You are on page 1of 15

Simulation Modelling Practice and Theory 94 (2019) 134–148

Contents lists available at ScienceDirect

Simulation Modelling Practice and Theory


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

New strategies for improvement of numerical model accuracy in


T
machining of nickel-based alloy
Farshid Jafariana, , Soroush Masoudib, Domenico Umbrelloc, Luigino Filicec

a
Faculty of Engineering, Mahallat Institute of Higher Education, Mahallat, Iran
b
Young Researchers and Elite Club, Najafabad Branch, Islamic Azad University, Najafabad, Iran
c
Department of Mechanical, Energy and Management Engineering, University of Calabria, Rende, CS 87036, Italy

ARTICLE INFO ABSTRACT

Keywords: Accuracy and reliability of the numerical results is one of the most important challenges in the
Machining simulation scientific community for finite element modeling (FEM) of cutting processes. It becomes more
Calibration strategy essential in machining of difficult-to-cut materials such as nickel-based alloys where high thermo-
Microstructure prediction mechanical loads are induced into the workpiece. In this paper, new strategies were adopted to
Grain size-based flow stress
improve accuracy of FE modeling of cutting process of Inconel718 superalloy. Firstly, a novel
hybrid strategy was established to simultaneously calibrate controllable simulation parameters. It
was implemented based on the design of experiment, intelligent systems and FEM of cutting
process. Using validation with experimental results, a great improvement was obtained for
prediction of cutting forces, maximum temperature, and chip geometry compared with numerical
results obtained previously by conventional calibration procedure. At the second stage of the
paper, a grain size-based flow stress was developed by implementation of advanced FORTRAN
user-subroutine in FE code to take into account the effect of changes in material properties during
the chip formation. According to the results, implementation of grain size based flow stress has a
significant effect on enhancement of simulation accuracy. Finally, it can be concluded that, the
innovated strategies presented in this paper provide fundamental and useful approaches to im-
prove precision in modeling of other manufacturing processes.

Abbreviations

FEM finite element method


ANN artificial neural network
GA genetic algorithm
DRX dynamic recrystallization
HRC Rockwell hardness measured on the C scale

1. Introduction

Nickel-based super alloys were created for the first time in the 1940s. These super alloys are utilized in conditions with severe
mechanical and thermal loads. Having high strength and hardness at high temperatures, high melting temperature and high


Corresponding author.
E-mail address: fjafarian@mahllat.ac.ir (F. Jafarian).

https://doi.org/10.1016/j.simpat.2019.02.006
Received 29 July 2018; Received in revised form 5 February 2019; Accepted 23 February 2019
Available online 25 February 2019
1569-190X/ © 2019 Elsevier B.V. All rights reserved.
F. Jafarian, et al. Simulation Modelling Practice and Theory 94 (2019) 134–148

Nomenclatures C Johnson–Cook (J-C) model's constant


n Johnson–Cook (J-C) model's constant
m shear friction factor m Johnson–Cook (J-C) model's constant
µ coulomb coefficient Vc cutting speed
h heat transfer coefficient af feed rate
D damage value constant Z Zener–Hollomon parameter
σ flow stress R universal gas constant (8.3145 J/Kmol)
ɛ plastic strain Q released energy (116.7KJ/Mol)
0 reference plastic strain (s−1) d grain size
strain rate (s−1), d0 initial size of grain
T workpiece temperature (°C) a material constant
Tmelt Melting Point k material constant
T0 room temperature (T0 = 25 °C) C0 material constant
F initial material hardness constant C1 material constant
G initial material hardness constant ɛCr critical strain
A Johnson–Cook (J-C) model's constant b a* d0
B Johnson–Cook (J-C) model's constant

resistance to corrosion, wear, creep, thermal fatigue and thermal shock are key features of nickel-base super alloys [1]. In spite of
these superior properties, machining of nickel-based super alloys is very difficult and confronted with many technical and economical
problems mainly due to rapid hardening. Among nickel-based alloys, Inconel718 is the most widely employed alloy and it is utilized
in the gas turbines parts in the aerospace industries [2–3]. In recent years, many empirical studies have been conducted to investigate
the machinability of this superalloy [4]. Also there has been considerable interest in the FE simulation of Inconel718 cutting in recent
years [5]. By using FE simulations, a significant reduction in the time and cost of experimental tests could be achieved. In addition,
some output parameters which are difficult to measure can easily be extracted using the FE models [6].
Material model is the most important and influential factor in accuracy of FEM of machining processes. The Johnson–Cook (J-C)
model is one of the most appropriate models for simulation of machining processes that describe behavior of material at the different
strain, strain rate and temperature [7]. In recent years, many research studies have been conducted to propose a suitable material
model for machining of Inconel718 alloy. Demange et al. [8] investigated the effect of material model on the simulation results of
Inconel718 alloy. Ozel et al. [9] developed a modified material model to consider temperature-dependent flow softening effect.
Predicted cutting forces and chip thickness were reported to have acceptable accuracy. Del Prete et al. [10] proposed a modified
material model to take into account the effect of initial material hardness value on the J–C material model. Using this model, it is
possible to model machining process of Inconel718 at the different hardness value with the same J–C material constants. Klocke et al.
[11] conducted a numerical inverse methodology to determine J–C material constants. The comparison of the predicted temperatures
with the measured ones indicated that the temperature was predicted with an acceptable accuracy. Jafarian et al. [12] compared the
effect of different material models found in the literature with machining experiments and finally, the most suitable J–C model was
recognized for machining simulation of Inconel 718. In another study conducted by Jafarian et al. [13], the new strategy based on the
evolutionary optimization algorithm was introduced to identify the new J–C material constants. It was shown that, the precision of
the new material model is comparable with that of other efficient J–C material models.
In the aforementioned studies, the effect of dynamic events (variation in material properties) during the process was not con-
sidered, while it is dominant event in real machining. Recently, new material models based on the grain size variation (grain size-
based flow stress) have been introduced for machining of titanium and waspaloy [14,15]. In these material models, grain size
variation in the workpiece during the process is incorporated into the material model to provide more realistic and more precise
simulation compared to conventional material models. To implement this material model, the microstructure changes should be
firstly modeled using the customized FE model. In this respect, dynamic recrystallization criterion is mostly taken into account for
modeling of grain size and microhardness variation. According to the mentioned advantages, this material model is deserved to be
developed for machining simulation of Inconel718.
In addition to the material model, other parameters are also effective on machining simulation results. They include frictional and
thermal conditions at tool-chip interface and chip fracture criterion. The sliding and sticking friction conditions at the tool-chip
interface are defined using the coulomb coefficient (μ) and shear friction factor (m), respectively. The thermal condition in cutting
simulations is defined by the heat transfer coefficient (h) [16]. Chip fracture criterion completes the material behavior and its effect is
visible in chip geometry and modeling of serrated chip formation. The effect of these parameters on the machining outputs including
cutting forces, temperature and chip geometry have been studied numerically by researchers [17]. Ceretti et al. [18] conducted
experimental validation to investigate temperature distribution in orthogonal cutting simulations. They defined the heat transfer
coefficient as a function of the pressure and temperature at tool-chip interface. Filice et al. [19] investigated the effect of different
friction conditions on cutting force and compared the numerical results with corresponding experimental ones. Ozel et al. [20]
defined the shear friction factor (m) as a function of pressure. Umbrello et al. [21] selected a constant value for the friction condition

