This article was downloaded by: [Kondayya D]

On: 04 April 2013, At: 02:21

Publisher: Taylor & Francis
Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: Mortimer House,
37-41 Mortimer Street, London W1T 3JH, UK

International Journal of Computer Integrated

Manufacturing
Publication details, including instructions for authors and subscription information:
http://www.tandfonline.com/loi/tcim20

An integrated evolutionary approach for modelling and

optimisation of CNC end milling process
a

D. Kondayya & A. Gopala Krishna

Department of Mechanical Engineering, Sreenidhi Institute of Science and Technology,

Hyderabad, 501301, Andhra Pradesh, India
b

Department of Mechanical Engineering, University College of Engineering, Jawaharlal

Nehru Technological University, Kakinada, 533003, Andhra Pradesh, India
Version of record first published: 14 May 2012.

To cite this article: D. Kondayya & A. Gopala Krishna (2012): An integrated evolutionary approach for modelling and
optimisation of CNC end milling process, International Journal of Computer Integrated Manufacturing, 25:11, 1069-1084
To link to this article: http://dx.doi.org/10.1080/0951192X.2012.684718

PLEASE SCROLL DOWN FOR ARTICLE

Full terms and conditions of use: http://www.tandfonline.com/page/terms-and-conditions
This article may be used for research, teaching, and private study purposes. Any substantial or systematic
reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any form to
anyone is expressly forbidden.
The publisher does not give any warranty express or implied or make any representation that the contents
will be complete or accurate or up to date. The accuracy of any instructions, formulae, and drug doses should
be independently verified with primary sources. The publisher shall not be liable for any loss, actions, claims,
proceedings, demand, or costs or damages whatsoever or howsoever caused arising directly or indirectly in
connection with or arising out of the use of this material.

International Journal of Computer Integrated Manufacturing

Vol. 25, No. 11, November 2012, 10691084

An integrated evolutionary approach for modelling and optimisation of CNC end milling process
D. Kondayyaa* and A. Gopala Krishnab
a

Department of Mechanical Engineering, Sreenidhi Institute of Science and Technology, Hyderabad 501301, Andhra Pradesh, India;
Department of Mechanical Engineering, University College of Engineering, Jawaharlal Nehru Technological University, Kakinada
533003, Andhra Pradesh, India

Downloaded by [Kondayya D] at 02:21 04 April 2013

(Received 8 August 2011; nal version received 1 April 2012)

In this research, a novel integrated evolutionary based approach is presented for the modelling and multi-objective
optimisation of a machining process. Computer numerical control end milling process has been considered in the
present work as it nds signicant applications in diversied engineering industries. Firstly, genetic programming
(GP) has been proposed for explicit formulation of non-linear relations between the machining parameters (spindle
speed, feed and depth of cut) and the performance measures of interest (material removal rate and tool wear) using
experimental data. Genetic programming approach optimises the complexity and size of the model during the
evolutionary process itself and hence this technique has the potential to identify the true models avoiding the
problems of conventional methods. Central composite second-order rotatable design had been utilised to plan the
experiments and the eect of machining parameters on the performance measures is also reported. In the second
part, as the chosen responses are conicting in nature, a multi-objective optimisation problem has been formulated.
A non-dominated sorting genetic algorithm-II (NSGA-II) has been used to simultaneously optimise the objective
functions. The Pareto-optimal set generated is useful for process planning which is a critical link in computerintegrated manufacturing (CIM).
Keywords: end milling; modelling; genetic programming; multi-objective optimisation; NSGA-II

1. Introduction
Machining processes nd extensive applications in
most of the engineering industries such as aerospace,
automotive, defence and tool and die industries. In this
work, one of the most widely used machining
processes, computer numerical control (CNC) end
milling is considered for investigation. End milling is a
complex and costly shape machining process used for
obtaining proles, slots, engraving, surface contouring
and pockets on nished components. This machining
process has become indispensable in industry and is
continuously nding further applications with the
developments of new cutting tool materials.
Computer numerical control machining being an
expensive process (Tolouei-rad et al. 1997), there can
be a big payo from even small increase in performance. Hence, the selection of optimum process
parameters is highly essential to reduce the machining
cost and increase the production rate. Optimisation
through full enumerations is not possible owing to the
complex nature of the machining process and complicated interaction between the variables of machining.
Moreover, the said procedures neither lead to optimal
use of the machines nor the quality of surface
generated. Though considerable number of researchers

*Corresponding author. Email: [email protected]

ISSN 0951-192X print/ISSN 1362-3052 online
2012 Taylor & Francis
http://dx.doi.org/10.1080/0951192X.2012.684718
http://www.tandfonline.com

carried out various investigations for improving the

process performance, proper selection of machining
parameters for best process performance is nevertheless a challenging job. Owing to the highly
complicated interactions between process parameters,
current analytical models and analyses cannot provide
accurate prediction for better quality control and
higher throughput. Therefore, an ecient method is
needed to determine the optimal machining
parameters.
In case of CNC milling, material removal rate
(MRR) and tool wear (TW) are the most important
response parameters. Specically in rough machining
operations, these responses directly aect the economy
of the process. While material removal rate indicates
the productivity, tool wear is a measure of the quality
of machined components. As CNC milling, is a
complex process (Wang et al. 2006), it is very dicult
to determine the optimal machining parameters for the
best machining performance. Moreover, the performance measures, namely MRR and TW, are conicting in nature as it is desirable to have higher MRR
with lower value of TW.
The overall objective of this research is to apply a
new methodology for modelling and optimisation of

1070

D. Kondayya and A. Gopala Krishna

the non-linear and complex CNC end milling process.

