Results in Engineering 11 (2021) 100251

Contents lists available at ScienceDirect

Results in Engineering
journal homepage: www.sciencedirect.com/journal/results-in-engineering

Optimization of CNC turning parameters using face centred CCD approach

in RSM and ANN-genetic algorithm for AISI 4340 alloy steel
A. Johnson Santhosh a, *, Amanuel Diriba Tura a, Iyasu Tafese Jiregna a,
Wendimu Fanta Gemechu a, N. Ashok b, Murugan Ponnusamy a
a
Faculty of Mechanical Engineering, Jimma Institute of Technology, Jimma University, Jimma, Ethiopia
b
Faculty of Mechanical & Production Engineering, Arba Minch University, Ethiopia

A R T I C L E I N F O A B S T R A C T

Keywords: AISI 4340 alloy steel is used in a wide range of industrial applications due to its improved hardness, toughness,
Alloy steel fatigue, and wear resistance. The surface roughness of AISI 4340 alloy steel components in a typical CNC ma­
CNC Parameters chine is minimized using CNC machining parameters such as feed rate, rotational speed, and depth of cut. The
Surface roughness
Coded and Actual Empirical model was generated using Face Centred Central Composite Design (CCD) approach
Response surface methodology
in Response Surface Methodology (RSM) to forecast the predicted values. The machining parameters interactions
Artificial neural network
Genetic algorithm are studied using three-dimensional surface plots, and optimal process parameters are predicted with the
Optimization desirability graph. The Artificial Neural Network (ANN) approach is employed to increase the coefficient of
regression (R2) and to get the well-trained best fitness model for the Genetic Algorithm (GA). The confirmation
test results explore experimental surface roughness value, and the percentage of error is less in the Genetic
Algorithm than Response Surface Methodology for AISI 4340 alloy steel components. As a result, this research
suggests using a combination of Artificial Neural Network (ANN) and Genetic Algorithm (GA) methodology to
find the best machining process parameters and get a good performance response in practical applications.

1. Introduction maximum tool—chip interface temperature, and microhardness varia­

tions after machining using ANOVA statistical analysis [8]. Marlon et al.
Manufacturing industries are started to use alloy steel material due to worked on AISI 4340 steel for optimum quenching parameters to
its low cost and enriched mechanical properties [1]. AISI 4340 alloy improve the microstructural and mechanical properties using Design of
steel is low alloy steel with chromium, nickel, and molybdenum that can Experiments (DoE) [9].
be heat treated. In heat-treated conditions, it has high toughness and The product’s surface texture plays a significant role in appearance,
strength [2]. In addition to carbon, alloy steels contain other alloying function, and reliability in industrial and consumer applications [10].
elements that increase hardness, durability, fatigue resistance, and wear Surface roughness directly reduces the aesthetic value and price of the
resistance [3,4]. Power transmission gears and shafts, aircraft landing product [11]. Good surface quality is required for cost-effectiveness,
gear, and other structural components are all made of AISI 4340 alloy functionality, and appearance [12]. Surface roughness can be mini­
steel. Due to a combination of high mechanical strength, ductility, and mized by implementing the optimal process parameter combinations in
stiffness, forged steel crankshafts are used in commercial and military the machining environment [13,14]. The importance of surface rough­
aircraft, automotive systems, forged hydraulic and other machine tool ness in product functionality is discussed by the various literature [15,
applications, and forged steel crankshafts [5–7]. 16].
Cydas selected the AISI 4340 steel to study the performance of Achieving good machining Quality is an essential requirement in any
various cutting tools concerning hardness for different parameters such manufacturing industry, and it depends upon the optimal values of
as Cutting speed, feed rate, depth of cut, workpiece hardness, and tool process parameters of the selected work environment. Some of the
types for output responses in Surface roughness, tool flank wear, process parameters can be controllable, and some process parameters

* Corresponding author. Faculty of Mechanical Engineering, P.O.378, Jimma Institute of Technology, Jimma, Ethiopia. Tel.: +251 472115547; Fax.: +251
471111450.
E-mail addresses: [email protected], [email protected] (A.J. Santhosh).

https://doi.org/10.1016/j.rineng.2021.100251
Received 20 May 2021; Received in revised form 27 June 2021; Accepted 6 July 2021
Available online 28 July 2021
2590-1230/© 2021 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license
(http://creativecommons.org/licenses/by-nc-nd/4.0/).
A.J. Santhosh et al. Results in Engineering 11 (2021) 100251

Fig. 1. CNC Turning machining operation.