135
F. Jafarian, et al. Simulation Modelling Practice and Theory 94 (2019) 134–148

by the trial and error procedure to calibrate cutting forces. Karpat and Ozel [22] defined the shear friction factor as a function of
shear stress. Klocke et al. [11] considered a high heat transfer coefficient in all machining conditions in order to quickly reach the
thermal steady state conditions. Jafarian et al. [23] conducted experimental and numerical investigation for calibration of thermal
conditions in 3D machining of Inconel718 alloy. Using the trial and error calibration procedure, the heat transfer coefficient was
found as a function of machining parameters. In addition, many studies have been carried out for modeling of serrated chip formation
using the different fracture criteria. Umbrello et al. [24] used the Cockcraft and Latham's fracture criterion for different material
hardnesses and satisfactory agreement with experimental results of chip morphology was reported. It should be noted that, for all of
the previous studies the simulation parameters were separately calibrated only for one output, while each parameter can affect
several outputs during the process. For example, the chip fraction criterion simultaneously influences chip geometry and cutting
force. Similarly, the friction condition also changes both cutting forces and tool/chip interface temperature. Since these parameters
have remarkable influence on accuracy of simulation results, the simultaneous calibration of simulation parameters is highly ne-
cessary in order to improve precision of the numerical results.
Reviewing the literature shows that a comprehensive research for simultaneous calibration of simulation parameters is not
available. Unfortunately, the trial and error strategy is mostly employed for calibration of simulation parameters in machining
processes. In fact, it is really important to establish a new strategy that is able to simultaneously calibrate all the machining simu-
lation parameters compared with experimental ones. In addition, the J–C material model is often taken into account for modeling of
chip formation and dynamic recrystallization during the process is neglected. Owing to the fact that material behavior during the
large plastic deformations is highly influenced by microstructure changes, neglecting the dynamic events (such as grain refinement)
in material model leads to the unavoidable limitations in precision of material model. Unfortunately, this item has not been ex-
tensively applied to modeling of machining processes due to the some limitations such as development of customized FE model.
According to the limitations mentioned at the above paragraph, at the present study, new methods were established for improving
accuracy in FE simulation of cutting process. Firstly, a novel hybrid strategy was proposed based on the design of experiment,
intelligent systems (including predictive and optimizing methods) and FE simulation to simultaneously calibrate all of the simulation
parameters in machining of Inconel718.To evaluate performance of the approach, the new results were successfully compared with
several corresponding experimental ones and previous numerical results obtained by conventional calibration procedure. In the
second section of the paper, a modified J–C material model (grain size-based flow stress) was presented to take into account grain size
variation during the process into the material model. The effect of this new material model on precision of the numerical results was
investigated for prediction and validation of microstructure changes.

2. FE modeling of orthogonal cutting process

In order to model and simulate orthogonal cutting process of Inconel718, DEFORM-2D v10.0 software was used. The update-
Lagrangian code with remeshing method was employed for simulations. In this method, a high density mesh at the tool-chip interface
was utilized, which leads to higher simulation accuracy. The workpiece was modeled as a plastic material with isotropic hardening
and was meshed with 15,000 quadrilateral elements. In order to improve the simulation accuracy, in the region near the cutting area,
a finer mesh (the size of each side: 1 µm) was used. In the simulations, tool was modeled to be a rigid body and was meshed with 8000
elements. For defining friction condition, a hybrid model based on the sticking-sliding area was utilized. In the model, the sliding and
sticking areas were defined with coulomb coefficient (µ), and shear friction factor (m), respectively. The thermal condition at tool-
chip interface was controlled by the heat transfer coefficient (h). The Cockcroft and Latham's fracture criterion was implemented for
simulation of serrated chip formation. In this criterion, the damage value constant (D) was taken into account for calibration of chip
geometry. The reader can refer to Ref. [12] for further information. In this section of the paper, to define the material behavior in
cutting simulation of Inconel718, the hardness-based flow stress was utilized. The model's constants were determined in the previous
study [13] as the most appropriate model for the cutting simulation of Inconel718. The initial workpiece hardness is defined using the
following equations:
F (HRC) = 2.008HRC2) 141.97 + 2305.4 (1)

G (HRC) = 0.292(HRC2) + 28.72 700.3 (2)


m
T Troom
= (A + F + G + B n) 1 + Cln 1
0 Tmelt Troom (3)
where, Rockwell hardness measured on the C scale, σ is flow stress, ɛ is plastic strain, 0 is reference plastic strain (s−1), is strain rate
(s−1), T is workpiece temperature (°C), Tmelt is melting point (1300 °C for Inconel 718) and T0 is room temperature (T0 = 25 °C). A, B,
C, n and m are the Johnson–Cook (J-C) model's constants and are selected as the most appropriate ones according to the study
performed by Jafarian et al. [13]. The F and G are parameters that define the workpiece initial hardness in the material model.

