You are on page 1of 11

JMEPEG ASM International

DOI: 10.1007/s11665-014-1081-x 1059-9495/$19.00

Tensile Behavior of T91 Steel Over a Wide Range


of Temperatures and Strain-Rate Up To 104 s1
M. Scapin, L. Peroni, C. Fichera, and A. Cambriani

(Submitted February 7, 2014; in revised form April 2, 2014)

High chromium ferritic/martensitic steel T91 (9% Cr, 1% Mo), on account of its radiation resistance, is a
candidate material for nuclear reactor applications. Its joining by an impact method to create a cold joint is
tested in the realm of scoping tests toward the safe operation of nuclear fuels, encapsulated in representative
T91 materials. Hitherto, T91 mechanical characterization at high strain rates is relatively unknown, par-
ticularly, in relation to impact joining and also to nuclear accidents. In this study, the mechanical char-
acterization of T91 steel was performed in tension by varying the strain-rate (1023 up to 104 s21) and
temperature (20-800°C) on dog-bone specimens, using standard testing machines or Hopkinson Bar ap-
parati. As expected, the material is both temperature and strain-rate sensitive and different sets of
parameters for the Johnson-Cook strength model were extracted via a numerical inverse procedure, in
order to obtain the most suitable set to be used in this field of applications.

T91 shows superior radiation damage and swelling tolerance


Keywords high strain-rate, high temperature, Hopkinson Bar,
inverse method, Johnson-Cook, nuclear applications, as well as good resistance to high temperature creep and
solid-state welding, T91 steel corrosion (Ref 12). Today T91 is mainly used in conventional
fossil fuel power plants components such as boiler super
heaters and re-heaters tubes or furnace tubes for petrochemical
units where resistance to high temperature creep is required.
A key issue for the technological application envisaged for
1. Introduction T91 in the nuclear field is its joining and welding techniques. In
fact, joining of T91 with conventional fusion welding processes
The investigation of the materials behavior at high strain- such as TIG or laser welding requires the application of a Post
rates is an important aspect in different design engineering Welding Heat Treatment (PWHT) to retrieve its original
problems. In this sense, many studies were performed in the characteristics and to obtain strong and ductile joints. To avoid
automotive field for crashworthiness applications and aviation PWHT, non-conventional solid-state joining techniques such as
industries (see e.g., Ref 1-4). The knowledge of the material explosive welding, Electro Magnetic Pulse Technology
behavior at high strain-rates and even at high temperatures is (EMPT), friction stir welding, and diffusion bonding have
strongly related to design and safety analysis also in nuclear been investigated in the framework of the 7th Euratom
fields, as, e.g., reported in Ref 5-11. Aim of this work is the Framework Program research collaborative project ‘‘Genera-
investigation of the tensile behavior of high chromium ferritic/ tion IV and Transmutation Materials’’ (GETMAT) (Ref 13). In
martensitic steel T91 (9% Cr, 1% Mo), which, among the particular, explosive welding and EMPT processes create a
alternative materials for cladding, is currently being considered metallurgical bond by means of a high velocity impact between
for operating conditions beyond current nuclear industry two metal parts that, depending on the process and on the
experience. The role of the cladding is key to the safe operation relevant setup, are deformed and accelerated by electromag-
of nuclear reactors. AIM1 austenitic steels have been favored netic forces or explosive energy (Ref 14, 15). The JRC-ITU is
until now, but other alternatives [ferritic/martensitic or oxide pursuing a cold welding route, based on EMPT. In the EMPT
dispersion strengthened (ODS) steels] are considered as they process, the plastic deformation, acceleration, and subsequent
exhibit better resistance to neutron damage. Although the JRC- impact occur in a time frame of tenths of a microsecond,
ITUÕs programmes focus largely on fuel safety, it was involving deformations that result in strain-rates close to the
recognized that experimental safety performance tests in the 103-104 s1 range. Similar orders of magnitudes are encoun-
future will need to be performed with representative cladding tered in explosive welding. In order to optimize the process
materials, necessitating an investigation of suitable welding parameters, numerical modeling is extensively used in both
technologies. techniques by means of research and commercially available
numerical simulation packages. The implementation of the
correct material strength model into the numerical codes,
M. Scapin, L. Peroni, and C. Fichera Department of Mechanical and accounting for material behavior at high strain rates, is therefore
Aerospace Engineering, Politecnico di Torino, Corso Duca degli
Abruzzi, 24, 10129 Torino, Italy; and A. Cambriani, European
of paramount importance for the reliability of the model.
Commission, Directorate General Joint Research Centre, Institute for Today, knowledge about the behavior of T91 material at
Transuranium Elements (JRC-ITU), P.O. Box 2340, 76125 Karlsruhe, high strain rates is scarce if existent at all. To this aim,
Germany. Contact e-mail: martina.scapin@polito.it. mechanical characterization of T91 at high strain rates and