With this aim, accurate prediction models to estimate
MRR and TW were developed from the experimental
data using a potential evolutionary algorithm called
genetic programming (GP). Subsequently, the developed models were used for optimisation of the process.
As the chosen objectives, MRR and TW, are conicting in nature, the problem was formulated as a multiobjective optimisation problem. A popular evolutionary algorithm, non-dominated sorting genetic algorithm II (NSGA-II), was then used to solve and
thereby retrieve the multiple optimal sets of input
variables.

Downloaded by [Kondayya D] at 02:21 04 April 2013

2.

Literature survey

Many approaches have been proposed for end milling

to select the appropriate process parameters. A review
of literature is presented here:
Machinability database systems were in use for
long time for the selection of appropriate cutting
conditions. Model building in machinability database
systems was studied using regression based and
articial neural networks (ANNs) techniques
(Choudhury and El-baradie 1996, Wong and Hamouda 2003).
Optimum cutting parameters for multi-pass milling
utilising the two mathematical techniques dynamic
programming and geometric programming were determined by I.hsan Sonmeza et al. (1999). Shunmugam
et al. (2000) used genetic algorithm to optimise the
machining parameters for multi-pass milling operations. El-Mounayri et al. (2002) performed modelling
and optimisation of at end milling process using
articial neural networks (ANNs) and particle swarm
optimisation (PSO) respectively. Linear regression
models were developed for cutting force using ANNs,
while a production cost objective function was used to
optimise the variables, spindle speed and feed rate.
Tandon et al. (2002) applied the evolutionary computation technique, PSO to optimise multiple machining
parameters simultaneously for the case of pocketmilling on CNC milling machine. A hybrid approach
based on genetic algorithm (GA) and simulated
annealing (SA) was developed by Wang et al. (2004)
to select the optimal machining parameters for plain
milling process, the obtained results demonstrate that
the hybrid method is superior to those using GA alone.
Baskar et al. (2005) demonstrated the eciency of
several optimisation procedures such as GA, tabu
search, ant colony algorithm and particle swarm
optimisation for the optimisation of machining parameters for milling operation. Oktem et al. (2006)
optimised the cutting parameters for minimum surface
roughness in end milling mould surfaces of a

component used in biomedical applications by coupling ANN and GA. Yih-fong (2007) applied Taguchi
methods coupled with principal component analysis in
the process optimisation of the high-speed CNC
milling. Krain et al. (2007) experimentally optimised
the tool life, tool wear and productivity when end
milling Inconel 718. For optimisation they considered
the eects of fed rate, radial depth of cut, tool material
and geometry. They concluded that no single tool
material or geometry gave the best overall performance. Palanisamy et al. (2008) used regression and
ANN techniques for developing models to predict tool
wear on AISI 1020 steel using a carbide cutter in a
universal milling machine. In their work, ank wear
was taken as the response variable measured during
milling, while cutting speed, feed and depth of cut as
input parameters. Bala Murugan et al. (2009) carried
out experimental investigations for machinability
study of hardened steel and obtained optimum process
parameters by grey relational analysis. The machining
process parameters considered were cutting speed,
feed, depth of cut and width of cut and the multiple
responses that were optimised were volume of material
removed, surface nish, tool wear and tool life. Oktem
(2009) used ANNs to model surface roughness and
applied GA to optimise the cutting parameters when
end milling of AISI 1040 steel material with TiAlN
solid carbide tools under wet condition. Lu et al.
(2009) applied grey relational analysis coupled with
principal component analysis for optimisation of
cutting parameters for rough cutting in high speed
end milling on SKD61 tool steel. Modelling results for
surface roughness in end milling of 6061 aluminium
using combined adaptive neuro-fuzzy inference system
and genetic algorithms were presented by Samanta
(2009). Ding et al. (2011) investigated the main eects
and interaction eects and optimised the cutting
parameters for desirable surface roughness in end
milling of hardened AISI H13 steel with PVD coated
carbide insert. The best surface roughness was
achieved with high cutting speed, small axial depth of
cut, high feed and small radial depth of cut. Azlan
et al. (2011) applied an integrated approach to search
for optimal cutting conditions leading to the minimum
value of surface roughness in end milling of Ti-6AL-4V
alloy. Regression equations were developed for surface
roughness which were then optimised by conventional
GA and conventional simulated annealing (SA) algorithms separately and then by an integrated GA and
SA technique.
3. Critique on literature survey
The literature survey reveals that specic eorts were
devoted to determine the most accurate empirical