cannot control since it depends on various environmental factors. Poor

selection of value and implementation of the machining industries
process parameters results in poor surface finish, low fatigue life,
defected components, tool failures, and performance reduction in the
machines. So the proper selection of process parameters and their
optimal value is a vital task in the machining-related manufacturing
industries.
In the CNC turning operation, machining and geometric tool pa­
rameters play a prominent part [17]. Surface roughness and machining
time can be optimized by MasterCAM software for the CNC milling
process [18]. Arunnath and Haja selected the standard cylindrical
workpiece to reduce the surface roughness by optimizing the CNC
machining parameters such as depth of cut, cutting speed, feed rate
using ANOVA in Taguchi Methodology [19]. Camposeco-Negrete states
Fig. 2. Sample Stepped Shafts used in this study.
that the Taguchi technique minimizes surface roughness and energy
consumption in a CNC turning machine [20]. Nagendra et al. optimized
the surface roughness in Fused Deposition Modelling (FDM) parameters Table 3
in Additive Manufacturing (AM) using L27 orthogonal array using Experimental Data for surface roughness response.
ANOVA [21]. Campatelli et al. studied the behavior of CNC milling Std Run Factor 1 Factor 2 Factor 3 Response
process parameters such as cutting speed, feed rate, radial and axial
A: Feed B: Rotational C: Depth of Surface
depth of cut using Response of surface methodology (RSM) [22]. Raykar Rate Speed cut Roughness
et al. work on Al 7075 material to optimize surface roughness, power,
mm/Rev rpm mm μm
and time with low cutting speed, feed rate, and depth of cut using
Taguchi Methodology [23]. Qureshi, Sorte, and Teli researched CNC 7 1 0.15 1250 0.6 1.69
18 2 0.3 1000 0.4 1.47
turning parameters (spindle speed, feed rate, and depth of cut) for AISI
1 3 0.15 750 0.2 1.91
P20 steel with TiN-coated tungsten carbide insert to optimize the surface 20 4 0.3 1000 0.4 1.47
roughness using ANOVA in the Taguchi method [24]. 9 5 0.15 1000 0.4 1.77
Optimization techniques like Particle Swarm Optimization (PSO), 5 6 0.15 750 0.6 1.84
Artificial Neural Network (ANN), 2K Factorial Design Method are 14 7 0.3 1000 0.6 1.52
6 8 0.45 750 0.6 1.38
employed to solve the various engineering optimization problems such
3 9 0.15 1250 0.2 1.73
as heat sink for thyristor [25], the synthesis of ultrafiltration asymmetric 12 10 0.3 1250 0.4 1.56
membranes based on organic polymers [26], soil classifications [27] and 15 11 0.3 1000 0.4 1.47
flotation parameters for copper recovery from sulfide flotation tailings 19 12 0.3 1000 0.4 1.47
11 13 0.3 750 0.4 1.62
[28].
13 14 0.3 1000 0.2 1.65
Balasubramanian K investigated optimum CNC machining process 2 15 0.45 750 0.2 1.66
parameters for stir-casted aluminum metal matrix composites using 17 16 0.3 1000 0.4 1.47
Response Surface Methodology (RSM). The mathematical relationship 8 17 0.45 1250 0.6 1.32
and interaction between the machining variables and optimum process 4 18 0.45 1250 0.2 1.61
16 19 0.3 1000 0.4 1.47
parameters are determined [29]. Palanisamy et al. optimized the
10 20 0.45 1000 0.4 1.37
machining parameters using Response Surface Methodology (RSM), and

Table 1
Chemical composition of AISI 4340 alloy steel.
C (%) Mn(%) P (%) S (%) Si (%) Ni (%) Cr (%) Mo (%) Balance

0.38–0.43 0.6–0.8 0.035 0.040 0.15 - 0.3 1.65 - 2.0 0.7–0.9 0.2 – Fe
0.3

Table 2
Mechanical properties of AISI 4340 alloy steel.
Hardness Tensile strength (MPa) Yield strength (MPa) Bulk modulus (GPa) Shear modulus (GPa) Elastic modulus (GPa) Poisson’s ratio

217 BHN 745 470 140 80 190–210 0.27–0.30

2
A.J. Santhosh et al. Results in Engineering 11 (2021) 100251

Table 4
ANOVA result for Surface Roughness.
Source Sum of Squares DOF Mean Squares F-value P-value Remarks (%)

Model 0.470812273 9 0.052312475 30.13 4.37842E-06 significant

A-Feed Rate 0.256 1 0.256 147.44 2.61407E-07 63.32
B-Rotational Speed 0.025 1 0.025 14.40 0.003517027 6.18
C-Depth of cut 0.06561 1 0.06561 37.79 0.000108724 16.23
AB 0.00605 1 0.00605 3.48 0.091506501 1.50
AC 0.02645 1 0.02645 15.23 0.002946376 6.54
BC 5E-05 1 5E-05 0.03 0.868631707 0.01
A2 0.004602273 1 0.004602273 2.65 0.134563997 1.14
B2 0.010202273 1 0.010202273 5.88 0.035813458 2.52
C2 0.008596023 1 0.008596023 4.95 0.050262492 2.13
Residual 0.017362727 10 0.001736273 0.43
Cor Total 0.488175 19
Std Deviation Mean CV % R2 Adjusted R2 Predicted Adeq. Precision
R2
0.041668606 1.5725 2.649832 0.9644 0.9324 0.8022 19.7528

