You are on page 1of 15

Original Article

Proc IMechE Part C:


J Mechanical Engineering Science
Comparative studies of the constitutive 2022, Vol. 236(7) 3636–3650
© The Author(s) 2021
models for tungsten heavy alloy Article reuse guidelines:
sagepub.com/journals-permissions
DOI: 10.1177/09544062211045168
(95W-3.5Ni-1.5Fe) at high strain rates journals.sagepub.com/home/pic

Xiuwen Lai, Zhanjiang Wang  and Na Qin

Abstract
The plastic behaviors’ description of a tungsten heavy alloy (95W-3.5Ni-1.5Fe) at temperatures of 298–773 K and strain rates
of 0.001–11,000 s1 is systematically studied based on four constitutive models, that is, Zerilli-Armstrong model, modified
Zerilli-Armstrong model, Mechanical Threshold Stress model, and modified Mechanical Threshold Stress model. The quasi-
static compression experiments using an electronic universal testing machine and the dynamic compression experiments
using a split Hopkinson pressure bar apparatus are employed to obtain the true stress–strain curves at a total of three
temperatures (298 K, 573 K, and 773 K) and a wide range of strain rates (0.001–11,000 s1). The parameters of the four
constitutive models are obtained by the above fundamental experimental data and Grey Wolf Optimizer. The correlation
coefficient and average absolute relative error are used to evaluate the predicted performance of these models. Modified
Mechanical Threshold Stress model is found to have the highest predicted performance in describing the flow stress of the
95W-3.5Ni-1.5Fe alloy. Eventually, two compression experiments whose loading conditions are not in the fundamental
experiments are conducted to validate the four models.

Keywords
Tungsten heavy alloys, Zerilli-Armstrong model, modified Zerilli-Armstrong model, mechanical threshold stress model,
modified mechanical threshold stress model, grey wolf optimizer

Introduction Threshold Stress (MTS) model21-27 belong to physical


models. Only more accurate constitutive models can make
Tungsten heavy alloys (WHAs) have the advantages of high the finite element simulation more effective and precise.
densities (16–18.5 g/cm),1 high melting point,2,3 and wonderful Therefore, many constitutive models have been built. Yadav
properties such as high strength, high fracture toughness, and Ramesh28 studied 91 wt% tungsten (91W) under very
good corrosion resistance, and so on.4,5 WHAs are composite high strain rates of 103-105 s1 at room temperature, and they
materials where body-centered cubic (BCC) tungsten par- established JC model and ZA model for 91W. Chen and
ticles are randomly embedded in a face-centered cubic (FCC) Gray24 studied the tantalum–tungsten alloys under strain
matrix. Tungsten particles are of high hardness and low rates of 0.001–2500 s1 and temperatures of 77–873 K, and
ductility, while the matrix exhibits a good ductility but a low they established JC model, ZA model, and MTS model for
strength.6-8 Tungsten-based alloys are mainly prepared by tantalum–tungsten alloys. Dümmer et al.25 studied the effects
liquid phase sintering, in which the main element, tungsten, of strain rates (quasi-static strain rate at 3×103 s1 and
is sintered with a small amount of Ni, Fe, Cu, Co., etc. After dynamic strain rates from 103 to 4×103 s1) on plastic flow
sintered, WHAs with different microstructures and mechanical in polycrystalline tungsten (PW). The flow stress of PW
properties can be obtained.9-11 The most common WHAs, was described by ZA model and MTS model, respectively.
named W-Ni-Fe, contain 85 to 98 wt% tungsten with Ni:Fe Lee et al.29,30 illustrated the quasi-static (0.01–1 s1) and
weight ratios between 6:4 and 9:1.12
In the last thirty years, many researchers used experiments
and the finite element method (FEM) to study the plastic
Department of Mechanical Engineering, Southwest Jiaotong University,
behaviors of tungsten and its alloys at different temperatures
Chengdu, China
and strain rates. The constitutive models used in the FEM are
built based on experimental data. These constitutive models Corresponding authors:
are usually classified as phenomenological models and physical Zhanjiang Wang, Department of Mechanical Engineering, Southwest
models. Phenomenological models are based on macro Jiaotong University, No.111, North Section 1, 2nd Ring Road, Chengdu
610031, China.
parameters’ effects on flow stress, while physical models Email: wangzhanjiang001@gmail.com
are based on micro-deformation mechanisms.13 Johnson-
Na Qin, Department of Mechanical Engineering, Southwest Jiaotong
Cook (JC) model,14 Khan-Huang-Liang (KHL) model,15,16 University, NO.111 , Northbound 1, 2nd Ring Road, Chengdu 610031,
and Arrhenius model17 belong to phenomenological models, China.
while Zerilli-Armstrong (ZA) model18-20 and Mechanical Email: tinaspirit@163.com
Lai et al. 3637