International Journal of Computer Integrated Manufacturing

Downloaded by [Kondayya D] at 02:21 04 April 2013

models for process performances such as surface

roughness, material removal rate and tool wear. These
models were utilised as objective functions and
optimised to obtain the machining conditions.
The other main ndings of literature survey are:
(1) The predominant modelling technique for
establishing the process models were regression
based techniques and ANNs.
(2) Multi-objective optimisation for multi pass
milling has been performed considering production time and production cost as objective
functions.
(3) Eorts were mainly focused on optimisation of
only single performance characteristic in CNC
end milling.
(4) No published work available on multi-objective
optimisation of the conicting objectives, MRR
and TW.
In statistical-based techniques like regression analysis, a prediction model has to be determined in
advance and a set of coecients have to be found. The
pre-specied size and shape of the model means that
these are not adequate to capture the complex relation
between inuencing variables and output parameters.
Although ANNs have also been used in the literature
for modelling, they have the drawback of not being
able to quantify explicitly the relationships between
inputs and outputs (Baykaso
glu 2008). Consequently,
great care must be taken with the ANN approach to
prevent over-tting (Benyamin and Daniel 2002).
However, the accuracy and possibility of determining
the global optimum solution depends on the type of
modelling technique used to express the objective
function and constraints as functions of the decision
variables (Jain et al. 2007). Therefore, eective,
ecient and economic utilisation of the CNC milling
process necessitates an accurate modelling and optimisation procedure.

as compared to traditional approaches is in the model

structure denition which is not apriori. Therefore, an
advantage of this approach is that fewer assumptions
have to be made regarding the nal form of the model
as the algorithm can evolve the model structure from
elementary building blocks. The generated model helps
directly to obtain an interpretation of the parameters
aecting the process. More details of this methodology
are discussed in section 5. The models developed by
GP were subsequently used for optimisation.
In the current study, the optimisation problem of
CNC milling was explicitly formulated as a multiobjective optimisation problem, as the determination
of the optimal machining conditions involves a conict
between maximising MRR and minimising the TW. It
can be noted that the classical optimisation methods
(gradient descent, weighted sum methods, goal programming, minmax methods, etc.) are not ecient for
handling multi-objective optimisation problems. The
aforesaid methods do not nd multiple solutions in a
single run, and therefore it is necessary for them to be
applied as many times as the number of desired Paretooptimal solutions (Sardinas et al. 2006). In addition,
classical methods fail when the objective function
becomes discontinuous. The above-mentioned diculties of classical optimisation methods are eliminated in
evolutionary algorithms, as they can nd the multiple
solutions in a single run. As a result, one of the best
evolutionary approaches, the NSGA-II is proposed in
this paper for multi-objective optimisation of CNC
milling. Genetic algorithm (GA) based multi-objective
optimisation methodologies have been widely used in
the literature to nd Pareto-optimal solutions (Fonseca
and Fleming 1993, Horn et al. 1994, Zitzler and Thiele
1998). In particular, NSGA-II has proven its eectiveness and eciency in nding well-distributed and wellconverged sets of near Pareto-optimal solutions (Deb
2002). The proposed methodology of integrating GP
and NSGA-II is depicted in Figure 1.
5.

4.

Proposed methodology

In this paper, a novel approach using GP is presented

for modelling of material removal rate and tool wear.
Genetic programming is a technique pioneered by
Koza (1992) and belongs to the class of genetic or
evolutionary algorithms. Since its introduction in the
early 1990s GP has a number of signicant achievements in tackling industrial scale modelling, data
analysis, search and optimisation problems (Riolo
et al. 2007). The capabilities of GP in capturing the
empirical relationship between input variables and
output features are of primary signicance for modelling machining processes. The distinctive aspect of GP

1071

Modelling using GP

The Evolutionary Algorithms (EAs) are stochastic

search methods which combine important characteristics like robustness, versatility and simplicity. The
ability of EAs to search for complete and global
solutions for a given problem makes them powerful
problem solving tools. EAs exploit not only the
information contained in each individual, but also
the information of the population as a whole. Genetic
programming is one such EA to generate an optimum
model structure with interaction terms or higher order
terms. Genetic programming is a fairly recent EA
method compared to other three variants of evolutionary computation viz., genetic algorithms,

Downloaded by [Kondayya D] at 02:21 04 April 2013

1072

Figure 1.

D. Kondayya and A. Gopala Krishna

Proposed methodology.

evolutionary strategies and evolutionary programming

(Angeline 1995).
The cardinal principles of GP were originally
proposed by Koza (1992). Since then this technique
has found widespread applications in diversied elds
such as industrial robotics (Dolinsky et al. 2007),
chemical process modelling (Gree and Aldrich 1998),
prediction of shear strength of beams (Ashour et al.
2003), etc. However, it is noteworthy that the potential
of this tool has not been exploited in the eld of
manufacturing and few applications in the eld of
manufacturing can be found in the literature (Nastran
and Balic 2002, Brezocnik et al. 2004, Kovacic et al.
2007). Therefore, in the present work, the proposed
method is used to model the machining process.
Genetic Programming principles are primarily
derived from GAs, and hence share most of GAs
properties. The main dierence between GP and GAs
lies in the individual representations. In GAs population of individuals is represented as strings whereas in
GP individuals are represented as tree structures. The
tree like structures of GP, which are in fact solutions to
problems, have hierarchical compositions of terminals
and functions. Terminals are input variables appropriate to a particular problem domain and user
species a number of functions that manipulate
terminals. The primitive functions typically include:

Figure 2.

