Engineering Science and Technology, an International Journal xxx (2016) xxx–xxx

Contents lists available at ScienceDirect

Engineering Science and Technology,

an International Journal
journal homepage: www.elsevier.com/locate/jestch

Full Length Article

Modeling and prediction of cutting forces during the turning of red brass
(C23000) using ANN and regression analysis
M. Hanief ⇑, M.F. Wani, M.S. Charoo
Mechanical Engineering Department, National Institute of Technology, Srinagar, Jammu & Kashmir 190006, India

a r t i c l e i n f o a b s t r a c t

Article history: The life of a cutting tool is greatly influenced by the forces acting on it during a cutting operation. A
Received 23 August 2016 machining operation is a complex process. It is very difficult to develop a comprehensive model involving
Revised 21 October 2016 all the parameters. The present study aims to develop a model to investigate the effects of cutting param-
Accepted 31 October 2016
eters (speed, depth of cut and feed rate) on the cutting forces during the turning operation of red brass
Available online xxxx
(C23000) using high speed steel (HSS) tool. The experimental results are based on full factorial design
methodology to increase the reliability and confidence limit of the data. Artificial neural network and
Keywords:
multiple regression approaches were used to model the cutting forces on the basis of cutting parameters.
ANN
Brass
In order to check the adequacy of the regression model, analysis of variance (ANOVA) was used. It was
Cutting forces clear from the ANOVA that the regression model is capable to predict the cutting forces with high accu-
Regression racy. However, ANN model was found to be more accurate than the regression model.
Turning Ó 2016 Karabuk University. Publishing services 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/).

1. Introduction Dorlin et al. [5] studied the geometrical modeling of tool-

workpiece interaction and its effects on the cutting forces during
The cutting forces are a result of extreme conditions at the tool- turning. The analysis focused on convex contact radius between
workpiece interface. The interaction can be directly related to the the machined part and the tool. The experiments were based on
tool wear and in worst cases to the failure of the tool. Conse- cylindrical and face turning of Ti6Al4V titanium alloy. It was
quently, the tool wear and cutting forces are related to each other observed that the contact radius had significant effect on the cut-
[1,2]. Thus, it is necessary to carry out the optimization of cutting ting forces and the cutting forces increase with the increase in
process to evaluate the optimal values of cutting parameters to the radius. Xie et al. [6] studied cutting force and cutting temper-
determine the performance and useful life of the cutting tool. Sur- ature during the turning of titanium alloy using micro-grooved tool
inder et al. [3] investigated the cutting forces (tangential and feed under dry conditions. The objective of the study was to estimate
force) in turning of unidirectional glass fiber reinforced plastics the influence of shape and size of micro groove on the temperature
(UD-GFRP). The process parameters of cutting tool (nose radius, and force in dry turning. The micro-grooved tool decreases cutting
rake angle, cutting speed, feed rate, depth of cut and cutting envi- temperature by 103 °C, while as the shear angle increases with
ronment) were investigated using Taguchi robust design method- decreasing micro-groove depth. Philip et al. [7] studied the effects
ology. The relative significance of parameters was studied using of cutting speed and feed rate on tool wear, surface roughness and
ANOVA. The tangential force was found to decrease with decrease cutting force on nitrogen alloyed duplex stainless steel in a dry
in tool nose radius, feed rate and depth of cut and increase with the turning process, using Taguchi method. The results revealed that
cutting speed. Cascona et al. [4] developed mechanistic model for the feed had the most significant influence on the cutting forces.
prediction of cutting forces in turning of non-axis-symmetric parts. The cutting speed was found to be the most significant parameter
This study presents a mechanistic model for predicting the orthog- affecting the tool wear. Shear force, ploughing force and particle
onal turning forces (in 3 directions), torque and power consump- fracture force were considered by Sikder et al. [8] to estimate the
tion along the machining path of non-axis-symmetric parts. cutting force during the machining of metal matrix composites
(MMCs). The chip formation force, ploughing force and fracture
force were obtained by Johnson-Cook model, slip line filed theory
⇑ Corresponding author.
and Griffith’s theory respectively. The results showed good
E-mail addresses: [email protected] (M. Hanief), [email protected] (M.F. Wani),
agreement between the predicted and experimental values of the
[email protected] (M.S. Charoo).
cutting forces. Two body abrasion and three body rolling due to
Peer review under responsibility of Karabuk University.

http://dx.doi.org/10.1016/j.jestch.2016.10.019
2215-0986/Ó 2016 Karabuk University. Publishing services 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/).

