Original Article

Proc IMechE Part B:

J Engineering Manufacture
226(8) 1287–1297
Application of genetic algorithms Ó IMechE 2012
Reprints and permissions:
for optimizing the Johnson–Cook sagepub.co.uk/journalsPermissions.nav
DOI: 10.1177/0954405412447735

constitutive model parameters pib.sagepub.com

when simulating the titanium alloy

Ti-6Al-4V machining process

Guang Chen, Chengzu Ren, Wei Yu, Xiaoyong Yang and Lifeng Zhang

Abstract
Finite element simulation of metal machining requires accurate constitutive models to characterize the material stress–
strain response in plastic deformation processes. An optimization methodology using genetic algorithms was developed
to determine the Johnson–Cook material model for Ti–6Al–4V alloy. The optimization of the parameters resulted in
lower errors between the calculated flow stress and the experimental values obtained through the split Hopkinson pres-
sure bar tests at different temperatures (ranging from 25 °C to 900 °C) and strain rates (2000 and 2500s21). Optimized
Johnson–Cook constitutive parameters were used to calculate the flow stress under various conditions. The calculated
results showed excellent agreement with the experimental values, with errors lower than 4%. In addition, Ti–6Al–4V
alloy orthogonal cutting experiments were carried out to validate the finite element simulation results. The experimental
chip morphology was compared with the simulation results obtained by the optimized Johnson–Cook model (M2) and
the original Johnson–Cook model (M1). The simulated results (including chip morphology and cutting force) were
affected by the flow stress model. Comparison of the experimental and simulated results revealed that the optimized
Johnson–Cook model can provide relatively good prediction results for the titanium alloy machining process, especially
for chip morphology prediction.

Keywords
Titanium alloy, Johnson–Cook constitutive model, optimization, machining simulation

Date received: 31 August 2011; accepted: 17 April 2012

Introduction to obtain experimentally, such as stress, strain, tem-

perature, etc. Metal cutting always involves large strain
Titanium alloys have become popular materials in (1–10), high strain rate (103–106 s21) (Jaspers and
aerospace and medical industries because of their com- Dautzenberg3) and a wide range of temperatures. A
bined properties, such as high stiffness, low density, material constitutive equation plays a key role in char-
excellent corrosion and oxidation resistance.1 Titanium acterizing the deformation behavior in machining.
alloys are difficult-to-machine materials because of the Several material models have been developed to deter-
high flow stress at elevated temperatures and the sensi- mine the deformation behavior under high strain rate
tivity of these materials to the processing parameters of conditions. Cockroft and Latham4 developed the
the forming process.2 Therefore, understanding the
machining mechanics and improving machining effi-
ciency have become increasingly important for the Key Laboratory of Advanced Ceramic and Machining Technique (Tianjin
industrial application of titanium alloy. The finite ele- University), Ministry of Education, China
ment (FE) method has recently emerged as an essential
modeling methodology because of its efficiency and Corresponding author:
Chengzu Ren, Key Laboratory of Advanced Ceramic and Machining
usefulness to decrease the cost of experiments in Technique (Tianjin University), Ministry of Education, Tianjin 300072,
machining. This method is also a convenient technique China.
to acquire the machining parameters that are difficult Email: [email protected]

Downloaded from pib.sagepub.com at HOWARD UNIV UNDERGRAD LIBRARY on January 26, 2015
1288 Proc IMechE Part B: J Engineering Manufacture 226(8)