operation for AISI 4140 Steel to optimize the roughness value using
Response Surface Methodology (RSM) [37,38].
Several researchers have used various methods such as Genetic Al­
gorithm (GA), Symbiotic Organisms Search (SOS), Simulated Annealing
(SA), Firefly Algorithm, Particle Swarm Optimization (PSO), Differential
Evolution (DE), and ant colony optimization (ACO) to optimize surface
roughness for different process parameters in CNC machining [39].
Chew Ying Nee et al. optimize the surface roughness for Co28Cr6Mo
medical alloy by considering the optimal effect of rotational speed, feed
rate, depth of cut, and tool tip radius in CNC lathe machining using
Differential Evolution (DE), and results are compared with the Response
Surface Methodology (RSM) [40]. Surface Roughness of Fused Deposi­
tion Modelling (FDM) fabricated product is improved using Response
Surface Methodology (RSM) and Meta-Heuristic algorithms by opti­
mized machining process parameters to achieve a good surface finish.
Meta-heuristic algorithms show good performance compared to RSM
[41]. Congbo Li et al. integrated the Taguchi method, response surface
method (RSM), and multi-objective particle swarm optimization algo­
rithm (MOPSO) to save energy in CNC machining by optimizing the
machining parameters [42]. Sanjeevi et al. investigated CNC milling
machining process parameters such as Speed, Feed Rate, and Depth of
Cut to reduce the surface roughness using Artificial Neural Network
(ANN) for Al6061 material [43]. An Artificial Neural Network can
improve the predicted Surface roughness value with repeated training
on three layers (Input, Output & Hidden layers) [44]. During the fin­
ishing & semi-finishing operation in the CNC turning process, an Arti­
Fig. 3. Shows a comparison of experimental and RSM expected values. ficial Neural Network (ANN) is combined with a Genetic Algorithm (GA)
to minimize residual surface stress in Inconel superalloy material [45].
an overlay plot with the best desirability function shows the optimal Thangarasu et al. optimized the CNC milling parameters such as Spindle
process parameters. Also, predicted values from the empirical models speed, feed rate, and Depth of cut for optimal surface finish and Material
are very near to the practical values [30]. Yahya explained how the Removal Rate (MRR) using Taguchi-based Box-Behnken RSM (Response
cutting speeds, amperes, and arc voltages parameters affect surface Surface Methodology) method and Multi-Objective Genetic Algorithm
roughness and hardness in the CNC plasma machine [31]. (MOGA) [46]. Selection of the optimum process parameters using a
Camposeco-Negrete states low feed rate, high cutting speed, and cutting genetic algorithm provides exemplary performance in CNC turning
depth provides good surface finish and energy savings in Response operation [47].
Surface Methodology (RSM) [32]. Adem et al. considered drilling pro­ Literature reveals that most researchers are not focused on CNC
cess parameters to improve surface and hole quality in CNC milling machining-related process parameters using Central Composite Design
machine for AISI 304 stainless steel using Taguchi and Response Surface (CCD) approach in RSM and the Algorithmic optimization approach for
Methodology (RSM). The error rate is less in RSM compared to Taguchi’s AISI 4340 alloy steel. Hence, this study focuses on optimizing CNC
Technique [33]. Response Surface Methodology (RSM) was employed to turning process parameters using the Face-Centred CCD approach in
determine the optimal turning process parameters using developed Response Surface Methodology (RSM), Artificial Neural Network
empirical models [34]. CNC turning machining process parameters were (ANN), and Genetic Algorithm (GA) to minimize the surface roughness
optimized using the Central composite design (CCD) method to reduce for the selected material.
the surface residual stress in pure iron material [35]. In the CNC end
milling process, RSM predicts the optimum surface roughness value and 2. Materials and methods
other required output responses for medium carbon steel material [36].
Radha Krishnan et al. selected different machining process parameters The experiments are conducted in a Small Scale Industry using the
such as cutting speed, depth of cut, and feed rate for CNC grinding Gdweiler Unit urn 300HD TM model CNC Turning Machine with a

3
A.J. Santhosh et al. Results in Engineering 11 (2021) 100251

are shown in Tables 1 and 2.

Although the material’s hardness is tested randomly using the Brinell
Hardness tester before proceeding with the experiments for the confir­
mation of mechanical property and all the sample specimens are well
satisfied in the range of 217 BHN value.
A typical Mitutoyo Surface Roughness tester is used to check the
output response for each run. High-Speed Tool steel is used for turning
AISI 4340 alloy steel which can be used in various products. Based on
the various literature studies and industrial environments, the following
machining parameters with three-level ranges are used in RSM to
identify optimum values. For this study, three levels of machining pa­
rameters were used. 1) Feed Rate (0.15, 0.3 and 0.45 mm/rev), 2)
Rotational Speed (750, 100, 1250 rpm) and 3) Depth of Cut (0.2, 0.4 and
0.6 mm). Central Composite Design (CCD) is employed to implement
RSM in this research. Later, the Genetic Algorithm is executed to obtain
the best fitness value through the Artificial Neural Network (ANN).
Each factor is coded from low level (− 1) to High level (+1) for the
selected range. In the Face-Centred CCD approach, the Alpha value is 1
for 20 Runs. In which 14 runs are defined for Non-Centre Points and the
remaining six runs having Centre points. The experimental works are
carried out for various diameters of the stepped shafts shown in Fig. 2
under different experimental setups. Observed output response for
selected CNC input machining parameter setup for 20 runs is shown in
Table 3.

2.1. ANOVA results for responses

The effect of the input parameters on the surface roughness (Ra)

Response was investigated using the Analysis of Variance (ANOVA). The
ANOVA results for surface roughness Ra are presented in Table 4. Feed
Rate (A) has the most significant impact on Surface Roughness (Ra),
followed by Depth of Cut (C). The adjusted R2 value of 0.9324 is rela­
tively similar to the expected coefficient of determination R2 value of
0.8022. The value of adequate precision is 19.753, which is higher than
four and means that there is enough signal.
The percentage share of each model relative to each sum of squares is
shown in the remarks column of Table 4. The percentage share is a rough
estimate of each model’s relative value, but it’s a good starting point
[48,49]. The Adequate precision is greater than 4, indicating a favour­
able signal, and a ratio of 19.753 indicates a sufficient positive signal.
The Regression Coefficient (R2 > 0.9644) indicates that the models built
in the experimental analysis are relatively strong, and therefore the re­
lationships are acceptable and a quadratic model is suggested.

2.2. Evaluation of regression mathematical model

Equation (1) provides the association between the measured re­

sponses and the input process variables.
Response Function, Y = f (ABC) + ε (1)

where A = Feed Rate (mm/rev), B = Rotational Speed (rpm), C = Depth

of Cut (mm) ε is the unidentified error.
Based on the regression technique, polynomial (second-order)
mathematical models for the response of surface roughness (Ra) have
been established as shown in Equation (2).