dynamic (800-4000 s1) flow stress of 92.5W-5.25Ni-2.25Fe


at different temperatures of 298–1373 K by Arrhenius model
and ZA model, respectively. Lennon and Ramesh31 studied
the plastic behaviors of PW under strain rates of 103–
4×103 s1 and temperatures of 300 K–1023 K by JC model
and ZA model. Xu and Huang32 established seven consti-
tutive models (JC, KHL, PB, NNL, ZA, VA, and RK) to
describe the thermomechanical behaviors of 93W-4.9Ni-
2.1Fe under strain rates of 103–3×103 s1 and tempera-
tures of 173 K–873 K. Wang et al.33 used six constitutive
models (Arrhenius, JC, modified-JC, ZA, modified-ZA, and
KHL) to describe the thermomechanical behaviors of pure
tungsten under quasi-static conditions and elevated tem-
peratures of 1523 K–1823 K.
It is found that the studied strain rates are mostly below
4000 s1 except for the study of Yadav and Ramesh,28 and
the researchers mentioned above tend to study the individual
effects of high strain rates and temperatures on the plastic
behaviors of tungsten and its alloys except for Chen and
Gray,24 Lee et al.,29,30 and Xu and Huang.32 However, there
are coupling effects of high strain rates and temperatures in
the actual processing. For example, in elliptical vibration
cutting processing, materials are intensely deformed under
high strain rates at various temperatures.4,34 Therefore, the
previous studies on the coupling effects of high strain rates
and temperatures on the plastic behaviors of tungsten and its
alloys are insufficient, especially for the tungsten alloy whose
tungsten content is more than 95 wt%. Moreover, the pre-
dicted performance of physical models is generally better
than that of phenomenological models.32 Among the
physical models, ZA model is widely used because of its easy
expression, easy fitting of parameters, and good predicted
performance. However, MTS model, which also belongs to
physical models, is rarely studied because MTS model
considers more complex micro-deformation mechanisms
and has many parameters that are not easy to fit. Aiming at Figure 1. The typical microstructure of 95W and the statistical
results of grain size.
the problem that the parameters of MTS model are difficult to
fit, an optimization algorithm can be used to obtain the model
parameters according to some researches.35,36 Therefore, in shows the typical microstructure of 95W and the statistical
this paper, the two typical physical models, that is, ZA and results of grain size, which obey normal distribution. The ex-
MTS models, are studied. The fundamental quasi-static and pectation X , standard deviation S, and the quantity of tungsten
dynamic compression experiments and Grey Wolf Opti- particles Q are 28.5 μm, 10.553 μm, and 700, respectively.
mizer37 (GWO) are employed to obtain the parameters of ZA The quasi-static compression experiments are performed
and MTS models. Further, a modified ZA (M-ZA) model and on an electronic universal testing machine (EUTM) at a
a modified MTS (M-MTS) model are proposed to increase 0.18 mm/min loading rate. The specimens of the quasi-static
the predicted performance of the original ZA and MTS compression experiments are made into cylinders, which are
models. These two modified models’ parameters are also 4 mm in diameter and 3 mm in length, that is, the strain rate is
obtained by the fundamental compression experimental data 0.001 s1. The dynamic compression experiments are per-
and GWO. Subsequently, comparative studies of the four formed using a split Hopkinson pressure bar (SHPB) at
constitutive models are conducted based on expressions and temperatures of 298–773 K and strain rates of 2500–
the agreement between predicted curves and the fundamental 11,000 s1. The specimens of SHPB tests are also made into
experimental data. Eventually, two compression experiments cylinders, which are 4 mm in diameter and 3 mm in length. All
whose loading conditions are not in the fundamental ex- specimens are obtained by wire electro-discharge machining
periments are conducted to validate the four models. and polished on both sides to reduce end friction on the
specimens. Figure 2 shows the schematic diagram of SHPB.

Experimental procedures and results The flow characteristics and


Material and experiments deformation mechanisms
The material studied in this paper is 95W-3.5Ni-1.5Fe alloy Because the gas gun of SHPB can only provide five dif-
(95W), which is prepared by liquid phase sintering. Figure 1 ferent pressures of 0.2–0.6 MPa when conducting the
3638 Proc IMechE Part C: J Mechanical Engineering Science 236(7)

Figure 2. The schematic diagram of split Hopkinson pressure bar.

Figure 3. The results of quasi-static and dynamic compression experiments: (a) quasi-static condition, (b) 0.2 MPa, (c) 0.3 MPa, (d) 0.4 MPa,
(e) 0.5 MPa, and (f) 0.6 MPa.

dynamic compression experiments, the experimental data and the experimental data are also based on the true strain
under the same pressure and three different temperatures are before 0.25.
put together for study. Figure 3 shows a total of 16 results of
quasi-static and dynamic compression experiments. Three
repeated experiments are conducted for each loading Evaluation method of constitutive models
condition. It is seen that there is a rapid decrease of the flow The correlation coefficient (Rr) and average absolute rel-
stress when the true strain exceeds about 0.25 under high ative error (AARE) are employed to assess the predicted
strain rates. This phenomenon appears obviously when the performance of the four models quantitatively. The ex-
strain rate exceeds 5000 s1. Lee et al.29 found a similar pressions are written as follows
phenomenon in the 92.5W flow stress curves at a fixed
strain rate of 4000 s1. They made a metallographic ob- N  
1 X 
Ei  Pi  × 100
servation on a specimen tested under 4000 s1 and 298 K, AAREð%Þ ¼  (1)
N i¼1 Ei 
and they found that there are macrocracks propagating in
the Fe-Ni matrix and microcracks at the interface between   
PN
tungsten grains and matrix, and finally, microcrack formation i¼1 Ei  E Pi  P
between contiguous tungsten grains. Hence, it’s reasonable Rr ¼ sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
 2  2ffi (2)
to speculate that the specimens will cause and accumulate PN PN
damage when the true strain exceeds 0.25. Based on this i¼1 Ei  E i¼1 Pi  P

assumption, the following constitutive models are estab-


lished by the experimental data before the true strain reaches where N is the number of data points, Ei is the experimental
0.25. Similarly, the comparisons between predicted curves value, Pi is the predicted value, E stands for the average
Lai et al. 3639

Figure 5. Hierarchy of grey wolves in Grey Wolf Optimizer.

mimicking the leadership hierarchy and hunting mechanism


Figure 4. The selected points from the experimental and
of the grey wolves. The leadership hierarchy of the grey
predicted flow stress curves for a certain loading condition wolves is defined as four types, that is, alpha (α), beta (β),
(Schematic diagram). delta (δ), and omega (ω), as shown in Figure 5.
The first three levels, α, β, and δ, stand for the dominant
grey wolf, the superior grey wolf, and the middle grey wolf,
value of all experimental values, and P stands for the average
respectively. The lowest level ω is regarded as the re-
value of all predicted values.
maining grey wolves. The three main steps of hunting,
searching for prey, encircling prey, and attacking prey, are
Constitutive parameter employed to describe the hunting mechanism of the grey
optimization process wolves. When the grey wolves hunt, the positions of ω level
are updated by the positions of α, β, and δ levels. In this
Mathematical form of optimization paper, the specific steps of GWO are as follows:
The process of constitutive parameter optimization in this
paper is a minimization problem. The mathematical form of Step 1. Positions initialization. The maximum number of
the minimization problem is as follows iterations is set as 1000, and 30 sets of constitutive pa-
rameters are randomly generated in the range of initial
N 
   