Cockroft–Latham model according to the density of to determine the material constants for AZ61 magne-
strain energy. Johnson and Cook5 presented the sium alloy by minimizing the errors between calculated
Johnson–Cook (J–C) model, considering the influence and experimental results.
of strain, strain rate and temperature on flow stress. Very few studies have focused on the optimization
Zerilli and Armstrong6 developed the Zerilli– of constitutive model parameters under high strain rate
Armstrong model based on the dislocation mechanics conditions. Lin and Yang19 applied a GA-based multi-
theory. Recently, Calamaz et al.1 presented the hyper- objective optimization technique to determine the vis-
bolic tangent (TANH) model considering the effect of coplastic constitutive equations for titanium alloy with
strain softening. Some of these models are considered low strain rates (5 3 1025–1 3 10–3 s21) at 927 °C.
as semi-empirical constitutive models that are widely Recently, Özel and Karpat20 applied the cooperative
used in existing FE simulation of metal cutting pro- particle swarm optimization (CPSO) method to identify
cesses.7 The material parameters of the J–C model, one the J–C parameters by SHPB tests in Lee and Lin,21
of the most widely used models,8 are commonly considering the influence of different temperatures at a
obtained through split Hopkinson pressure bar (SHPB) strain rate of 2000 s21. Although optimization tech-
tests. This experimental technique is commonly used to niques are currently widely used to obtain a material
assess the deformation behavior at strain rates ranging constitutive model, very few studies have reported
from 102 s21 to 104 s21,9 so the J–C model parameters on the application of model optimization results. As
obtained by SHPB tests can be applied to simulate the J–C model is commonly applied to characterize the
machining processes under high strain rate conditions. material deformation behavior in machining, it is nec-
Many studies have applied the J–C model in machin- essary to use machining simulation to validate the opti-
ing simulation. Umbrello et al.10 used the J–C constitu- mized J–C constitutive parameters.
tive equation with different J–C parameters to simulate There is a similarity in the deformation behaviors of
AISI 316L steel machining and found that the residual SHPB tests and machining under high strain rate condi-
stresses are very sensitive to the J–C model parameters. tions. Therefore, a GA-based optimization technique is
Özel11 applied the J–C model in a three-dimensional FE presented in the current study to calculate the J–C para-
model to simulate the AISI 4340 turning process and meters according to the SHPB tests under different high
investigated the influence of the micro-geometry edge on strain rates and temperatures. Afterwards, the opti-
forces, stress and tool wear of polycrystalline cubic boron mized J–C model parameters are applied in machining
nitride (PCBN) inserts. Chen et al.12 presented a ductile simulations to validate the optimization results.
failure FE model based on the J–C constitutive model to
simulate Ti–6Al–4V alloy high-speed machining.
The material parameters are crucial to the accuracy Johnson–Cook model
of the J–C model. The J–C parameters are usually To model the plastic behavior of Ti–6Al–4V alloy, the
fitted according to the strain–stress curves obtained J–C constitutive model was employed, which can be
through SHPB tests.13 Seo et al.9 determined the J–C represented by the following equation5
parameters using SHPB tests at 1400 s21 strain rate     
and from room temperature to 1000 °C. Lee and Lin14 n e_ T  Troom m
 = ½A + BðeÞ  1 + C ln
s  1
fitted J–C model parameters using SHPB tests to study e_0 Tmelt  Troom
the high-temperature deformation behavior of Ti–6Al– ð1Þ
4V alloy under different strain rates. However, the fit-
ting method for the J–C model parameters was not dis- where e is the equivalent plastic strain, e_ is the equiva-
cussed in the article. According to the study of lent plastic strain rate, e_0 is the reference strain rate,
Karpat,15 the flow stress calculated from the J–C model Troom is the room temperature (25 °C), Tmelt is the melt-
parameters listed in Lee and Lin14 do not agree very ing temperature of workpiece material (1660 °C for
well with the experimental data at a strain rate of Ti–6Al–4V in this work). The parameter A is the quasi-
800 s21 and at room temperature. static yield strength of the material (MPa), B is the
Optimization technologies were developed to deter- hardening modulus (MPa), n is the hardening index, C
mine constitutive model parameters using various is the strain rate sensitive coefficient and m is the
objective functions to assess the difference between thermo-softening index.
experimental and computed data. Dusunceli et al.16 Some J–C parameter sets for Ti–6Al–4V alloy are
applied the genetic algorithm (GA) optimization tech- listed in Table 1. Although some of these sets are based
nique to determine the parameters for the viscoplasti- on the same SHPB test data, the calculated J–C para-
city model based on overstress (VBO model) by meters vary in a wide range owing to the possible dif-
minimizing the errors between experimental and simu- ferences in microstructures and/or the mathematical
lated stress. Besides, Cao and Lin17 developed an opti- techniques applied in calculation.15 Different material
mization technique focused on objective function to parameters will lead to different simulation results
determine material constants in creep/viscoplastic con- when the J–C model is applied in machining simula-
stitutive equations based on experimental data. tion. Umbrello et al.10 studied the influence of J–C
Similarly, Anaraki et al.18 presented an inverse method parameters on the results of AISI 316L steel machining

Downloaded from pib.sagepub.com at HOWARD UNIV UNDERGRAD LIBRARY on January 26, 2015
Chen et al. 1289

Table 1. J–C model parameters for Ti–6Al–4V alloy.

Source A/MPa B/MPa n C m e_0 /s–1 Method

Seo et al.9 in 2005 997.9 653.1 0.45 0.0198 0.7 1 SHPB

Lee and Lin14 in 1998 724.7 683.1 0.47 0.035 1 1025 SHPB
Özel and Karpat20 in 2007 987.8 761.5 0.41433 0.01516 1.516 2000 Orthogonal cutting
Lee and Lin21 in 1998 782.7 498.4 0.28 0.028 1 1025 SHPB
Meyer and Kleponis22 in 2001 862.5 331.2 0.34 0.012 0.8 1 SHPB
Pittalà and Monno23 in 2011 0 1508 0.049 0.067 0.71 1000 milling

SHPB: split Hopkinson pressure bar.