Please cite this article in press as: M. Hanief et al., Modeling and prediction of cutting forces during the turning of red brass (C23000) using ANN and
regression analysis, Eng. Sci. Tech., Int. J. (2016), http://dx.doi.org/10.1016/j.jestch.2016.10.019
2 M. Hanief et al. / Engineering Science and Technology, an International Journal xxx (2016) xxx–xxx

Nomenclature

fx thrust force (N) Df degrees of freedom

fy cutting force (N) SS sum of squares
fz feed force (N) MSE mean square error
fR, resultant force (N) MAPE mean absolute % error
v cutting speed (m/min) u transfer function
f feed rate (mm/rev) ti target values
d depth of cut (mm) oi observed values
k, c1, c2, c3 Model parameters R2 coefficient of determination

reinforcements in composites was studied by Uday et al. [9] during Although several materials such as steels, aluminum alloys,
the machining of Al/SiCp composite. Theory of oblique cutting was composites etc. have been investigated during the turning to
used for predicting the cutting forces during the machining of this develop the model for cutting forces. But very few models and
composite. With the assumption that 40% of the finer reinforce- investigations are devoted to brasses. The intention of the present
ment particles contribute to abrasion at tool-chip interface, the work is to develop a cutting force predictive model and to investi-
model was found to be accurate. Similarly for 60% of coarser rein- gate the influence of cutting parameters on the cutting forces dur-
forcement contributing to the abrasion, the model yielded good ing the turning of red brass (C23000) using regression analysis and
results. Pramanik et al. [10] proposed a model for predicting the ANN (see Fig. 1).
cutting forces during the cutting of aluminum-based SiC/Al2O3 par-
ticle reinforced MMCs. Three factors, chip formation force, plough- 2. Experimentation
ing force and particle fracture force were considered to be the force
generation mechanisms. Merchant’s analysis was used to obtain In this study, red brass (C23000) cylindrical bars of diameter
the chip formation force, while as, slip line field theory of plasticity 30 cm as work piece material and HSS tool were used. The
and the Griffith theory of fracture were used to formulated the experiments were performed under dry conditions on a (5HP and
matrix ploughing deformation and particle fracture. It was con- 45–2000 rpm range) Kiloshkar make lathe. The tool post was fitted
cluded that the force due to chip formation is much higher than with a dynamometer for measuring three components of the
those due to ploughing and particle fracture. Joardar et al. [11] cutting force, namely feed force ðf z Þ, radial thrust force ðf x Þ and
studied the influence of cutting speed, depth of cut and weight per- tangential (main) cutting force ðf y Þ The forces were recorded in a
centage of SiCp on the cutting forces during the turning of alu- digital computer which was interfaced with the dynamometer. A
minum MMC (aluminum alloy reinforced with silicon carbide full factorial design methodology was adopted and in total 27
particles) under dry conditions, using response surface methodol- (33) experiments were performed. Feed rate, cutting speed and
ogy. Cutting speed was found to be the most significant factor DOC were chosen for the study. The experimental details of
influencing the cutting forces. Shoba et al. [12] investigated the machining process are given in Table 1. The resultant force was cal-
influence of machining parameters (cutting speed, feed rate and qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2 2 2
depth of cut) on the cutting forces during the turning of hybrid culated by f R¼ f x þ f y þ f z .
composites. Different percentages of SiC (0, 2, 4, 6 and 8%) by
weight and rice hush ash were used in reinforced composite spec- 3. Results and discussion
imens. The comparison of reinforced and unreinforced specimens
revealed that cutting forces decrease with the increase in weight 3.1. Regression model
percentage of the reinforcement. This trend was attributed to the
dislocation densities produced from the mismatch between the The model of predicted cutting force, F R can be expressed as
reinforcement and the matrix. Fountas et al. [13] investigated the Eq. (1).
influence of cutting speed and feed-rate on the cutting forces dur-
F R ¼ kv 1 f 2 d 3
c c c
ing the turning of PA66 GF-30 Glass Fiber Reinforced Polyamide ð1Þ
using carbide cutting tool. It was concluded that the soft comput- where k, c1, c2 and c3 are model parameters
ing techniques can be effectively used to predict the cutting force By logarithmic transformation Eq. (1) can be written as
components. Vaxevenidis et al. [14] also concluded from their
research that ANN can be effectively used to predict the cutting
ln F R ¼ ln k þ c1 lnv þ c2 lnf þ c3 lnd ð2Þ
forces and surface roughness while investigating the turning of
AISI D6 tool steel, Ti6Al4V ELI and CuZn39Pb3 brass under dry cut-
ting environment with spindle speed, feed rate and depth of cut as
input; and surface roughness and cutting forces as outputs. Foun-
tas et al. [15] conducted a series of 5 axis machining experiments
in CAM environment to simulate operations using an L27 orthogo-
nal array. Four machining parameters namely tool type, stepover,
Work-piece
lead angle and tilt angle as inputs and surface deviation and
machining time were selected as the outputs. Similar investigation
was conducted by Vaxevanidis et al. [16] while turning the Ti-6Al-
4V alloy with input as spindle speed, the feed rate and the depth of
cut; and outputs as cutting force and the centre line average sur- Tool
face roughness. The methodology was found to be robust enough
to predict optimal values for quality objectives. Fig. 1. Direction of forces in turning operation.

