of
precision and accuracy of 0.001 mm (approximately 0.1% of the 0.9% by measuring the average thickness throughout a sheet 508
thickness). The 38DL measures the round-trip transit time for the by 610 mm in size and weighing it.
sound waves which allows the computation of the sound velocity
as follows: 2.3. Resonant frequency damping analysis (RF)
Measured Thickness(mm)
Velocity(mm/s) = (2) Resonant frequency damping analysis (RF) is a nondestructive
Round Trip Time(s)/2
method to measure elastic moduli using the resonant frequencies of
a specimen and the damping of these frequencies (Sever, 2011). All
 Z. Chen et al. / Journal of Materials Processing Technology 227 (2016) 227–243 231

RF measurements presented here were conducted by Joris Bracke

[IMCE] using commercially-available IMCE “RFDA Basic” measuring
equipment.
Rectangular specimens 55 mm × 20 mm were prepared at OSU
using electro-discharge machining and specimen dimensions were
measured using a micrometer having a precision and accuracy of
0.001 mm (Mitutoyo, 2008). In Belgium each specimen was posi-
tioned on two thin wires at the nodes of the first mode of flexural
vibration. The vibration was excited by a sharp hammer blow
at an anti-node and detected with a high resolution microphone
near another node. The recorded signal was converted to an FFT-
spectrum from which the Young’s modulus was calculated (Bollen,
2006). The equations for the determination of Young’s and shear
modulus are as follows (ASTM E 1876, ISO 12680-1, ENV 843-2):

mff2 l3
E = 0.9465( )( )T (6)
w t3
where ff is the flexural frequency, m is the mass, l is the length, w
is the width, t the thickness, and T is the correction coefficient.

2.4. Mechanical testing

Standard ASTM E-8 room-temperature tensile tests were per-

formed in the RD and TD directions at a nominal strain rate of
10−3 /s (the end displacement rate was controlled at 0.08 mm/s).
Testing was done using an MTS 810 tensile frame, equipped with a
100 kN MTS Model 318.10 force transducer (MTS, 2006). For each
material, two tests were repeated to estimate the reproducibil-
ity of results. Strain measurements were taken using an Epsilon
Class A1 extensometer (3452-0200-030-ST), 2” gage, +30%/−10%.
The control type was displacement-controlled. Data was collected
at a sampling frequency of 10 Hz up to 0.005 strain and at 1 Hz
subsequently.
Similar tests except with loading-unloading-loading cycles
added (LUL tests) were performed with the same settings and
equipment using an automated control program constructed for
this purpose. Data was collected at a sampling frequency of 10 Hz in Fig. 4. Definition of analytical procedures used to determine slopes in a typical
the approximate elastic regions, i.e. at the start of the test, as in the unloading-loading cycle: (a) displacement rate reversal points, (b) stress ranges used
standard tensile tests, and from er − 0.005 to er + 0.005, (where er is to define E1 , E2 , E3 and E4 .

the engineering strain at which the displacement rate is reversed)

and 1 Hz outside of those regions. The control program was set to the difference between the current true strain  and the preceding
reverse the displacement direction (to a compressive direction) at reversal strain r .
a fixed displacement corresponding approximately to the desired The lower-right inset in Fig. 4(a) illustrates the procedure used
engineering strain. When the load dropped to 50 kN, the direction of to define the reversal strain for unloading, 1 . The last two data
displacement was again reversed to the original tension direction. points acquired before displacement reversal/unloading began (i.e.
Three types of LUL tests were conducted. In the multi-cycle ones, at the reversal time, which is known) are used to define a line, as are
reversals were programmed to correspond to engineering strains the first two acquired data points following the reversal/unloading.
of 2%, 4%, 6%, 8%, and 10%. In the two types of single-cycle tests, the The intersection of the two lines defines 1 (which is approximately
reversal strains er were set to eu /2 or eu , where eu is the engineer- the true strain equivalent of er , the engineering strain at the rever-
ing strain at maximum load, i.e. the uniform elongation. As with the sal, which was controlled by the end displacement of the specimen).
tensile tests, all LUL tests were repeated at least twice to estimate For the unloading leg of the LUL cycle,  = | − 1 |.
the test-to-test scatter. The upper-left inset in Fig. 4(a) illustrates the procedure used to
define the reversal strain for reloading, 2 . This represents the strain
2.5. Mechanical modulus analysis for the end of unloading and the start of reloading. The last two
unloading points before the new displacement reversal/reloading
A typical L-U-L cycle is shown in Fig. 4. This is the entirety of the are used to construct a line, the intercept of which at zero stress
data acquired for the 4th LUL cycle of DP980-1, i.e. at r = 0.08. The defines 2 . For the reloading leg of the LUL cycle,  = | − 2 |.
figure emphasizes the discrete nature of the data collected on this For numerical analysis of reloading, it is also necessary to limit
highly-expanded scale. In view of the data-gathering intervals, it is the range of data retained representing nominally elastic reload-
necessary to know the incremental strains relative to the displace- ing, that is, before steady-state plastic deformation is re-attained
ment reversal strain.  is defined here as the absolute value of after the pair of displacement reversals.2 For the current work, that

1 2
This model of extensometer has Class A resolution and accuracy in the range of A similar problem arises at the initial loading at the start of a tensile test or LUL
extensions used in the current work. test, as will be discussed later.
232 Z. Chen et al. / Journal of Materials Processing Technology 227 (2016) 227–243

Table 3
Ultrasonically-measured (US) elastic constants and test-to-test scatters (half of standard deviation), all in GPa unless shown otherwise. All ± values are less than 1.0 GPa
unless shown differently.