functions are defined in the present study in terms of a

particular form of difference between the calculated
and experimental data for a stress–strain curve.
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
X
n1 X m1
f1 (x) = (sci, j  sei, j )2 ð2Þ
j=1 i=1
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
X
n2 Xm2
f2 (x) = (sck, l  sek, l )2 ð3Þ
l=1 k=1

where f1(x) and f2(x) are the sums of error functions

under two strain rates, 2000 and 2500 s21, respectively.
x = [A, B, n, C, m] represents the design variables, i.e.
the J–C parameters. m1 is the number of stress–strain
data for each temperature j at strain rate of 2000 s21,
Figure 1. The flow stress calculated by J–C parameters of Lee whereas n1 is the number of temperature setting consid-
and Lin14 and the SHPB test results at 800 s21 and 25 °C. ered; while m2 is the number of stress–strain data for
each temperature l at strain rate of 2500 s21, whereas n2
is the corresponding number of temperature values con-
simulation. Besides, Umbrello24 investigated the effect sidered. sci, j and sei, j are the computational and experi-
of J–C model parameters on the predicted segmented mental stresses, respectively, for the same strain level i
chip morphology in Ti–6Al–4V alloy orthogonal cut- and temperature j at strain rate of 2000 s21; sck, l and
ting simulation. These previous studies have demon- sek, l are the computational and experimental stresses,
strated that material constitutive parameters have a respectively, for the same strain level k and temperature
significant impact on simulation results, such as resi- l at strain rate of 2500 s21. In addition, sci, j and sck, l are
dual stresses, forces and chip morphology. Lee and calculated using the J–C model equation (1); sei, j and
Lin14 performed the SHPB tests at different tempera- sek, l are obtained using the experimental results of Lee
tures and strain rates, and they developed a set of J–C and Lin.14 Lee and Lin14 applied the form of
parameters according to the test results. However, the
stress calculated by the J–C parameters presented in s = A + Ben ð4Þ
the article did not agree very well with the experimental
data, i.e. the stress at 800 s21 strain rate and at 25 °C to represent the measured stress–strain relationship
(Figure 1). Thus, the main objective of this study is to under each strain rate and temperature condition.
acquire a set of accurate J–C model parameters using Therefore, sei, j and sek, l are obtained by providing the
an optimization method according to the experimental strain data into equation (4) at specific strain rates and
results of SHPB tests in Lee and Lin.14 temperatures.
The ultimate objective is to acquire a set of appro-
priate parameters to minimize f1(x) and f2(x) simul-
Determination of constitutive parameters taneously. This is a multiple objective optimization
by genetic algorithm problem. However, there is no obvious conflict in opti-
mizing the stress–strain responses at different tempera-
Objective function tures and strain rates. The stress–strain responses at
Optimization techniques can be applied to identify different strain rates can be considered as a set of
material parameters by minimizing a particular norm curves, which are similar to those at different tempera-
of the errors between the calculated and the experimen- tures. Hence, for simplicity, this multiple objective opti-
tal results without any conversion.25 Two objective mization problem is translated into a simple objective

Downloaded from pib.sagepub.com at HOWARD UNIV UNDERGRAD LIBRARY on January 26, 2015
1290 Proc IMechE Part B: J Engineering Manufacture 226(8)

problem, and the penalty function method26 is also 2. The initial population is generated randomly.
induced. The ultimate objective function is given by 3. Each individual xi in the population is evaluated
by the fitness function.
f(x) = c1 f1 (x) + c2 f2 (x) ð5Þ 4. The optimization procedure is terminated if the fit-
function(A, B, n, C, m) = Minff(x)g ð6Þ ness function related stopping criterion (the
weighted average change of the fitness function
where c1 and c2 are weight coefficients for f1(x) and
value is less than function tolerance over a speci-
f2(x).
fied generations) is satisfied, i.e. the mean fitness
of the population does not evolve after a finite
Computation procedure number of generations. Otherwise, if gi . gm, then
Compared with the conventional optimization meth- optimization is terminated.
ods, the GA has the advantage of solving the 5. If the stopping criteria are not satisfied, then the
constrained non-linear optimization problem. This genetic operators, such as reproduction, crossover
technique is a stochastic search method based on evolu- and mutation on the population are executed
tion and genetics, exploiting the concept of survival of (Figure 3). The individuals in population Pi are
the fittest.27 A fitness function F(x) derived from the ranked according to the values of fitness function in
objective function is used in the successive genetic oper- the reproduction process. Some parent individuals
ations. Individuals with high fitness will survive to the with high fitness values are selected as elite children
next generation. In the current study, optimization is a (EC), which are automatically survived to the next
minimization problem, so the fitness is defined as generation Pi + 1 (Figure 3). In addition, there are
two other types of children, namely, crossover chil-
F(x) = C0  f(x) ð7Þ dren (CC) and mutation children (MC), which are
where C0 is a predefined constant to ensure that the fit- both produced by parent individuals. The individu-
ness function has a positive value. als that have high fitness values are usually selected
The schematic of the GA optimization procedure is as parent individuals in the GA optimization. The
shown in Figure 2. The step-by-step optimization pro- CC are created by combing the chromosome (e.g.
cess is outlined as follows. binary string) of the parent individuals. The muta-
tion operator is used to reduce the probability of a
1. The design variables of the optimization are cho- premature loss of genetic material caused by repro-
sen and the domains of material parameters are duction and crossover.19 The MC are created by
inputted. The population size k, maximum number changing one variable on a chromosome, selected at
of generation gm, crossover probability lc and random, from a single parent individual.
mutation probability lm are defined before the 6. The children population is generated by reproduc-
optimization program is started. tion, crossover and mutation operations. The number
of generations is set as gi = gi + 1. Step 3 is repeated
until one of the termination criteria is satisfied.