1 X 
σ exp i  σ y ðxÞ i
 × 100,
values as the initial positions of 30 grey wolves.
Minimize f ðxÞ ¼  
X2RL N i¼1  σ exp i 
(3) Step 2. Social hierarchy. The AARE is regarded as the
T
x ¼ ðx1 ,x2 ,…,xL Þ subject to dj ≤ xj ≤ ej ðj ¼ 1,2,/,LÞ objective function. The fitness of each grey wolf is cal-
culated in the population, and the best three wolves with the
where the components xj of x are called design variables
lowest three AARE values are labeled by α, β, and δ. The
(constitutive parameters), L is the number of design vari-
remaining wolves are labeled by ω.
ables (the number of constitutive parameters), f ðxÞ stands
for the objective function (AARE), dj stands for the mini-
Step 3. Encircling prey. During the hunt, the grey wolves
mum value of xj, ej stands for the maximum value of xj, and encircle prey. The encircling process can be expressed by a
σ exp and σ y are the flow stress from experiments and mathematical model as follows37
constitutive model calculation, respectively.
Because of the deviations of experimental flow stress Xðkþ1Þ ¼ XðkÞ
p A
ðkÞ
 DðkÞ
curves before the true strain of 0.05, the evaluations of  ðkÞ  (4)
constitutive models’ predicted performance are based on ¼ XpðkÞ  2aðkÞ  r1  aðkÞ  DðkÞ
the true strain from 0.05 to 0.25. In this range of the true
where k presents the latest iteration, Xp indicates the po-
strain, 21 pairs of data points from experimental and pre-
sition vector of the prey, X stands the position vector of a
dicted flow stress curves are evenly selected for each
grey wolf, A ¼ 2a  r1  a, r1 is a random vector between
loading condition, as shown in Figure 4. Therefore, a total
[0, 1], the components of a is linearly decreased from 2 to 0
of 336 pairs of data points (N=336) are used to evaluate the
over the process of iterations, and D is calculated by
constitutive models’ predicted performance for 16 loading
equation (5)
conditions (in Figure 3).
 
 
DðkÞ ¼ CðkÞ  XpðkÞ  XðkÞ 
  (5)
Grey Wolf Optimizer  ðkÞ 
¼ 2r2  XpðkÞ  XðkÞ 
Grey Wolf Optimizer, which is inspired by the behaviors of
the grey wolf predation, is a new meta-heuristic to solve the where D stands for a vector depending on the position of
optimization problems. This optimizer is proposed by the prey (Xp), C ¼ 2r2 , r2 is also the random vector between
Mirjalili et al.37 This optimization algorithm is based on [0, 1].
3640 Proc IMechE Part C: J Mechanical Engineering Science 236(7)

Table 1. Material properties of 95W.39

Properties Values

Density, ρ 17500 kg/m3


Young’s modulus, E 372 GPa
Poisson’s ratio, ν 0.28
Specific heat, Cp 134 J/kg/K

 ðkÞ   
 ðkÞ ðkÞ 
DαðkÞ ¼ C1  XαðkÞ  XðkÞ , Dβ ¼ C2  Xβ  XðkÞ ,
ðkÞ

 ðkÞ ðkÞ 
Dδ ¼ C3  Xδ  XðkÞ 
ðkÞ

(8)

When the maximum number of iterations is reached, the


optimization is completed, and Xα is the optimal solution,
which is used for the corresponding constitutive model.
Figure 6 shows the flow chart of GWO to search for the
optimal constitutive parameters.

Constitutive models and predicted results


All experimental data are converted to the true stress–strain
curves. The plastic strain is calculated by εp ¼ ε  σ=E, in
which ε is the true strain, σ is the true stress, and E is
Young’s Modulus of 95W. During the dynamic tests, a
temperature rise due to plastic deformation of 95W should
be considered by equation (9)
Z
η
ΔT ¼ σdεp (9)
ρCp
Figure 6. The flow chart of Grey Wolf Optimizer to search for
the optimal constitutive parameters. where the fraction η of plastic work converted into heat is
considered as 0.9, ρ is the density of 95W, and Cp is the
specific heat of 95W. The material properties of 95W are
Step 4. Hunting. The position of the prey is recognized by listed in Table 1.
the grey wolves, then the grey wolves encircle the prey,
and eventually, the grey wolves attack the prey. This
process is mainly guided by the three grey wolf leaders Zerilli-Armstrong model
(α, β, δ), the remaining grey wolves (ω) follow the three Zerilli-Armstrong model based on dislocation-mechanics is
leaders. However, the search space of the problem is al- a physical model, which considers the effects of strain
ways abstract. There is no idea about the position of the hardening, strain rate hardening, and thermal softening on
prey (optimal solution). Therefore, it’s supposed that the the flow stress. This model has different forms of expres-
three grey wolf leaders (the best three candidate solutions) sions for different crystal structure materials. Zerilli and
have a better ability to search for the potential location Armstrong18 proposed BCC material ZA model expressed
of prey. Then the three leaders are retained in the cur- as equation (10) and FCC material ZA model expressed as
rent population at each iteration, and the remaining equation (11)
grey wolves’ positions are updated by the three leaders’ 
positions. The following expressions are used for σ y ¼ c0 þ c1 exp c3 T þ c4 T ln_εp þ c5 εnp (10)
updating37,38

ðkÞ ðkÞ ðkÞ
ðkþ1Þ X þ X2 þ X3 σ y ¼ c0 þc2 exp c3 T þ c4 T ln_εp ε1=2 (11)
X ¼ 1 (6) p
3
where c0 ¼ σ 0 þ kl 1=2 is athermal stress, σ 0 is an addi-
where X1, X2, and X3 are calculated by equation (7) tional portion of stress related to solutes and the original
ðkÞ ðkÞ ðkÞ ðkÞ ðkÞ ðkÞ dislocation density, k stands for the microstructural stress
X1 ¼ XαðkÞ  A1  DαðkÞ , X2 ¼ Xβ  A2  Dβ ,
strength, l stands for the average grain diameter, c1-c5 and n
ðkÞ ðkÞ ðkÞ ðkÞ
X3 ¼ Xδ  A3  Dδ are the constitutive parameters. The formulation of FCC
(7) material considers that thermal softening and strain rate
hardening are related to strain. In contrast, BCC material
where Dα, Dβ, Dδ are calculated by equation (8) formulation considers that thermal softening and strain rate
Lai et al. 3641

Table 2. The ranges of initial values for ZA model parameters of 95W.

c1 (MPa) c3 (K1) c4 (K1) c5 (MPa) n


4 5 4 4 4
[0, 10 ] [-5×10 , 0.2] [-1×10 , 2.5×10 ] [0, 10 ] [0.1, 0.6]

Table 3. Zerilli-Armstrong model parameters of 95W.

c0 (MPa) c1 (MPa) c3 (K1) c4 (K1) c5 (MPa) n

0 1457 1.8703×103 1.6296×104 1147 0.293

Figure 9. Evaluation results of ZA model.