Material Eus G G Gbar G/Gbar

N-N N-R N-45 N-T

0 45 90

DP980-1 211 82 82 76 −3 81 −3%

DP980-2 216 85 85 76 −5 83 −6%
DP980-3 213 84 84 79 −3 83 −3%
DP600-1 210 82 82 76 −3 80 −4%
DP600-2 208 81 81 76 −3 80 −3%
DP600-3 215 82 83 78 −3 82 −4%
HSLA-1 209 81 81 76 −3 80 −3%
HSLA-2 211 82 82 75 −3 80 −4%
HSLA-3 210 82 82 76 −3 80 −4%
IF-1 200 76 76 72 −2 75 −3%
IF-2 196 ± 1 74 75 70 −3 73 −4%
IF-3 198 74 74 71 −2 73 −3%
DP980 Avg. 213 ± 1 84 ± 1 84 ± 1 77 ± 1 −3 82 ± 1 −4%
DP600 Avg. 211 ± 2 82 ± 1 82 ± 1 77 ± 1 −3 81 ± 1 −4%
HSLA Avg. 210 ± 1 82 ± 1 81 ± 0 76 ± 0 −3 80 ± 0 −4%
IF Avg. 198 ± 1 74 ± 1 75 ± 1 71 ± 1 −2 74 ± 1 −3%
Non-IF Avg. 211 ± 1 83 ± 1 82 ± 1 77 ± 1 −3 81 ± 1 −4%
All Avg. 208 ± 3 80 ± 2 81 ± 2 75 ± 2 −3 79 ± 2 −4%

Key:
N-R: ND propagation direction, RD excitation direction.
G: Shear modulus.
Gbar = (G0 − 2 × G45 + G90 )/2.
G = (G0 + 2 × G45 +G90 )/4.
±: Indicate one-half of the standard deviation for repeated tests for single alloy.
For grade averages, ± represents standard deviation among 3 average values, one for each alloy in the grade.

limit is taken to be 1 . Therefore, the range of || for loading and 3. Results and Discussion
reloading legs is identical: 0 <  < |1 − 2 |.
The purpose of the current work is to determine the vari- 3.1. Ultrasonic pulse-echo results (US)
ability of nominally elastic, mechanically-measured properties
across suppliers of a single grade, as well as across grades, The elastic constant measurements and test-to-test scatter for
of automotive steels. With this in mind, and in view of the ultrasonic method are summarized in Table 3. The average
the marked nonlinear stress–strain behavior of steels in this test-to-test scatter is less than ±1 GPa (<0.5%), while the estimated
regime, best-fit linear analysis over substantial stress ranges error of the density measurement used to compute elastic moduli
at the start and end of unloading/reloading legs was selected is approximately 0.5%. The combined scatter is therefore expected
in order to balance accuracy/reproducibility with stress reso- to be approximately ±1% or +/− = 2 GPa. With these scatter values
lution, favoring the former over the latter. That is, substantial in mind, Table 3 provides several summary results:
strain ranges were selected to ensure a sufficient number of
data points to arrive at a representative best-fit line slope
1) The variation of ultrasonically-measured shear moduli for in-
rather than trying to maximize resolution by computing slopes
plane directions is less than 4% (one exceptional material shows
from adjacent data points (thus having correspondingly larger
6%, most are 3% or less) indicating a high degree of in-plane
scatters). The ranges specified in Fig. 4(b) were chosen based
plastic isotropy.
on analysis that showed that the least-squares slopes were
2) All of the ultrasonically measured elastic constants are the same
nearly constant (within the experimental scatter) throughout the
among the suppliers for each grade of material.
range selected, but began to vary significantly for larger stress
3) All of the non-IF steels have the same ultrasonically-measured
ranges.
elastic moduli. The Young’s modulus (N–N direction) is 211 GPa.
Four slopes of the stress–strain curves, E1 –E4 , were computed
4) The IF steels have an ultrasonically-measured Young’s modulus
during unloading and reloading legs, as defined in Fig. 4(b). E1 is the
(N–N) of 198 GPa, approximately 5% lower than the non-IF steels
(normally largest) slope of the unloading stress–strain data over the
first one-third of the unloading stress range, i.e. from  u to 2 u /3,
where  u is the stress at the displacement reversal/start of unload- While all of the materials show substantial in-plane isotropy, the
ing. The range chosen to define E1 , i.e.  u /3, is consistent with the lower through-thickness modulus of the IF steels indicates a differ-
range of linearity observed and used in Sun and Wagoner (2011). E2 ent texture affecting that property. Because only through-thickness
is the (normally smallest) slope of the unloading stress–strain data Young’s moduli can be measured accurately with the ultrasonic
over the last one-quarter of the unloading stress range, i.e. from technique, additional measurements of the in-plane Young’s mod-
 u /4 to 0. This smaller stress range was adopted for the final seg- ulus were undertaken using the resonant frequency method, as
ment of the unloading leg because of the higher curvature observed shown in the next section.
there and the corresponding smaller range of apparent linearity. The US-measured Young’s moduli were unaffected by tensile
Slopes E3 and E4 are corresponding constructions for the reload- deformation. After multi-cycle LUL testing to the uniform elonga-
ing leg of the LUL cycle, as shown in Fig. 4(b). As will be shown tion for each alloy, the average absolute change was 0.3 GPa, while
more clearly later, E1 and E3 are similar while E2 and E4 are similar, the average signed change was 0.01 GPa, both significantly smaller
consistent with Sun and Wagoner’s (2011) observations. than the estimated test-to-test scatter. These results show that:
 Z. Chen et al. / Journal of Materials Processing Technology 227 (2016) 227–243 233

Table 4
Resonant-frequency (RF) measured elastic constants. All units are GPa unless otherwise indicated. All values have ± less than 1 GPa.

Materials ERD ETD GRD G45 G90 G Gbar

DP980-1 201 210 80 82 80 -2 81

DP980-2 222 80
DP980-3 203 82
DP600-1 196 204 81 79 81 2 80
DP600-2 211 84
DP600-3 217 90
HSLA-1 200 215 81 80 82 1.5 81
HSLA-2 202 79
HSLA-3 201 80
IF-1 203 210 82 82 82 0 82
IF 2 2012 84
IF 3 211 83
DP980 Avg. 209 ± 6 81 ± 1
DP600 Avg. 208 ± 6 85 ± 3
HSLA Avg. 201 ± 1 80 ± 1
IF Avg. 209 ± 3 83 ± 1
Non-IF Avg. 206 ± 2 82 ± 1 81
All Average 207 ± 4 210 ± 3 82 ± 2 81 ± 1 81 ± 1 81

Key:
E: Young’s modulus; G: shear modulus; : Poisson Ratio.
± : Indicate one-half of the standard deviation for repeated tests for single alloy.
For repeated tests. For grade averages, ± represents standard deviation among.
3 average values, one for each alloy in the grade.