The GA technique can overcome the difficulty of

choosing the correct initial values for the constants that
exist in traditional optimization techniques.19 The

Figure 3. The genetic operators of the GA optimization.

Figure 2. The schematic of GA optimization procedure. EC: elite children; MC: mutation children; CC: crossover children.

Downloaded from pib.sagepub.com at HOWARD UNIV UNDERGRAD LIBRARY on January 26, 2015
Chen et al. 1291

domains of design variables (J–C parameters) and the ten generations compared with subsequent generations.
variables of optimization are specified at the beginning: Afterwards, the value of the objective function
the population size k = 1000, maximum number of gen- decreases gradually toward a stable value, until it
eration gm = 100, crossover probability lc = 0.8 and reaches 542.6 when the optimization is terminated.
mutation probability lm = 0.06. The domains of J–C That is, the specified error (calculated by equations (2)
parameters are defined as 700 MPa 4 A 4 1000 MPa, to (5)) between the experimental and calculated stress
600 MPa 4 B 4 1000 MPa, 0 4 n 4 1, 0 4 C 4 1 and using the optimized J–C model M2 (Table 2) is 542.6,
0.5 4 m 4 1.5. The initial population is generated whereas the error between the experimental and calcu-
within the domains. The GA procedure begins with the lated stress using the J–C model M1 (Table 2) is
initial population that is composed of random individu- 7836.2. Therefore, the flow stress calculated using J–C
als, i.e. the design variables. Then, the optimal design model M2 is closer to the experimental flow stress than
variables are determined by computing the fitness func- that calculated using model M1 at the specified tem-
tion, generation by generation until one of the termina- peratures and strain rates as mentioned above.
tion criteria is reached. The experimental and calculated flow stress at differ-
ent temperatures is shown in Figure 6(a) and (b). The
Optimization results calculated values are obtained using the optimized J–C
The objective function is calculated using equations (2) model M2. The errors between experimental and calcu-
to (6) to minimize the difference between the experi- lated stress (given by equation (8)) at different tempera-
mental and calculated stresses. The experimental data tures are shown in Figure 7(a) and (b). The calculated
are obtained by SHPB tests from Lee and Lin.14 The flow stress shows excellent agreement with the experi-
stress–strain data at different temperatures (25 °C, mental values, except for a few data at the deformation
300 °C, 500 °C, 700 °C, 900 °C and 1100 °C) are consid- initiation when e \ 0.01. The error is lower than 4%
ered in the objective function. The strain rate in at a strain rate of 2000 s21, whereas the error is lower
machining processes is high and varies with cutting than 3% at a strain rate of 2500 s21, when e 5 0.01.
speeds. Therefore, two different strain rates, 2000 and
sei, j  sci, j
2500 s21, are also involved in the objective function. Errori, j = 3 100% ð8Þ
The referenced experimental flow stress curves are sei, j
shown in Figure 4. Although the flow stress is affected
In order to see the optimized results at different
by temperature and strain rate, the effect of tempera-
strain rates (2000 and 2500 s21), the flow stress calcu-
ture is more pronounced than that of strain rate.
lated by the J–C model with two sets of parameters
The flow stress of the J–C model is calculated using
(M1 and M2 in Table 2) and the experimental results
equation (1). The objective function of the GA-optimized
obtained by the SHPB test at 300 °C and 900 °C are
result is shown in Figure 5. The design variables (M2)
shown in Figure 8(a) and (b). A strain rate hardening
obtained by the GA optimization and the fitted J–C
phenomenon is evident, both in the experimental and
model (M1) given by Lee and Lin14 are listed in Table 2.
calculated curves. A higher strain rate leads to higher
Using the GA technique, the value of the objective
stress at identical temperatures. However, the sensitiv-
function decreases rapidly within the first ten genera-
ity of the strain rate hardening is weak at the two afore-
tions, i.e. the difference between the experimental and
mentioned strain rates, especially for the calculated
calculated flow stress is minimized markedly in the first
results. For instance, the maximum distance of

Figure 4. Stress–strain responses based on the data of Lee and Figure 5. The objective function result of the optimization
Lin14 at different temperatures and strain rates. procedure.

Downloaded from pib.sagepub.com at HOWARD UNIV UNDERGRAD LIBRARY on January 26, 2015
1292 Proc IMechE Part B: J Engineering Manufacture 226(8)

Table 2. Comparison of the two sets of J–C model parameters.