Example of GP procedure.

arithmetic operations ( , , *, /), Boolean operations

(AND, OR, NOT), logical operations (IFTHEN
ELSE), and non-linear functions (sin, cos, tan, exp,
log). Compatibility between the functions and terminals must be ensured in order to pass information
impeccably between each other (Sette and Boullart
2001). Figure 2 shows a typical example of GP
procedure. Representation of an individual in GP is
shown in Figure 3 which corresponds to the expression

International Journal of Computer Integrated Manufacturing

Downloaded by [Kondayya D] at 02:21 04 April 2013

of (x72)+y/(z2). The set of functions in the representation are { , , *, /} whereas the set of terminals are
{x, y, z}.
5.1. Initial population
Genetic programming algorithm begins with a set of
randomly created individuals called initial population.
Each individual is a potential solution represented as a
tree. Each tree is constructed by random compositions
of the sets of functions and terminals. The shape of a
tree has an inuence on its evolution and both sparse
and bushy trees should be present in the initial
population. To ensure this, a ramped half-and-half
method, suggested by Koza (1992) is usually used in
GP. In this method, half of the trees in the population
are generated as full trees and other half as random
trees. Once the population is generated, a suitable
tness function should be given for evaluating the
tness value of each individual. Then, a set of
individuals with better tness value will be selected
and used to generate new population of next generation by the predened genetic operators while the
population size is kept constant.
5.2.

Genetic operators

The purpose of genetic operators is to evolve

individual trees. The main operators generally include
reproduction, crossover and mutation. Reproduction
involves copying of the selected individual into the next
generation population. In general, about 10% of the
population is selected for simple reproduction and
90% is selected through crossover. Crossover is usually
the most important genetic operator in GP. Its
application produces two children trees from two
parent trees by exchanging randomly selected sub-trees
of each parent. Both parents are selected using one of

Figure 3.

Representation of an individual in GP.

1073

the stochastic selection methods such as tness

proportional selection or tournament selection.
The cross-over operator is illustrated in Figure 4.
To avoid excessive growth of the branches of the
program trees, a maximum depth value is usually
established. If the crossover operation produces an
ospring of impermissible depth, this ospring is
disregarded and its parent is reproduced as it is.
Koza (1992) uses the default maximum depth of 17.
Similar to GAs, GP uses the mutation operator in
order to avoid falling into the local optimal solution.
Mutation is typically a replacement of a randomly
selected single terminal of an individual with a new
randomly generated terminal. While mutation plays an
important role in maintaining genetic diversity in the
population, most new individuals in a particular
generation result from crossover. The mutation operator is illustrated in Figure 5.
5.3. Result designation and termination criteria
Execution of the aforementioned three genetic operators constitutes one generation and the procedure is
repeated until a termination criterion is met. The single
individual with the best value of tness over all the
generations is designated as the result of a run. The
termination criterion can be either a xed number of
generations or specied quality of the solution. The
number of runs required for a satisfactory solution
depends on the complexity of the problem under
consideration.

Figure 4.

Illustration of cross-over operator.

1074
6.

D. Kondayya and A. Gopala Krishna

Optimisation using NSGA-II

Downloaded by [Kondayya D] at 02:21 04 April 2013

Multi-objective evolutionary algorithms (MOEAs) are

optimisation methods that make the best use of the
features of multipoint search of EAs, and can obtain
Pareto approximation set at the same time. Elitist
Non-dominated Sorting Genetic Algorithm (NSGAII) proposed by Deb et al. (2002) is a well known and
an excellent implementation of MOEA.
NSGA-II improves upon the original version by
incorporating the following main features:
(1) At each generation, the best solutions found are
preserved and included in the following generation using an elite-preserving operator.
(2) A fast algorithm is used to sort the nondominated fronts.
(3) A two level ranking method is used to assign
the eective tness of solutions during the
selection process.
6.1. Principle of NSGA-II
Initially, a random parent population Pgt (the subscript t indicates the generation) of size N is created.
The population is sorted based on the non-domination

Figure 5.

Illustration of mutation operator.

Figure 6.

