Professional Documents
Culture Documents
a r t i c l e i n f o a b s t r a c t
Article history: In this present study, an experimental investigation has been done to predict surface roughness by taking
Received 10 July 2020 into consideration of the following parameters: cutting speed, rate of feed, cutting depth and nose radius
Received in revised form 11 March 2021 in the hard turning of Duplex 2205 (ASTM A276) round bar material utilizing carbide tip tool material.
Accepted 6 May 2021
Machining of duplex stainless Steel is considered as an inefficient machinable material, low removal rate
Available online xxxx
and requires more exertion to machine the item. For experimental purposes, the 4-factor and 5-level
design matrix is structured using the Design-Expert software. The statistical method is used to analyze
Keywords:
the variance using ANOVA technique to examine the unique performance character.The statistical
Cutting speed
Rate of feed
method is employed to analysis of variance (ANOVA) in order to examine the performance uniqueness.
Cutting depth A second-order mathematical model is created according to the Design of Experiments (DoE) of
Nose radius Response Surface Methodology (RSM) to anticipate surface roughness. In view of cutting test, the cutting
Duplex stainless steel speed is the parameter which has the greatest impact on the machining activity. The nose radius is a
ANOVA minor parameter that has a less impact on machining activity. For the best possible results, the cutting
RSM parameters leading to surface roughness, the Artificial Neural Network (ANN) model is trained and tested
Artificial neural network (ANN) using MAT Lab. For this condition, the ANN suggests the utmost least predicted value of surface rough-
ness. The results of the tests confirming the expected values proved to be in perfect agreement with the
experimental values. Finally, efficiency of three different ANN transfer function were compared to choose
the best predicted values. The Log-sigmoid is the most suitable transfer function that provides minimum
surface roughness.
Ó 2021 Elsevier Ltd. All rights reserved.
Selection and peer-review under responsibility of the scientific committee of the International Confer-
ence on Advances in Design, Materials and Manufacturing.
https://doi.org/10.1016/j.matpr.2021.05.118
2214-7853/Ó 2021 Elsevier Ltd. All rights reserved.
Selection and peer-review under responsibility of the scientific committee of the International Conference on Advances in Design, Materials and Manufacturing.
Please cite this article as: M. Gopal, Prediction of surface roughness in turning of duplex stainless steel (DSS) using response surface methodology (RSM)
and artificial neural network (ANN), Materials Today: Proceedings, https://doi.org/10.1016/j.matpr.2021.05.118
M. Gopal Materials Today: Proceedings xxx (xxxx) xxx
Table 1
Process parameters and their levels.
Table 2
Experimental Values with Responses.
Run Factor 1 Factor 2 Factor 3 Factor 4 Experimental value (observed) Predicted by RSM Predicted by ANN
Cutting speed (tc) Rate of feed (fz) Cutting depth (ap) Nose radius (rn) Surface roughness Surface roughness Surface roughness
(m/min) (mm/rev) (mm) (mm) (lm) (lm) (lm)
1 120 0.4 3.5 1.6 2.47 2.49 2.4998
2 100 0.5 3 1.2 3.01 2.99 3.0828
3 140 0.5 3 1.2 2.01 2.00 2.0808
4 100 0.3 4 2.4 2.86 2.87 2.7478
5 120 0.4 4.5 1.6 2.74 2.73 2.7705
6 100 0.3 3 2.4 2.91 2.92 2.8744
7 140 0.5 4 1.2 2.29 2.3 2.2920
8 140 0.3 4 2.4 2.43 2.44 2.5580
9 120 0.4 3.5 1.6 2.53 2.49 2.4998
10 120 0.4 3.5 1.6 2.52 2.49 2.4998
11 140 0.5 4 2.4 2.21 2.22 2.2767
12 120 0.4 3.5 1.6 2.45 2.49 2.4998
13 80 0.4 3.5 1.6 3.39 3.40 3.2712
14 120 0.4 3.5 1.6 2.52 2.49 2.4998
15 100 0.5 4 1.2 3.34 3.35 3.1585
16 120 0.4 3.5 1.6 2.47 2.49 2.4998
17 140 0.3 3 1.2 2.17 2.16 2.1379
18 120 0.6 3.5 1.6 2.69 2.68 2.6709
19 140 0.3 4 1.2 2.51 2.5 2.5026
20 100 0.3 3 1.2 2.86 2.85 2.9770
21 140 0.5 3 2.4 2.31 2.32 2.3065
22 100 0.5 4 2.4 2.97 2.96 3.0503
23 120 0.4 2.5 1.6 2.31 2.32 2.3097
24 100 0.5 3 2.4 3.01 3.02 3.0597
25 120 0.2 3.5 1.6 2.74 2.75 2.7772
26 100 0.3 4 1.2 3.23 3.22 3.2301
27 160 0.4 3.5 1.6 1.88 1.86 2.1649
28 120 0.4 3.5 2.8 2.72 2.70 2.7061
29 140 0.3 3 2.4 2.54 2.53 2.5604
30 120 0.4 3.5 0.8 2.7 2.72 2.7337
surface methodology (RSM). The result point out that the cutting researcher proposed that rate of feed is the most impacting vari-