Model A/MPa B/MPa n C m e_0 /s21

M1: Lee and Lin14 in 1998 724.7 683.1 0.47 0.035 1 1025
M2: Optimized result 888.5 600 0.21 0.007 1.012 1025

Figure 6. Comparison of experimental14 and calculated flow stress obtained by optimized J–C model M2 at different temperatures:
(a) strain rate = 2000 s21; (b) strain rate = 2500 s21.
J–C: Johnson–Cook.

Figure 7. Errors between experimental14 and calculated flow stress obtained by optimized J–C model M2 at different
temperatures: (a) strain rate = 2000 s21; (b) strain rate = 2500 s21.

experimental stress at the two different strain rates is Ti–6Al–4V alloy machining simulation
20 MPa, whereas that of calculated stress using M2 is for model parameters validation
2 MPa, at a temperature of 900 °C. This phenomenon
may be related to the objective function of the GA pro- Ti–6Al–4V alloy orthogonal cutting experiments
cedure, which aims to minimize the errors between A set of FE simulation and titanium alloy orthogonal
experimental and calculated stresses at the two different cutting experiments were developed to evaluate the
strain rates. On the other hand, the stress calculated by accuracy of the optimized J–C model parameters in
model M2 is closer to the experimental results than that describing material deformation behavior of the machin-
calculated by M1 at temperatures of 300 °C and 900 °C ing process. In this work, three different tests (Test1:
(Figure 8). v = 60 m/min, f = 80 mm/rev; Test2: v = 80 m/min,

Downloaded from pib.sagepub.com at HOWARD UNIV UNDERGRAD LIBRARY on January 26, 2015
Chen et al. 1293

Figure 8. Comparison of experimental14 and calculated flow stress obtained by J–C models M1 and M2 at different strain rates: (a)
T = 573 K; (b) T = 1173 K.

f = 100 mm/rev; and Test3: v = 130 m/min, f = 100 mm/ comparing the experimental and simulated results. The
rev) were performed for a Ti–6Al–4V alloy orthogonal geometry of the 2D FE model is shown in Figure 10.
dry cutting experiment and FE simulation. The workpiece was meshed with 8000 plane strain ther-
The orthogonal cutting experiments were performed mally coupled quadrilateral elements (type CPE4RT),
on a vertical machining center type NEXUS410B-HS. and the tool was meshed with 255 elements. The char-
The workpiece was composed of Ti–6Al–4V alloy disks acteristic dimensions of the tool were in accordance
100 mm in diameter and 2.03 mm thick. The experiment with the experimental condition: rake angle a = 3°,
set-up is shown in Figure 9. The cutting tool was clearance angle b = 8° and edge radius r = 5 mm. The
an uncoated carbide insert type TPGN 110304 workpiece was fixed, meanwhile, a velocity along the X
(Kennametal Inc.) with 3° rake angle, 8° clearance direction was exerted on the tool to accomplish the cut-
angle and 0° inclination angle. The cutting edge radius ting process.
was approximately 5 mm. The cutting forces were moni- The J–C plastic material model (equation (1)) and
tored by a dynamometer (Kistler 9257A) and a charge an energy density-based ductile failure criterion, which
amplifier (Kistler 5070A). A new insert was used for was illustrated in a previous work,12 were applied to
each test to avoid tool wear in the cutting process. the FE model in the cutting simulation. Besides the
energy density failure criterion, a shear failure criterion
FE modeling is assigned to a separation layer to separate the chip
from the workpiece (Figure 10). The failure parameters
A two-dimensional (2D) FE model was developed to
were identical to those of the previous work.12
validate the optimized J–C model parameters by
The cutting conditions of the simulation were in
accordance with those in the orthogonal cutting

Figure 9. Schematic diagram of the orthogonal cutting

experiment set-up. Figure 10. The geometry of the cutting FE model (f = 80 mm/rev).

Downloaded from pib.sagepub.com at HOWARD UNIV UNDERGRAD LIBRARY on January 26, 2015
1294 Proc IMechE Part B: J Engineering Manufacture 226(8)

Figure 11. Comparison of experimental chip morphology (Exp.) and simulation results (Sim.) obtained by J–C models M1 and M2
in various cutting conditions.