Procedural steps of NSGA-II.

principle. Each solution is assigned a tness (i.e. rank)

equal to its non domination level (1 is the best level, 2
is the next-best level, and so on). Thus, maximisation
of tness is assumed. At rst, the usual binary
tournament selection, recombination, and mutation
operators are used to create an ospring population
Qgt of size N. Since elitism is introduced by comparing
current population with the previously best found nondominated solutions, the procedure is dierent after
the initial generation. The procedural steps are
illustrated in Figure 6. First, a combined population
Rgt Pgt U Qgt is formed. The population Rgt is of size
2N. Then, a fast non-domination sorting algorithm is
used to rank the solutions according to their dominance rank and organise fronts of equal rank.
In this ranking method, an individual, k, is
randomly chosen from the population Rgt and inserted
in an intermediate set named F1. Then, another
solution k0 is drawn from Rgt and compared to all
individuals from F1. If k0 dominates k, k0 enters F1 and
k is deleted, but if k dominates k0 , then k0 is deleted and
k stays in F1. Continuing this comparison for all
individuals, F1 will consist of all non-dominated
individuals of Rgt, the Pareto front by denition.
Now the rst Pareto front is removed from the original
population and the same procedure is iteratively
continued to identify other layers of Pareto fronts {Fi
i 1,2,3 . . . }. Subsequently, individual solutions within
each front are ranked according to a density measure
using the crowding operator. This operator, as pictured
in Figure 7, measures the diversity of each individual
by measuring half of the perimeter of the rectangle that
encloses a solution in the objective function space and
assigning innite distance to the extreme points of the
Pareto-front. The next generation, Pgt 1, which has
the same size as the rst generation is lled with
consecutive Pareto fronts {Fi i 1,2,3 . . . } until no full
Pareto front can be fully accommodated anymore.

Downloaded by [Kondayya D] at 02:21 04 April 2013

International Journal of Computer Integrated Manufacturing

Then, the solutions in the next Pareto layer are sorted
in descending order according to their distance assignment and the empty spaces in the proceeding generation are lled with higher ranked solutions.
The next ospring population, Qgt 1, is created by
using the crowded tournament selection operator. Two
attributes can be considered for each individual
solution: First, a non-domination rank (equal to the
Pareto layer rank) and second, a crowding distance. In
the tournament selection, competitions are set up
between individuals. The tournament is won by that
individual having a better non-dominated rank (lies on
an outer Pareto front). If both individuals are on the
same Pareto front, ties are broken by the crowded
distance and the tournament is won by the one which
is least crowded. The procedure outlined in Figure 7 is
repeated until the termination criterion is met and the
best-known Pareto front is saved in an archive. The
solutions in this archive are the Pareto optimal
solutions of the problem under consideration.
7.

Experimental details

The machining experiments were conducted on a

powerful and precise 3-axis CNC vertical machining
centre (model: AGNI BMV45 made by Bharat Fritz
Werner Ltd.) employing a continuously variable
spindle speed up to a maximum of 6000 rpm and
with a maximum spindle power of 5.5 kW. The feed
rates can be set up to a maximum of 10m/min.
Workpieces of AISI 52100 steel of size 200 mm 6 300
mm 6 25 mm with hardness 50 HRC after heat
treatment were prepared and utilised. All of the
experiments were performed under dry conditions
along the 300 mm length in conventional milling
mode, as recommended by the tool supplier for the

Figure 7.

Illustration of Crowded comparision operator.

1075

specic work material. Coated carbide inserts (ISO:

APKT11T308-PM) were used to machine the tool
steel. The inserts have rake angle of 08 and clearance
angle of 118. End milling inserts with this geometry are
suited for semi-dry and dry machining operations of
steels. A commercially available double end mill insert
holder of type EMP01-020-G20-AP1102 of 20 mm
diameter and overall length 100 mm was utilised for
experiments. Figures 8 and 9 show the CNC set-up of
the experiment and view of the machining zone.
CNC end milling involves several control variables
such as surface cutting speed, spindle speed, axial
depth of cut, radial depth of cut, feed and radial
engagement of tool. The workpiece and cutting tool
material also inuence the performance of the process.
However, based on the literature survey and the trial
experiments, the variables, namely, spindle speed (x1),
feed (x2), and axial depth of cut (x3) were considered
as the decision (control) variables and the MRR and
the tool wear were considered as the output responses.
Characterisation of wear on cutting tool insert is
specied primarily by ank wear and its progressive
growth (Urbanski et al. 2000). Flank wear evaluation
of insert was made o line by a tool makers
microscope (make: Mitutoyo TM500) with 30X
magnication and 1 mm resolution. The admissible
wear was ascertained in accordance with ISO
8688:1989 standard and measured by taking the
average ank wear for two inserts for each experiment.
Each experiment was repeated twice and a new insert
with same specication was used every time and
machining was stopped after ve passes for tool wear
measurement. Machining time is noted at the end of
each pass and the material removal rate is calculated
by loss of weight method. The weight of work piece is
measured accurately by digital balance with 0.001 mm
accuracy and the material removal rate is expressed in
g/min.
Experimental designs constitute powerful methodology for accumulating and analysing information
about a given process rapidly and eciently from a
small number of experiments, thereby minimising the
experimental costs. The GP models for material
removal rate and tool wear were trained by a dataset
constructed using design of experiment based rotatable
central composite design (CCD). The 27 runs CCD
method is chosen since the method provides a wider
covering region of parameter space and good consideration of variables interaction in the model. The
ranges of machining parameters used were as follows:
spindle speed (x1) of 900, 1200 and 1500 RPM, feed
rates (x2) of 30, 45, 60 mm/min and axial depth of cut
(x3) of 0.4, 0.5, 0.6mm. The radial depth of cut was
maintained constant at half of the cutter diameter. The
measured values of tool wear and material removal

Downloaded by [Kondayya D] at 02:21 04 April 2013

1076

Figure 8.