Y = β0 + β1 A + β2 B + β3 C + β4 AB + β5 AC + β6 BC + β7 A2 + β8 B2 + β9 C2
(2)
Fig. 4. (a)–(c) Surface plots for surface roughness Ra. The mathematical Expression as shown in equation (2) is developed
based on regression coefficients for surface roughness responses and it is
maximum spindle of 4000 rpm and 7.5 KW motor power shown in Fig. 1. given in equation (3).
According to the application, AISI 4340 alloy steel’s raw material with In terms of coded factors, coded Equation (3) can be used to predict
different diameters is brought from the ISO steel fabricating vendor. The the outcome for each factor’s given levels. High levels of the variables
chemical composition and mechanical properties of AISI 4340 alloy steel are coded as +1 by default, while low levels are coded as − 1. By
comparing the factor coefficients, the coded Equation (3) can determine

4
A.J. Santhosh et al. Results in Engineering 11 (2021) 100251

Fig. 5. Optimal Process parameters using RSM.

the relative factors effect.

Table 5
Learning factors selected for ANN. Surface Roughness, Ra ​ = + 1.49 − 0.16A − 0.05B − 0.081C + 0.0275AB
Network type Feed forward neural network − 0.0575AC + 0.0025BC + 0.0409A2 + 0.0609B2 + 0.0559C2
Training function Train Levenberg-Marquardt (LM) (3)
Adaption Learning function LEARNGD (Gradient descent)
Actual Equation (4) can predict the response for each factor’s given
Performance function Mean square error
Network topology 3-14-1 levels in terms of actual factors in the original units. Actual Equation (4)
Transfer function TANSIG in terms of actual factors can be used to make predictions about the
Number of Hidden Layers 1 response for each factor’s given levels and regression coefficients are
Number of Neurons 14 scaled to the original unit of each parameter.
Training method Back-propagation
Number of Epochs 1000 Surface Roughness, Ra = + 3.54745 − 2.12424A − 0.02389B − 0.998182C
+ 0.000733AB − 1.9166AC + 0.00005BC + 1.81818A2 + 9.74e− 7 B2
+ 1.3977C2
Table 6
Output of ANN along with the comparison of RSM & Measured surface (4)
roughness. For 20 runs, Fig. 3 compares real experimental values to expected
RUN Measured Predicted Predicted values determined by the RSM mathematical model. It demonstrates
Surface Roughness Surface Roughness RSM Surface Roughness ANN that the majority of the expected outcomes are in fair accordance with
(μm) (μm) (μm)
the real experimental outcomes.
1 1.69 1.71286 1.6816 Interaction Effect of machining parameters on Surface Roughness
2 1.47 1.49364 1.4725 (Ra).
3 1.91 1.91486 1.8966
4 1.47 1.49364 1.4725
Fig. 4 (a) – (c) illustrates the interactive effects of machining pa­
5 1.77 1.69455 1.7523 rameters on surface roughness Ra for textured inserts using three-
6 1.84 1.86286 1.8256 dimensional surface plots based on the established mathematical
7 1.52 1.46855 1.4266 models. Fig. 4 (a)–(c) shows Predicted versus Actual values of surface
8 1.38 1.49364 1.38
finish has a minimum deviation between the values. It confirms that a
9 1.73 1.75486 1.7202
10 1.56 1.50455 1.5335 developed mathematical model is excellent and helpful to predict the
11 1.47 1.49364 1.4725 optimized machining process parameters. While generating the inter­
12 1.47 1.49364 1.4725 action graphs and Response surface plots between two variables, one
13 1.62 1.60455 1.6366 variable is kept constant. Depth of cut is marked as 0.4 mm when an
14 1.65 1.63055 1.6548
15 1.66 1.65486 1.6719
interaction between A &B. Rotation speed is marked as 1000 rpm when
16 1.47 1.49364 1.4725 between A & C. Feed rate is kept as 0.3 mm when an interaction between
17 1.32 1.33286 1.3388 B & C. A higher feed rate produces a good surface finish.
18 1.61 1.60486 1.598 Fig. 4 (a) shows that the higher feed rate tends to result in an
19 1.47 1.49364 1.4725
excellent surface finish. It’s also been discovered that integrating a faster
20 1.37 1.37455 1.3457
rotational speed with a faster feed rate produces the desired surface
finish. It enhances the cutting process’s performance and reduces the
vibration produced during the machining process, lowering surface
roughness. Fig. 4 (b) and (c) indicated that increasing the Feed rate

5
A.J. Santhosh et al. Results in Engineering 11 (2021) 100251

Fig. 6. ANN trained Model for the best fitness.

Table 7 Table 8
Optimization results obtained for RSM and ANN. Selected parameter functions and parameter value in Genetic algorithm.
Methods Statistical Measure of Optimized Predicted Surface Population type Double vector
Regression Model Parameter Values Roughness, Ra
Population size 50
RSM R2 = 0.9644 Feed Rate = 1.31697 μm Creation Function Feasible Population
Adjusted R2 = 0.9324 0.4438 mm/rev Fitness Scaling Function Rank
Predicted R2 = 0.8022 Speed = 1181.46 Selection Function Tournament
rpm Tournament size 4
Depth of cut = Reproduction Default values
0.5813 mm Elite Count 2.5 (0.05* Population size)
ANN R2 = 0.97248 Feed Rate = 1.27871 μm Crossover Fraction 0.8
Adjusted R2 = 0.9557 0.4434 mm/rev Mutation Function Adaptive Feasible
Predicted R2 = 0.8458 Speed = 1036.6 Crossover Function Constraint Dependent
rpm Number of Generations 300 (100*No. of input parameters)
Depth of cut = Function Tolerance 1e-6
0.57705 mm Constraint Tolerance 1e-3

increases the Surface Roughness (Ra) significantly. (A) = 0.443804 mm/rev, Rotational Speed (B) = 1181.46 rpm, Depth of
In this study, optimization target is set to “In range,” and the solution Cut = 0.581316 mm for minimum surface roughness value of 1.31697
destination is set to “Minimize.” The desirability function’s predicted μm under Desirability = 1. These optimum values are implemented in a
output is in the form of “smaller-is-better” characteristics. real-time machining environment for a confirmation test and compared
Fig. 5 shows the optimal process parameters values such as Feed Rate with the Genetic Algorithm results obtained in the next step.