Please cite this article in press as: M. Hanief et al., Modeling and prediction of cutting forces during the turning of red brass (C23000) using ANN and
regression analysis, Eng. Sci. Tech., Int. J. (2016), http://dx.doi.org/10.1016/j.jestch.2016.10.019
 M. Hanief et al. / Engineering Science and Technology, an International Journal xxx (2016) xxx–xxx 3

Table 1
Experimental data obtained from turning operation.

S. No Speed (m/min) Feed (mm/rev) Depth of cut (mm) f x ðNÞ f y ðNÞ f z ðNÞ f R ðNÞ

1 840 0.40 0.10 1.8640 4.2310 2.2150 5.1266

2 840 0.40 0.13 1.9946 4.1226 3.6500 5.8563
3 840 0.40 0.16 3.1940 5.3203 3.9776 7.3708
4 1280 0.40 0.10 2.3749 4.7795 3.2778 6.2632
5 1280 0.40 0.13 4.0120 5.0208 4.5283 7.8619
6 1280 0.40 0.16 3.9443 6.4434 5.0099 9.0650
7 1000 0.40 0.10 3.4921 4.2817 3.7600 6.6832
8 1000 0.40 0.13 3.6121 5.0568 3.1600 6.9717
9 1000 0.40 0.16 4.0353 5.8314 3.1893 7.7756
10 840 0.80 0.10 3.8638 6.3423 3.1522 8.0678
11 840 0.80 0.13 2.5223 7.7341 2.1280 8.4087
12 840 0.80 0.16 2.8110 9.1173 3.0937 10.0298
13 1280 0.80 0.10 2.4642 7.6636 2.9444 8.5716
14 1280 0.80 0.13 3.1269 7.4498 2.6920 8.5161
15 1280 0.80 0.16 2.7561 8.8065 3.7529 9.9617
16 1000 0.80 0.10 4.3497 6.7663 3.8258 8.9073
17 1000 0.80 0.13 4.3497 6.7663 3.8258 8.9073
18 1000 0.80 0.16 1.7883 7.6812 2.5603 8.2918
19 840 0.12 0.10 2.8105 1.6157 2.4624 4.0710
20 840 0.12 0.13 1.8123 1.6314 2.2874 3.3434
21 840 0.12 0.16 1.2031 2.7263 2.7559 4.0590
22 1280 0.12 0.10 2.1616 2.5057 4.2470 5.3841
23 1280 0.12 0.13 1.3419 1.2452 1.5682 2.4105
24 1280 0.12 0.16 1.9665 2.8219 3.4265 4.8550
25 1000 0.12 0.10 1.7347 2.0911 2.8472 3.9355
26 1000 0.12 0.13 1.3625 2.3763 2.8540 3.9558
27 1000 0.12 0.16 1.7977 2.8515 4.1100 5.3155

or
Table 2
Y r ¼ K þ c1 V þ c2 F þ c3 D ð3Þ
ANOVA analysis.