D. Kondayya and A. Gopala Krishna

CNC set-up of the experiment.

rate for 27 experiments conducted as per CCD are

shown in Table 1. This constitutes the training dataset
for the GP algorithm. In addition, the developed
model is validated to ensure the generalisation
capability of the predicted model for unseen cases
and the data set used for validation is presented in
Table 2.
8.

Implementation issues

GP, being a stochastic search technique, makes no

prior assumptions about the actual model form. The
structure and complexity of the model evolve automatically. The populations of models of initial tree
generation for MRR and tool wear are rst initialised
using the ramped half-and-half method. The terminal
set T and the function set F were dened as: T (x1,
x2, x3, <) and F (, 7, *, /). The terminal set consists
of all input variables of the CNC end mill process and
a random ephemeral constant < 2 (7200, 200). The
function set contains addition, subtraction, multiplication and protected division. The genetic programming run is controlled by many parameters of which
the two major numerical parameters are the population size and the maximum number of evolutionary
generations. These two parameters depend on the
diculty of the problem involved. Other minor
numerical parameters include the probability of crossover, reproduction and mutation (Koza 1992). Therefore, some preliminary experiments were carried out to

obtain some estimates of workable parameters. These

preliminary test runs in the GP system were executed
for the output parameters independently. Based on
these experiments, the nal parameters used to
generate the models are given in Table 3.
Evolutionary algorithms are generally robust to
variations of control parameters and some guidelines
are provided (Koza 1992) for choosing the control
parameters of standard GP. Population size and
number of generations were optimally set to 500 and
40 respectively as low population size did not show
better results while still large population size appeared
to yield little advantage in terms of the improvement of
tness. Also there was increase in computation times
with large population size and higher number of
generations. GP being probabilistic in application, it
is important to study the problem by carrying out
multiple runs of the algorithm, and analyse the results
before arriving at conclusions, thus, multiple runs of
the algorithm were conducted. After the investigation
of various alternative models, the following expressions for MRR and TW were found to have the best
tness value.








MRR 0:9896x3
x2 x3 2 22:46=x1 0:24x2 x3




x2 x3 1  x3
1






x2  24= x3 2  0:997x2 64:85








 160 x1 36 =x3
89 :
2

TW x2 x3



1077

International Journal of Computer Integrated Manufacturing

Downloaded by [Kondayya D] at 02:21 04 April 2013

Table 1.
S.
no.

Speed (x1)
(RPM)

Feed (x2)
(mm/min)

Depth of
cut (x3)
(mm)

MRR
(g/min)

TW
(mm)

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27

1500
900
1500
1200
1200
900
1500
1200
900
1200
900
1200
1200
1500
1500
900
1500
900
900
1500
900
1500
1200
1500
1200
900
1200

30
30
60
45
30
30
30
60
45
45
45
30
45
45
60
45
45
60
30
30
60
60
60
45
60
60
30

0.6
0.4
0.5
0.6
0.5
0.6
0.4
0.4
0.4
0.5
0.5
0.4
0.4
0.6
0.6
0.6
0.4
0.6
0.5
0.5
0.5
0.4
0.5
0.5
0.6
0.4
0.6

5.587
3.487
8.494
7.87
4.888
5.587
3.492
5.924
4.744
6.641
6.741
3.497
4.744
7.645
9.479
7.59
4.844
9.479
4.888
4.988
8.376
5.925
8.294
6.654
9.579
5.624
5.594

0.163
0.09
0.209
0.199
0.13
0.139
0.115
0.145
0.151
0.174
0.149
0.115
0.148
0.207
0.225
0.175
0.11
0.225
0.13
0.116
0.189
0.175
0.202
0.188
0.235
0.173
0.143

Table 2.
Figure 9.

View of machining zone.

The tness measure while developing the above

models is considered as the correlation coecient, R2
as this produces an accurate solutions in fewer
generations. The algorithm run was continued until a
value as close as possible to the maximum value of R2
was obtained. The correlation coecient value lies in
the range [0, 1] and measures the way in which the
predicted values and actual values vary together.
Therefore credit is given to the solutions that are close
to the correct form. Figures 10 and 11 show the
convergence graphs of the models predicted by the
proposed algorithm. The tness measures for MRR
and TW have gradually improved with the number of
generations and nally converged to 0.9980 and
0.8950, respectively. On the basis of the high values
of R2 it can be said that the models are adequate in
representing the process. The normal probability plots
of the residuals for the output responses are also
shown in Figures 12 and 13. A check on these plots
reveals that the residuals are located on a straight line,
which means that the errors are distributed normally
and the obtained models are fairly well tted with the
observed values.