Figure 7. The convergence curve of ZA model’s AARE.

Figure 8. Comparisons between fundamental experimental data and predicted curves of ZA model under different loading conditions: (a)
quasi-static condition, (b) 0.2 MPa, (c) 0.3 MPa, (d) 0.4 MPa, (e) 0.5 MPa, and (f) 0.6 MPa.

hardening are independent of strain. Some researchers, such parameters are listed in Table 2. The constitutive parameters
as Xu and Huang,32 used BCC material ZA model to de- of ZA model are presented in Table 3. Among those pa-
scribe the flow stress of WHAs because tungsten as the rameters, c0 is obtained from Dümmer et al.25 Other pa-
main element is BCC structure. In this section, BCC ma- rameters are obtained by fundamental experimental data
terial ZA model is also employed to describe the flow and the optimization of GWO. The convergence curve of
stress for 95W. The ranges of initial values for ZA model the objective function (AARE) of ZA model is shown in
3642 Proc IMechE Part C: J Mechanical Engineering Science 236(7)

Table 4. Modified ZA model parameters of 95W.

c0 (MPa) c1 (MPa) c3 (K1) c4 (K1) c5 (MPa) c6 (K1) n


3 4 5
0 1457 1.8703×10 1.6296×10 1147 1.5110×10 0.293

Table 5. The ranges of initial values for p1-p6.

p1 (K1) p2 (K1) p3 (K) p4 (K) p5 p6

[-104, 105] [104, 102] [0,103] [-103, 10] [8, 2×104] [-0.02, 3×103]

Figure 7. It is seen that the convergence rate of GWO is very 1stopt. The detailed relations between the parameter c7 and
fast, and the AARE can generally converge to a good value different dynamic loading conditions are shown in
within 1000 iterations. Therefore, the number of iterations Figure 10. The convergence curve of the objective function
of GWO in the following sections is all set as 1000 iter- (AARE) of M-ZA model is shown in Figure 11. The results
ations. The results of comparisons between fundamental of comparisons between fundamental experimental data
experimental data and predicted curves of ZA model for and predicted curves of M-ZA model for different loading
different loading conditions are plotted in Figure 8. In conditions are plotted in Figure 12. The AARE and Rr of M-
Figure 9, there are a total of 336 pairs of selected points ZA model, as shown in Figure 13, are 2.9823% and 0.9581,
from the experimental and predicted flow stress curves, respectively.
which are the same as the selected points described in
Section 4.1. These data points are used to calculate the
AARE and Rr. The AARE and Rr of ZA model, as shown in Mechanical threshold stress model
Figure 9, are 3.5204% and 0.9363, respectively. The flow stress σ y of MTS model is considered as a function
of a mechanical threshold stress b
σ , which is the reference
Modified ZA model stress or the flow stress at 0 K. Two parts are used to
The Fe and Ni matrix of WHAs is FCC structure. Zerilli and characterize the mechanical threshold stress, as shown in
Armstrong20 developed a HCP material ZA model con- equation (14)21
sidering the Peierls stress type interactions (predominant
b
σ¼b
σa þ b
σt (14)
in BCC) and the intersection of forest dislocations type
interactions (predominant in FCC). Based on their study, where bσ a and b
σ t are the athermal portion and the thermal
an M-ZA model considering the effects of FCC structure portion, respectively. The former describes the rate-
on 95W is developed to improve the predicted perfor- independent interactions of dislocations with long-range
mance of ZA model. The formulation of M-ZA model is barriers such as grain boundaries, and the latter describes
written as follows the rate-dependent interactions of dislocations with short-
 range obstacles such as forest dislocations, which can be
σ y ¼ c0 þc1 exp c3 T þ c4 T ln_εp overcome by the thermal activation. The flow stress related
 (12) to the mechanical threshold stress can be expressed as
þc5 exp c6 T þ c7 T ln_εp εnp
follows21,24
where c0-c5 and n are the same as those used in ZA model,  μb
σt
σ a þ s ε_ p ,T
σy ¼ σa þ σt ¼ b (15)
c6 and c7 are the constitutive parameters of M-ZA model. μ0
Among those parameters, firstly, c6 and c7 are considered as
two constants, which are obtained by GWO, and the pa- where σ a ≡ bσ a is a constant, the factor sð_εp ,T Þ includes the
rameters c0-c6 and n are presented in Table 4. Then, c7 is effects of strain rates and temperatures, which specifies the
modified as a variable related to temperatures and strain ratio between the flow stress and the mechanical threshold
rates, which is expressed by equation (13) as follows stress, and this factor is smaller than one for thermally
activated controlled glide because the stress required to
tan1 ½ðT  p3 Þ=p4 
c7 ¼ p1 þp2 0:5 þ force a dislocation past an obstacle is reduced by the
π contribution of the thermal activation energy,21 μ0 is the
( h . i)
tan 1
ln_εp  p5 p6
(13) shear modulus at 0 K, μ is the shear modulus at the current
0:5 þ temperature, which is defined by equation (16),22,24 σ t is
π the thermal portion of the flow stress. In the thermally
activated glide regime, the interaction kinetics for short-
where the fitting constants p1-p6 are obtained by GWO range obstacles are described by an Arrhenius expression
when parameters c0-c6 and n are fixed as constants. The as equation (17)21,24
ranges of initial values for p1-p6 are listed in Table 5. The
constants p1-p6 are presented in Table 6. Equation (13) is Dr
μ ¼ μ0  (16)
found by using Levenberg-Marquardt algorithm in software expðTr =T Þ  1
Lai et al. 3643