Source of variation Df SS MS F where ln F R ¼ Yr; lnk ¼ K; lnv ¼ V; lnf ¼ F; lnd ¼ D.

Regression 3 108.254 36.080 52.59 The model parameters in Eq. (3) were evaluated by using least
Residual 23 15.789 00.686 square method.
Total 26 124.000 The data presented in Table 1 was used for developing the
regression model. Accordingly Eq. (4) was obtained for estimating
the cutting force.

Fig. 2. Comparison of experimental and predicted values of cutting force by regression model.

Please cite this article in press as: M. Hanief et al., Modeling and prediction of cutting forces during the turning of red brass (C23000) using ANN and
regression analysis, Eng. Sci. Tech., Int. J. (2016), http://dx.doi.org/10.1016/j.jestch.2016.10.019
4 M. Hanief et al. / Engineering Science and Technology, an International Journal xxx (2016) xxx–xxx

Fig. 3. (a–i) Cutting force versus cutting parameters.

Please cite this article in press as: M. Hanief et al., Modeling and prediction of cutting forces during the turning of red brass (C23000) using ANN and
regression analysis, Eng. Sci. Tech., Int. J. (2016), http://dx.doi.org/10.1016/j.jestch.2016.10.019
 M. Hanief et al. / Engineering Science and Technology, an International Journal xxx (2016) xxx–xxx 5

W W
∑ ∑ output
input
bias bias

hidden layer output layer

Fig. 4. ANN structure used for modeling.

Table 3 are presented in Table 2. Based on the F-ratio of the predictive

Training parameters of ANN model.
model and that of the experimental results, the model was found
No. of neurons on the layer i/p;1, hidden:2 and o/p:1 to be adequate. The experimental and predicted values of regres-
Initial weight and biases 1 to 1 sion model are depicted in Fig. 2. It is obvious from Fig. 2 that there
Activation function Tansig
Learning rate 0.05
is a close agreement between the experimental values and that of
Momentum constant 0.95 the predicted values.
From Fig. 3(a–i) it is clear that the cutting force strongly depends
on the feed rate. At (a) v = 840 m/min and (b) 1000 m/min, the
F R ¼ 3:9093v 0:2226 f
0:4145 0:3008
d ð4Þ effect of feed rate on force is almost same but at (c) v = 1280
In order to determine the influence of each parameter on the m/min the effect of feed rate on force is prominent as the velocity
cutting force ANOVA was employed. The results of the ANOVA is also high. Similarly at constant feed rate the effect of depth of

Fig. 5. ANN output.

Please cite this article in press as: M. Hanief et al., Modeling and prediction of cutting forces during the turning of red brass (C23000) using ANN and
regression analysis, Eng. Sci. Tech., Int. J. (2016), http://dx.doi.org/10.1016/j.jestch.2016.10.019
6 M. Hanief et al. / Engineering Science and Technology, an International Journal xxx (2016) xxx–xxx

Fig. 6. Comparison of experimental and predicted values of cutting force by ANN model.

Table 4 weights computed during this stage are used to make decisions
Statistical parameters of output of ANN. for the evaluation of output. The MATLAB toolbox was used for
Sample Size MSE R2 ANN training, validation and testing in this investigation. The
Training 19 3.77217  10 24
1.00 parameters used for ANN are presented in Table. 3. Several inde-
Validation 4 7.39120  102 0.999 pendent runs were performed to achieve a satisfactory solution,
Testing 4 1.70073  101 0.999 with different initial random weights. The error during the learning
All values 27 3.6146  102 0.996 process was calculated by
X
N
jt i  oi j2
cut is prominent which is clear from (d) f = 0.10 mm, (e) i¼1
f = 0.13 mm and (f) = 0.16 mm. As stated earlier the velocity has MSE ¼ ð6Þ
N
least effect on the force, which is evident from (g) d = 0.12 mm,
The weights between hidden layer and output are adjusted and
(h) d = 0.40 mm and (i) d = 0.80 mm.
are again calculated using
3.2. ANN model Dwji ðnÞ ¼ aDwji ðn  1Þ þ gdj ðnÞyi ðnÞ ð7Þ