1) Physical moduli do not change appreciable in the tensile strain in-plane anisotropy determined from US tests is confirmed
ranges employed here. by the RF tests (4% vs. 5%). Both indicate substantial in-plane
2) The test-to-test scatter of ultrasonic modulus measurement for isotropy.
a single specimen is on the order of 0.3 GPa, only about one-third 3) Non-IF and IF steels have the same RD elastic modulus mag-
of the estimated overall uncertainty from density measurement, nitudes from RF measurements. The consistent difference in
thickness measurement, material variability, and so on. these two groups from US testing presumably arises because
of through-thickness texture differences that are not reflected
These results are consistent with the common knowledge that in the in-plane Young’s modulus variation.
physical Young’s moduli are related to texture and that texture
evolves little in the tensile strain range. For example, Luo and Taking both types of physical measurements in the RD into
Ghosh (2003) found that the RF-measured Young’s moduli in the account, if one adopts a standard, uniform physically measured
RD changed less than 1% up to 18% strain. Young’s modulus of 208 GPa ± 6 GPa, nearly all of the measure-
ments in all directions by both methods are accounted for. (The
3.2. Resonant frequency (RF) damping results absolute largest range encompassing all individual measurements
of both types is 196–222 GPa.) As will be seen shortly, the variations
Young’s modulus values measured using the RF method are of mechanically-measured slopes are several times larger than this
shown in Table 4. The RF method allows measurement of the narrow range.
Young’s modulus in directions directly of interest for comparison
with in-plane mechanical tests, but, because it was only possible to 3.3. Mechanical testing results
perform limited tests remotely it was difficult to obtain a complete
estimate of scatter and accuracy. Fig. 5 present summarize the RD mechanical testing results. For
Repeated tests on the same specimens showed that test-to-test comparison, the physically-measured Young’s moduli (EUS and ERF ,
scatter is negligible, but the preparation of the specimens and the where the US measurement is for the N direction and the RF one is
accuracy of the cut shape, its dimensions, and its mass all influence in the R direction) are represented by horizontal lines.
the final modulus value obtained. Taking individual uncertainties The average test-to-test scatter for the mechanically-measured
into account (particular on specimen dimensions and density) pro- moduli in each figure is reported numerically at the upper left
duces an estimated accuracy and scatter of 1–2%. for each material excluding zero pre-strain cases. It corresponds
The RF results shown in Table 4 confer the following summary approximately to the size of the points used to represent each test.
results: As discussed below, the scatter for zero pre-strain cases are larger
than these average values and are shown separately on each graph
1) Taken for all 12 steels, Young’s modulus in the RD is 207 ± 4 GPa. as error bars. The test-to-test scatter of the physically-measured
(These numbers are determined by averaging the averages for moduli is shown on the labels for the horizontal lines; they are
each grade. The differences for individual steels within a grade approximately the width of the lines shown.
are larger, for example DP 980 shows a range of ERF values In order to consistently apply the procedures of Fig. 4(b) for
from 201 GPa to 222 GPa, depending on the supplier, the average determining the loadings slopes E3 , E4 , and Echord at zero pre-
standard deviation is ±6 GPa.) Therefore, except for supplier- strain, it is necessary to identify a value of the material hardness,
to-supplier differences, there is essentially no difference in the equivalent to the unloading stress,  u ×  u would conventionally
Young’s modulus among these grades. be considered to be the yield stress at 0.2% offset. This is the pro-
2) For the 4 steels that were tested in both RD and TD, the aver- cedure followed for the zero pre-strain points presented in Fig. 5.
age difference between TD (higher) and RD (lower) was 10 GPa, However, it is well known that this definition is a convenience,
ranging from 8 to 15 GPa. Thus, the general magnitude of the not a true indication of first plastic flow or hardness. In order
234 Z. Chen et al. / Journal of Materials Processing Technology 227 (2016) 227–243

to allow for other definitions, the scatter range shown on Fig. 5 are consistent with any other part of LUL tests, even when applying
for E values at zero-pre-strain corresponds to a definition of yield a minimal definition of yield stress.
(first plastic flow) of 0.2% (point shown) to 0.02% (limit of error bar A few generalizations are apparent in looking at the mechanical
shown), i.e. an order of magnitude smaller than the conventionally- testing results in the 12 Fig. 5 as a whole:
adopted limit. The results show that all of the initial loading curves
“bend over” earlier and faster than the corresponding ones after
any of the pre-strains used in the current work. Therefore, lower 1) The initial loading and unloading slopes, E1 and E3 , are generally
chord moduli and final moduli are found upon initial loading than indistinguishable from each other at all pre-strains.
2) At zero pre-strain, E3 is equal to the physically-measured
Young’s modulus although the mechanical measurements

Fig. 5. Variation of various slopes/measures of Young’s modulus (ERF , EUS , E1 , E2, E3, E4, Echord ) with plastic pre-strain determined from multi-cycle (solid points) and single-
cycle (open points) LUL tests, ultrasonic and RFDA methods for 12 tested steels: (a) DP980-1 (b) DP980-2 (c) DP980-3 (d) DP600-1 (e) DP600-2 (f) DP600-3 (g) HSLA-1 (h)
HSLA-2 (i) HSLA-3 (j) IF-1 (k) IF-2 (l) IF-3.
 Z. Chen et al. / Journal of Materials Processing Technology 227 (2016) 227–243 235

Fig. 5. (Continued).

during original loading are more scattered than all such mea- strain 0.077 is of 8 GPa, Table 5. E2 and E4 , as well as Echord (which
surements after any nonzero value of pre-strain. may be usefully considered a kind of average between E1 /E3
3) For all nonzero pre-strains tested here, E1 and E3 are nearly con- and E2 /E4 ), show a small but significant decline with pre-strain
stant. Without better resolution at very small pre-strains (less (contrary to E1 and E3 ). By material grade, the average decrease
than 0.02), the critical pre-strain for the transition from an initial, of E2 or E4 over the tested pre-strain range (excluding zero pre-
physical, modulus to a lower, but nearly pre-strain — indepen- strain) is as follows:
dent one, is not well delineated. Thus, the parameters for the
exponential decay equation proposed by Yoshida et al. (2002), DP980: 12 GPa,
Eq. (1), cannot be identified accurately. DP600: 15 GPa,
4) The final unloading and reloading slopes, E2 and E4 , respectively, HSLA: 14 GPa,
differ slightly from each other; the average difference for pre- IF: 3 GPa.
236 Z. Chen et al. / Journal of Materials Processing Technology 227 (2016) 227–243

Table 5
Mechanically-measured slopes from LUL tests at prestrain = 0.077, RD. All units are GPa, unless otherwise indicated (e.g. “%”).