Journal of Materials Engineering and Performance


derivation of a reliable material strength model are a necessary In this work, the mechanical characterization of T91 steel
step to allow the application of valid numerical models for was performed under static and dynamic loading conditions at
parameter optimization of joining processes such as EMPT and different temperatures. The experimental tests were performed
explosive welding. in tensile loading conditions on small dog-bone specimens,
T91 mechanical characterization at high strain rates could machined from bar-formed samples. A series of tests at room
also bring valuable support to safety analysis for the future temperature were performed at different strain-rates in order to
reactor systems, especially when transient and accident reactor obtain information on the strain-rate sensitivity of the material.
conditions are evaluated. Rapid transients and severe accidents The nominal strain-rates of the tests were 103, 101, 101, 103,
for a nuclear reactors using T91 as fuel cladding or in-core and 104 s1. In addition, a series of tests at different
structural material could result in dynamic loading of the temperatures in static loading conditions were performed to
structure within a timescale of few milliseconds, with a gain information on the thermal softening of the material. The
consequent abrupt release of mechanical energy to the reactor nominal temperatures of the tests were 20, 200, 400, 600, and
vessel, in-core piping, and components (Ref 16). The evolution 800C. The specimen was heated using an induction coil
and consequences of such accidents involving rapid transients system especially designed to concentrate the heat flux in the
are generally assessed by dedicated numerical codes which gage length. The specimen temperature was monitored using
must account for the response of the core structural material to thermocouples directly welded on the specimen surface.
the relevant loading conditions. Therefore, knowledge about Finally, the experimental data were used in an optimization
T91Õs response to rapid loading is necessary to update the procedure based on a numerical inverse method (presented in
computer code material properties database for conducting detail in section 5) to extract the material model parameter of
safety analysis evaluations. the J-C model. The numerical simulations were performed in
The experimental determination of the relationship between LS-DYNA (Ref 24) and the optimization was managed by the
stress and strain for a material necessitates a test campaign in software LS-OPT (Ref 25).
which different loading conditions are investigated. The final
goal is the identification of work hardening, strain-rate
sensitivity, and thermal softening parameters to correctly
reproduce the experimental material response with a specific
2. Material and Applications
material model. If the material model is empirical, the model
parameters are usually obtained by fitting the experimental data The nominal composition of commercial T91 steel accord-
analytically. With this standard approach, the quality of the ing to ASTM standards is shown in Table 1, and the T91 quasi-
results could be affected by geometrical effects that lead to non- static mechanical properties at room temperature are summa-
uniform stress-strain field within the specimen, and thermo- rized in Table 2.
mechanical coupling in case of high strain-rates, when the Heat treatment conditions in the production of T91 include
thermal softening effects become more relevant. In contrast, a normalization at 1040C, cooling in air and tempering at 730C
numerical inverse method is more useful to extract material for 1 h to allow homogenous carbide precipitation within the
strength parameters from experimental results in all the cases in martensitic structure. It is the carbide precipitation that provides
which the stress and strain fields are not correctly described or improved creep rupture strength. The typical microstructure of
predictable with an analytical model. The use of a numerical these materials is that of tempered martensite (Ref 12).
inverse method for the identification of material parameters is A metallographic section of the T91 material used for the
now widespread thanks to the large computing power available tests is showed in Fig. 1. The line across the picture is the solid-
at lower cost. In recent years, different authors applied a state joint obtained with the EMPT process.
combined numerical and experimental technique to extract the The superior mechanical properties of T91 rely on the
material model parameters, usually implemented in commercial creation and maintenance throughout its service life of a
FE codes, starting from compressive or tensile results, obtained tempered martensitic microstructure. Conventional fusion
under different loading conditions. The Johnson-Cook (J-C) welding processes destroy this condition and require an
(Ref 17) model is one of the most used, and the literature appropriate PWHT to recreate the microstructure.
includes many studies, in which original or modified formu-
lations are used. One of the most important works on this topic
is by Zhao (Ref 18), in which a modified formulation of the J-C
model is used to fit the experimental data varying the strain- Table 2 T91 mechanical properties at room temperature
rate. Other examples exist (Ref 8, 19-23), in which a combined
numerical and experimental technique was adopted to obtain Yield Tensile Minimum Maximum
strength, strength, longitudinal hardness
the parameters of J-C-like models in terms of hardening, strain-
MPa MPa elongation, % HB
rate, and thermal sensitivity starting from standard compression
or tension tests performed at different strain-rates and/or Minimum 415 Minimum 585 20 250
temperatures.

Table 1 Nominal composition of commercial T91 steel (wt.%)


C Mn P S Si Cr Mo V Nb N Al Ni

Min 0.08 0.30 0.20 8.00 0.85 0.18 0.06 0.030


Max 0.12 0.60 0.020 0.010 0.50 9.50 1.05 0.25 0.10 0.070 0.040 0.40

Journal of Materials Engineering and Performance


Fig. 1 Tempered martensite microstructure of a T91 specimen (a); sketch of the specimens used for tests at different temperatures and strain-
rate up to 103 s1 and the miniaturized specimens used for high strain-rate tests at 104 s1 (b)

on the specimen dimension are available, but the chosen


geometry has also been used by other researchers, e.g.,
(Ref 5, 26-28). A single specimen geometry was used for all
the tests, to avoid the influence of geometry and dimension and
to have the possibility to directly compare the results coming
from different loading conditions.
A second set of 5 samples (see Fig. 1) was produced with
compact dimensions 1.5 mm diameter of the reduced section
(instead of 3 mm) and 1.5 mm of gage length (instead of
5 mm) in order to reach very high strain-rates (104 s1). Also in
this case both quasi-static and dynamic tests were performed on
samples with the same geometry.