3. Experiments

In order to verify the simulation results, the results reported in the previous study were utilized [13]. Experiments were conducted
on a CNC machine using a carbide tool (DNMG ISO SDNMG150616-Sandvik) with 10 µm cutting edge radius. The orthogonal ma-
chining tests with 2 mm depth of cut (disk thickness) were conducted under dry condition. The tool holder provided a −4° top rake

136
F. Jafarian, et al. Simulation Modelling Practice and Theory 94 (2019) 134–148

angle and 6° clearance angle. Before machining, all workpieces were solution treated and aged to a hardness value of 45 HRC. The
orthogonal cutting tests were carried out to investigate the effect of cutting speed and feed rate at 9 testing conditions (ID_1 to ID_9).
In all experiments, cutting forces, maximum temperature and chip geometry were measured. The thermocouple (K-type) was em-
bedded between the insert and tool-holder. The local temperature in this location was used to determine maximum machining
temperature generated in cutting edge by means of inverse numerical methodology [13]. Each experiment was repeated three times
to validate the results. Fig. 1 shows the experimental orthogonal cutting tests utilized for measuring cutting forces, maximum
temperature, and chip geometry. The different levels considered for each parameter are presented in Table 1.

4. New hybrid strategy for calibration of simulation parameters

As mentioned earlier, the simulation parameters (m, µ, h, D constants) are one of the most important factors influence simulation
results. Each parameter directly affects two or more process outputs (including cutting forces, maximum temperature, and chip
morphology). In fact, point by point calibration of the simulation parameters versus each process output is not adequate to provide
excellent calibration. Therefore, in order to improve simulation accuracy as far as possible, it is really important to simultaneously
calibrate simulation parameters. Fig. 2 shows the flowchart of the applied strategy. The purpose of this strategy is to find the optimal
simulation parameters (m, µ, h, D constants) for each machining to obtain the minimum total error for prediction of process outputs.
To implement and verify the proposed strategy, 9 different testing conditions were considered. In this regard, 6 and 3 testing
conditions were employed, respectively, for testing (Testing_Exp) and validation (Validation_Exp) of the method.
The proposed strategy was implemented in 3 steps. The first step is based on the design of experiments, numerical simulations,
comparing with experiments and extracting the total simulation error. It should be noted that simulations were carried out according
to the 6 experimental tests (Testing_Exp). In the second step, intelligent systems were used for finding the optimal parameters. Using
the Artificial Neural Network (ANN) model, total simulation errors were firstly determined as a function of the simulation para-
meters. Then, the optimal simulation parameters for minimum simulation errors were extracted by combination of ANN and opti-
mization algorithm. In the third step, a new series of numerical simulations was performed according to the optimal simulation
parameters found by the previous step for both testing and validation machining conditions to evaluate efficiency of the proposed
method. Finally, the average error was reported for the validation error and testing error and then, they were compared with the ones
obtained by conventional calibration in our previous study. To provide more information about the adopted strategy, the obtained
results for each section are explained in detail.

4.1. Step 1

In this step, a combined effect of the simulation parameters, including sliding and sticking friction coefficients, heat transfer
coefficient and chip fracture criterion were investigated on the simulation outputs including cutting forces, chip geometry and
maximum temperature. The experimental results (at different feed rates and cutting speeds) were considered to verify the simula-
tions. The Taguchi method (mixed level design) was employed for design of experiments in order to consider a wide range of
simulation parameters under different machining conditions. Regarding the input variables, the number of four simulation para-
meters (including constants D, m, µ, and h) and one machining parameter (cutting speed) were selected in three levels and another
machining parameter (feed rate) was considered in two levels (0.050 and 0.1 mm/rev). In fact, the machining tests with feed rate of
0.075 mm/rev (3 tests in 9 machining tests in Table 1) were used as validation tests in step 3. The range of simulation parameters
were selected based on the values determined by trial-and error strategy at the previous studies [12,13]. In these studies, each
simulation constant was calibrated separately by comparing the specific numerical process output with corresponding experiment.
A total number of 36 experiments at different parameters (D, m, µ, h) and 6 different machining conditions (ID_1, ID_4, ID_7, ID_3,
ID_6, and ID_9 in Table 1) were performed in this step. In addition, the three remaining tests (ID_2, ID_5, and ID_8) were kept for the
final validation (step 3). The results obtained from the simulation were compared with the corresponding experimental ones and the

Fig. 1. Experimental tests: (a) orthogonal cutting process (disk machining) (b) embedded thermocouple in tool holder (c) chip morphology [13].