Material E1 E2 E3 E4 E13 E24 E13 and E13 % E24 and E24 % (E13 − E24 )%

DP980-1 181 ± 3 118 ± 0 184 ± 1 109 ± ± 2 183 ± 2 113 ± 1 −3 −2% 9 8% 47%

DP980-2 180 ± 1 135 ± 1 192 ± 1 125 ± 1 186 ± 1 130 ± 1 −12 −7% 10 7% 36%
DP980-3 190 ± 3 136 ± 2 192 ± 3 131 ± 3 191 ± 3 133 ± 3 −2 −1% 6 4% 36%
DP600-1 178 ± 1 124 ± 3 178 ± 3 116 ± 1 178 ± 2 120 ± 2 0 0% 8 7% 39%
DP600-2 183 ± 1 123 ± 2 183 ± 0 115 ± 3 183 ± 1 119 ± 3 0 0% 8 7% 42%
DP600-3 183 ± 2 141 ± 3 182 ± 3 131 ± 2 183 ± 3 136 ± 3 0 0% 10 7% 29%
HSLA-1 188 ± 0 156 ± 2 186 ± 1 145 ± 2 187 ± 1 151 ± 2 2 1% 11 7% 21%
HSLA-2 189 ± 4 155 ± 4 196 ± 3 139 ± 2 192 ± 4 147 ± 3 −6 −3% 16 11% 27%
HSLA-3 185 ± 4 155 ± 1 190 ± 2 142 ± 2 188 ± 3 148 ± ± 2 −5 −3% 12 8% 23%
IF-1 183 ± 3 139 ± 4 176 ± 5 142 ± 2 180 ± 4 140 ± 3 7 4% −2 −2% 24%
IF-2 197 ± 4 146 ± 2 185 ± 2 141 ± 4 191 ± 3 144 ± 3 12 6% 5 3% 29%
IF-3 188 ± 1 138±3 182 ± 5 134 ± 3 185 ± 3 136 ± 3 6 3% 3 3% 30%
Average* 185 ± 3 139 ± 7 185 ± 3 131 ± 6 185 ± 3 135 ± 7 0 0% 8 6% 32%
Avg. DP980 184 ± 3 129 ± 5 189 ± 3 121 ± 5 187 ± 3 125 ± 5 −6 −3% 8 6% 39%
Avg. DP600 181 ± 2 129 ± 5 181 ± 2 121 ± 5 184 ± 2 124 ± 5 −1 0% 7 7% 37%
Avg. HSLA 187 ± 1 155 ± 1 190 ± 3 142 ± 2 184 ± 2 135 ± 2 1 −2% 9 9% 24%
Avg. IF 189 ± 4 141 ± 2 181 ± 3 139 ± 2 189 ± 4 149 ± 2 −3 4% 13 1% 28%

Key:
E13 = (E1 + E3 )/2.
E24 = (E2 + E4 )/2.
E13 and E13 % = E1 − E3 and (E1 − E3 )/E13 × 100%.
E24 and E24 % = E2 − E4 and (E2 − E4 )/E24 × 100%.
(E13 − E24 )% = (E13 − E24 )/(E13 + E24 )/2.

These results can be summarized by noting that the change of resent the extremes in terms of differences between RD and TD
slope of the stress–strain response increases with pre-strain and behavior with E/SD of 0.1 and 2.8, respectively.
with flow stress of the material, thus the initial loading or unloading Fig. 6 compare the RD and TD results for DP980-1 and DP980-2
occurs near the physical modulus but the slope is reduced more for for the various pre-strains. Even for the materials exhibiting signif-
pre-straining and stronger materials. icant differences between RD and TD, the differences were only2-3
times that of the test-to-test scatter. Therefore, any differences
5) By approximately an order of magnitude, the dominant vari- between RD and TD for the 12 tested steels can be neglected with
ation of slope is the change of from the initial loading or little error.
unloading to the final loading or unloading, that is, between As shown in Fig. 5, the loading and unloading legs of an LUL test
E1 /E3 and E2 /E4 . It is natural that this change is larger for larger have similar slopes for all steels tested here. That is, E1 is nearly
stress/strain ranges, even if the shape of the curve in the initial identical to E3 , while E2 and E4 differ only modestly — by ∼8 GPa as
stress/strain range remains the same. At a pre-strain of 0.08, the compared with ∼50 GPa difference between E1 /E3 and E2 /E4 . Figs.
average difference between E1 /E3 and E2 /E4 is 51 GPa, depend- 7 and 8 compare in more detail the loading and unloading legs in
ing primarily on the strength of the material, i.e. the flow stress terms of the  − , Fig. 7, or ET − , Fig. 8. DP980-1, Figs. 7 and
before unloading, or the yield stress at zero pre-strain, Table 5. 8, is representative of the materials with the most similar unloading
6) There is no distinguishable difference between the measured and reloading slopes as shown in Table 5. DP980-2, Figs. 7 and 8,
stress–strain responses for single-cycle tests and corresponding is the material tested here with the most dissimilar unloading and
multi-cycle tests. This can be verified in Fig. 5 by comparing the reloading slopes as shown in Table 5.
trends for the multi-cycle tests (solid points) and the single- DP 980-1 is typical of most of the alloys, which exhibit consis-
cycle tests at similar pre-strains (open points). This observation tent loading and unloading shapes while DP980-2 exhibits the most
confirms that of Sun and Wagoner (2011). difference between loading and unloading. Fig. 7(a) and (b) present
stress–strain, i.e. (), whereas Fig. 8(a) and (b) present the
The foregoing results represent the RD tests only. Careful exami- same information in terms of E(), where E represents the slope
nation shows that the differences between RD and TD tests, Table 6, of the () data. The slopes were calculated from the ()
are small, showing all the same trends and justifying the same con- data points as follows: the incremental tangent modulus ET was
clusions. For brevity, only the TD results for pre-strain 0.077 are obtained by taking a moving window of 6 adjacent data points and
presented in Table 6. With three materials excepted, namely DP fitting a least-squares line through those points, the slope of which
980-23 , DP600-2, and HSLA-2, the mechanically-measured RD and is ET corresponding to the average  for those points.
TD moduli are the same within the combined test-to-test scatter. Figs. 7 and 8 show that, to a close approximation, the load-
The last column in Table 6 represents the average (among E1 –E4 ) ing and unloading legs show the same development of curvature,
difference between moduli measured in RD and TD, E being pre- that is, they can be represented by the same functional depen-
sented as a fraction of the average combined scatter of RD and TD dence, although the precision of the similarity varies from material
testing for the same four modulus measures. Only the 3 materials to material. In view of the reduced test-to-test scatter of unload-
(in red/italics) show average differences more than 50% greater4 ing legs as compared with reloading ones, the unloading legs are
than the test-to-test scatters. Note that DP980-1 and DP980-2 rep- preferable for representing the material behavior in a meaningful
way.
Fig. 9 present the effect of pre-strain on the details of ()
and Et () for DP 980-1 as a representative of most of the alloys
3
DP980-2 is noted as exceptional in several measurements among the materials
tested here. There is no discernible difference in the shape of curves
tested here, including the highest RF-measured modulus by a wide margin, 222 GPa,
as shown in Table 4.
(Fig. 9(a)) or the slopes (Fig. 9(b)) among the various pre-strains.
4
The 50% threshold value is an arbitrary one adopted in view of the limited The only first-order difference is simply the larger stress and strain
number of tests (2 for each case) used to establish test-to-test scatter values.
 Z. Chen et al. / Journal of Materials Processing Technology 227 (2016) 227–243 237

Table 6
Difference between mechanically measured slopes, TD-RD, for LUL tests at prestrain = 0.077. All units are GPa.