Due to the fact that the ANN has a capacity to solve non-linear where Dwji ðnÞ is the change in weights, a is the momentum coeffi-
problems, it has been widely used by the researchers. Thus, in pre- cient, dj is the error-scaled by the signal slope, g is learning rate
sent work ANN was also employed to model and predict the cut- parameter and yi(n) is the output. After satisfactory training the
ting forces during turning of red brass. Depending upon the experimental data, not presented in the previous stage was used
complexity and nature of the problem, an ANN model has several for testing the network. The statistical methods (R2 and MAPE) were
layers. In general, ANN has an input layer, hidden layers and output used for comparing the results obtained from the models.
PN !
layer. The input and output layers constitute the first and last lay- 2
2 i¼1 ðt i  oi Þ
ers respectively. The hidden layers process the data received from R ¼1 PN 2 ð8Þ
the input layer. Similarly, the next hidden layer computes the out- i¼1 oi

put and the last layer processes this output to produce the final The mean absolute percentage deviation (MAPD), also known as
result. The final results are computed by hidden and output layer mean absolute percentage error (MAPE) was estimated to measure
using transfer functions. of accuracy of the model and is defined as
In this paper a transfer function – tansig was used, as is given in
N  
Eq. (5). 1X 
t i  oi 
MAPE ¼  ð9Þ
en  en N i¼1 oi 
u ¼ n n ð5Þ
e þe where N is the number of experiments
The structure of the ANN model is shown in the Fig. 4. In order to develop ANN model, the network was trained by
The first step in the ANN is training. An input is fed to the ANN using a set of experimental values. After successful training the
along with the target outputs and the weights are set randomly, network was used to predict the cutting force for validation and
initially. The satisfactory level of performance is achieved by min- testing. The ANN results are presented in Fig. 5. It is clear from
imizing the global error using back propagation algorithm. Back the Fig. 5 that all the experimental and predicted values during
propagation algorithm is a learning technique that adjusts weights training coincide perfectly on the regression line which make the
in ANN by propagating weight changes from the output to input R2 = 1 in training. The R2 was found to be equal to 0.99690 and
neurons in backward direction. The training of the network is 0.99962 for validation and testing, respectively. At this stage the
stopped when the desired level of performance is achieved. The training was stopped and all the 27 experimental values were used

Please cite this article in press as: M. Hanief et al., Modeling and prediction of cutting forces during the turning of red brass (C23000) using ANN and
regression analysis, Eng. Sci. Tech., Int. J. (2016), http://dx.doi.org/10.1016/j.jestch.2016.10.019
 M. Hanief et al. / Engineering Science and Technology, an International Journal xxx (2016) xxx–xxx 7