137
F. Jafarian, et al. Simulation Modelling Practice and Theory 94 (2019) 134–148

Table 1
Cutting conditions and experimental results [13].
ID number Cutting speed Feed rate Cutting force Thrust force Chip peak Chip valley Chip pitch T-maximum workpiece
[m/min] [mm/rev] [N] [N] [µm] [µm] [µm] [°C]

ID-1 50 0.050 505.3 482.7 139.1 96.8 77.4 554.00


ID-2 50 0.075 537.5 478.1 114.8 92 50 684.00
ID-3 50 0.100 748.9 615.1 179.3 132.4 79.2 630.00
ID-4 60 0.050 439.1 458.4 100.8 70.7 72.6 865.00
ID-5 60 0.075 614.1 568.7 148.3 107.8 89.5 985.00
ID-6 60 0.100 665.9 496.3 189.9 122.1 112.9 962.00
ID-7 70 0.050 453 483.5 125.8 88.5 61.4 657.00
ID-8 70 0.075 510.4 496.8 133.6 93 71.4 747.00
ID-9 70 0.100 653.1 488 172.9 124.9 85.1 705.00

Fig. 2. Flowchart of new hybrid strategy for calibration of simulation parameters.

average percentage error (for all of the outputs) was taken into account as the total error. The input parameters and simulation errors
are presented in Table 2. In this table, ID number indicates the machining parameters and levels of cutting speed and feed rate are
given in Table 1. Fig. 3 shows an example of temperature distribution and cutting force curve (for 1 mm depth of cut) in experiment
number 16. As shown in Fig. 3(a), steady state thermal condition at the workpiece is achieved after the simulation of chip formation
with enough length of cut. Fig. 3(b) shows small fluctuations in cutting force due to the serrated chip formation.

4.2. Step 2

Artificial neural network models are able to detect the nonlinear correlation between input and outputs parameters. In fact, an
ANN model is able to predict desired outputs for specific input values [25,26]. In recent years, soft computing methods such as ANN
and GA have been widely applied to predict machining process outputs [27,28]. Accuracy of the prediction is the greatest challenge
in ANN models. For this reason, many researchers have proposed new methods to train ANN and enhance performance of the models.
It has been shown that, training the ANN models using the evolutionary optimization algorithms is one of the most suitable ap-
proaches and it will be more efficient when few number of training data is available [28,29]. Therefore, this method of training is
suitable to be employed for estimation of manufacturing processes. More information about this method is available in [30].

138
F. Jafarian, et al. Simulation Modelling Practice and Theory 94 (2019) 134–148

Table 2
Design of experiment and simulations results.
Input simulation and machining parameters Simulation error compared with experiment (in %)
Number ID number m µ D h Temperature Chip geometry Average cutting forces Overall error

01 ID-1 0.6 0.3 100 100 44 34 22 33


02 ID-4 0.8 0.5 200 10,000 15 11 13 13
03 ID-7 1 0.7 300 1,000,000 21 23 10 18
04 ID-1 0.6 0.3 100 10,000 15 35 19 23
05 ID-4 0.8 0.5 200 1,000,000 16 07 15 13
06 ID-7 1 0.7 300 100 43 32 14 30
07 ID-1 0.6 0.5 300 100 56 23 13 31
08 ID-4 0.8 0.7 100 10,000 19 17 09 15
09 ID-7 1 0.3 200 1,000,000 11 28 13 17
10 ID-1 0.6 0.7 200 100 56 22 15 31
11 ID-4 0.8 0.3 300 10,000 15 17 16 16
12 ID-7 1 0.5 100 1,000,000 01 31 12 14
13 ID-1 0.8 0.7 100 1,000,000 19 36 17 24
14 ID-4 1 0.3 200 100 01 09 11 07
15 ID-7 0.6 0.5 300 10,000 17 30 17 21
16 ID-1 0.8 0.7 200 100 54 21 13 29
17 ID-4 1 0.3 300 10,000 14 14 17 15
18 ID-7 0.6 0.5 100 1,000,000 07 24 09 13
19 ID-3 0.8 0.3 300 1,000,000 09 24 20 18
20 ID-6 1 0.5 100 100 04 23 07 12
21 ID-9 0.6 0.7 200 10,000 11 10 05 09
22 ID-3 0.8 0.5 300 1,000,000 11 09 19 13
23 ID-6 1 0.7 100 100 05 27 12 15
24 ID-9 0.6 0.3 200 10,000 07 14 04 08
25 ID-3 1 0.5 100 10,000 07 18 24 16
26 ID-6 0.6 0.7 200 1,000,000 21 15 06 14
27 ID-9 0.8 0.3 300 100 35 16 09 20
28 ID-3 1 0.5 200 10,000 05 21 21 16
29 ID-6 0.6 0.7 300 1,000,000 01 25 11 12
30 ID-9 0.8 0.3 100 100 28 22 10 20
31 ID-3 1 0.7 300 10,000 11 14 18 15
32 ID-6 0.6 0.3 100 1,000,000 26 23 06 19
33 ID-9 0.8 0.5 200 100 37 18 08 21
34 ID-3 1 0.3 200 1,000,000 09 19 17 15
35 ID-6 0.6 0.5 300 100 01 26 10 12
36 ID-9 0.8 0.7 100 10,000 12 17 04 11

In this step, an ANN model was developed and the correlation between the input and output parameters was determined. First, the
best ANN structure was determined to be a network with one hidden layer and 9 neurons in the hidden layer. The nonlinear Tansig
and Logsig activation functions were selected for the hidden layer and output layer, respectively. Fig. 4 shows the structure of the ANN
model. There are two types of the input parameters including simulation parameters and machining parameters (fixed parameters).
The ANN model is trained so that the output value (overall error) is dependent on both of them. Therefore, it is possible to fix the
machining parameters and find the optimal simulation parameters (D, m, µ, h) that provide the minimum overall error at each specific
cutting conditions. Detailed information about results of this strategy is presented in the following.
After selecting the best topology, the ANN model was trained with genetic algorithm. In order to avoid overlearning of the ANN
model, the early stopping technique was used. The reader can refer to Ref. [29,30] for more information. The number of 30 test data
from Table 2 was taken into account to train the ANN model and 6 remaining data was used for testing the model. The average error
of training and testing was determined as 4.5% and 5.01%, respectively. The error values show that not only are the testing and
training errors in an acceptable range, but also the ANN model was trained without overlearning as the training and testing errors are
close to each other. Fig. 5 shows the predicted results by the ANN model versus the simulation results.
After training the ANN model, the combinational method of ANN-GA was utilized to find the optimal simulation parameters. In
this step, for each specific machining condition, the optimal simulation parameters were determined. To achieve this, the function
implemented by the ANN model was used as an objective function of the optimization algorithm. The simulation parameters (D, m, µ,
h) were considered as optimization variable parameters and the machining parameters (Vc, af) were considered as fixed optimization
variables. This procedure was repeated for all of the testing conditions (ID-number) and the optimal simulation parameters were
found at different cutting speeds and feed rates. The optimal simulation parameters found (for different machining conditions) using
the proposed strategy are presented in Table 3.