experiments. The J–C model M1 and the optimized segmented chip. In addition, the thermoplastic instabil-
model M2 (Table 2) were both applied in the cutting ity theory was also proved by the relationship of strain
simulation to compare the original and optimized J–C and temperature along shear bands in the Ti–6Al–4V
model parameters. Except for the changes of J–C alloy machining simulation.30
model parameters, the other parameters remained Chip morphology was characterized by the average
unchanged to study the effect of material parameters dimension of peak, valley and pitch (Figure 11(a)) to
on the machining simulation. compare the experimental chip morphology with the
simulation results using model M1 and M2. Each
dimension value was measured three times to reduce
Simulation results and discussion the measurement error. The measured average dimen-
The predicted and experimental chip morphology is sions of the segmented chip are given in Table 3.
shown in Figure 11, in which Figure 11(a)–(c) are pre- The average peak dimensions of M1 and M2 agree
dicted using model M1, whereas Figure 11(a’)–(c’) are well with the experimental results in Tests 1 to 3, with
predicted using model M2. relative errors lower than 15%, and they are similar
Obvious segmented chip morphology exists both in with each other. The errors between the predicted (M1
the predicted and experimental results. The predicted and M2) and experimental valley dimensions (in Tests 1
temperature distribution is illustrated in the simulation to 3) are also lower than 15%, except for M1 (20.7% in
results. Many studies have discussed the mechanism of Test 1). The predicted valley of the M2 model is closer
segmented chip formation. Calamaz et al.1 classified to the experimental results than that of the M1 model
the mechanism of chip formation into two groups: 1) in Tests 1 and 2. In addition, the average valley dimen-
the thermoplastic instability; 2) the crack initiation and sion of M1 is smaller than that of M2. This phenom-
propagation. Hua and Shivpuri28 developed the tita- enon may be caused by the material plastic flow along
nium alloy machining simulation and stated that crack the shear bands. The density of material ductile failure
initiation and propagation are the main causes of seg- energy Ge defined by equation (9) is constant during the
mented chip. On the other hand, Komanduri and material failure process. Therefore, if the stress of M1
Hou29 developed a thermal model for the thermoplastic (sy1) is larger than that of M2 (sy2) (Figure 8), then the
instability and stated that thermoplastic instability variation of plastic strain along the shear band De1 in
caused by thermal softening along the adiabatic shear M1 will be smaller than De2 in M2. Hence, model M1 is
band is one of the main factors that leads to a easier to reach the state of complete failure from failure

Downloaded from pib.sagepub.com at HOWARD UNIV UNDERGRAD LIBRARY on January 26, 2015
Chen et al. 1295

Table 3. Comparison of measured chip dimensions in experiments and simulations.

Cutting condition Model Peak (mm) Valley (mm) Pitch (mm)

Test1 Exp. 123.5 92.3 37.5

v = 60 m/min Sim. M1 (error) 109.8 (11.1%) 73.2 (20.7%) 44.9 (–19.7%)
f = 80 mm/rev Sim. M2 (error) 108.3 (12.3%) 81.2 (12%) 37.4 (0.26%)
Test2 Exp. 144 101.3 79.4
v = 80 m/min Sim. M1 (error) 134.5 (6.6%) 94 (7.2%) 53.8 (32.2%)
f = 100 mm/rev Sim. M2 (error) 133.2 (7.5%) 98.8 (2.5%) 52.7 (33.6%)
Test3 Exp. 128 82.2 76.6
v = 130 m/min Sim. M1 (error) 125.8 (1.7%) 89 (–8.3%) 44.7 (41.6%)
f = 100 mm/rev Sim. M2 (error) 131.2 (–2.5%) 93 (–13.1%) 57 (25.59%)

initiation, which will lead to a larger plastic deforma- respectively, the errors between M2 and experiment are
tion along the shear band in M1, inducing a relatively lower than 15% (12.9%, 13.3% and 10.2% for Tests 1
small chip valley dimension to 3, respectively). As the measurement error and the
ð epl differences of material physical behavior exist in the FE
f
Ge = sy de ð9Þ model and actual machining, the predicted forces can be
eoi considered as reasonable for model M1 and M2.

Comparing with the peak and valley dimensions, larger

relative errors exist for the pitch dimension, as illustrated Conclusions
in the results of Tests 2 and 3. The pitch dimensions of
model M2 are closer to the experimental results than A genetic algorithm was proposed to minimize the errors
those of M1, except Test 2. Although there is no distinct between the flow stress obtained through the J–C consti-
difference between M1 and M2 in the prediction of peak tutive model parameters and the SHPB experimental
and valley dimensions, the pitch sizes of M2 are closer results in Lee and Lin14 at different temperatures (25 °C
to the experimental results than those of M1. Thus, the to 1100 °C) and high strain rates (2000 and 2500 s21). In
optimized J–C model M2 is more appropriate to be addition, Ti–6Al–4V alloy orthogonal cutting experi-
applied in the titanium alloy machining simulation. ments and the corresponding FE simulations were per-
The predicted (M1 and M2) and experimental princi- formed to validate the optimized J–C model parameters.
ple cutting forces of Tests 1 to 3 are shown in Figure 12. In summary, the following conclusions can be drawn.
The predicted forces of M1 are larger than the experi-
mental results, contrary to the relatively small predicted 1. The GA technique can be applied to minimize the
forces of M2. Based on the stress–strain curves of M1 difference between experimental and calculated
and M2 models (Figure 8), higher stress will lead to a material flow stress.
higher cutting force in the same FE model under the 2. The flow stress obtained through the optimized J–C
same strain condition. Although the predicted forces model M2 is closer to the experimental data than
using M1 are closer to the experimental results, the that of model M1. The percentage differences
errors of Tests 1 to 3 are 6.8%, 5.7% and 7.9%, between the flow stress of the J–C model M2 and
the experimental data at 2000 and 2500 s21 are
lower than 4% and 3%, respectively. Therefore, the
flow stress obtained through the J–C model at dif-
ferent temperatures and strain rates can characterize
the material flow stress obtained in the SHPB tests.
3. The experimental chip morphology and the FE
simulation results obtained by the two sets of the
J–C model (M1 and M2) showed no distinct differ-
ence between models M1 and M2 for peak and val-
ley dimensions. In addition, the valley dimension
of the segmented chip is greatly dependent on the
J–C model parameters. At a constant density of
material ductile failure energy in the material fail-
ure process, higher stress will lead to a shorter fail-
ure process, and thus cause a smaller valley
dimension. The predicted pitch dimensions of the
J–C model M2 are closer to the experimental
Figure 12. The predicted (with model M1 and M2) and results than those of M1. Thus, the GA-optimized
experimental principle cutting forces. material parameters M2 are more appropriate for