Fig. 2 Schematic of a typical EMPT setup for T91 fuel pin closing 3.1 Tests Varying the Strain-Rate
The low strain-rate tests were performed on a standard
electro-mechanical testing machine Zwick Z-100 (maximum
Among the cold joining processes available, impact joining load 100 kN, maximum travel speed 5 mm/s). The medium
such as EMPT appears particularly suitable for the envisaged strain-rate tests were performed on a standard servo-hydraulic
applications of T91 in the nuclear field. In this regard, JRC-ITU testing machine Dartec HA100 (maximum load 100 kN;
is exploring the feasibility of EMPT applied to joining of T91 maximum speed 100 mm/s). The high strain-rate tests were
nuclear fuel cladding. In depth knowledge of T91 response to performed using a Split Hopkinson Bar setup for direct tension.
impact loading is therefore a basic requirement for the The standard Hopkinson Bar setup is widely used to
understanding and the optimization of EMPT joining activities characterize material at strain-rate in the order of magnitude of
carried on at JRC-ITU. In this step, a specially formed conical 103 s1. Several configurations can be adopted and in this work
end plug is inserted in the fuel pin, and the wall of the cladding the arrangement for direct tensile tests with two different setups
is accelerated in the high electromagnetic field onto the conical was used. One setup covering the 103 s1 range with standard
plug to create a weld. The time-dependent EM field and the specimen dimensions (3 mm diameter) and a second setup
associated Lorentz forces are created by discharging a bank of covering the 104 s1 range with miniaturized specimen dimen-
capacitor through a coil. The schematic of a typical EMPT sions (1.5 mm diameter). The 103 s1 range setup consists of a
setup is showed in Fig. 2. gas-gun, an impactor and input and output bars (see Fig. 3).
The gas-gun is 1.5 m long and is driven with compressed air.
The striker bar is a 750-mm long tube made from glass
reinforced nylon. The input bar is made from martensitic high
3. Experimental Setup strength stainless steel (17-4PH) 10 mm in diameter and 6.8 m
long, with an anvil at one end. The output bar has the same
The experimental test campaign was performed on dog-bone characteristics of the input bar and has a length of 3.4 m. The
specimens with a gage diameter of 3 mm and a gage length of anvil at the outer end of the input bar is hit by the striker which
5 mm, as reported in Fig. 1. In general, for Hopkinson Bar is pneumatically accelerated. A tensile stress wave ri (incident
tests, there are no specific standards, as well as no specifications wave) is generated and propagates along the input bar toward

Journal of Materials Engineering and Performance


Fig. 3 Scheme of the Hopkinson Bar setup used to perform tensile tests (direct tension up to 103 s1)

Fig. 4 Diagrams of the waves recorded in terms of force vs. time for tests at 103 s1 (a) and 104 s1 (b)

the specimen. Due to the difference of the sound velocity in the the specimen to ensure a uniaxial stress state. Decreasing the
materials of the input and striker bars, the impact against the diameter of the specimen inevitably reduces the level of force
anvil produces a pulse which has a time duration of about on the output bar and concomitantly a low output signal to
600 ls, which corresponds to a length of about 3 m in the steel noise ratio is obtained. Furthermore, the specimen has to be
bars. When the wave reaches the specimen, the wave is partly large enough to be representative of the material under testing.
reflected back into the input bar (reflected wave rr) and partly A widely used technique requires the miniaturization of the
transmitted to the output bar (transmitted wave rt). The waves setup: in this way it is possible to increase the strain-rate,
ri, rr, and rt are measured by means of semiconductor strain- keeping at the same time the test design simple with the
gages (KYOWA KSP-1-350-E). With respect to traditional possibility to directly compare the results with those obtained at
resistance strain-gages, this type of gage provides signals with a lower strain-rate. The benefits of developing miniaturized setup
very low noise level. The strain-gages are located at a distance for compression tests have been reported (Ref 28).
of 1700 mm from the specimen on the input bar and 200 mm In this work, the authors developed a miniaturized setup for
behind the specimen in the output bar. The signals were tensile tests, which consists of a gas-gun 1 m long which uses
acquired with a NATIONAL INSTRUMENTS PC-6133 acqui- compressed air. The impactor was a tube made from aluminum
sition board at a sampling rate of 3 MHz with a maximum and it is 150 mm long. The input bar was made from
resolution of 150 lV (14 bits in the range ±1.25 V). martensitic high strength stainless steel (17-4PH) with a
In order to extend the range in strain-rate, different types of 10 mm of diameter and 2 m long, with an anvil at one end.
tests can be used, such as Hopkinson Bar, Taylor and Flyer The output bar had a diameter of 6 mm with a length of
Impact tests. In general, very high strain-rates using a 500 mm. The strain-gages were located at 250 mm on the input
Hopkinson Bar setup can be achieved in two ways. One bar and 50 mm on the output bar from the specimen.
possibility is to increase the speed of the striker bar, but this In Fig. 4, the waves (in terms of force vs. time) recorded
also increases the stress level in the bar, which is limited by the during the tests with the two Hopkinson setups are shown.
yield strength of its material. Another possibility is the Since in both setups, the measuring points are located far from
reduction of the specimen dimensions, but also here there are the specimen ends, the forces and displacements at the actual
some restrictions. The reduction of the length of the sample can ends of the specimen necessitate shifting of the recorded signals
only be achieved by maintaining the length to diameter ratio of forward or backward, toward the interface with the specimen.