139
F. Jafarian, et al. Simulation Modelling Practice and Theory 94 (2019) 134–148

Fig. 3. Simulation results for test number 16 (a) temperature distribution and (b) cutting force.

Fig. 4. The topology of the ANN model.

140
F. Jafarian, et al. Simulation Modelling Practice and Theory 94 (2019) 134–148

Fig. 5. Predicted results by the ANN model versus the numerical errors.

Table 3
Optimal simulation parameters at different machining conditions (Testing_Exp).
ID Number Cutting conditions Optimal simulation parameters
Vc af m µ D h

ID-1 50 0.050 1 0.39 197 226,122


ID-3 50 0.100 1 0.70 212 419,015
ID-4 60 0.050 1 0.55 133 256,822
ID-6 60 0.100 1 0.70 170 498,957
ID-7 70 0.050 1 0.65 100 287,398
ID-9 70 0.100 1 0.70 116 606,863

Table 4
The optimal parameters for validation simulation (Validation_Exp).
ID Vc af m µ D h

ID-2 50 0.075 1 0.65 180 317,316


ID-5 60 0.075 1 0.70 144 347,136
ID-8 70 0.075 1 0.70 110 410,820

4.3. Step 3

After determination of optimal simulation parameters, effectiveness of the method was examined in this step. In order to evaluate
performance of the method, the new simulations (according to the new set-up) were firstly carried out for testing conditions of the
previous step (Testing_Exp) and for new testing conditions that have not been utilized previously (Validation_Exp). In this step, a new
series of simulations based on the optimal simulation parameters (reported in Table 4) were carried out. The obtained results were
compared with corresponding experimental ones and total error was reported for each testing conditions as a new simulation error.
At the previous study, the trial and error method was used for calibration of simulation parameters for these machining conditions
[13]. It was very time-consuming procedure to separately calibrate each constant and at the same time keep other outputs in a
desirable precision. The results obtained at the present study were compared with the corresponding results of the previous study and
were reported in Fig. 6. The results indicate that the applied strategy in this research leads to improvement of simulation accuracy for
all the outputs.
In order to validate the accuracy of the proposed strategy, the strategy was applied to the new validation experiments
(Validation_Exp). Optimization of simulation parameters were performed again and the results were compared with the experimental
ones. The optimal parameters for validation simulation were presented in Table 4 and the comparison between new and old si-
mulation errors for Validation_Exp was reported in Fig. 7.
The average total error of simulations for validation and testing experiments were compared with the previous results in Fig. 8. As
can be seen, the simulation accuracy was improved. In fact, it can be mentioned that simultaneous calibration of simulation para-
meters including sliding and sticking friction coefficients, heat transfer coefficient and fracture criterion lead to the improvement of

141
F. Jafarian, et al. Simulation Modelling Practice and Theory 94 (2019) 134–148

Fig. 6. Comparing the new and old simulation errors for (Testing_EXP).

simulation accuracy. Therefore, the proposed strategy can be implemented as a reliable approach to determine the optimal simulation
parameters and can be utilized for future works along this filed.

5. Correlation between simulation parameters and process conditions

Using the proposed strategy, not only were the optimal simulation parameters determined simultaneously, but also it provides an
excellent opportunity to investigate the effect of cutting parameters on optimal simulation parameters. In more detail, it has been
previously shown that, friction, heat transfer coefficient and fracture criterion are not constant and they should change dependent on
the process condition [10,24]. The constructed ANN-GA model is able to find correlation between the simulation parameters (D, m, µ,
h) and the process conditions (Vc and af) and it was reported in Fig. 9. The shear friction value (m) was found constant (m = 1) for all
9 machining tests. The results presented in Fig. 9 shows that, for low cutting speed (50 m/min), by increasing the feed rate, µ is also
increased significantly. In contrast, it was almost a constant value (µ = 0.7) at the higher cutting speeds. Increase in feed rate and
cutting speed resulted in a higher heat transfer coefficient. It was also found that cutting speed has more effect on damage value
compared to feed rate. In addition, higher cutting speed led to significant decrease in damage value (D), while an increase in feed rate
resulted in a slight increase in D (except for V = 50 m/min).

Fig. 7. Comparing new and old simulation errors for (Validation_Exp).

142
F. Jafarian, et al. Simulation Modelling Practice and Theory 94 (2019) 134–148

Fig. 8. The comparison between total average error for new and old simulations.

6. Considering dynamic phenomenon for improvement of material model precision

As shown at the previous section, the simulation parameters such as frictional and thermal conditions as well as fracture criterion
are effective factors in accuracy of the numerical results. In addition to them, the material model plays a critical role in accuracy of
numerical models. Since cutting process is exposed to severe plastic deformation, the precision of material model becomes more
essential. Therefore, in the following, a new material model is proposed based on the microstructure changes in material during the
process. In order to implement this model, the microstructure change of material was firstly modeled using the dynamic re-
crystallization mechanism. After that, the effect of grain size variation during the process was taken into account at each stage of
simulation using the new material model. In the following, the implemented strategy is introduced.