6
A.J. Santhosh et al. Results in Engineering 11 (2021) 100251

the smaller network dataset in this study. Train Levenberg-Marquardt

(TrainLM) is often the fastest backpropagation algorithm in the simu­
lation software and is highly recommended as a first-choice supervised
algorithm, although it does require more memory than other algorithms
for bigger data sets. This method is tailored for functions of the type
sum-of-squared-error. That makes it to be very fast when training neural
networks measured on that kind of errors. The following Hyperbolic
tangent sigmoid transfer is used in ANN training. It runs faster than the
MATLAB implementation of tanh, but the results can have very small
numerical differences.

Function a = tansig(n) = 2/(1 + exp(-2*n))-1 (5)

A validation data set containing 20 new data combinations are used

to assess the prediction of levels and values of parameters. The network
is trained by changing the default training parameters to achieve an
excellent R2 range. ANN is stopped training when the error, i.e., the
difference between the desired output and the expected output is below
Fig. 7. Best fitness generated for minimum surface roughness. a designated threshold value or the number of iterations or epochs is
above a designated threshold value. Training a neural network can be
terminated by setting the values of the following parameters: Minimum
Gradient Magnitude (min_grad) = 1e-07, Maximum Number of Valida­
tion Increases (max_fail) = 6, Maximum Training Time (time) = Infinite,
Minimum Performance Value (goal) = 0, and Maximum Number of
Training Epochs (Iterations) = 1000.
Output Responses of ANN along with the comparison of RSM &
Measured surface roughness are given in Table 6.
In Fig. 6, comparison graphs between observed and expected values
are shown. After the repeated training, the ANN model offers the best
fitness of R2 = 0.97248 (Fig. 6), which is much better than RSM results.
The predicted values of the ANN model are considered to be in good
agreement with the experimental values. As a result, this fitness model
can find the best process parameters for the best surface fitness in the
genetic algorithm. Coded equation (6) and actual equation (7) are
generated for best fitness of R2 = 0.97248 in ANN model. The actual
equation (7) is implemented in the Genetic Algorithm to obtain the
better-optimized parameter values.
It is possible to optimize using developed RSM model GA and pre­
vious works are done in this method. But, in this study ANN model is
used to achieve more accurate results compared with the developed
RSM model. Statistical Measure of Regression Model in ANN shows a
Fig. 8. Before & after machined Stepped Shafts for confirmation tests. better Predicted R2 (0.8458) value compared with the RSM model Pre­
dicted R2 (0.8022) value. Optimization results obtained for RSM and
2.3. Artificial neural network (ANN) ANN are shown in Table 7. Also, the objective function (predicted sur­
face roughness) value is minimized in the ANN model.
An input layer, a hidden layer, and an output layer are often used to The Coded Equation of ANN Model,
create Artificial Neural Network modelling. The feed rate, rotational
speed, and depth of cut process parameters are all controlled by the Surface Roughness, Ra ​ = + 1.49 − 0.1542A − 0.0539B − 0.0889C
input layer. The number of neurons used to improve the output value’s + 0.0257AB − 0.0552AC + 0.0081BC + 0.0435A2 + 0.0796B2
precision was filled into the hidden layer [50]. ANN architecture and + 0.0352C2 (6)
learning factors selected for this study are given in Table 5. One hidden
layer feed-forward backpropagation network, 14 number of neurons and The Actual Equation of ANN Model,
1000 number of Epochs are selected in this study to achieve the opti­
mum output range. The output response surface roughness assigns to the
output layer. The Levenberg-Marquardt algorithm is employed to train

Table 9
Confirmation Tests for RSM & GA for optimum process parameters.
Methods Optimized Parameter Values Surface Roughness, Ra (μm) Sample Sample Sample Average (μm)
1 (μm) 2 (μm) 3 (μm)

RSM Feed Rate = 0.4438 mm/rev Experiment 1.3869 1.3525 1.4055 1.381633
Speed = 1181.46 rpm Predicted 1.31697 1.31697 1.31697 1.31697
Depth of cut = 0.5813 mm % of error 1.39 2.63 6.30 3.44
Genetic Feed Rate = 0.4321 mm/rev Experiment 1.3263 1.3045 1.2872 1.306
Algorithm Speed = 1192.3 rpm Predicted 1.257665 1.257665 1.257665 1.257665
Depth of cut = 0.5738 mm % of error 5.17 1.39 2.29 2.95