Journal of Materials Engineering and Performance


Accordingly, the timing of incident, reflected and transmitted the middle of the specimens to avoid any undesired influences
waves, was performed on the basis that equilibrium conditions at the neck. Before performing the tests, a calibration procedure
were reached in terms of force at both ends of the specimen (see was performed to correlate the temperature in the middle of the
Fig. 4). Moreover, an algorithm for the wave dispersion gage length with the temperature where the thermocouples are
correction was applied (Ref 29). welded and to monitor the uniformity of the heating. Five
thermocouples were used in the calibration procedure to obtain
3.2 Tests Varying the Temperature the temperature distribution along the gage length of the
specimen (Fig. 5). Given the small dimension of the specimen
In order to completely characterize the material behavior, it
and the high power of the heating induction system, the
is important to perform tests at different temperatures, permit-
maximum difference in temperature between the middle of the
ting the description of the thermal softening effects on the
specimen and the position of the control thermocouple T1 was
material strength. Such tests were limited to quasi-static tests
in the order of magnitude of 2%, registered at 200C. In more
(103 s1) at 20, 200, 400, 500, 600, and 800C. The heating
details, the absolute error in temperature was about 5C in all
of the specimen was performed using an induction coil system,
the investigated range of temperature. Similarly, the maximum
designed to concentrate the heat flux in the gage length of the
difference in the gage length, from T2 to T4, was less than 1%.
specimen. The temperature was controlled via a Proportional-
In a real test, only one thermocouple at position 1 was used for
Integral-Derivative controller (PID controller), commonly used
the temperature control. Each specimen was kept for about
in closed control loop, which adjusts the command signal to the
5 min at the desired temperature before performing the test, in
heating system on the basis of the measurements obtained from
order to reach uniform temperature.
thermocouples directly welded on the specimen surface far
from the gage length. The thermocouples were not welded on

4. Experimental Results

For each loading condition, at least two tests were


performed in order to evaluate the repeatability of the results.
The engineering stress-strain curves of the material are given in
Fig. 6 as a function of temperature and strain rate (the levels of
strain-rate reported in the next figures represent the nominal
values of the tests).
The material, as expected, is both strain-rate and tempera-
ture sensitive. By varying the strain-rate, the material strength
increases due to the dynamic hardening: the initial oscillations
observed in high strain-rate tests are not related to a material
behavior (yielding), but are typical signal aberrations in
Hopkinson Bar tensile tests (Ref 30). In test at 103 s1, it is
possible to control the source of error, such as misalignment,
initial clearance, etc.; in miniaturized setup 104 s1,the problem
is amplified (higher level of acceleration) and more difficult to
Fig. 5 Scheme of the location of the thermocouples (T) for the control. Comparing the quasi-static data obtained from the two
temperature calibration (a); details of the testing system used in qua- geometries it appeared evident that they produce the same
si-static up to medium strain-rate at high temperature (b) results until necking starts. After this, the stress versus strain is

Fig. 6 Experimental results in terms of engineering stress vs. engineering strain: (a) varying the temperature at strain rate 103 s1; (b) varying
the strain-rate at room temperature. (b) Two curves are labeled as 103 s1: one refers to the results obtained from the specimen with D = 3 mm
and L = 5 mm (solid line), the other refers to the results obtained from the specimen with D = 1.5 mm and L = 1.5 mm (dashed line)