6.1. FE modeling of microstructure changes

In this research, the mechanical recrystallization mechanism was utilized to simulate microstructural changes in orthogonal
cutting of Inconel718. The Zener–Hollomon equation was used for modeling of recrystallization and subsequently for simulation of
grain size changes in material [31]. Also, the Hall–Petch equation was employed for modeling microhardness variation in workpiece
[32]. In the following, Zener–Hollomon and Hall–Petch equations are introduced:
Q
Z = exp
RT (4)

d = ad 0 Zm (5)

d= bZ m (6)
In which, Z is the Zener–Hollomon parameter, is strain rate, R is the universal gas constant (8.3145 J/Kmol), T is temperature of
the material and Q is the released energy (116.7 KJ/mol). Also, d0 is the initial size of grain, and a and m are the constants of material
and b is equal to ad0. These constants were obtained using the calibration process carried out in the previous study [13]. Micro-
hardness variations which depend on the new grain size of material were calculated using the Hall-Petch equation as follows:
HV = C0 + C1 d 0.5 (7)
where C0 and C1 are constants of material and d is the grain size. The C0 and C1were considered 387 and 298.2, respectively, in
simulations. When strain in material becomes more than a critical level which is called the critical strain (ɛCr), the dynamic re-
crystallization (DRX) will be accrued. Enhancement of critical strain depends on the workpiece material and process type and it is
considered as an important parameter to control depth of the recrystallized layer. In fact, the critical strain must be small enough to
permit starting the recrystallization and also should be high enough to stop it. In this study, the following equation for the critical
strain in cutting simulation of Inconel718 was used [13]. In this equation, the critical strain is introduced as a function of the
temperature and strain rate of material as below:

143
F. Jafarian, et al. Simulation Modelling Practice and Theory 94 (2019) 134–148

Fig. 9. Relationship between (a) μ, (b) h, and (c) D, and cutting parameters.

0.1293 5759.863
Cr = 0.00234 × × exp /107
T + 273 (8)

At each stage of simulation by extracting the temperature and strain rate, the value of Zener–Hollomon parameter and critical
strain were calculated. If the strain rate in material exceeds the critical strain, DRX accrues and the new gain size is calculated. Based
on the dimension of the new grains, the microhardness of material was also calculated using the Hall–Petch equation. If the strain
does not reach the critical level, grain refinement and hardness variation will not report. It should be noted that the subroutine
program was developed so that the mentioned strategy can be implemented for each element during the process.

144
F. Jafarian, et al. Simulation Modelling Practice and Theory 94 (2019) 134–148

Fig. 10. The subroutine flowchart for implementation of the grain size-based flow stress.

6.2. Development of grain size-based flow stress

In conventional material models such as the Johnson–Cook model, which is employed to simulate cutting process, only the effect
of changes in temperature, strain and strain rates is taken into account and any changes in the mechanical properties of material
during the process is neglected. In a cutting process, severe hardening and grain size changes occur in the superficial layers of
material, hence the modeling of dynamic behavior of material during the process is very important and influential and neglecting this
item leads to limitation in accuracy of simulation. Accordingly, in this paper, in order to increase the accuracy of results in the cutting
simulation of Inconel 718, a new material model based on microstructural variations was developed. The new material model was
developed based on the Johnson-Cook material model. To achieve this, an advanced FORTRAN subroutine program was developed in
DEFORM-2D software to implement the following material model:

m
T Troom
= (A + B n) 1 + Cln 1
0 Tmelt Troom (9)

K
A=a+
d (10)

The constants of these equations are similar to the constants utilized in Eq. (3). The only difference is constant A, which represents
the initial material hardness based on Eq. (10). In this equation, d is the grain size value (for each element). Also, a and k are material
constants extracted from literature [14]. In order to implement this material model, the microstructural changes were firstly modeled
using the Zener–Hollomon, Hall–Petch and critical strain equations. Then, the effect of grain size variation was applied to the
material model at each stage of simulations in accordance with Eqs. (9) and 10. Fig. 10 shows the flowchart of the subroutine that
describes development of grain size-based flow stress in numerical modeling of machining process.

145
F. Jafarian, et al. Simulation Modelling Practice and Theory 94 (2019) 134–148

Fig. 11. Customized FE model for prediction of grain size and microhardness variation for ID-8.

6.3. FE validation using the new and old material model

In this section, the microstructural change during the orthogonal cutting process of Inconel718 was modeled using the con-
ventional J–C material model and grain size-based flow stress. To achieve this, for both material models, the advanced FORTRAN
user-subroutine was implemented so that the first model only simulates the microstructure changes during the process and the second
model implements the effects of grain size variations into the material model to take into account dynamic events. The simulations
were performed at cutting conditions of ID-7, ID-8, and ID-9. FE simulation of grain size and microhardness variation in machining of
Inconel718 alloy (for ID_8) is given in Fig. 11. It is obvious that by considering grain size variation in material model, material
behavior during the process and consequently results of the simulation will be modified. In order to investigate and compare the
accuracy of these models, simulation results were compared with experimental results in terms of grain size and micro-hardness
changes at three different machining conditions. The results were presented in Fig. 12. It should be noted that the suitable region to
extract the results is always chosen based on the criteria for reaching the steady-state condition. The length at which the cutting
reaches the steady-state condition is highly dependent on the cutting speed (Vc), workpiece's stiffness (E) and flow stress. Based on
this, the region should be far enough from the chip root and the right edge of the workpiece (in the mid-length of the machined
surface), where the cutting reaches the steady-state condition. For more information in this regard, the reader can refer to Ref. [16].
As shown in this figure, the use of grain size-based flow stress leads to better prediction of microstructure changes at the different
cutting conditions. It shows that considering the dynamic events in the material model modify the material behavior during cutting
process and subsequently improve the accuracy of numerical results.