Table 6. The fitting constants p1-p6.

p1 (K1) p2 (K1) p3 (K) p4 (K) p5 p6


5 4
1.3093×10 9.1054×10 733.9443 0.7654 504.1685 40.9273

Figure 10. The relations between parameter c7 and different loading conditions (strain rates and temperatures).

sð_ε,T Þ between flow stress and mechanical threshold stress


is expressed as follows
8 "  . # 91=p
 σ t =μ < k b T ln ε_ r ε_ p
1=q
=
s ε_ p ,T ¼ . ¼ 1 
b
σ μ : g0 μb3 ;
t 0

(19)

The thermal portion σ t of the flow stress is composed of


two components as follows24-27
σt ¼ σi þ σe (20)

where σ i is the component of the flow stress due to intrinsic


barriers to thermally activated dislocation motion and
dislocation–dislocation interactions, σ e is the component of
the flow stress due to microstructural evolution with in-
creasing strain hardening. Therefore, equation (15) can
Figure 11. The convergence curve of M-ZA model’s AARE. eventually be expressed as follows27

where Dr and Tr are the fitting constants. σy ¼ σa þ σi þ σe


   μbσi  μbσe
ΔG ¼b σ a þ si ε_ p ,T þ se ε_ p ,T
ε_ p ¼ ε_ r exp (17) μ0 μ0
kb T
8 "  . #1=qi 91=pi
where ε_ r is a constant, kb stands for the Boltzmann constant, < kb T ln ε_ ri ε_ p = μb σi
ΔG stands for the free energy that can be expressed by a ¼b σa þ 1  (21)
: g0i μb 3 ; μ0
phenomenological relation as equation (18)21,24
" !p #q 8 "  . # 91=pe
σ t =μ < kb T ln ε_ re ε_ p
1=qe
= μb
σe
ΔG ¼ g0 μb 1 
3
. (18) þ 1
b
σ t μ0 : g0e μb 3 ; μ0

where g0 is regarded as a constant if the obstacle character where b σ i stands for a component of mechanical threshold
does not vary, b is the magnitude of the Burgers vector, and stress mainly due to intrinsic barriers to thermally activated
p and q are constants that characterize the shape of the dislocation motion, b σ e stands for a strain hardening com-
obstacle profile (ðp 2 ½0,1, q 2 ½1,2Þ). After combining and ponent of mechanical threshold stress. The strain hardening
rearranging the above equations (17) and (18), the ratio θ ¼ db σ e =dεp is written as
3644 Proc IMechE Part C: J Mechanical Engineering Science 236(7)

Figure 12. Comparisons between fundamental experimental data and predicted curves of M-ZA model under different loading conditions:
(a) quasi-static condition, (b) 0.2 MPa, (c) 0.3 MPa, (d) 0.4 MPa, (e) 0.5 MPa, and (f) 0.6 MPa.

Table 7. The ranges of initial values for σ a, b


σ i , g0i, σ 0es, g0es, and
θ 0.

σ a ðMPaÞ b
σ i ðMPaÞ g0i σ 0es ðMPaÞ g0es θ0 ðMPaÞ

[0, 100] [0, 104] [0, 1] [0, 103] [-0.8, 1] [0, 2×104]

ranges of initial values for the above six parameters are


listed in Table 7. Other parameters are obtained from Chen
and Gray24 and Dümmer et al.25 The constitutive param-
eters of MTS model are presented in Table 8. The con-
vergence curve of the objective function (AARE) of MTS
model is shown in Figure 14. The results of comparisons
between fundamental experimental data and predicted
curves of MTS model for different loading conditions are
plotted in Figure 15. The AARE and Rr of MTS model, as
shown in Figure 16, are 3.2361% and 0.9453, respectively.
Figure 13. Evaluation results of M-ZA model.

2 !3 Modified MTS model


γb
σe
6 tanh
bσes σa 7 The hardening rate θ0 is supposed to be a constant in the
6 7
θ ¼ θ0 6
6 1  7
7 (22) theoretical study of Follansbee and Kocks21; however, they
4 tanhðγÞ 5 found that θ0 has a definite strain rate dependence according
to the experimental data, and an empirical expression of θ0
related to strain rates is given as follows
where θ0 is the rate of change of the mechanical threshold 
stress with strain at zero strain,23 γ stands for linear variation θ0 ¼ a1 þa2 ln ε_ p þ a3 ε_ p (24)
of strain hardening rate with stress,24 the saturation threshold
stress b
σ es is related to strain rates and temperatures, which can where a1-a3 are fitting constants obtained by experimental
be expressed by following equation (23)21,24,27 data. In MTS model illustrated above, θ0 is considered as a
! ! constant, which is independent of strain rates and tem-
b
σ es kb T ε_ p peratures. However, it’s found that θ0 is not a constant by
ln ¼ ln (23)
b
σ 0es g0es μb3 ε_ res comparing predicted curves and experimental data, and θ0
is related not only to strain rates but also to temperatures. To
where bσ 0es , g0es , ε_ res are constants. Among MTS model improve the predicted performance of MTS model, an M-
parameters, ε_ ri ¼ ε_ re ¼ 4:5 × 104 s1 and ε_ res ¼ 106 s1 MTS model is developed by considering that θ0 depends on
are set by experience. σ a, b σ i , g0i, σ 0es, g0es, and θ0 are both strain rates and temperatures. The expression of θ0 can
obtained by fundamental experimental data and GWO. The be written as follows
Lai et al. 3645

Table 8. Mechanical Threshold Stress model parameters of 95W24,25.

σ a ðMPaÞ μ0 ðGPaÞ Dr ðGPaÞ Tr ðKÞ kb =b3 ðMPa=KÞ

35 159.5 33.69 1217 0.671


b
σ i ðMPaÞ ε_ ri ðs1 Þ g0i qi pi
1924 4.5×104 8.5123×102 1.5 0.5
ε_ re ðs1 Þ g0e qe pe γ
4.5×104 1.6 1.0 2/3 1.0
σ 0es ðMPaÞ g0es ε_ res ðs1 Þ θ0 ðMPaÞ
452 3.8709×102 106 6632

Figure 16. Evaluation results of mechanical threshold stress


Figure 14. The convergence curve of MTS model’s AARE. model.