7
A.J. Santhosh et al. Results in Engineering 11 (2021) 100251

Surface Roughness, Ra = + 3.8253 − 2.13728A − 0.003032B − 0.759961C surface roughness (Ra) of AISI 4340 alloy steel components for the effect
of three different process parameter’s such as Feed Rate (mm/rev),
+ 0.000684AB − 1.83958AC + 0.000163BC + 1.93475 + 1.27331e− 6 B2
Speed (rpm), Depth of Cut (mm) in a typical CNC turning machine using
+ 0.880795C2 Face-Centred Central Composite Design (CCD) approach in RSM and
(7) Genetic Algorithm through Artificial Neural network (ANN). The pre­
sent work effectively minimizes the surface roughness of AISI 4340 alloy
2.4. Optimization using genetic algorithm steel by implementing the optimal CNC turning process parameters with
the help of RSM and ANN-Genetic Algorithm in the real time
Nowadays, instead of using traditional approaches, researchers manufacturing environment. The following conclusions are drawn from
integrate an artificial neural network with a genetic algorithm to this work:
incorporate different forms of single or multi-objective optimization
problems in the engineering field to achieve the best fitness values • AISI 4340 alloy steel having good mechanical and microstructural
[51–53]. Optimum machining input parameters and best output properties. Experimental results show a good effect on the surface
response are achieved with the latest algorithmic approaches, i.e., finish in the CNC machining environment, and hence it can be used
artificial intelligent techniques [54]. Genetic algorithm (GA) is a widely for various industrial applications due to its enhanced properties.
used natural-based evolutionary optimization technique, a branch of • Face-Centred Central Composite Design (CCD) approach in RSM
meta-heuristic algorithms. Search space is very high in the Genetic al­ provides a good coefficient of regression (R2 > 0.9644) and adequate
gorithm (GA) to obtain the best fitness value. precision (19.753).
Genetic Algorithm is searching the optimized results from the • ANOVA reveals that the feed rate is the most essential factor. From
various population of points. In each iteration, the genetic algorithm three-dimensional surface plots, machining variables’ interaction
generates a population of points and the random number of generations effect on surface roughness (Ra) shows that a higher feed rate de­
whereas the traditional convention algorithm generates only a single livers a better surface finish. It’s also been discovered that combining
point and the next point is selected by the deterministic rule. The other a faster rotational speed with a faster feed rate results in a significant
optimization algorithm’s ensures only local minimized parameter surface finish.
values. But, Genetic Algorithm works on the criteria to find a global • Statistical Measure of Regression Model in ANN shows a better
minimum value for the given objective function and ensures that output Predicted R2 (0.8458) value compared with the RSM model Pre­
is the converged value. Hence, Genetic Algorithm is an efficient tool for dicted R2 (0.8022) value.
optimizing process parameters to find out the best fitted optimum so­ • The Artificial Neural Network (ANN) approach provides the best
lution from the global search based on the given objective function to fitness model to implement the genetic algorithm.
minimize the surface roughness in this study. • The confirmation test results explore experimental surface roughness
Hence, setting up the lower and upper boundary ranges is an value, and the percentage of error is less in the Genetic Algorithm
essential task in the Matlab Optimtool module. Here, the selected pa­ than Response Surface Methodology for AISI 4340 alloy steel
rameters lower and upper boundary conditions are Feed Rate from 0.15 components.
to 0.45 (mm/rev), Rotational Speed from 750 to 1250 rpm, and Depth of • A combination of Artificial Neural Network and Genetic Algorithm
Cut from 0.2 to 0.6 mm. The well-trained ANN fitness function given in methodology is recommended to obtain the optimal machining
equation (7) is used in the Genetic Algorithm (GA) to search for the best process parameters and get a good output response in practical
fitness surface roughness values. The best fitness value is obtained by environments.
varying the Genetic algorithm (GA) parameter functions in trial & error.
Table 8 shows the primary parameter functions selected for the Genetic Credit author statement
algorithm. Fig. 7 shows the best fitness graph to achieve minimum
surface roughness. After achieving mean fitness of 1.2576648 (μm), the A.Johnson Santhosh: Investigation; Data curation; Conceptualiza­
model is terminated in 180 iterations, and the following optimum tion; Methodology; Resources; Writing - original draft. Amanuel Diriba
machining parameters are obtained: Feed Rate: 0.4321mm/Rev, Rota­ Tura: Software; Data curation; Formal analysis; Resources. Iyasu Tafese
tion speed: 1192.3 rpm, Depth of cut: 0.5738 mm. These optimum Jiregna: Software; Data curation; Formal analysis; Resources. Wen­
values are implemented in the CNC machine for a confirmation test. dimu Fanta Gemechu: Software; Data curation; Formal analysis; Re­
sources. N. Ashok: Data curation; Formal analysis; Resources. Murugan
2.5. Confirmation test Ponnusamy: Validation; Supervision.

The confirmation experiments validate the predicted results from the Declaration of competing interest
RSM and GA by implementing the optimal process parameters in a CNC
turning machine for stepped shafts, shown in Fig. 8. An average of three The authors declare that they have no known competing financial
samples is taken for the confirmation study. Three trials of surface interests or personal relationships that could have appeared to influence
roughness measurements on three different surfaces of each sample are the work reported in this paper.
carried out. The average is tabulated along with the percentage of de­
viation from the predicted results in Table 9.
Acknowledgements
Both the methods show good performance on the samples after
implementing the optimum machining parameters obtained earlier. It
The authors are grateful to Goldwin Industries, Coimbatore for the
confirms both methods can be implemented in a machining environment support granted during the development of this work.
to get the optimum process parameters. Experimental surface roughness
value on the samples and the percentage of error is less in Genetic Al­
References
gorithm than Response Surface Methodology for AISI 4340 alloy steel
components. [1] L.H. Van Vlack, Elements of Materials Science and Engineering, sixth ed., Pearson,
1989.
3. Conclusions [2] R.O. Ritchie, B. Francis, W.L. Server, Evaluation of toughness in AISI 4340 alloy
steel austenitized at low and high temperatures, Metall Trans A (1976).
[3] W.-S. Lee, T.-T. Su, Mechanical properties and microstructural features of AISI
In this study, the experimental investigation is carried out on the 4340 high-strength alloy steel under quenched and tempered conditions, J. Mater.