force is the influential parameter compared to cutting depth and able compared to other parameters [8]. The analytical research is
hardness of the coating material, rate of feed and cutting material carried out with duplex stainless steel of the standard grade EN
hardness affect surface unevenness. The author assesses the 1.4462 and super EN 1.4410 DSS. The experiment is performed
machinability of low-alloy high-strength AISI 4340 steel and cold by maintaining at constant cutting speed during facing operation
AISI D2 steel as a working material at different hardness levels. [9]. The experimental investigation is done [10] to predict surface
The results specified that the surface finish increases at larger cut- roughness and high corrosion resistance by considering the cutting
ting speed and rate of feed [5].The non-linear regression analysis speed and cooling conditions at low and high fluid pressure cool-
with logarithmic data transformation strategy is performed to ing. Long tool life, good surface finish and better corrosion resis-
examine surface roughness in turning activity by considering work tance are achieved under high fluid pressure cooling conditions.
material hardness, cutting time, rate of feed, spindle speed, tool The experiment is conducted using super duplex stainless steel
point angle and cutting depth as cutting parameters [6]. Numerous to measure surface roughness and residual stress, feed rate, nose
investigations have been carried out by the researches to predict radius, rake angle and cutting speed were chosen as experimental
surface irregularities as important output parameters. factors [11]. An experiment [12] was conducted using TiC and TiCN
cemented carbide cutting tools to process duplex stainless steels of
1.1. Duplex stainless steel different grades in dry turning. The outcome result of the experi-
ment is that when the cutting speed increases, the surface rough-
The statistical model using is developed by Krolczyk et al. [7] on ness decreases automatically. A study is conducted by considering
duplex stainless steel using turning processes to measure surface cutting speed and feed speed as parameters using carbon steel and
roughness, as a result, the feed rate is the fundamental variable duplex stainless steel materials to measure residual stress, micro-
affecting surface roughness. A turning operation is carried out structure, micro-hardness, and surface roughness during milling
using nitrogen alloyed duplex stainless steel by using the Taguchi operations. The results show that the higher cutting speed and
method to predict surface roughness and cutting force. The low feed enhance the quality of the machined surface. [13]. An
2
M. Gopal Materials Today: Proceedings xxx (xxxx) xxx
Table 3
ANOVA Table for prediction of surface roughness.
Table 4
experiment is carried out using duplex stainless steel alloyed with Optimum values of network parameter.
nitrogen; optimization is performed using the Taguchi method to
Sl. No Parameter Values
reduce surface roughness. The author suggests that the rate of feed
and spindle speed is the main affecting parameter affecting surface 1 Number of input layer 1
2 Number of input layer unit 4
roughness [14]. The study has been made while turning of DSS rod
3 Number of Hidden layer 1
using WC-Co coated carbide inserts. The desirability function as 4 Number of Hidden layer unit 5
well as the Taguchi method approach is used to correlate surface 5 Number of output layer 1
roughness and material withdrawal rate. The results showed that 6 Number of output layer unit 1
7 Number of Epochs 1000
rate of feed is the most significant factor affecting machining
8 Algorithm Back propagation
[15]. The analysis is performed using speed, feed and depth of 9 Learning rule Gradient descent rule
cut to measure cutting speed, surface roughness and material
removal rate. The Taguchi experimental design methodology is
employed to measure for practical trial. [16]. Vegetable oils - coco-
nut oil and neem oil are used for cooling when turning stainless steel
duplex (DSS) using a multi-layered carbide insert using the ultra-
sonic emulsification method rather than using inorganic biodegrad-
able mineral oil. The input parameter is spindle speed, feed rate,
cutting depth and type of cutting fluid. The Taguchi method is used
for machining, results the coconut oil based cutting fluid shows bet-
ter performance to improve the surface roughness [17].