Downloaded from pib.sagepub.com at HOWARD UNIV UNDERGRAD LIBRARY on January 26, 2015
1296 Proc IMechE Part B: J Engineering Manufacture 226(8)

the simulation of the titanium alloy machining 10. Umbrello D, Saoubi RM and Outeiro JC. The influence
process than model M1. of Johnson–Cook material constants on finite element
4. The percentage errors between the predicted (M1 simulation of machining of AISI 316L steel. Int J Mach
and M2) and experimental principle cutting forces Tools Manuf 2007; 47: 462–470.
are lower than 15%. Hence, the predicted forces 11. Özel T. Computational modelling of 3D turning: Influ-
can be regarded as reasonable for both M1 and M2 ence of edge micro-geometry on forces, stresses, friction
and tool wear in PcBN tooling. J Mater Process Technol
models.
2009; 209: 5167–5177.
5. The optimized J–C model M2, which considers dif- 12. Chen G, Ren CZ, Yang XY, et al. Finite element simula-
ferent temperatures (25 °C to 1100 °C) and strain tion of high speed machining of titanium alloy (Ti-6Al-
rates (2000 and 2500 s21), can be used to character- 4V) based on ductile failure model. Int J Adv Manuf Tech
ize the Ti–6Al–4V alloy deformation mechanism in 2011; 56: 1027–1038.
machining simulation. 13. Sun J and Guo YB. Material flow stress and failure in
multiscale machining titanium alloy Ti-6Al-4V. Int J Adv
Manuf Tech 2009; 41: 651–659.
Funding
14. Lee WS and Lin CF. Plastic deformation and fracture
This research was supported by the National High behaviour of Ti–6Al–4V alloy loaded with high strain
Technology Research and Development (863) Program rate under various temperatures. Mater Sci Eng A 1998;
of China [Grant No. 2008AA042509]. 241: 48–59.
15. Karpat Y. Temperature dependent flow softening of tita-
nium alloy Ti6Al4V: An investigation using finite element
Acknowledgments simulation of machining. J Mater Process Technol 2011;
The authors gratefully thank Mr Yang Lijian for his 211(4): 737–749.
support in developing the optimization model. 16. Dusunceli N, Colak OU and Filiz C. Determination of
material parameters of a viscoplastic model by genetic
algorithm. Mater Des 2010; 31: 1250–1255.
Conflict of interest 17. Cao J and Lin J. A study on formulation of objective
The authors declare that there is no conflict of interest. functions for determining material models. Int J Mech
Sci 2008; 50(2): 193–204.
18. Anaraki MT, Sanjari M and Akbarzadeh A. Modeling of
References high temperature rheological behavior of AZ61 Mg-alloy
1. Calamaz M, Coupard D and Girot F. A new material using inverse method and ANN. Mater Des 2008; 29:
model for 2D numerical simulation of serrated chip for- 1701–1706.
mation when machining titanium alloy Ti–6Al–4V. Int J 19. Lin J and Yang J. GA-based multiple objective optimisa-
Mach Tools Manuf 2008; 48: 275–288. tion for determining viscoplastic constitutive equations for
2. Luo J, Li M, Li X, et al. Constitutive model for high superplastic alloys. Int J Plasticity 1999; 15(11): 1181–1196.
temperature deformation of titanium alloys using inter- 20. Özel T and Karpat Y. Identification of constitutive mate-
nal state variables. Mech Mater 2010; 42: 157–165. rial model parameters for high-strain rate metal cutting
3. Jaspers SPFC and Dautzenberg JH. Material behaviour conditions using evolutionary computational algorithms.
in metal cutting: strains, strain rates and temperatures in Mater Manuf Process 2007; 22: 659–667.
chip formation. J Mater Process Technol 2002; 121(1): 21. Lee WS and Lin CF. High-temperature deformation beha-
123–135. viour of Ti6Al4V alloy evaluated by high strain-rate com-
4. Cockroft MG and Latham DJ. Ductility and workability pression tests. J Mater Process Technol 1998; 75: 127–136.
of metals. J Inst Met 1968; 96: 33–39. 22. Meyer HW and Kleponis DS. Modeling the high strain
5. Johnson GR and Cook WH. A constitutive model and rate behavior of titanium undergoing ballistic impact and
data for metals subjected to large strains, high strain penetration. Int J Impact Eng 2001; 26: 509–521.
rates and high temperature. In: Proceedings of the 7th 23. Pittalà GM and Monno M. A new approach to the
international symposium on ballistics. Netherlands: The prediction of temperature of the workpiece of face milling
Hague, 1983, pp.541–547. operations of Ti-6Al-4V. Appl Therm Eng 2011; 31: 173–
6. Zerilli FJ and Armstrong RW. Dislocation-mechanics- 180.
based constitutive relations for material dynamics calcu- 24. Umbrello D. Finite element simulation of conventional
lations. J Appl Phys 1987; 61(5): 1816–1825. and high speed machining of Ti6Al4V alloy. J Mater Pro-
7. Özel T and Zeren E. Determination of work material cess Technol 2008; 196: 79–87.
flow stress and friction for FEA of machining using 25. Gelin JC and Ghouati O. An inverse solution procedure
orthogonal cutting tests. J Mater Process Technol 2004; for material parameters identification in large plastic defor-
153–154: 1019–1025. mations. Commun Numer Meth Eng 1996; 12: 161–173.
8. Lai XM, Li HT, Li CF, et al. Modelling and analysis of 26. Hu YD. Practical multi-objective optimization. 1st ed.
micro scale milling considering size effect, micro cutter Shang Hai: Shang Hai Science and Technology Press,
edge radius and minimum chip thickness. Int J Mach 1990, in Chinese.
Tools Manuf 2008; 48: 1–14. 27. DeJong K. Evolution computation: recent developments
9. Seo S, Min O and Yang H. Constitutive equation for Ti– and open issues. In: Proceeding of the evolution algorithms
6Al–4V at high temperatures measured using the SHPB in engineering computer science. Finland, Chichester:
technique. Int J Impact Eng 2005; 31: 735–754. Wiley, 1999, pp.43–54.