8
A.J. Santhosh et al. Results in Engineering 11 (2021) 100251

Process. Technol. 87 (1999) 198–206. Available from: https://www.sciencedirect. textured tungsten carbide inserts, in: Mater Manuf Process, vol. 34, Taylor &
com/science/article/pii/S0924013698003513. Francis, 2019, pp. 502–510, https://doi.org/10.1080/10426914.2019.1566612.
[4] S. Chinchanikar, S.K. Choudhury, Investigations on machinability aspects of Available from:.
hardened AISI 4340 steel at different levels of hardness using coated carbide tools, [31] Y.H. Çelik, Investigating the effects of cutting parameters on materials cut in CNC
Int J Refract Met Hard Mater (2013). plasma, Mater. Manuf. Process. 28 (2013) 1053–1060.
[5] A. Mehrabi, H. Sharifi, M. Asadi Asadabad, R. Amini, A. Rajaee, Improvement of [32] C. Camposeco-Negrete, Optimization of cutting parameters using Response Surface
AISI 4340 steel properties by intermediate quenching – microstructure, mechanical Method for minimizing energy consumption and maximizing cutting quality in
properties, and fractography, Int. J. Mater. Res. (2020) 111. turning of AISI 6061 T6 aluminum, J. Clean. Prod. (2015).
[6] S. Boakye-Yiadom, M.N. Bassim, Effect of prior heat treatment on the dynamic [33] A. Çiçek, T. Kıvak, E. Ekici, Optimization of drilling parameters using Taguchi
impact behavior of 4340 steel and formation of adiabatic shear bands, Mater. Sci. technique and response surface methodology (RSM) in drilling of AISI 304 steel
Eng., A (2011). with cryogenically treated HSS drills, J. Intell. Manuf. (2015).
[7] M.A.S. Torres, H.J.C. Voorwald, An evaluation of shot peening, residual stress and [34] B. Davoodi, A.H. Tazehkandi, Experimental investigation and optimization of
stress relaxation on the fatigue life of AISI 4340 steel, Int. J. Fatig. (2002). cutting parameters in dry and wet machining of aluminum alloy 5083 in order to
[8] U. Çydaş, Machinability evaluation in hard turning of AISI 4340 steel with different remove cutting fluid, J. Clean. Prod. (2014).
cutting tools using statistical techniques, Proc. Inst. Mech. Eng. Part B J Eng Manuf [35] J. Luo, Y. Sun, Optimization of process parameters for the minimization of surface
(2010). residual stress in turning pure iron material using central composite design, in:
[9] M.M. Nunes, E.M. Da Silva, R.A. Renzetti, T.G. Brito, Analysis of quenching Meas J Int Meas Confed, vol. 163, Elsevier Ltd, 2020, p. 108001, https://doi.org/
parameters in AISI 4340 steel by using design of experiments, Mater. Res. 22 10.1016/j.measurement.2020.108001. Available from:.
(2018) 1–7. [36] N. Masmiati, A.A.D. Sarhan, M.A.N. Hassan, M. Hamdi, Optimization of cutting
[10] P.R.S.S. Rahul Tukaram Bhole, Optimization of drilling process parameters for AISI conditions for minimum residual stress, cutting force and surface roughness in end
316L by using Taguchi method, Int J Sci Res Sci Eng Technol (2016). milling of S50C medium carbon steel, Meas J Int Meas Confed (2016).
[11] J. Boér, P. Blaga, Reducing production costs by monitoring the roughness of raw [37] R.K. Beemaraj, R. Aravindh, M. Barathkumar, K. Gowtham, R. Hariharan,
product surfaces, Procedia Manuf (2018). Prediction of Surface Roughness (AISI 4140 Steel) in Cylindrical Grinding
[12] M.S. Alsoufi, A.E. Elsayed, Surface roughness quality and dimensional accuracy—a Operation by RSM, vol. 9, 2018.
comprehensive analysis of 100% infill printed parts fabricated by a personal/ [38] Br, Review of surface roughness prediction in machining process by using various
desktop cost-effective FDM 3D printer, Mater. Sci. Appl. (2018). parameters, Int J Recent TRENDS Eng Res 6 (2020) 7–12, 0.
[13] G.M. Krolczyk, R.W. Maruda, J.B. Krolczyk, P. Nieslony, S. Wojciechowski, [39] M. Abdullahi, M.A. Ngadi, S.M. Abdulhamid, Symbiotic Organism Search
S. Legutko, Parametric and nonparametric description of the surface topography in optimization based task scheduling in cloud computing environment, Futur Gener
the dry and MQCL cutting conditions, Meas J Int Meas Confed (2018). Comput Syst, 2016.
[14] N.M. Vaxevanidis, N.A. Fountas, A. Koutsomichalis, J.D. Kechagias, Experimental [40] C.Y. Nee, M.S. Saad, A. Mohd Nor, M.Z. Zakaria, M.E. Baharudin, Optimal process
investigation of machinability parameters in turning of CuZn39Pb3 brass alloy, parameters for minimizing the surface roughness in CNC lathe machining of
Procedia Struct Integr (2018). Co28Cr6Mo medical alloy using differential evolution, Int J Adv Manuf Technol.
[15] P.M. Pandey, K. Thrimurthulu, N.V. Reddy, Optimal part deposition orientation in The International Journal of Advanced Manufacturing Technology 97 (2018)
FDM by using a multicriteria genetic algorithm, Int. J. Prod. Res. (2004). 1541–1555.
[16] T. Nancharaiah, D.R. Raju, V.R. Raju, An experimental investigation on surface [41] M.S. Saad, A.M. Nor, M.E. Baharudin, M.Z. Zakaria, A.F. Aiman, Optimization of
quality and dimensional accuracy of FDM components, Int. J. Emerg. Technol. surface roughness in FDM 3D printer using response surface methodology, particle
(2010). swarm optimization, and symbiotic organism search algorithms, Int J Adv Manuf
[17] A.K. Sahoo, S. Pradhan, Modeling and optimization of Al/SiCp MMC machining Technol. The International Journal of Advanced Manufacturing Technology 105