Journal of Materials Engineering and Performance


strongly dependent on the geometry and the results differ. The In FE codes, usually, the J-C model could take account of
material strength gradually decreases with increasing the thermal softening essentially due to heat conversion of plastic
temperature and the shape of the stress-strain curve changes work occurring at high strain-rate deformations. For
dramatically at 800C, where the material behaves like a e_ 0 ‡ 102 s1 both thermal conduction and convection can be
perfectly plastic material. The results obtained are in accor- neglected and thermal softening can be evaluated under the
dance with those expected for this kind of material (Ref 17). assumption of adiabatic conditions. Given this last hypothesis
The strain at failure increases with strain-rate and temperature. and the further assumption of uniform stress, strain, and
The high strain-rate tests were recorded using a high-speed temperature fields, the temperature can be analytically com-
camera (Photron SA5: maximum resolution 1024 9 1024, puted as a function of plastic work.
maximum frame-rate 106 frame/s) and the sequences obtained Before starting with the description of the optimization
for two tests performed nominally at 103 s1 are reported in results, it is important to underline that the J-C model is a
Fig. 7. The images are acquired at 25,000 frame/s. Analyzing purely empirical model and it is necessary to decide which of
the images indicates that the phase of necking took a long time its parameters should be considered as optimization variables
during the test. Moreover, also in high dynamic loading without any physical interpretation. For what concerns the
condition, the necking area was localized in the middle of the strain-hardening effect, three parameters (A, B, and n) are
specimen which reached an equilibrium status during the available to fit the data and different procedures can be used: all
dynamic test and the deformed shape was comparable with that the strain-hardening coefficients can be optimized or one of
of a quasi-static test. them can be fixed a priori and the other two optimized. For
what concerns the temperature sensitivity, usually, the only J-C
temperature parameter that is considered as an optimization
variable is m. This approach could lead to a poor fit, but an
5. Constitutive Model and Optimization Procedure improvement can be achieved if Tm is also considered as an
optimization variable.
The specimen geometry could affect the results and their Similarly, for the strain-rate sensitivity, only the parameter C
interpretation as early necking can occur (see Fig. 7). Moreover is usually considered as an optimization variable, while e_ 0 is set
non-uniform stress, strain, strain-rate, and temperature distri- equal to 1 s1, as in the original formulation. In LS-DYNA this
butions could develop inside the specimen. To correctly choice would mean that the strain-rate influence is neglected for
compute all these effects, the real strength model of the all the numerical simulations of tests with strain-rates less than
material has to be extracted using a numerical procedure based unity. If also e_ 0 is used as optimization variable, an improve-
on an inverse method for optimization. ment in the experimental data fit could be obtained.
The main objective of such an inverse optimization method The optimization of the parameters was performed with a
is the determination of a selected set of unknown parameters in dedicated algorithm included in the software LS-OPT that
a numerical model. Starting from a trial point, the unknown manages the parameter variation strategy, runs the numerical
parameters are estimated iteratively by comparing experimen- simulation, performed in LS-DYNA, analyzes the results, and
tally measured with numerically computed quantities for the extracts the optimum set of parameters. All the simulations
same material test conditions. The great advantage of this were performed with an explicit integration method and in
procedure is that no hypothesis about the internal specimen order to reduce the computational time for low strain-rate tests
stress-strain, temperature, or strain-rate fields is made. In fact, the deformation rate was increased with respect to the
the comparison is made in terms of macroscopic quantities that, experimental conditions (see the following discussion for
in general, are force and displacement. The main disadvantage details).
of inverse methods is the high computational times that these As first step, for each experimental curve, the numerical
algorithms need in an iterative procedure using many FEM interpolation of the experimental data (with the strain hardening
simulations. The number of iterations increases dramatically part of the model) was performed, which basically is equivalent
when the degrees of freedom of the problem grow or the trial to the calculation of true stress-true strain data starting from
parameters are far from optimum. experimental force-stroke curve (see Fig. 8). By performing
For the material under investigation, the widely used J-C this calculation numerically (instead of a standard analytical
model was chosen (Ref 17), which expresses the flow stress as approach), it is possible to account of the change in shape of the
specimen during necking. The deformation mechanism of the
  e_ pl
 
T  Tr
m 
specimen is supposed to be a function only of the stress-strain
ry ¼ A þ Benpl 1 þ C ln 1 ;
e_ 0 Tm  Tr relation of the material and not of the test speed. After an
ðEq 1Þ analysis of the experimentally deformed shapes, this hypothesis
could be considered as reasonable (i.e., inertia effects could be
where A is the elastic limit strength, B and n are the work neglected). The failure condition of each curve was numerically
hardening parameters, C is the strain-rate sensitivity coeffi- obtained in correspondence to the stroke at which the specimen
cient, and m describes the thermal softening. The thermal ef- failure occurred: for each curve, the maximum strain is the
fects are also described as a function of Tr, a reference equivalent plastic strain at failure (average value in the necking
temperature at which there are no thermal effects and Tm that section). The quality of the procedure was confirmed, as the
is the melting temperature at which the material mechanical results obtained from the models of the two geometries (3-mm
strength is zero. In the LS-DYNA formulation (Ref 24), e_ 0 diameter specimens and 1.5-mm diameter specimens) in quasi-
represents the strain-rate threshold, i.e., the highest strain- static tests produced the same J-C model (see Fig. 8), i.e., if the
rate for which the strain-rate effects on the flow stress are material properties are based on the same strength model, they
neglected. must be geometry independent.

Journal of Materials Engineering and Performance


Fig. 7 Sequence of images taken with a high speed camera at 25,000 f/s for tests at a strain rate of 103 s1 compared with the results of
numerical analysis (see section 5)

The thermal softening and the strain-rate sensitivities of the mentioned above, both m and Tm were considered as fitting
material were analyzed starting from the diagrams of Fig. 8. variables. The parameters obtained at this stage were fixed in all
The stress values (obtained at the fixed value of strain of 20%) the subsequent analyses. Similarly, it is possible to obtain a first
were normalized with respect to the stress value obtained for estimation of J-C model parameters (C and e_ 0 ), by performing a
quasi-static test at room temperature. For a complete analysis, linear piecewise interpolation of the data.
also others strain levels were analyzed (until 100% of strain). In Fig. 9, the stress data and their interpolation are
However, the 20% data were considered as representative of the compared with literature data (Ref 17) for two different types
behavior over a wider range in strain, also because, in of steel. For the strain-rate sensitivity, the stress level is
accordance with the J-C formulation, the behavior is the same saturated at the quasi-static value for strain-rate less than e_ 0 ,
at each strain (both in temperature and strain-rate). For the accordingly to the implementation of the J-C model in
evaluation of the temperature sensitivity, the data were LS-DYNA.
analytically fitted with the J-C expression for the thermal These preliminary results are affected by two different types
softening. In order to get the best level of accuracy, as of approximation:

Journal of Materials Engineering and Performance


Fig. 8 Results of the first optimization procedure in terms of equivalent stress vs. equivalent plastic strain: (a) varying the temperature; (b)
varying the strain-rate