7. Conclusion

The development of accurate and reliable FE model for cutting processes has received considerable attention from the scientific
community in recent years. In machining of difficult-to-cut materials such as nickel based alloys high thermo-mechanical loads are
induced into the workpiece and FE modeling of the process is encountered with several limitations. To overcome this problem,
controllable simulation parameters such as thermal and frictional conditions at tool-chip interface and chip fracture criterion are
mostly calibrated by conventional methods such as trial and error procedure. Not only is it time-consuming procedure, but also it is
extremely difficult to calibrate them at the same time for all of the outputs. Apart from this, as far as the material model is concerned,
it has great influence on precision of cutting process simulation. Using the conventional J–C material model, the dynamic events
during the process is neglected that leads to some unavoidable limitations. Based on the mentioned limitations, new strategies were
developed in this paper to improve accuracy of FE modeling of cutting process of Inconel718 superalloy.

• Firstly, a novel hybrid strategy was established to simultaneously calibrate controllable simulation parameters. In this respect,
after identification of optimal parameters, the numerical results were successfully validated with several corresponding experi-
mental results.
• Using this strategy, it was also possible to understand the effect of process conditions on variation of optimal simulation para-
meters. On this basis, the effect of cutting speed and feed rate was investigated on frictional and thermal conditions at tool-chip
interface and chip fracture criterion.
• At the second stage of the paper, a grain size-based flow stress was developed to take into account the effect of changes in material
properties during chip formation. To do this, an advanced sure subroutine was developed in the FE code and the microstructural

146
F. Jafarian, et al. Simulation Modelling Practice and Theory 94 (2019) 134–148

change (including grain size and microhardness variation) was initially modeled as new outputs by dynamic recrystallization
(DRX) criterion. Then, the grain size variation was incorporated into the new material model to take into account dynamic events
during the process.
• Experimental results of the grain size and hardness variation were compared with corresponding numerical results obtained using

Fig. 12. The comparison between experimental results of grain size and microhardness and numerical results with J–C and new material model at
cutting conditions of ID-7, ID-8, and ID-9.

147
F. Jafarian, et al. Simulation Modelling Practice and Theory 94 (2019) 134–148

the J–C and new material models. It was shown that implementation of a grain size based flow stress significantly increased the
simulation accuracy.
• Finally, it can be concluded that, innovated strategies presented in this paper provide fundamental and useful approaches for
improving the precision of cutting process simulation. These approaches can also be extended to numerical modeling of other
manufacturing processes in future studies.