Material E1 E2 E3 E4 S.D., RD S.D., TD Combined S.D. Avg. TD-RD (TD-RD)/combined

DP980-1 −1 2 2 2 4 5 10 1 0
DP980-2 29 17 24 17 3 5 8 23 3
DP980-3 -1 3 6 4 3 3 6 3 0
DP600-1 -2 7 11 1 2 3 5 5 1
DP600-2 1 6 14 5 2 5 7 7 1
DP600-3 4 10 9 13 3 6 8 8 1
HSLA-1 −9 3 17 −8 4 6 10 4 0
HSLA-2 13 9 13 12 6 9 14 12 1
HSLA-3 6 7 9 6 6 13 20 7 0
IF-1 9 19 38 13 4 15 19 22 1
IF-2 -1 12 9 13 4 11 15 7 0
IF-3 6 21 15 14 2 7 9 14 2
Average 5 10 14 8 3 6 9 9 1
SD 10 6 9 7 3 4 7 8 1
Avg. DP980 9 7 11 7 3 3 6 9 1
SD 17 8 11 8 2 2 3 12 4
Avg. DP600 1 8 11 6 2 3 5 7 1
SD 3 2 2 6 3 3 6 3 0
Avg. HSLA 3 6 13 4 4 8 12 8 1
SD 11 3 4 10 4 8 11 6 1
Avg. IF 5 17 21 13 4 10 14 14 1
SD 5 5 16 1 3 6 9 8 1

Key:
S.D., RD: Standard deviation among the 3 average moduli for each alloy in rolling direction.
S.D., TD: Standard deviation among the 3 average moduli for each alloy in transverse direction.
Combined S.D. = (S.D., RD) + (S.D., TD).
Avg. TD-RD = (S.D., TD) − (S.D., RD).
(TD-RD)/combined = (Avg. TD-RD)/(combined S.D.).

Fig. 7. Comparison of loading and unloading legs in terms of  vs.  as defined
Fig. 6. Comparison of RD and TD variations of various slopes/measures of Young’s
in the text: (a) DP 980-1 (b) DP 980-2.
modulus (ERF , EUS , E1 , E2, E3, E4, Echord ) with plastic pre-strain determined from multi-
cycle (solid points) and single-cycle (open points) LUL tests, ultrasonic and RFDA
methods for 2 tested steels: (a) DP 980-1 (b) DP 980-2.
238 Z. Chen et al. / Journal of Materials Processing Technology 227 (2016) 227–243

Fig. 9. The role of pre-strain on nominally elastic unloading and reloading of DP

980-1: (a)  vs. , (b) ET vs. .
Fig. 8. Comparison of loading and unloading legs in terms of ET vs. : (a) DP 980-1
(b) DP 980-2.
strains, with very little deviation or experimental scatter. This is
astounding because each point and the corresponding error bars
range for the larger pre-strains, which produces a larger reduction represent the full range of scatter6 for three materials from three
of slope/modulus by continuing the curved, common shape. suppliers, each using proprietary processing and chemistry. Fur-
Fig. 10 are similar to Fig. 9, except that they represent the mate- thermore, the effects of changing the strength of the material (i.e.
rial here with largest change of () and Et () with pre-strain,  u ) is the same whether the change is accomplished by strain hard-
i.e. DP980-2. Even in this case, the similarity is striking such that ening (i.e. dislocation multiplication) or microstructural hardening
the differences could confidently be ignored without a significant (e.g. increasing volume fractions of martensite among DP grades).
penalty in accuracy. In view of Figs. 11(b) and 9(a), which shows similar ()
Detailed plots as presented above can obscure the simplifying, except for the range, the other single characterization parame-
overall trends of the so-called “modulus effect.” For clarity and per- ter of interest, Echord , would be expected to evolve similarly for
spective, it is useful to adopt a single parameter describing the material having similar strengths. In fact, this is demonstrated in
magnitude of the stress–strain nonlinearity in the nominally elastic Fig. 12(a). The two low-strength alloys IF and HSLA, with ultimate
range. For this purpose, a parameter, denoted by ı here, is defined as tensile strengths of 278–300 MPa and 449–485 MPa, respectively
shown in Fig. 11(a).5 As for other hysteretic phenomena, the width show identical, but lesser reductions of chord modulus with pre-
of the loop, along with the unloading stress  u , is related to the strain as compared with DP600 and DP980, with ultimate tensile
energy dissipated. Fig. 11(b) shows the loop widths for all of the strengths of 636–693 MPa and 960–1036 MPa, which also show
multi-cycle tests performed here, as a function of unloading stress, identical behavior. The equations and best-fit parameters shown
which is a proxy for the material’s current strength or hardness. on Fig. 12(a) are of the Yoshida type, Eq. (1) as well as the linear
Fig. 11(b) shows that there is a “universal,” single curve (a type. It shows that the standard deviations by the Yoshida and lin-
nonlinear fitting equation shown on Fig. 11b, with parameters ear fit are the same compared with the experimental data, that
a = 14970 MPa−0.74 , b = −0.74, E0 = 208000 MPa) representing all of is, <S.D.> = 3 GPa for HSLA and IF steels and <S.D.> = 2 GPa for DP
the multi-cycle unloading experiments for all materials and all pre- 980 and DP 600 steels. Therefore, both of Yoshida and linear fits
work well to capture the variation of chord modulus as prestrains,

5
While Fig. 11(a) shows two components of ␦, one related to deviations from the
6
chord modulus during unloading and one during reloading, there is no significant The scatter bars extend in two directions because the unloading stresses differ
difference with that “summed difference” and a simple “maximum width” of the for the suppliers of a given grade, depending on its tensile hardening behavior, as
overall loop at a single stress level. well as in the loop widths varying from material to material.
 Z. Chen et al. / Journal of Materials Processing Technology 227 (2016) 227–243 239