Fig. 9 Temperature (a) and strain-rate (b) sensitivities

– the influence of the non-uniformity of strain-rate and tem- velocity profiles (i.e., the real strain rate time history). If the
perature in the specimen is neglected. In this way, each different objectives conflict, no single solution can be consid-
experimental result is reproduced by a stress-strain (pure ered optimum with respect to all the objectives (Ref 31). The
hardening) relation and the obtained strength model could optimal solution is such that any further attempts to optimize on
be considered representative of an average behavior of the a single objective lead to worse results for the other(s) (Ref 32).
material at the nominal strain-rate and/or temperature con- Mathematically, the MOO unconstrained problem is defined
dition of the test (i.e., strain-rate and temperature are con- as follows:
sidered constant in the specimen);
– the self heating of the material due to the adiabatic condi- min FðU1 ; U2 ; . . . ; UN Þ; ðEq 2Þ
tion (conversion of mechanical work into heat) in high
strain-rate tests is ignored. where F represents the multi-objective function and
Uk ¼ Uk ðx1 ; x2 ; . . . ; xn Þ ðk ¼ 1; . . . ; N Þ are the various objec-
To overcome these issues, a Multi-Objective Optimization tive functions with xi (i = 1,…,n) the n design variables. In
(MOO) procedure was performed, in which all the tests at particular, the MOO function is defined as
different strain-rates were simultaneously simulated. The final
goal is the determination of the global set of J-C model X
N
parameters, which reproduces as best as possible all the tests. F¼ xk Uk ; ðEq 3Þ
Both the thermal softening and the strain-rate parameters k¼1
should be estimated considering the variation of the corre-
sponding properties on the basis of multiple data curves. All the where xk are the weights to assign to each single objective
tests at strain-rate lower than 102 s1 were assumed and function. It is important to note that each objective function has
simulated as isothermal, while the other tests were considered a target. For the analysis performed in this work, all the weights
to be adiabatic. The dynamic tests were simulated with the real xk are set to unity, so all the objectives are equally important.

Journal of Materials Engineering and Performance


Table 3 J-C parameters obtained as results of the MOO procedure: the parameters fixed a priori for each case
are reported in italic
A, MPa B, MPa n e_ 0 (s21) C Tm, K m rRMSE, %

SET1 446 601 0.186 10.5 0.0388 1142 1.35 3.58


SET2A 446 511 0.221 103 0.0219 1142 1.35 6.50
SET2B 475 545 0.221 1 0.0219 1142 1.35 6.50
SET3 446 592 0.174 6.27 0.0272 … … 4.40

Fig. 10 Results of the optimization (SET1): comparison between numerical and experimental force vs. displacement curve for quasi-static, low
and high strain-rate tests

For the present study, the single objective function was the Three different procedures (SET 1, 2A and 2B, and 3) were
Mean Squared Error (MSE) Uk, defined as (Ref 25) used and the results are summarized in Table 3.
  For SET1, the optimization was performed on the strain-
1X P
fp ðxÞ  Gp 2 hardening and strain-rate parts of the model, while the thermal
Uk ¼ WP ; ðEq 4Þ
P p¼1 sp softening parameters were fixed to those obtained analytically
on the basis of quasi-static curves varying the temperature
where P is the number of points (about 100 for this analysis) (m = 1.352 and Tm = 1141.8 K). The parameter A was fixed to
in which the MSE is calculated, Gp, varying p, are the that obtained for the quasi-static test at room temperature
values on the target curve G and fp(x) the corresponding com- (A = 445.7 MPa). The variables for the optimization were B, n,
ponents of the computed curve f. Wp and sp are scale func- C, and e_ 0 . The optimized set of parameters produced a medium
tions for each point p and finally x is the design vector. percentage rRMSE (relative root means square error) of 3.6%.
Several definitions of the MSE are possible by varying the In this case, the J-C model was used in accordance with the
value of the scale function sp. In this analysis, the relative LS-DYNA formulation, i.e., e_ 0 was considered as a threshold
MSE is considered with sp constant and equal to the maxi- under which any strain-rate effects are neglected. This model
mum of G. could be applied in case of dynamic loading conditions with