References

[1] G.R. Thellaputta, P.S. Chandra, C. Rao, Machinability of nickel based superalloys: a review, Mater. Today 4 (2) (2017) 3712–3721 https://doi.org/10.1016/j.
matpr.2017.02.266.
[2] F. Jafarian, H. Amirabadi, J. Sadri, H.R. Banooie, Simultaneous optimizing residual stress and surface roughness in turning of Inconel718 superalloy, Mater.
Manuf. Processes 29 (3) (2014) 337–343.
[3] H. Chávez-García, K.K. Castillo-Villar, Simulation-based model for the optimization of machining parameters in a metal-cutting operation, Simul. Model. Pract.
Theory 84 (2018) 204–221 https://doi.org/10.1016/j.simpat.2018.02.008.
[4] A. Kortabarria, I. Armentia, P. Arrazola, Sensitivity analysis of material input data influence on machining induced residual stress prediction in Inconel 718,
Simul. Model. Pract. Theory 63 (2016) 47–57 https://doi.org/10.1016/j.simpat.2016.02.005.
[5] Y.M. Arısoy, C. Guo, B. Kaftanoğlu, T. Özel, Investigations on microstructural changes in machining of Inconel 100 alloy using face turning experiments and 3D
finite element simulations, Int. J. Mech. Sci. 107 (2016) 80–92 https://doi.org/10.1016/j.ijmecsci.2016.01.009.
[6] F. Jafarian, H. Amirabadi, J. Sadri, Integration of finite element simulation and intelligent methods for evaluation of thermo-mechanical loads during hard
turning process, Proc. Inst. Mech. Eng. Part B 227 (2) (2013) 235–248 https://doi.org/10.1177/0954405412466995.
[7] M. Agmell, A. Ahadi, J. Zhou, R. Peng, V. Bushlya, J.-E. Ståhl, Modeling subsurface deformation induced by machining of Inconel 718, Mach. Sci. Technol. 21 (1)
(2017) 103–120 https://doi.org/10.1080/10910344.2016.1260432.
[8] J.J. DeMange, V. Prakash, J.M. Pereira, Effects of material microstructure on blunt projectile penetration of a nickel-based super alloy, Int. J. Impact Eng. 36 (8)
(2009) 1027–1043 https://doi.org/10.1016/j.ijimpeng.2009.01.007.
[9] T. Ozel, I. Llanos, J. Soriano, P.-J. Arrazola, 3D finite element modelling of chip formation process for machining Inconel 718: comparison of FE software
predictions, Mach. Sci. Technol. 15 (1) (2011) 21–46 https://doi.org/10.1080/10910344.2011.557950.
[10] A. Del Prete, L. Filice, D. Umbrello, Numerical simulation of machining nickel-based alloys, Procedia CIRP 8 (2013) 540–545 https://doi.org/10.1016/j.procir.
2013.06.147.
[11] F. Klocke, D. Lung, S. Buchkremer, Inverse identification of the constitutive equation of Inconel718 and AISI 1045 from FE machining simulations, Procedia CIRP
8 (2013) 212–217 https://doi.org/10.1016/j.procir.2013.06.091.
[12] F. Jafarian, M.I. Ciaran, P.J. Arrazola, L. Filice, D. Umbrello, H. Amirabadi, Effects of the flow stress in finite element simulation of machining Inconel 718 alloy,
Key Eng. Mater. 611 (2014) 1210–1216 https://doi.org/10.4028/www.scientific.net/KEM.611-612.1210.
[13] F. Jafarian, M.I. Ciaran, D. Umbrello, P. Arrazola, L. Filice, H. Amirabadi, Finite element simulation of machining Inconel718 alloy including microstructure
changes, Int. J. Mech. Sci. 88 (2014) 110–121 https://doi.org/10.1016/j.ijmecsci.2014.08.007.
[14] S. Chatterjee, S.S. Mahapatra, K. Abhishek, Simulation and optimization of machining parameters in drilling of titanium alloys, Simul. Modell. Pract. Theory 62
(2016) 31–48 https://doi.org/10.1016/j.simpat.2015.12.004.
[15] M. Ahmadi, Y. Karpat, O. Acar, Y.E. Kalay, Microstructure effects on process outputs in micro scale milling of heat treated Ti6Al4V titanium alloys, J. Mater.
Process. Technol. 252 (2018) 333–347 https://doi.org/10.1016/j.jmatprotec.2017.09.042.
[16] M. Sadeghifar, R. Sedaghati, W. Jomaa, V. Songmene, A comprehensive review of finite element modeling of orthogonal machining process: chip formation and
surface integrity predictions, Int. J. Adv. Manuf. Technol. 96 (9–12) (2018) 3747–3791 https://doi.org/10.1007/s00170-018-1759-6.
[17] T. Özel, Computational modelling of 3D turning: influence of edge micro-geometry on forces, stresses, friction and tool wear in PcBN tooling, J. Mater. Process.
Technol. 209 (11) (2009) 5167–5177 https://doi.org/10.1016/j.jmatprotec.2009.03.002.
[18] E. Ceretti, L. Filice, D. Umbrello, F. Micari, ALE simulation of orthogonal cutting: a new approach to model heat transfer phenomena at the tool-chip interface,
CIRP Ann. Manuf. Technol. 56 (1) (2007) 69–72 https://doi.org/10.1016/j.cirp.2007.05.019.
[19] L. Filice, F. Micari, S. Rizzuti, D. Umbrello, A critical analysis on the friction modelling in orthogonal machining, Int. J. Mach. Tools Manuf. 47 (3–4) (2007)
709–714 https://doi.org/10.1016/j.ijmachtools.2006.05.007.
[20] T. Özel, Y. Karpat, A. Srivastava, Hard turning with variable micro-geometry PcBN tools, CIRP Ann. Manuf. Technol. 57 (1) (2008) 73–76 https://doi.org/10.
1016/j.cirp.2008.03.063.
[21] D. Umbrello, G. Ambrogio, L. Filice, R. Shivpuri, A hybrid finite element method–artificial neural network approach for predicting residual stresses and the
optimal cutting conditions during hard turning of AISI 52100 bearing steel, Mater. Des. 29 (4) (2008) 873–883 https://doi.org/10.1016/j.matdes.2007.03.004.
[22] Y. Karpat, T. Ozel, Process simulations for 3D turning using uniform and variable microgeometry PCBN tools, Int. J. Mach. Mach. Mater. 4 (1) (2008) 26–38
https://doi.org/10.1504/IJMMM.2008.020908.
[23] F. Jafarian, S. Masoudi, H. Soleimani, D. Umbrello, Experimental and numerical investigation of thermal loads in Inocnel 718 machining, Mater. Manuf.
Processes 33 (9) (2018) 1020–1029 https://doi.org/10.1080/10426914.2018.1424907.
[24] D. Umbrello, J. Hua, R. Shivpuri, Hardness-based flow stress and fracture models for numerical simulation of hard machining AISI 52100 bearing steel, Mater.
Sci. Eng. 374 (1–2) (2004) 90–100 https://doi.org/10.1016/j.msea.2004.01.012.
[25] R.S. Pawade, S.S. Joshi, Multi-objective optimization of surface roughness and cutting forces in high-speed turning of Inconel718 using Taguchi grey relational
analysis (TGRA), Int. J. Adv. Manuf. Technol. 56 (1–4) (2011) 47–62 https://doi.org/10.1007/s00170-011-3183-z.
[26] S. Masoudi, M. Sima, M. Tolouei-Rad, Comparative study of ann and anfis models for predicting temperature in machining, J. Eng. Sci. Technol. 13 (1) (2018)
211–225.
[27] F. Jafarian, H. Amirabadi, M. Fattahi, Improving surface integrity in finish machining of Inconel718 alloy using intelligent systems, Int. J. Adv. Manuf. Technol.
71 (5–8) (2014) 817–827 https://doi.org/10.1007/s00170-013-5528-2.
[28] A.S. Mihăiţă, L. Dupont, M. Camargo, Multi-objective traffic signal optimization using 3D mesoscopic simulation and evolutionary algorithms, Simul. Model.
Pract. Theory 86 (2018) 120–138 https://doi.org/10.1016/j.simpat.2018.05.005.
[29] F. Jafarian, M. Taghipour, H. Amirabadi, Application of artificial neural network and optimization algorithms for optimizing surface roughness, tool life and
cutting forces in turning operation, J. Mech. Sci. Technol. 27 (5) (2013) 1469–1477 https://doi.org/10.1007/s12206-013-0327-0.
[30] F. Jafarian, D. Umbrello, S. Golpayegani, Z. Darake, Experimental investigation to optimize tool life and surface roughness in Inconel718 machining, Mater.
Manuf. Processes 31 (13) (2016) 1683–1691 https://doi.org/10.1080/10426914.2015.1090592.
[31] F.-l. Sui, Y. Zuo, X.-h. Liu, L.-q. Chen, Microstructure analysis on IN 718 alloy round rod by FEM in the hot continuous rolling process, Appl. Math. Model. 37
(20–21) (2013) 8776–8784 https://doi.org/10.1016/j.apm.2013.04.005.
[32] G. Hughes, S. Smith, C. Pande, H. Johnson, R. Armstrong, Hall–Petch strengthening for the microhardness of twelve nanometer grain diameter electrodeposited
nickel, Scr. Metall. 20 (1) (1986) 93–97 https://doi.org/10.1016/0036-9748(86)90219-X.

148

You might also like