Figure 15. Comparisons between fundamental experimental data and predicted curves of MTS model under different loading conditions:
(a) quasi-static condition, (b) 0.2 MPa, (c) 0.3 MPa, (d) 0.4 MPa, (e) 0.5 MPa, and (f) 0.6 MPa.

tan1 ½ðT  q3 Þ=q4  where the fitting constants q1-q6 are obtained by GWO
θ0 ¼ q1 þ q2 0:5 þ when other parameters are fixed as constants. The ranges
π
( h . i) (25)
of initial values for q1-q6 are listed in Table 9. The con-
tan1 ln_εp  q5 q6 stants q1-q6 are listed in Table 10. The other constitutive
0:5 þ parameters of M-MTS model are the same as those used in
π
MTS model listed in Table 8. Equation (25) is found by
3646 Proc IMechE Part C: J Mechanical Engineering Science 236(7)

Table 9. The ranges of initial values for q1-q6. plastic strain when temperatures and strain rates change,
which illustrates that ZA model is not suitable for materials
q1 (MPa) q2 (MPa) q3 (K) q4 (K) q5 q6
whose strain hardening behaviors depend on temperatures
3 4 3 4
[10 , 10 ] [10 , 10 ] [100, 850] [-10, 10] [7, 9] [-1, 1] and strain rates. However, Figure 8 shows that strain

Table 10. The fitting constants q1-q6.

q1 (MPa) q2 (MPa) q3 (K) q4 (K) q5 q6

6073.6382 4794.1411 404.5134 3.6525 8.3170 1.9540×102

Figure 17. The relations between parameter θ0 and different loading conditions (strain rates and temperatures).

using Levenberg-Marquardt algorithm in software 1stopt.


The detailed relations between the parameter θ0 and dif-
ferent dynamic loading conditions are shown in Figure 17.
The convergence curve of the objective function (AARE)
of M-MTS model is shown in Figure 18. The results of
comparisons between fundamental experimental data and
predicted curves of M-MTS model for different loading
conditions are plotted in Figure 19. The AARE and Rr of
M-MTS model, as shown in Figure 20, are 2.5367% and
0.9608, respectively.

Discussion
Comparisons based on the
fundamental experiments
Comparisons of the four constitutive models’ predicted
performance based on expressions and the agreement be-
tween predicted curves and the fundamental experimental Figure 18. The convergence curve of M-MTS model’s AARE.
data are conducted in this part. The evaluation results of all M-MTS: a modified MTS; AARE: average absolute relative
error.
constitutive models in describing fundamental experimental
data are shown in Figure 21.
The four models’ expressions all consider strain hard- hardening behaviors of 95W have some differences when
ening, strain rate hardening, and thermal softening. How- the loading conditions are at different temperatures and
ever, strain hardening is the focus of discussion. Firstly, the strain rates. Hence, the predicted performance of ZA model
expressions of the original ZA model and M-ZA model are is the worst of the four models. The AARE and Rr of ZA
analyzed. The form of ZA model expression is relatively model are 3.5204% and 0.9363, respectively. Therefore, an
simpler than other physical constitutive models. The M-ZA model is developed to increase the predicted per-
number of parameters of ZA model is only six. The process formance of ZA model. The influences of FCC matrix are
to fit ZA model’s parameters is relatively convenient. The taken into consideration in the expression of M-ZA model
parameters c5 and n, which are used to describe strain by a combination of BCC material ZA model expressed as
hardening behaviors, are two constants in ZA model’s equation (10) and FCC material ZA model expressed as
expression. It means that strain hardening is constant at each equation (11). The final expression of M-ZA model is
Lai et al. 3647

Figure 19. Comparisons between fundamental experimental data and predicted curves of M-MTS model under different loading
conditions: (a) quasi-static condition, (b) 0.2 MPa, (c) 0.3 MPa, (d) 0.4 MPa, (e) 0.5 MPa, and (f) 0.6 MPa.

Figure 20. Evaluation results of modified MTS model.

written as equation (12). It is seen that equation (12) can


degenerate as equation (10) when c6=c7=0. Similarly,
equation (12) can degenerate as equation (11) when c1=0
and n=1/2. Therefore, M-ZA model considers both BCC
crystal structure and FCC crystal structure in 95W. Strain
hardening of M-ZA model is related to temperatures and
strain rates by the term c5 expðc6 T þ c7 T ln_εp Þεnp in
equation (12), and equation (13) shows that c7 is regarded as
a variable depending on temperatures and strain rates to
improve M-ZA model’s performance in describing strain
hardening behaviors of 95W. The AARE and Rr of M-ZA
model are 2.9823% and 0.9581, respectively. Compared
with ZA model’s predicted performance, M-ZA model’s
predicted performance has a distinct improvement. As for
the original MTS model, the expression consists of three Figure 21. The predicted performance of the four models in
parts, that is, σ a ≡ b
σ a (the athermal portion), σ i (the thermal describing fundamental experimental data: (a) AARE and (b) Rr.
portion of the flow stress due to intrinsic barriers to thermally
activated dislocation motion and dislocation–dislocation inter- hardening of the mechanical threshold stress is represented by
actions), and σ e (the component of the flow stress due to mi- θ ¼ db σ e =dεp ¼ θ0 ½1  tanhðγb
σ e =ðb
σ es  σ a ÞÞ=tanhðγÞ. θ0 is
crostructural evolution with increasing strain hardening). Strain the rate of change of the mechanical threshold stress with strain at
3648 Proc IMechE Part C: J Mechanical Engineering Science 236(7)

Figure 22. The predicted performance of the four models in Figure 23. The predicted performance of the four models in
describing the quasi-static compression experiment conducted describing the dynamic compression experiment conducted
under 298 K and 0.1 s1 (The AARE is calculated by this quasi- under 473 K and 3300 s1 (The AARE is calculated by this dynamic
static loading condition). loading condition).