Training dataset.

Validation dataset.

S.
no.

Speed
(x1)
(RPM)

Feed (x2)
(mm/min)

Depth of cut
(x3) (mm)

MRR
(g/min)

TW
(mm)

1
2
3
4
5
6
7
8
9
10

1300
1000
1100
1300
1100
1300
1000
1000
1000
1100

35
35
50
40
35
35
35
50
40
40

0.55
0.42
0.55
0.45
0.45
0.32
0.55
0.45
0.32
0.45

6.1
4.429
7.988
5.68
4.89
2.832
5.99
6.461
3.221
5.34

0.159
0.123
0.19
0.146
0.123
0.097
0.158
0.167
0.098
0.143

Table 3.

GP control parameters.

Terminal set:
Function set:
Population size:
Number of generations (max.):
Number of independent runs:
Crossover probability (%):
Mutation probability (%):
Reproduction probability (%):
Selection method:
Fitness measure:

{x1, x2, x3}

{ , 7 ,*, /}
500
55
10
85
5
10
Tournament
R2

Downloaded by [Kondayya D] at 02:21 04 April 2013

1078

D. Kondayya and A. Gopala Krishna

Figure 10.

Convergence graph of MRR model.

Figure 11.

Convergence graph of Tool wear model.

Figure 13. Probability plot of the residuals for the output

response Tool wear.

Figure 14. Comparison of the predicted model and the

experimental value for the validation datasets of MRR.

8.1. Eects of cutting parmeters on the responses

Surface plots have been drawn using MINITAB for
the convenience of understanding the surface eects
and selecting the best combinations of cutting parameters. The MRR and tool wear variation for dierent
combinations of cutting parameters are shown in
Figures 16 and 17.
Figure 12. Probability plot of the residuals for the output
response MRR.

8.1.1.

A comparison of the predicted models and the

experimental values for the validation datasets of
MRR and TW are shown in Figures 14 and 15
respectively. Very high values of R2 for MRR and TW
of validation sets are obtained and found to be 0.9989
and 0.9988 respectively. These indicate that the
developed models satisfactorily represent the outputs.

Figure 16a shows how the material removal rate

depends on the feed (x2) and depth of cut (x3) in the
case when the spindle speed (x1) of 1200 RPM is kept
constant. It can be seen that even though both factors
have inuence on MRR the feed is a more dominant
factor. The combination of high value of work piece feed
and the depth of cut provide the highest material
removal rate. The inuence of the spindle speed and

MRR

Downloaded by [Kondayya D] at 02:21 04 April 2013

International Journal of Computer Integrated Manufacturing

Figure 15.

Comparison of the predicted model and the experimental value for the validation datasets of TW.

Figure 16.

Surface plots of speed, feed and depth of cut on MRR.

1079

Downloaded by [Kondayya D] at 02:21 04 April 2013

1080

D. Kondayya and A. Gopala Krishna

depth of cut on the MRR for the constant feed of 45

(mm/min) is shown in Figure 16b. Both the factors show
similar intensity of inuence on MRR. The MRR
decreases if the cutting speed increases. The depth of cut
has the opposite eect, that is, MRR decreases when
depth of cut decreases. Figure 16c shows the inuence of
the spindle speed and feed on the MRR for the constant
depth of cut of 0.5mm. Inherent to milling machines is
the cutting speed which is independent of table feed.
Thus the MRR cannot be increased by increasing the
cutting speed (Kopac and Krajnik 2007). It is obvious
from the gure that MRR increases only marginally
when the speed increases. The feed is again, by far, a
more inuential factor.

adhesion of work piece material onto the tool ank

face, thus hastening the tool wear. From Figure 17b it
is clear from the contour surface that, wear is
maximum (about 0.20 mm) when x1 and x2 are at
their higher limits and is minimum (about 0.136 mm)
when x1 and x2 are at their lower limits. The Figure 17c
shows the disposition of tool wear with input parameters. In this gure, it could be generally inferred that
as the spindle speed, feed, and depth-of-cut increase,
the tool wear increases. These parameters clearly
aect the tool wear. It is evident from the contour
surface that wear is maximum (about 0.20 mm) when
x2 and x1 are at their higher limits and is minimum
(about 0.14 mm) when x2 and x1 are at the lower limit.

8.1.2. Tool wear

9.

Figure 17a shows the distribution of tool wear with

input parameters feed and depth of cut. It is evident
from the contour surface that tool wear is maximum
(about 0.24 mm) when x2 and x3 are at their higher
limits and is minimum (about 0.10 mm) when x2 and x3
are at their lower limits. At lower values of depth of
cut, fewer work piece material holds to the ank than
at larger depth of cut. As the forces and heat generated
during the machining process are higher at larger depth
of cut, it is inferred that the higher temperature and the
higher force are the major reasons that cause the

The two objective functions considered in this study

are:

Figure 17.