Journal of Materials Engineering and Performance


self-heating (e.g., impacts) and for quasi-static loading condi- temperature dependence is not considered, was obtained and
tions at different temperatures. the results are also satisfactory. Furthermore, a deterioration in
For SET2A and SET2B, the procedure was similar to that the model accuracy was observed if the strain-rate threshold
used for SET1, but in this case the strain-rate threshold was e_ 0 was fixed, as usually performed in many engineering
fixed equal to the lowest strain-rate level of the tests (103 s1) applications.
in case 2A and equal to 1 s1 in case 2B. The parameter A was
fixed to that obtained for the quasi-static test at room
temperature for case 2A and properly scaled for case 2B, in
accordance with the variation of the strain-rate threshold. The
variables for the optimization were B, n, and C. The optimized References
sets of parameters produce a medium percentage error of 6.5% 1. W. Wang, M. Li, C. He, X. Wei, D. Wang, and H. Du, Experimental
and much more precise results with respect to the previous set Study on High Strain Rate Behavior of High Strength 600–1000 MPa
were obtained for the low strain-rate loading condition. The set Dual Phase Steels and 1200 MPa Fully Martensitic Steels, Mater. Des.,
2013, 47, p 510–521. doi:10.1016/j.matdes.2012.12.068
2B could be suitable for numerical simulations performed in FE
2. A.K. Paul, A. Raj, P. Biswas, G. Manikandan, and R.K. Verma, Tensile
codes in which the J-C model is implemented in its original Flow Behavior of Ultra Low Carbon, Low Carbon and Micro Alloyed
formulation ( e_ 0 = 1 s1 without threshold, i.e., different from Steel Sheets for Auto Application Under Low to Intermediate Strain
LS-DYNA). Rate, Mater. Des., 2014, 57, p 211–217. doi:10.1016/j.matdes.
In case of SET3, the effect of the self-heating was included 2013.12.047
in the strain-rate sensitivity, i.e., the thermal part of the J-C 3. F. Feng, S. Huang, Z. Meng, J. Hu, Y. Lei, M. Zhou, D. Wu, and Z.
Yang, Experimental Study on Tensile Property of AZ31B Magnesium
model was considered to be unity. The parameter A was fixed Alloy at Different High Strain Rates and Temperatures, Mater. Des.,
for the single optimization of the quasi-static test at room 2014, 57, p 10–20. doi:10.1016/j.matdes.2013.12.031
temperature (A = 445.7 MPa). The variables for the optimiza- 4. D. Samantaray, A. Patel, U. Borah, S.K. Albert, and A.K. Bhaduri,
tion were B, n, C, and e_ 0 . The optimized set of parameters Constitutive Flow Behavior of IFAC-1 Austenitic Stainless Steel
produced a medium percentage error of 4.4% and should be Depicting Strain Saturation Over a Wide Range of Strain Rates and
Temperatures, Mater. Des., 2014, 56, p 565–571. doi:10.1016/j.matdes.
used in dynamic loading conditions. 2013.11.053
By the evaluation of the rRMSE, it is possible to conclude 5. E. Cadoni, M. Dotta, D. Forni, and P. Spätig, Strain-Rate Behavior in
that the best set of parameters is the SET1, for which the Tension of the Tempered Martensitic Reduced Activation Steel
comparison between experimental and numerical results in Eurofer97, J. Nucl. Mater, 2011, 414(3), p 360–366. doi:10.1016/j.
terms of force versus displacement is reported in Fig. 10. The jnucmat.2011.05.002
6. G. Solomos, C. Albertini, K. Labibes, V. Pizzinato, and B. Viaccoz,
greatest error lies in the quasi-static test, but for low and high Strain rate Effects in Nuclear Steels at Room and Higher Temperatures,
strain-rate tests, the model is able to reproduce the material Nucl. Eng. Des., 2004, 229(2–3), p 139–149. doi:10.1016/j.nucengdes.
behavior with a good level of accuracy. 2003.10.006
7. A.K. Dureja, S.K. Sinha, D.N. Pawaskar, P. Seshu, J.K. Chakravartty,
and R.K. Sinha, Modelling Flow and Work Hardening Behaviour of
Cold Worked Zr-2.5Nb Pressure Tube Material in the Temperature
6. Conclusions Range of 30–600C, Nucl. Eng. Des., 2013, doi:10.1016/j.nucengdes.
2013.08.006
8. P. Spatig, R. Bonadé, G.R. Odette, J.W. Rensman, E.N. Campitelli, and
In this work, the investigation of the mechanical response of
P. Mueller, Plastic Flow Properties and Fracture Toughness Character-
T91 steel was performed at different temperatures and strain- ization of Unirradiated and Irradiated Tempered Martensitic Steels, J.
rates, in order to improve the analysis of innovative joining Nucl. Mater., 2007, 367–370, p 527–538. doi:10.1016/j.jnucmat.
techniques and the evaluation of structural response to dynamic 2007.03.038
loading resulting from reactor accident conditions. 9. M. Scapin, L. Peroni, and M. Peroni, Parameters Identification in
Strain-Rate and Thermal Sensitive Visco-plastic Material Model for an
The mechanical tests at different strain-rates were performed
Alumina Dispersion Strengthened Copper, Int. J. Impact Eng., 2012,
from 103 to 104 s1. In order to reach such high strain-rate 40–41, p 58–67. doi:10.1016/j.ijimpeng.2011.10.002
tests, a miniaturized Hopkinson Bar arrangement for direct 10. G.P. Skoro, J.R.J. Bennett, T.R. Edgecock, and C.N. Booth, Yield
tensile tests was developed. The tests varying the temperature Strength of Molybdenum, Tantalum and Tungsten at High Strain Rates
(from room temperature to 800C) were performed in quasi- and Very High Temperatures, J. Nucl. Mater., 2012, 426(1–3), p 45–51.
doi:10.1016/j.jnucmat.2012.03.044
static loading condition and the specimen was heated using an
11. W.-S. Lee, C.-F. Lin, T.-H. Chen, and W.-Z. Luo, High Temperature
induction coil system. The experimental results clearly indicate Deformation and Fracture Behaviour of 316L Stainless Steel Under
that the material is both temperature and strain-rate sensitive High Strain Rate Loading, J. Nucl. Mater., 2012, 420(1–3), p 226–234.
and the J-C visco-plastic model was chosen to reproduce doi:10.1016/j.jnucmat.2011.10.005
the data using a numerical inverse procedure of optimization. 12. R.L. Klueh and A.T. Nelson, Ferritic/Martensitic Steels For Next-
With this aim, a numerical inverse MOO was performed in Generation Reactors, J. Nucl. Mater., 2007, 371, p 37–52
13. C. Fazio et al., European Cross-Cutting Research on Structural
order to extract a set of model parameters able to reproduce the
Materials for Generation IV and Transmutation Systems, J. Nucl.
behavior of the material under different loading conditions (at Mater., 2009, 392, p 316–323
different strain-rates), simultaneously. Different configurations 14. Y. Zhang et al., Application of High Velocity Impact Welding at Varied
were evaluated and the best results were obtained with: Different Length Scales, J. Mater. Process. Technol., 2011, 211(5),
A = 445.7 MPa, B = 601.1 MPa, n = 0.186, e_ 0 =10.52 s1, p 944–952
C = 0.0388, Tm = 1141.8 K, and m = 1.352. This set of 15. V. Psyk et al., Electromagnetic Forming: A review, J. Mater. Process.
Technol., 2011, 211, p 787–829
parameters is able to reproduce the behavior of the material
16. T. Malmberg, Dynamic Plastic Behaviour of Metals, Kernforschung
up to strain-rates in the order of 104 s1 with a good level of Zentrum Karlsruhe, Report No. KFK 2023, September 1974
accuracy. For the modeling of high strain-rate behavior also a 17. G.R. Johnson and W.A. Cook, A Constitutive Model and Data for
simplified formulation of the J-C model, in which the Metals Subjected to Large Strains, High Strain Rates and High