zero strain, which is regarded as a constant in the original MTS also on the ability to describe other experimental data (the
model. However, Strain hardening of MTS model is still related loading conditions are not in the fundamental experiments).
to temperatures and strain rates by the term se ð_εp ,T Þμbσ e =μ0 . The four models’ evaluation results in describing another
Therefore, the predicted performance of MTS model is slightly quasi-static compression experiment conducted under
better than it of ZA model. The AARE and Rr of MTS model 298 K and 0.1 s1 are shown in Figure 22.
are 3.2361% and 0.9453, respectively. Subsequently, an M- It is observed that MTS model has the lowest AARE and
MTS model is developed to increase the predicted perfor- M-MTS model has the second lowest AARE in describing the
mance of MTS model. In the expression of M-MTS model, quasi-static compression experiment conducted under 298 K
θ0 is considered as a variable depending on temperatures and and 0.1 s1. ZA model and M-ZA model have similar predicted
strain rates to increase the ability to describe strain hardening performance in this evaluation. All models can well predict
behaviors of 95W. θ0 is expressed as equation (25). The experimental data, and all AARE values are lower than 5%.
AARE and Rr of M-MTS model are 2.5367% and 0.9608, The four models’ evaluation results in describing an-
respectively. Hence, it is concluded that M-MTS model has other dynamic compression experiment conducted under
the highest predicted performance in describing fundamental 473 K and 3300 s1 are shown in Figure 23.
experimental data, and M-ZA model also has the excellent It is obvious that the predicted curves of the four models
predicted performance that is closest to the predicted per- have good agreement with experimental data. The AARE
formance of M-MTS model. Since the Rr values of all models values of all models are less than 4%. M-MTS model has
are very high (higher than 0.93) and close to each other, the the best predicted performance in describing the dynamic
following discussion is mainly based on the AARE. compression experiment conducted under 473 K and
3300 s1. Therefore, it is concluded that even if the loading
conditions are not in the fundamental experiments, the four
Constitutive models validation
models can still have good predicted performance, which
The applicability of a constitutive model depends not only also validates these models, and M-MTS model is still the
on the ability to describe fundamental experimental data but best model to describe the experimental data.
Lai et al. 3649

Conclusions shear band of tungsten heavy alloy fabricated through hot-


hydrostatic extrusion. Mater Sci Eng A 2008; 487(1–2):
The fundamental experiments at temperatures of 298– 235–242.
773 K and strain rates of 0.001–11,000 s1 are systemat- 2. Sharma V, Namburu SAS, Lalwani P, et al. Constitutive
ically performed. The fundamental experimental data and modelling and processing map analysis of tungsten heavy
GWO are used to establish the four constitutive models (ZA alloy (92.5W-5.25Ni-2.25Fe) at elevated temperatures. Int J
model, M-ZA model, MTS model, and M-MTS model). Refract Met Hard Mater 2018; 76: 168–179.
These constitutive models are employed to describe the 3. Bresciani LM, Manes A, Romano TA, et al. Numerical
flow stress of 95W, and comparative studies of these modelling to reproduce fragmentation of a tungsten heavy
constitutive models are conducted. The main conclusions alloy projectile impacting a ceramic tile: adaptive solid mesh
are listed as follows: to the SPH technique and the cohesive law. Int J Impact Eng
2016; 87: 3–13.
1. Grey Wolf Optimizer is very suitable for estimating 4. Pan Y, Kang R, Dong Z, et al. On-line prediction of ultrasonic
the parameters of the constitutive models based on elliptical vibration cutting surface roughness of tungsten
fundamental experimental data, and GWO has no heavy alloy based on deep learning. J Intell Manuf 2020;
specific limitation on the number of parameters of DOI: 10.1007/s10845-020-01669-9.
the models, which is very flexible. The iterative 5. Zamora KMO, Sevillano JG and Prrez MF. Fracture
process of GWO generally has good convergence toughness of W heavy metal alloys. Mater Sci Eng A 1992;
and a faster convergence rate. 157(2): 151–160.
2. In describing fundamental experimental data, M- 6. Kim DK, Lee S and Baek WH. Microstructural study of
MTS model has the highest predicted perfor- adiabatic shear bands formed by high-speed impact in a
mance. The AARE and Rr of M-MTS model are tungsten heavy alloy penetrator. Mater Sci Eng A 1998;
2.5367% and 0.9608, respectively. M-ZA model has 249(1–2): 197–205.
the second best predicted performance. The AARE 7. Bahgat M, Paek MK and Pak JJ. Reduction investigation of
and Rr of M-ZA model are 2.9823% and 0.9581, WO3/NiO/Fe2O3 and synthesis of nanocrystalline ternary
respectively. The original ZA and MTS models have W-Ni-Fe alloy. J Alloys Compd 2009; 472(1–2): 314–318.
similar predicted performance, which is enough to 8. Debata M, Acharya TS, Sengupta P, et al. Effect of high
describe the flow stress of 95W but weaker than the energy ball milling on structure and properties of 95W-3.5Ni-
predicted performance of M-ZA model. The pre- 1.5Fe heavy alloys. Int J Refract Met Hard Mater 2017; 69:
dicted curves of the four models correlate well with 170–179.
the experimental data. The Rr values of all models 9. Das J, Rao GA and Pabi SK. Microstructure and mechanical
are higher than 0.93. properties of tungsten heavy alloys. Mater Sci Eng A 2010;
3. In the validation of the four models, there is good 527(29–30): 7841–7847.
agreement between the four models’ predicted 10. Upadhyaya A, Tiwari SK and Mishra P. Microwave sintering
curves and the experimental data. In describing the of W-Ni-Fe alloy. Scr Mater 2007; 56(1): 5–8.
quasi-static compression experiment conducted 11. Upadhyaya A. Processing strategy for consolidating tungsten
under 298 K and 0.1 s1, the AARE values of all heavy alloys for ordnance applications. Mater Chem Phys
models are lower than 5%. In describing the dy- 2001; 67(1–3): 101–110.
namic compression experiment conducted under 12. Islam SH, Tufail M and Qu XH. Mechanical properties
473 K and 3300 s1, the AARE values of all variation with test temperature for liquid phase sintered 95W-
models are lower than 4%. M-MTS model is still the 3.5Ni-1.5Fe alloys. Mater Sci Forum 2007; 561: 647–650.
best model to describe the experimental data. 13. Lin YC and Chen XM. A critical review of experimental
results and constitutive descriptions for metals and alloys in
Declaration of conflicting interests hot working. Mater Des 2011; 32(4): 1733–1759.
The author(s) declared no potential conflicts of interest with re- 14. Johnson GR and Cook WH. A constitutive model and data for
spect to the research, authorship, and/or publication of this article. metals subjected to large strains, high strain rates and high
temperatures. In: Proceedings of the 7th International Sym-
Funding posium on Ballistics, Den Haag, The Netherlands, 1983;
The author(s) disclosed receipt of the following financial support pp.541–547.
for the research, authorship, and/or publication of this article:The 15. Khan AS and Liang R. Behaviors of three BCC metal over a
authors would like to express their sincere gratitude to the support wide range of strain rates and temperatures: experiments and
provided by the National Natural Science Foundation of China modeling. Int J Plast 1999; 15(10): 1089–1109.
(Grant Nos. 51775457, 51875487). 16. Khan AS, Suh YS and Kazmi R. Quasi-static and dynamic
loading responses and constitutive modeling of titanium
ORCID iD alloys. Int J Plast 2004; 20(12): 2233–2248.
17. Jonas JJ, Sellars CM and Tegart WJMG. Strength and
Zhanjiang Wang  https://orcid.org/0000-0003-0818-2577 structure under hot-working conditions. Metall Rev 1969;
14(1): 1–24.
References 18. Zerilli FJ and Armstrong RW. Dislocation-mechanics-based
1. Liu J, Li S, Fan A, et al. Effect of fibrous orientation on constitutive relations for material dynamics calculations.
dynamic mechanical properties and susceptibility to adiabatic J Appl Phys 1987; 61(5): 1816–1825.
3650 Proc IMechE Part C: J Mechanical Engineering Science 236(7)