Fig. 11. Single-parameter characterization unloading-loading cycle magnitudes in

terms of the loop width ı, from multi-cycle LUL tests: (a) definition of ı, (b) variation
Fig. 10. The role of pre-strain on nominally elastic unloading and reloading of DP of ı with material hardness (unloading stress,  u ) for twelve steels at various pre-
980-2: (a)  vs. , (b) ET vs. . strains vs. curve fit.

excluding the case of p = 0. In Fig. 12(b), the initial slopes of E13

to the role of  u , such that they may be ignored while still taking
(defined as an average of E1 and E3 for each grade) and the final
into account the principal features of the “modulus effect”. Exam-
slopes E24 (defined as an average of E2 and E4 for each grade) are
ples of such variables include the number of cycles, the direction of
shown and fitted well by the Yoshida type (Eq. (1)). Thus, it also indi-
testing, loading vs. unloading.
cates that E1 value is almost identical with E3 (<S.D.> = 6 GPa), while
In view of the consistency of the results among steel grades and
E2 and E4 have a medium difference (<S.D.> = 30 GPa). In Fig. 12(c),
producers, it may not be necessary to perform any tests in order
(E13 –E24 ) for each grade at difference prestrains are presented. It is
to take the modulus effect into account profitably, thus greatly
interesting to find that the difference between E13 and E24 increases
improving simulations involving small strain behavior, such as for
as the strength levels. For example, DP 980 shows the obvious high-
springback and pillar crush. Instead, the results of the current work
est values while IF or HSLA give the lowest ones. It is hard to tell the
suggest that this approach likely extends to any bcc steel8 by know-
difference between IF and HSLA due to the scatters resulting from
ing only the flow stress of the material equivalent to  u here. Such a
its relatively low stress levels.
procedure would confer the majority of the improvements in such
Taken as a whole, the foregoing results convey a surprisingly
simulations without invoking a corresponding penalty in terms of
uniform and consistent picture of the modulus effect in a wide
inconvenience or cost.
variety of steels exhibiting a range of strengths and ductility and
As noted earlier the mechanical modulus measurements are
originating with various producers.7 The only first-order determi-
particularly consistent after any non-zero pre-strain and partic-
nant of the magnitude of deviations from handbook linear elastic
ularly for the unloading leg. For these reasons, if testing is to be
behavior was identified as the current strength of the material, here
undertaken to get the best accuracy for a given application it is
characterized by the unloading stress,  u . The other anticipated
recommended to use the unloading leg of an LUL test after some
variables have no effect or have effects that are clearly second-order
pre-strain (2% pre-strain is sufficient, but the lower limit of suf-
ficiency was not probed here). The original loading suffers from
7
Although it is beyond the ambit of this paper, it is interesting to note that the
surprising uniformity (strength-only-dependence to first order) of the nonlinearity
8
represented by the current results points to a single governing mechanism in spite of While the data in the present work applies to only the 4 alloys tested here,
the various hardening mechanisms represented. This seems to point to a mechanism they represent a wide range of hardening mechanisms, strengths, and ductilities.
of elastic interactions among dislocations and various obstacles, including forest Whether the results here can be extrapolated to other alloy groups, including non-
dislocations. While DP steels have strong stress concentrators at a coarse scale that ferrous ones, fcc-matrix steels, deformation-transforming steels, or twinning steels
might provide a continuum source of nonlinearity, there is no corresponding feature remains to be seen. Such applications are being studied now and will be the subject
in IF or HSLA steels. of a subsequent paper.
240 Z. Chen et al. / Journal of Materials Processing Technology 227 (2016) 227–243

Fig. 12. Summary plots of the variation of mechanical modulus measures (slopes) with pre-strain for multi-cycle tests for 12 steels: (a) chord modulus and analytical
representations thereof, (b) average unloading and reloading slopes, initial (E13 ) and final (E24 ), with scatter bars shown instead of individual data. (c) Difference between
initial and final slopes, E13 –E24 . E E13 and E24 are defined in the key for Table 5.

Table 7
Comparisons of three overall modulus measures, RD.

Material Ultrasonic RF Mechanical, Linear Fit Comparisons

Eus <S.D.> ERF <S.D.> E3 * ( = 0) <S.D.> E3 * ( = 0.077) <S.D.> E3 * ( = 0) E3 * ( = 0.077)

GPa GPa GPa GPa GPa GPa GPa GPa Eus /ERF E3 /Eus E3 /Eus E3 /ERF

DP980-1 211 0.8 204 0.2 206 4 184 1 104% 97% 87% 90%
DP980-2 216 0.6 222 0.1 203 3 192 2 97% 94% 89% 87%
DP980-3 213 1.2 203 0.0 202 1 192 5 105% 95% 90% 95%
DP600-1 210 0.5 201 0.0 201 7 178 6 105% 96% 85% 89%
DP600-2 208 0.6 211 0.1 198 1 183 0 99% 95% 88% 87%
DP600-3 215 0.1 217 0.0 200 3 182 6 99% 93% 85% 84%
HSLA-1 209 1.7 206 0.0 205 2 186 2 102% 98% 89% 90%
HSLA-2 211 0.6 202 0.0 210 7 196 6 104% 100% 93% 97%
HSLA-3 210 0.2 201 0.0 207 8 190 3 105% 98% 90% 94%
IF-1 200 1.4 208 0.1 200 12 176 9 96% 100% 88% 84%
IF-2 196 2.3 212 0.1 208.8 15 185 3 92% 107% 95% 87%
IF-3 198 0.3 211 0.1 201.6 15 182 9 94% 102% 92% 86%
Average 208 0.9 208 0.1 204 7 185 4 100% 98% 89% 89%
S.D. 7 0.7 7 0.1 4 5 6 3 5% 4% 3% 4%
DP980 Avg. 213 1 210 204 2 189 2 102% 95% 89% 90%
DP980 S.D. 2 0 11 2 1 5 2 4% 2% 2% 4%
DP600 Avg. 211 0 210 200 4 181 4 101% 95% 86% 86%
DP600 S.D. 3 0 8 2 3 3 3 3% 1% 2% 2%
HSLA Avg. 210 1 203 207 6 190 4 103% 99% 91% 94%
HSLA S.D. 1 1 3 3 3 5 2 2% 1% 2% 3%
IF Avg. 198 1 210 203 14 181 7 94% 103% 92% 86%
IF S.D. 2 1 2 5 2 5 3 2% 3% 3% 2%
 Z. Chen et al. / Journal of Materials Processing Technology 227 (2016) 227–243 241