2. Experimental design
Ra ¼ c tc k1 fz k2 ap k3 rn k4 ð1Þ
where,
Ra - Predicted surface roughness (lm).
tc - Cutting speed (m/rev).
fz - Rate of feed (mm/rev).
ap - Cutting depth (mm).
rn - Nose radius (mm).
Fig. 4. Interaction effect of rate of feed.
k1, k2, k3, k4, k5 - are the model parameters (to be estimated
from experimental data).
from the experimentally measured values. The selected parameters C - Response error.
are the cutting speed, feed speed, cutting depth and nose radius, The Second order mathematical equation can be written as
and their ranges have been determined. The upper level is (+2) Y ¼ b0 x0 þ b1 x1 þ b2 x2 þ b3 x3 þ b11 x12 þ b22 x22 þ b33 x32
and the lower level is (2) of all the four variables as shown in
Table 1. The intermediate levels of 0 for the set of variables are þ b44 x42 þ b12 x1 x22 þ b12 x1 x2 þ b13 x1 x3 þ b14 x1 x4
determined by interpolation. The surface roughness of the work- þ b23 x2 x3 þ b24 x2 x4 ð2Þ
piece material is the output response. The design matrix is selected
for experiments as a four-factor CCD with 30 sets of coded condi- where,
tions, as shown in Table 2. The Design Expert V12 software is b0 - constant,
employed to solve the coefficient of the regression model. b1, b2, b3 - linear term coefficient,
b11, b22 - quadratic term coefficient and
b12 - interaction term coefficient.
2.1. Experimentation The Design Expert V12 software is used to analyze the test read-
ing to predict surface roughness. A second-order quadratic model
The medium-duty Kirloskar turn master-35 is used to accom- is developed in accordance with the DOE methodological design
plish the experimental analysis. Duplex stainless steel 2205 (ASTM rules.The developed model is checked for its satisfactoriness using
A276) round bar material is utilized as work material. The shape of analysis of variance (ANOVA). The ANOVA analysis for the
the workpiece material is 20 mm in diameter and 70 mm in length. prediction of surface roughness is shown in Table 3. The Model
The work tool is a carbide insert (Tungaloy make) with a general F-value (462.78) implies the model is significant. There is only a
4
M. Gopal Materials Today: Proceedings xxx (xxxx) xxx
to the pure error. There is a 95.60% chance that a Lack of Fit F-value
this large could occur due to noise. A non-significant lack of fit is
good, so model to fit.
5
M. Gopal Materials Today: Proceedings xxx (xxxx) xxx
Table 5
Comparison of predicted (ANN) and experimental values.
Run Cutting Speed Rate of feed (fz) Cutting depth Nose Radius Experimental surface roughness Predicted by ANN - surface Error
(tc) (m/min) (mm/rev) (ap) (mm) (rn) (mm) value (observed) (Ra) (lm) roughness value (Ra) (lm) (%)
21 140 0.5 3 2.4 2.31 2.3065 0.1515
22 100 0.5 4 2.4 2.97 3.0503 2.7037
23 120 0.4 2.5 1.6 2.31 2.3097 0.0130
24 100 0.5 3 2.4 3.01 3.0597 1.6512
25 120 0.2 3.5 1.6 2.74 2.7772 1.3577
26 100 0.3 4 1.2 3.23 3.2301 0.0031
27 160 0.4 3.5 1.6 1.88 2.1649 13.1600
28 120 0.4 3.5 2.8 2.72 2.7061 0.5137
29 140 0.3 3 2.4 2.54 2.5604 0.8031
30 120 0.4 3.5 0.8 2.7 2.7337 1.2481
Average Error (%) 2.1605
The regression equation in terms of actual factors are given depth has a slight notable effect on surface roughness in the
below machining process. Fig. 5 shows the relationship between pre-
dicted value vs actual value.
SurfaceRoughness ¼ þ4:07990 0:029924 tc þ 0:795276
fz þ 0:700579 ap 0:338271 rn 4. Artificial neural network
0:040625 tc fz 0:001250 tc
An artificial neural network is a versatile framework that mod-
ap þ 0:006205 tc rn 0:100000
ify its structure depends on external or internal data that run
fz ap 0:138393 fz rn throughout the network, which can display difficult universal
0:341071 ap rn þ 0:000090 tc2 performance, determined by the associations between the process-
ing components and component parameters. Artificial neural
þ 5:59742 fz2 þ 0:033897 ap2
þ 0:231020 rn2 ð3Þ
In the present research, the input variable is used to build up
mathematical model using RSM. The determined value of the F
ratio is greater than the tabulated surface roughness value as
shown in Table 3; the model satisfies a desired 95% confidence
level. The values in Table 2 shows the experimental and predicted
values, ie error is at an acceptable level Table 2.