19. Zerilli FJ and Armstrong RW. Description of tantalum de- 30. Lee WS, Lin CF and Chang ST. Plastic flow of tungsten-
formation behavior by dislocation mechanics based consti- based composite under hot compression. J Mater Process
tutive relations. J Appl Phys 1990; 68(4): 1580–1591. Technol 2000; 100(1–3): 123–130.
20. Zerilli FJ and Armstrong RW. Constitutive relations for tita- 31. Lennon AM and Ramesh KT. The thermoviscoplastic re-
nium and Ti-6Al-4V. In: AIP conference proceedings, Seattle, sponse of polycrystalline tungsten in compression. Mater Sci
Washington (USA), 13 -18 Aug 1995, American Institute of Eng A 2000; 276(1–2): 9–21.
Physics, 1996, pp.315–318. DOI: https://doi.org/10.1063/1. 32. Xu Z and Huang F. Thermomechanical behavior and consti-
50713. tutive modeling of tungsten-based composite over wide tem-
21. Follansbee PS and Kocks UF. A constitutive description of perature and strain rate ranges. Int J Plast 2013; 40: 163–184.
the deformation of copper based on the use of the mechanical 33. Wang J, Zhao G, Chen L, et al. A comparative study of
threshold stress as an internal state variable. Acta Metall several constitutive models for powder metallurgy tungsten at
1988; 36(1): 81–93. elevated temperature. Mater Des 2016; 90(15): 91–100.
22. Follansbee PS and Gray GT. An analysis of the low tem- 34. Xie HB, Yang ZQ, Qin N, et al. Strain rate analyses during
perature, low and high strain-rate deformation of Ti-6Al-4V. elliptical vibration cutting of Inconel 718 using finite element
Metall Trans A 1989; 20(5): 863–874. analysis, Taguchi method, and ANOVA. Adv Manuf 2020;
23. Gourdin WH and Lassila DH. The mechanical behavior of pre- 8(3): 316–330.
shocked copper at strain rates of 10-3-104s-1 and temperatures 35. Lin YC, Wen DX, Chen MS, et al. A novel unified dislocation
of 25-400 °C. Mater Sci Eng A 1992; 151(1): 11–18. density-based model for hot deformation behavior of a
24. Chen SR and Gray GT. Constitutive behavior of tantalum and nickel-based superalloy under dynamic recrystallization
tantalum-tungsten alloys. Metall Mater Trans A 1996; 27(10): conditions. Appl Phys A 2016; 122: 805.
2994–3006. 36. Sagar CK, Chilukuri A and Priyadrashini A. Determination of
25. Dümmer T, Lasalvia JC, Ravichandran G, et al. Effect of Johnson Cook material model constants for 93% WHA and
strain rate on plastic flow and failure in polycrystalline Optimization using Genetic algorithm. Mater Today Proc
tungsten. Acta Metall 1998; 46(17): 6267–6290. 2018; 5(9): 18911–18919.
26. Goto DM, Bingert JF, Chen SR, et al. The mechanical 37. Mirjalili S, Mirjalili SM and Lewis A. Grey wolf optimizer.
threshold stress constitutive-strength model description of HY- Adv Eng Softw 2014; 69: 46–61.
100 steel. Metall Mater Trans A 2000; 31(8): 1985–1996. 38. Zhou J, Qiu Y, Zhu S, et al. Optimization of support vector
27. Banerjee B. The mechanical threshold stress model for machine through the use of metaheuristic algorithms in forecasting
various tempers of AISI 4340 steel. Int J Sol Struct 2007; TBM advance rate. Eng Appl Artif Intel 2021; 97: 104015.
44(3–4): 834–859. 39. Sagar CK, Priyadarshini A, Gupta AK, et al. Determination
28. Yadav S and Ramesh KT. The mechanical properties of of Johnson cook material model constants and their influ-
tungsten-based composites at very high strain rates. Mater Sci ence on machining simulations of tungsten heavy alloy. In:
Eng A 1995; 203(1–2): 140–153. ASME 2018 international mechanical engineering congress
29. Lee WS, Xiea GL and Lin CF. The strain rate and temperature and exposition, Pittsburgh, Pennsylvania, USA, 9–15 No-
dependence of the dynamic impact response of tungsten vember ,2018. American Society of Mechanical Engineers
composite. Mater Sci Eng A 1998; 257(2): 256–267. Digital Collection, 2018.

You might also like