Downloaded from pib.sagepub.com at HOWARD UNIV UNDERGRAD LIBRARY on January 26, 2015
Chen et al. 1297

28. Hua J and Shivpuri R. Prediction of chip morphology Pi individual population of the genetic
and segmentation during the machining of titanium algorithm
alloys. J Mater Process Technol 2004; 150: 124–133. r tool edge radius
29. Komanduri R and Hou ZB. On thermoplastic shear T material temperature
instability in the machining of a titanium alloy (Ti-6Al- Tmelt melting temperature of workpiece
4V). Metall Mater Trans A 2002; 33: 2995–3010. Troom workpiece transaction temperature
30. Chen G, Ren C, Yang X, et al. Evidence of thermoplastic v cutting speed
instability about segmented chip formation process for x design variables of the optimization
Ti-6Al-4V alloy based on finite element method. Proc
procedure
IMechE, Part C: J Mechanical Engineering Science 2011;
225: 1407–1417.
xi individual design variables of each
population

Appendix a rake angle

b clearance angle
Notations
e strain
A initial yield stress e equivalent plastic strain
B hardening modulus of J–C flow stress epl
f equivalent plastic strain at complete
model failure
C coefficient of strain rate of J–C flow stress eoi equivalent plastic strain at onset of failure
model e_ equivalent plastic strain rate
C0 predefined constant e_0 reference strain rate
c1, c2 weight coefficient for f1(x) and f2(x) lc crossover probability
F(x) fitness function of the genetic algorithm lm mutation probability
f1(x), f2(x) sums of the error functions s stress
Ge density of the ductile failure energy s flow stress at non-zero strain rate
gi number of generations sci, j computational stress for the same strain
gm maximum number of generation level i and temperature j at strain rate
k population size 2000 s21
M1 J–C model parameters in Lee and Lin14 sck, l computational stress for the same strain
M2 J–C model parameters obtained by GA level k and temperature l at strain rate
optimization 2500 s21
m thermo-softening index for J–C flow stress sei, j experimental stress for the same strain
model level i and temperature j at strain rate
m1, m2 number of stress–strain data at each 2000 s21
temperature with strain rate 2000 and sek, l experimental stress for the same strain
2500 s21, respectively level k and temperature l at strain rate
n hardening index for J–C flow stress model 2500 s21
n1, n2 number of temperature at strain rate 2000 sy Mises yield stress
and 2500 s21, respectively f(x) objective function of the genetic algorithm

Downloaded from pib.sagepub.com at HOWARD UNIV UNDERGRAD LIBRARY on January 26, 2015