6
M. Gopal Materials Today: Proceedings xxx (xxxx) xxx
Table 6
Prediction value of LOGSIG, TRANSIG and PURELIN transfer functions.
Run Factor 1 Factor 2 Factor 3 Factor 4 LOGSIG Predicted by TANSIG Predicted by PURELIN Predicted
ANN ANN by ANN
Cutting speed (tc) Rate of feed (fz) Cutting depth (ap) Nose radius (rn) Surface roughness Surface roughness Surface roughness
(m/min) (mm/rev) (mm) (mm) (lm) (lm) (lm)
1 120 0.4 3.5 1.6 2.4998 2.5040 2.6340
2 100 0.5 3 1.2 3.0828 3.1528 2.8978
3 140 0.5 3 1.2 2.0808 2.0376 2.1596
4 100 0.3 4 2.4 2.7478 2.8853 3.1196
5 120 0.4 4.5 1.6 2.7705 2.7877 2.7956
6 100 0.3 3 2.4 2.8744 2.8654 3.0094
7 140 0.5 4 1.2 2.2920 2.2089 2.2723
8 140 0.3 4 2.4 2.5580 2.5700 2.3859
9 120 0.4 3.5 1.6 2.4998 2.5040 2.6340
10 120 0.4 3.5 1.6 2.4998 2.5040 2.6340
11 140 0.5 4 2.4 2.2767 2.2442 2.3138
12 120 0.4 3.5 1.6 2.4998 2.5040 2.6340
13 80 0.4 3.5 1.6 3.2712 3.2769 3.2401
14 120 0.4 3.5 1.6 2.4998 2.5040 2.6340
15 100 0.5 4 1.2 3.1585 3.2389 3.0299
16 120 0.4 3.5 1.6 2.4998 2.5040 2.6340
17 140 0.3 3 1.2 2.1379 2.1550 2.2140
18 120 0.6 3.5 1.6 2.6709 2.6958 2.5502
19 140 0.3 4 1.2 2.5026 2.4168 2.3405
20 100 0.3 3 1.2 2.9770 2.8168 2.9687
Average 2.6120% 2.6188% 2.6551%
operation by using the carbide insert as cutting tool. The input [6] X. Wang, C.X. Feng, Development of empirical models for surface roughness
prediction in finish turning, J. Adv. Manuf. Technol. 20 (5) (2002) 348–356.
parameter is cutting speed, rate of feed, cutting speed and nose
[7] G. Krolczyk, S. Legutko, M. Gajek, Predicting the surface roughness in the dry
radius. The optimization is carried out using an artificial neuronal machining of duplex stainless steel (DSS), Metalurgija 52 (2) (2013) 259–262.
network (ANN). The cutting speed is the most noteworthy influ- [8] D.S. Philip, P. Chandramohan, M. Mohanraj, Optimization of surface roughness,
encing parameter compared to the other parameters. The surface cutting force and tool wear of nitrogen alloyed duplex stainless steel in a dry
turning process using Taguchi method, Measurement 49 (2014) 205–215.
roughness is low at a high cutting speed. The radius of the nose [9] D.K. Rastee, U. Heisel, R. Eisseler, S. Schmauder, Modeling and optimization of
of the tool must be between 1.2 mm and 2.4 mm for the best min- turning duplex stainless steels, J. Manuf. Processes 16 (4) (2014) 451–467.
imum surface roughness. The optimization carried out using Arti- [10] A.D.O.J. Carlos, A.E. Diniz, R. Bertazzoli, Correlating tool wear, surface
roughness and corrosion resistance in the turning process of super duplex
ficial Neural Network, the results shows a good agreement stainless steel, J. B Society Mech. Sci. Eng. 36 (4) (2014) 775–785.
between the experimental and ANN predicted data for the increase [11] E. Capello, P. Davoli, G. Bassanini, A. Bisi, Residual stresses and surface
in surface roughness and the error percentage of the ANN data is roughness in turning, Trans. ASME 121 (3) (1999) 346–351.