Journal of Materials Engineering and Performance


Temperatures, 7th International Symposium on Ballistics, 1983, 541– 25. N. Stander, W. Roux, T. Goel, T. Eggleston, and K. Craig, LS-OPT
547 UserÕs Manual: A Design Optimization and Probabilistic Analysis Tool
18. H. Zhao, A Constitutive Model for Metals Over a Large Range of for the Engineering Analyst, Version 4.0, 2009, Livermore Software
Strain Rates Identification for Mild-Steel and Aluminium Sheets, Technology Corporation (LSTC)
Mater. Sci. Eng. A, 1997, 230, p 95–97 26. T. Børvik, O.S. Hopperstad, S. dey, E.V. Pizzinato, M. Langseth, and
19. M. Sasso, G. Newaz, and D. Amodio, Material Characterization at C. Albertini, Strength and Ductility of Weldox 460 E Steel at High
High Strain Rate by Hopkinson Bar Tests and Finite Element Strain Rates, Elevated Temperatures and Various Stress Triaxialities,
Optimization, Mater. Sci. Eng. A, 2008, 487, p 289–300 Eng. Fract. Mech., 2005, 72, p 1071–1087. doi:10.1016/j.engfracmech.
20. M. Sedighi, M. Khandaei, and H. Shokrollahi, An Approach in 2004.07.007
Parametric Identification of High Strain Rate Constitutive Model Using 27. A.H. Clausen, T. Børvik, O.S. Hopperstad, and A. Benallal, Flow and
Hopkinson Pressure Bar Test Results, Mater. Sci. Eng. A, 2010, 527, Fracture Characteristics of Aluminium Alloy AA5083-H116 as Func-
p 3521–3528 tion of Strain Rate, Temperature and Triaxiality, Mater. Sci. Eng. A,
21. A.S. Milani, W. Daboussi, J.A. Nemes, and R.C. Abeyaratne, An 2004, 364(1–2), p 260–272. doi:10.1016/j.msea.2003.08.027
Improved Multi-objective Identification of Johnson-Cook Material 28. D. Jia and K.T. Ramesh, A Rigorous Assessment of the Benefits of
Parameters, Int. J. Impact Eng., 2009, 36, p 294–302 Miniaturization in the Kolsky Bar System, Exp. Mech., 2004, 44,
22. B. Langrand, P. Geoffroy, J.-L. Petitniot, J. Fabis, E. Markiewicz, and p 445–454. doi:10.1177/0014485104047608
P. drazetic, Identification Technique of Constitutive Model Parameters 29. W. Cheng and B. Song, Split Hopkinson (Kolsky) Bar: Design, Testing
for Crashworthiness Modeling, Aerosp. Sci. Technol., 1999, 4, and Applications, Mechanical Engineering Series Springer, New York,
p 215–227 2011, ISBN 978-1-4419-7981-0
23. J. Peirs, P. Verleysen, W. Van Paepegem, and J. Degrieck, Determining 30. M. Borsutzki et al., Recommendations for Dynamic Tensile Testing of
the Stress-Strain Behaviour at Large Strains from High Strain Rate Sheet Steels, 2005, International Iron and Steel Institute
Tensile and Shear Experiments, J. Impact Eng., 2011, 38, p 406–415. 31. T. Goel and N. Stander, Multi-Objective Optimization using LS-OPT,
doi:10.1016/j.ijimpeng.2011.01.004 2007, LS-DYNA Conference
24. B. Gladman, LS-DYNA Keywords UserÕs Manual, Version 971, Vol. 1, 32. S. Bandyopadhyay and S. Saha, Unsupervised Classification, 2013.
2007, Livermore Software Technology Corporation (LSTC) doi: 10.1007/978-3-642-32451-2_2

Journal of Materials Engineering and Performance

You might also like