using Taguchi approach. Measurement vol. 46, 2013, pp. 3064–3072. Available (2019) 5121–5137.
from: https://www.sciencedirect.com/science/article/pii/S0263224113002340. [42] C. Li, Q. Xiao, Y. Tang, L. Li, A method integrating Taguchi , RSM and MOPSO to
[18] P. Cardoso, J.P. Davim, Optimization of surface roughness in micromilling, Mater. CNC machining parameters optimization for energy saving 135 (2016) 263–275.
Manuf. Process. 25 (2010) 1115–1119. [43] R. Sanjeevi, R. Nagaraja, B.R. Krishnan, Materials Today : proceedings Vision-
[19] A. Arunnath, P.H.S. Masooth, Optimization of process parameters in CNC turning based surface roughness accuracy prediction in the CNC milling process ( Al6061 )
process on machining SCM440 steel by uncoated carbide and TiCN/Al2O3/TiN using ANN, in: Mater Today Proc, Elsevier Ltd., 2020, https://doi.org/10.1016/j.
coated carbide tool under dry conditions, Elsevier Ltd, Mater Today Proc (2020), matpr.2020.05.122. Available from:.
https://doi.org/10.1016/j.matpr.2020.10.699. Available from:. [44] B.R. Krishnan, V. Vijayan, T.P. Pillai, T. Sathish, Influence of surface roughness in
[20] C. Camposeco-Negrete, Optimization of cutting parameters for minimizing energy turning process — an analysis using artificial neural network, Trans. Can. Soc.
consumption in turning of AISI 6061 T6 using Taguchi methodology and ANOVA, Mech. Eng. (2019).
J. Clean. Prod. (2013). [45] F. Jafarian, H. Amirabadi, J. Sadri, Experimental measurement and optimization of
[21] J. Nagendra, M.K. Srinath, S. Sujeeth, K.S. Naresh, M.S. Ganesha Prasad, tensile residual stress in turning process of Inconel718 superalloy, Meas J Int Meas
Optimization of process parameters and evaluation of surface roughness for 3D Confed (2015).
printed nylon-aramid composite, Mater Today Proc (2020). Available from: [46] V.S. Thangarasu, G. Devaraj, R. Sivasubramanian, High speed CNC machining of
https://www.sciencedirect.com/science/article/pii/S2214785320382328. AISI 304 stainless steel ; Optimization of process parameters by MOGA 4 (2012)
[22] G. Campatelli, L. Lorenzini, A. Scippa, Optimization of process parameters using a 66–77.
Response Surface Method for minimizing power consumption in the milling of [47] T. Srikanth, A real coded genetic algorithm for optimization of cutting parameters
carbon steel, J. Clean. Prod. (2014). in turning, Int J Comput Sci Netw Secur 8 (2008).
[23] S.J. Raykar, D.M. D’Addona, A.M. Mane, Multi-objective optimization of high [48] M. Sushil, K. Vinod, K. Harmesh, Experimental investigation and optimization of
speed turning of Al 7075 using grey relational analysis, Procedia CIRP (2015). process parameters of Al/SiC MMCs finished by abrasive flow machining, in: Mater
[24] M. Vishnu Vardhan, G. Sankaraiah, M. Yohan, H. Jeevan Rao, Optimization of Manuf Process, vol. 30, Taylor & Francis, 2015, https://doi.org/10.1080/
parameters in CNC milling of P20 steel using response surface methodology and 10426914.2015.1004704, 902–11.
Taguchi method, Mater Today Proc 4 (2017), 9163–9, https://www.sciencedirect. [49] D.C. Mongomery, Montgomery: Design and Analysis of Experiments, John Willy
com/science/article/pii/S2214785317315705. Sons, 2017.
[25] V.S. Bhandari, S.H. Kulkarni, Optimization of heat sink for thyristor using particle [50] R. Sanjeevi, R. Nagaraja, B. Radha Krishnan, Vision-based surface roughness
swarm optimization, Results Eng (2019). accuracy prediction in the CNC milling process (Al6061) using ANN, Mater Today
[26] M. Berradi, O. Berradi, M. Chellouli, R. Hsissou, M. El Bouchti, M. El Gouri, et al., Proc (2020).
Optimization of the synthesis of ultrafiltration asymmetric membranes based on [51] F. Pettersson, A. Biswas, P.K. Sen, H. Saxén, N. Chakraborti, Analyzing leaching
organic polymers, Results Eng (2020). data for low-grade manganese ore using neural nets and multiobjective genetic
[27] S. Inazumi, S. Intui, A. Jotisankasa, S. Chaiprakaikeow, K. Kojima, Artificial algorithms, Mater. Manuf. Process. (2009).
intelligence system for supporting soil classification, Results Eng (2020). [52] O.B. Nakhjavani, M. Ghoreishi, Multi criteria optimization of laser percussion
[28] K.K. Brest, M.M. Henock, N. Guellord, M. Kimpiab, K.F. Kapiamba, Statistical drilling process using artificial neural network model combined with genetic
investigation of flotation parameters for copper recovery from sulfide flotation algorithm, Mater. Manuf. Process. (2006).
tailings, Results Eng (2021). [53] S. Pal, S.K. Pal, A.K. Samantaray, Determination of optimal pulse metal inert gas
[29] K. Balasubramanian, M. Nataraj, P. Duraisamy, Machinability analysis and welding parameters with a neuro-ga technique, Mater. Manuf. Process. (2010).
application of response surface approach on CNC turning of LM6/SiCp composites, [54] W. Paszkowicz, Genetic algorithms, a nature-inspired tool: survey of applications
in: Mater Manuf Process, vol. 34, Taylor & Francis, 2019, pp. 1389–1400, https:// in materials science and related fields, in: Mater Manuf Process, vol. 24, Taylor &
doi.org/10.1080/10426914.2019.1660787. Available from:. Francis, 2009, pp. 174–197, https://doi.org/10.1080/10426910802612270.
[30] D. Palanisamy, K. Balasubramanian, N. Manikandan, D. Arulkirubakaran,
R. Ramesh, Machinability analysis of high strength materials with Cryo-Treated