many unavoidable sources of scatter9 including specimen prepara-

tion, extensometer mounting and seating, specimen flatness, lash
in grips and load train, and unknown starting condition of the sheet
as provided.
However, it should also be noted that the initial loading mod-
ulus, even taking scatter into account, does appear to be higher
than all of the subsequent, non-zero pre-strain tests. In fact, within
the experimental scatter, the initial mechanically-measured load-
ing modulus is the same as the physically-measured moduli. This
is illustrated in summary Table 7. The mechanically-measured
Young’s modulus for loading (E1 ) is on average 100% of the
RF-measured modulus within 3% variation and is 98% of the
US-measured modulus within 2% variation. However, after 8% pre-
strain, Table 7 shows that the mechanically-measured moduli are
now reduced to 89% of the RF-measured modulus within 4% varia-
tion and is 89% of the US-measured modulus within 3% variation.
In order to provide a practical example of an application of the
current results and recommendations, refer again to Fig. 2. Fol- Fig. A1. Compression testing of automotive structural pillars.

lowing the recommendations that resulted from the current work,

the QPE model was fit using the data from the unloading leg of Acknowledgments
LUL tests of the Toyota pillar material after a pre-strain of 13%,
rather than from the initial loading as was done previously. The This work was supported primarily by SMDI, the Steel Marketing
results, represented by the green short-dashed lines of Fig. 2, show and Development Institute, with materials provided through their
remarkable agreement with the experiments, reducing the stan- auspices by ArcelorMittal, Nucor, ThyssenKrupp, and U. S. Steel.
dard deviation to within 4 kN, thus verifying the effectiveness of Initial support was provided by the Toyota Research Institute. The
the methods advocated. authors are grateful to Joris Bracke for RFDA modulus measure-
ments performed at IMCE n.v. in Belgium. Ken Kushner and Ross
Baldwin are gratefully acknowledged for assistance with mechani-
4. Conclusions cal testing in the Department of Materials Science and Engineering
at The Ohio State University. Prof. Dayong Li is also gratefully
Computer-controlled tensile tests and data acquisition were acknowledged for helping us with discussions and manuscript
used in conjunction with a Class A extensometer, special care in proofreading.
special preparation, and sensitive analytical techniques to measure
the small stress–strain response under reverse-loading conditions Appendix A. Toyota structural pillar testing
of 12 steels representing 4 diverse grades and 4 producers. These
mechanically-measured “moduli” were compared with ultrasoni- 10. Toyota structural pillar testing
cally and resonant-frequency measured ones.
The following conclusions were reached: Sheet-formed structural pillars, 600 mm long and with cross-
sections as shown in Fig. 1, were press- formed and spot welded
(40 mm spacing between 4 and 5 mm diameter welds) from 0.8 mm
1) Within a grade, there are no significant differences among pro- — thick SPC 44010 steel sheet at the Toyota Research Institute
ducers. (Toyota 2009). After forming and welding, the pillars were com-
2) Among grades, the only first-order determinant of nonlinearity pression tested and the force-displacement curves recorded as
is material strength. Other effects are nil or second-order: cyclic points in Fig. 2.11 Fig. A1 shows the pillars being compression tested.
deformation, testing direction, unloading vs. loading. The deformation of the specimens was simulated originally at
3) In view of the above, it is possible to use the principles identi- Toyota in 2009–2012 using LS-Dyna and standard industrial tech-
fied here to effectively represent the nonlinear behavior of steels niques to identify the material behavior. The material model was a
without testing. standard elastic-plastic one with strain hardening based on a ten-
4) The initial loading modulus at zero pre-strain is the same as the sile test of the original sheet material and Young’s modulus from
physically-measured ones. handbook values. The strain hardening was of the Swift type:
5) The initial modulus at nonzero pre-strain is lower than the
physically-measured ones but it changes little with pre-strain  = 782MPa(0.008 + 0.184
p ) (A.1)
beyond some critical pre-strain. The critical pre-strain was not where p is the plastic strain. Table A1 presents various other con-
determined precisely, but is less than 2%. stitutive parameters for each of the models used. The discrepancy
6) The scatter in testing results at zero pre-strain is much greater seen in Fig. 2 was noted between the simulations and experiment.
than after pre-strain. Therefore, it is recommended to use an Extended discussions of possible origins of the discrepancy were
unloading leg after pre-strain to determine the nonlinear behav- considered and discussed among Toyota and SMDI personnel and
ior. This method was demonstrated to be effective for one members.
example case.

10
SPC440 is a JIS designation for a Steel Plate Cold-rolled with an ultimate tensile
strength of 440 MPa.
9 11
The statement here refers to our laboratory testing, which was aimed at obtain- The raw displacements were derived and recorded from the machine crosshead
ing the best possible stress–strain behavior at low strains. Typical industrial practice, motion. These raw displacements were corrected for machine compliance by using
i.e. using lower-accuracy extensometers, stamped specimens, and settings aimed at a measured a 0.4 mm difference between specimen length change and machine-
reduced tensile testing times, would be expected to introduce much more scatter in recorded length change at the maximum load, and assuming a linear machine
the initial loading case, thus making the conclusion presented here more compelling. compliance with load.
242 Z. Chen et al. / Journal of Materials Processing Technology 227 (2016) 227–243

Table A1
Best-fit constitutive parameters of alternate models used for simulation.

Parameters (unit) Eq. number* Elastic–plastic Modified QPE1 Modified QPE2

E0 (GPa) 11 208 183 156

 (Poisson’s ratio) – 0.3 0.3 0.3
E1 (GPa) 11 – 104.7 120
b 14 – 407.2 310.1
A1 (MPa) 14 – 105.84 105.84
B1 14 – 0.681 0.681
D1 14 – 4.1 4.1

Key:
*
All parameter labels and equation numbers refer to Sun and Wagoner (2011) nomenclature.

Fig. A2. Baseline Abaqus FE model with boundary conditions and coordinate system used for simulation of automotive structural pillar.