to predict cutting forces from the ANN model. The comparison of [2] S.A. Tobias, Machine-Tool Vibrations, Blackie, London, 1965.
[3] Surinder Kumar, Meenu Gupta, P.S. Satsangi, Multiple-response optimization
the experimental values with that of the predicted values of the
of cutting forces in turning of UD-GFRP composite using Distance-Based Pareto
model are shown in Fig. 6. It is obvious from the Fig. 6 that the Genetic Algorithm approach, Eng. Sci. Technol. 18 (2015) 1–16.
error between the experimental and the predicted values has been [4] C. Itxaso, J.A. Sarasuaa, Mechanistic model for prediction of cutting forces in
reduced considerably. The mean square error in ANN model was turning of non-axisymmetric parts, Procedia CIRP 31 (2015) 435–440.
[5] T. Dorlin, F. Guillaume, J.P. Costes, Analysis and modeling of the contact radius
found to be 0.0059 (see Table 4). effect on the cutting forces in cylindrical and face turning of Ti6Al4V titanium
alloy, Procedia CIRP 31 (2015) 185–190.
[6] J. Xie, M.J. Luo, K.K. Wu, L.F. Yang, D.H. Li, Experimental study on cutting
4. Conclusions temperature and cutting force in dry turning of titanium alloy using micro-
grooved tool, Int. J. Mach. Tools Manuf. 73 (2013) 25–36.
In this paper a predictive model for determining the cutting [7] D. Philip, P. Selvaraj, P. Chandramohan, M. Mohanraj, Optimization of surface
roughness, cutting force and tool wear of nitrogen alloyed duplex stainless
forces was developed in turning operation of red brass and the fol-
steel in a dry turning process using Taguchi method, Measurement 49 (2014)
lowing conclusions were drawn from this investigation. 205–215.
[8] S. Sikder, H.A. Kishawy, Analytical model for force prediction when machining
metal matrix composite, Int. J. Mech. Sci. 59 (2012) 95–103.
(i) The cutting force increases with increase in all the three
[9] Uday A. Dabade, D. Dapkekar, S.S. Joshi, Modeling of chip–tool interface friction
parameters (feed, cutting speed and depth of cutting). to predict cutting forces in machining of Al/SiCp composites, Int. J. Mach. Tools
(ii) Out of three cutting parameters resultant cutting force was Manuf. 49 (9) (2009) 690–700.
found to be largely influenced by feed rate and least by [10] A. Pramanik, L.C. Zhang, J. Arsecularatne, Prediction of cutting forces in
machining of metal matrix composites, Int. J. Mach. Tools Manuf. 46 (14)
depth of cut. (2006) 1795–1803.
(iii) ANOVA results confirm that the regression model is ade- [11] H. Joardar, N.S. Das, G. Sutradhar, S. Singh, Application of response surface
quate and is capable to predict the cutting forces. methodology for determining cutting force model in turning of LM6/SiCP
metal matrix composite, Measurement 47 (2014) 452–464.
(iv) On the basis of statistical parameters, it was concluded that [12] C. Shoba, N. Ramanaiah, D. Nageswara Rao, Effect of reinforcement on the
the ANN model can predict the cutting forces more accu- cutting forces while machining metal matrix composites-an experimental
rately that the regression model. approach, Eng. Sci. Technol. 18 (2015) 658–663.
[13] N.A. Fountas, I. Ntziantzias, J. Kechagias, A. Koutsomichalis, J.P. Davim, N.M.
Vaxevanidis, Prediction of cutting forces during turning PA66 GF-30 glass fiber
reinforced polyamide by soft computing techniques, Mater. Sci. Forum 07
Acknowledgements (766) (2013) 37–58.
[14] N.M. Vaxevanidis, J.D. Kechagias, N.A. Fountas, D.E. Manolakos, Open Constr.
Build. Technol. J. 8 (2014) 389–399.
The authors would like to acknowledge and thank Mr. G. N. Lol- [15] N.A. Fountas, J. Kechagias, R. Benhadj-Djilali, C.I. Stergiou, N. M. Vaxevanidis,
puri and Mr. Irshad Ahmad Qadri of N. I. T. Srinagar, for providing Optimizing 5-axis sculptured surface finish machining through design of
experiments and neural networks, In: ASME 2014 12th Biennial Conference on
the workshop facility and technical support for the accomplish-
Engineering Systems Design and Analysis, ESDA 2014.
ment of this research work. [16] N.M. Vaxevanidis, N.A. Fountas, J.D. Kechagias, D.E. Manolakos, Optimization
of main cutting force and surface roughness in turning of TI-6AL-4V titanium
References alloy using design of experiments and artificial neural networks, 2014, pp.
2889–2906.

[1] L.N. Lopez de Lacalle, A. Gutienerrez, J.I. Llorente, J.A. Sanchez, J. Alboniga,
Using high pressure coolant in the drilling and turning of low machinability
alloys, Int J. Adv. Manuf. Technol. 16 (2000) 85–91.

Please cite this article in press as: M. Hanief et al., Modeling and prediction of cutting forces during the turning of red brass (C23000) using ANN and
regression analysis, Eng. Sci. Tech., Int. J. (2016), http://dx.doi.org/10.1016/j.jestch.2016.10.019