[12] D.S. Philip, P. Chandramohan, Influence of cutting speed, feed rate and bulk
<5%. Thus, the logsig transfer function ANN predictive model is texture on the surface finish of nitrogen alloyed duplex stainless steels during
found to be more effective in carrying out surface roughness pre- dry turning, Engineering 2 (06) (2010) 453–460.
dictions. So the predictive ANN model is found to be accomplished [13] W.B. Saı̈, N.B. Salah, J.L. Lebrun, Influence of machining by finishing milling on
surface characteristics, J. Mach. Tools Manuf. 41 (3) (2001) 443–450.
of better predictions of surface roughness. [14] D.S. Philip, P. Chandramohan, M. Mohanraj, P.K. Rajesh, Experimental
investigations on surface roughness, cutting force and tool wear of duplex
CRediT authorship contribution statement stainless steel in end milling using Taguchi method, I R Mech. Engg. 7 (6)
(2013) 1133–1141.
[15] P. Kumar, J.P. Misra, Optimization of duplex stainless steel dry turning
Mahesh Gopal: Conceptualization, Investigation, Methodology, parameters using desirability function, Mater. Today Proc. 26 (2020) 1580–
Validation, Visualization, Writing - original draft, Writing - review 1584.
[16] G. Dinde, G.S. Dhende, Multi-response Optimization of Process Parameters
& editing. During Wet Turning of Super Duplex Stainless Steel UNS S32760 Using
Taguchi-Grey Relational Analysis, in: Optimization Methods in Engineering,
Declaration of Competing Interest Springer, Singapore, 2020, pp. 417–428.
[17] D.A. Ghatge, R. Ramanujam, B.S. Reddy, M. Vignesh, Improvement of
machinability using eco-friendly cutting oil in turning duplex stainless steel,
The authors declare that they have no known competing finan- Mater. Today Proc. 5 (2018) 12303–12310.
cial interests or personal relationships that could have appeared [18] G. Mahesh, S. Muthu, S.R. Devadasan, Prediction of surface roughness of end
milling operation using genetic algorithm, Int. J. Adv. Manuf. Technol. 77 (1–4)
to influence the work reported in this paper.
(2015) 369–381.
[19] I. Asiltürk, M. Çunkasß, Modeling and prediction of surface roughness in turning
References operations using artificial neural network and multiple regression method,
Expert Syst. Appl. 38 (5) (2011) 5826–5832.
[1] D. Singh, P.V. Rao, A surface roughness prediction model for hard turning [20] F. Jafarian, M. Taghipour, H. Amirabadi, Application of artificial neural network
process, J. Adv. Manuf. Technol. 32 (11–12) (2007) 1115–1124. and optimization algorithms for optimizing surface roughness, tool life and
[2] Y.K. Chou, H. Song, Tool nose radius effects on finish hard turning, J. Mater. cutting forces in turning operation, J. Mech. Sci. Technol. 27 (5) (2013) 1469–
Process. Technol. 148 (2) (2004) 259–268. 1477.
[3] R.M. Sundaram, B.K. Lambert, Mathematical models to predict surface finish in [21] M. Pala, N. Caglar, M. Elmas, A. Cevik, M. Saribiyik, Dynamic soil–structure
fine turning of steel. Part I, J. Prod. Res. 19 (5) (1981) 547–556. interaction analysis of buildings by neural networks, Constr. Build. Mater. 22
[4] A. Hamdi, M.A. Yallese, K. Chaoui, T. Mabrouki, J.F. Rigal, Analysis of surface (3) (2008) 330–342.
roughness and cutting force components in hard turning with CBN tool: [22] R. Sarkar, S. Julai, S. Hossain, W.T. Chong, M.A. Rahman, Comparative study of
Prediction model and cutting conditions optimization, Measurement 45 (3) activation functions of NAR and NARX neural network for long-term wind
(2012) 344–353. speed forecasting in Malaysia, Math. Probl. Eng. 6403081 (2019).
[5] J.G. Lima, R.F. Avila, A.M. Abrao, M. Faustino, J. Paulo Davim, Hard turning: AISI [23] M. Dorofki, A.H. Elshafie, O. Jaafar, O.A. Karim, S. Mastura, Comparison of
4340 high strength low alloy steel and AISI D2 cold work tool steel, J. Mater. artificial neural network transfer functions abilities to simulate extreme runoff
Process. Technol. 169 (3) (2005) 388–395. data, Int. Proc. Chem. Biol. Environ. Eng. 33 (2012) 39–44.