The Wagoner group undertook an extensive examination of the

FE model starting in 2013 and performed a sensitivity study to iden-
tify possible sources of error in the FE model capable of producing
the differences with the experiments shown in Fig. 2. A similar FE
model using an identical material model was constructed for use
with Abaqus Standard, as illustrated in Fig. A2. The results were
generally indistinguishable from the Toyota/LS-Dyna ones.
Several geometric variations of the pillar were then introduced.
Among these, altered spot welds were introduced (up to half of
them some being removed altogether or all of them weakened).
The perfect uniformity along the length of the pillar was altered by
twisting along the longitudinal axes by 10 degrees and by introduc-
ing shallow dimples/wrinkles in critical areas. Non-parallel platen
loading was introduced, again by rotating one platen 1◦ . None of
these modified FE models altered the basic simulation results sig-
nificantly up to the maximum load point. For example, Fig. A3
shows six typical perturbations introduced into the modeling, and Fig. A3. Experimental data vs. FE results for the variously perturbed FE models.
it turned out that these had insignificant effect on FE results at all.
A modified Sun and Wagoner QPE model (“Modified QPE1” in
Table A1) was fit to the tensile data for the initial loading, with
parameters obtained as shown in Table A1. The modification was to In view of the current results and conclusions, which include
incorporate a measured initial slope of 183 GPa, which differs from using an unloading leg after significant pre-strain for a more accu-
the handbook value of 208 GPa as adopted by Sun and Wagoner. A rate measurement of nominally elastic material behavior, a new
linear region extending for ∼1/3 of the flow stress was maintained, determination of QPE parameters (“Modified QPE2”) was made
according to the procedure adopted by Sun and Wagoner, and the from unloading after a pre-strain of 0.13. The resulting material
remaining parameters were fit using the least-squares method. model, Table A1, was then used to perform a new simulation, with
The QPE simulation results (red long-dashed curve in Fig. 2) were results as shown in the green short-dashed line in Fig. 2. The sim-
substantially improved with respect to the experiments. The sim- ulation errors were each reduced by 81% relative to elastic–plastic
ulation errors were reduced by 55–56%. and by 58% additional relative to the Modified QPE1.
 Z. Chen et al. / Journal of Materials Processing Technology 227 (2016) 227–243 243

References Morestin, F., Boivin, M., Silva, C., 1996. Elasto plastic formulation using a kinematic
hardening model for springback analysis in sheet metal forming. J. Mater.
Agnew, S., Weertman, J., 1998. The influence of texture on the elastic properties of Process. Technol. 56, 619–630.
ultrafine-grain copper. Mater. Sci. Eng. A 242, 174–180. Olympus, 2011. Olympus, 38DL Plus User’s Manual, 201. Olympus Scientific
Andar, M.O., Kuwabara, T., Yonemura, S., Uenishi, A., 2010. Elastic – Plastic and Solutions Americas, 48 Woerd Avenue, Waltham, MA, pp. 2453 http://www.
Inelastic Characteristics of High Strength Steel Sheets under Biaxial Loading olympus-ims.com/en/38dl-plus/
and Unloading, vol. 50, 4. ISIJ International, pp. 613–619. Opbroek, E.G. 2009. World auto steel version 4.0, pp. 44–45.
Bollen, B., 2006. MANUAL: RFDA System 23, RFDA-MF, Version 6.3.1. IMCE nv, Pavlina, E.J., Levy, B.S., Van Tyne, C.J., 2008. The unloading modulus of ADKQ steel
Belgien. after uniaxial and near plane-strain plastic deformation. Int. J. Mod. Phys. B vol.
Bracke, J. IMCE nv, Slingerweg 52, B-3600 Genk, Belgium. <http://www.imce.eu/>. 22 (31 & 32), 6070–6075.
Cleveland, R.M., Ghosh, A.K., 2002. Inelastic effects on springback in metals. Int. J. Sever, T., 2011. Impulse excitation technique, IET. <https://www.google.com/
Plast. 18, 769–785. search?newwindow=1&safe=strict&espv=2&q=sever+resonant
Fei, D.Y., Hodgson, P., 2006. Experimental and numerical studies of springback in +frequency&oq=sever+resonant+freque&gs l=serp.3.0.33i21.7780.12930.0.
air v-bending process for cold rolled TRIP steels. Nucl. Eng. Des. 236, 13871.21.20.1.0.0.0.277.2868.0j11j5.16.0..0.. 1c. 1. 64. serp. 9. 12.2066.
1847–1851. erFlt7zk6AA/>.
Gandhi, U., Wagoner, R.H., 2013. Investigation of non-linearity and anisotropy in SMDI: The Steel Market Development Institute, 2000. Town Center, Suite 320.
elastic modulus of Steel. In: Auto/Steel Partnership Meeting, March 19, 2013 Southfield, MI, pp. 48075 http://www.smdisteel.org/
Southfield, Michigan. Sun, L., Wagoner, R.H., 2011. Complex unloading behavior: nature of the
Lee, M.G., Kim, C., Pavlina, E.J., Barlat, F., 2011. Advances in sheet forming-materials deformation and its consistent constitutive representation. Int. J. Plast. 27,
modeling, numerical simulation, and press technologies. J. Manuf. Sci. Eng. 1126–1144.
Trans. ASME 133, 61001. Toyota Research Institute, 2009. Toyota Technical Center. Ann Arbor, MI.
Lems, W., 1963. The change of Young’s modulus after deformation at low Wagoner, R.H., Wang, J.F., Li, M., 2006. “Springback,” Chapter in ASM Handbook.
temperature and its recovery. Ph.D Dissertation. Delf, 1963. 14B: Metalworking. Sheet Forming, pp, pp. 733–755.
Luo, L.M., Ghosh, A.K., 2003. Elastic and inelastic recovery after plastic deformation Yang, M., Akiyama, Y., Sasaki, T., 2004. Evaluation of change in material properties
of DQSK steel sheet. J. Eng. Mater. Technol. 125, 237–246. due to plastic deformation. J. Mater. Process. Technol. 151, 232–236.
Mitutoyo, 2008. Coolant Proof Micrometer Quantumike, Model: No. 293-185, 630. Yoshida, F., Uemori, T., Fujiwara, K., 2002. Elastic-plastic behavior of steel sheets
Mitutoyo America Corporation, Aurora, Illinois, pp. 978–6495 under in-plane cyclic tension –compression at large strain. Int. J. Plast. 18,
www.mituyoto.com. 659–663.
Morestin, F., Boivin, M., 1996. On the necessity of taking into account the variation Zang, S.L., Cheng, G., Wei, G.J., Chen, F., Dong, W., Zhang, K., 2006. A new model to
in the Young modulus with plastic strain in elastic-plastic software. Nucl. Eng. describe effect of plastic deformation on elastic modulus of aluminum alloy.
Des., 107–116. Trans. Nonferrous Met. Soc. China 16, 1314–1318.