Formulation of multi-objective optimisation

(1) maximisation of material removal rate and

(2) minimisation of tool wear
which are given by Equations (1) and (2),
respectively.
The two objective functions are optimised subject
to the feasible bounds of input variables. The optimisation problem is dened as follows:

Surface plots of speed, feed and depth of cut on Tool wear.

International Journal of Computer Integrated Manufacturing

Maximise








x2 x3 2 22:46=x1 = 0:24x2 x3




3
x2 x3 1  x3

MRR 0:9896x3

Minimise








TW x2 x3 = x2  24= x3 2  0:997x2








64:85  160 x1 36 =x3
89 :

Subject to
900  x1  1500
30  x2  60

Downloaded by [Kondayya D] at 02:21 04 April 2013

0:4  x3  0:6

10. Results and discussions

The source code of the proposed optimisation algorithm is based on the description provided by Deb
et al. (2002). For the current work, the code was run on
a Pentium IV processor using Microsoft VC programming language on a Windows XP platform.
Tournament selection, Simulated Binary Crossover
(SBX) and polynomial mutation operators were
selected as the genetic operators of the real-coded
NSGA-II algorithm. NSGA-II starts with a random
generated population. To limit the eect of randomness on the results, the algorithm had to be run a
number of times. Analysing the results of various
simulations on number of test cases revealed that, after
running the algorithm for about 15 times, no
signicant improvement could be obtained for the
Pareto solutions. The control parameters required for
implementation of the algorithm are listed in Table 4.
For achieving better convergence, 350 generations
were used in the study. The algorithm found the
Pareto optimal front of conicting objective functions
with good diversity of solutions, as shown in Figure 18.

Table 4.

The optimal input variables and their corresponding

objective function values are presented in Table 5.
Since none of the solutions in the non-dominated set is
absolutely better than any other, any one of them is an
acceptable solution. The choice of one solution over
the other depends on the requirement of the process
engineer. By analysing the Pareto front, some decisions
could be taken, depending upon specic requirements
of the process. Analysing point 1 which is at the
extreme of the front, the highest value of material
removal rate (maximum productivity) is achieved but
the highest value of tool wear (worst surface quality) is
achieved at this point too. On the lower side of the
front at point 1 a component can be machined with
low tool wear (best surface quality) but with minimum
MRR (minimum productivity). All the other points are
intermediate cases. They must be employed when a
certain response factor is established beforehand. For
example, to maintain accuracy of product, if the
machining condition can allow a tool wear of
1.6 mm, the process engineer can choose the parameter
setting to obtain maximum MRR at the specied value
of tool wear. As can be observed from the graph, no
solution in the front is better than any other as all of
them are non-dominated solutions. The choice of a
solution has to be made purely based on production
requirements. To further illustrate the advantage of the
proposed methodology, some examples are explained
as follows.
From the experimental results of Table 1, the
parameters listed in the ninth experiment lead to the
TW value of 0.151 mm and the MRR of 4.744 g/min.
After optimisation, it can be noted that the MRR is
increased to 5.831 g/min for the same surface nish (S.
no. 42, Table 5), with a 23% increase in MRR. In
another instance, from Table 1, rst experiment, the set
of input variables leads to the MRR of 5.587 g/min
and the TW value of 0.163 mm. After optimisation,
the TW value is reduced to 0.1487 (Sl No. 13, Table 5)

NSGA-II Control parameters.

Population size (N):

Number of generations (ngen):
Selection strategy:
Crossover probability (pc):
Mutation probability (pm):
Distribution index for crossover
operator: (nc)
Distribution index for mutation
operator: (nm)
Population size (N):

100
350
Tournament
0.85
0.15
20
20
50

1081

Figure 18.

Pareto optimal front.

1082

D. Kondayya and A. Gopala Krishna

Downloaded by [Kondayya D] at 02:21 04 April 2013

Table 5.

Final optimal solutions.

S.
no

Spindle
Speed
(RPM)

Feed
(mm/min)

Depth
of cut
(mm)

MRR
(g/min)

Tool
wear
(mm)

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50

931.433
906.919
933.998
922.752
903.403
905.414
900.798
928.707
906.845
929.41
904.941
910.858
906.549
933.277
907.317
905.422
935.146
901.12
933.709
929.195
938.027
935.205
904.108
900.148
901.683
931.34
911.594
907.317
932.903
904.71
900.261
928.964
933.998
920.798
935.205
907.586
936.665
904.03
905.456
933.709
901.143
900.309
905.451
900.014
903.628
934.749
902.692
930.785
931.433
935.714

50.734
50.541
36.642
41.747
50.566
45.905
45.538
41.783
37.079
36.783
41.357
41.663
36.899
37.107
50.912
45.630
51.310
45.752
50.277
45.836
46.394
51.082
37.362
50.660
51.343
41.747
50.480
51.122
37.572
37.189
45.735
45.978
41.293
50.221
50.541
37.420
45.741
43.960
42.002
50.177
50.386
37.190
42.131
41.233
41.266
50.605
45.769
50.846
50.734
36.642

0.5012
0.5146
0.5224
0.5008
0.5159
0.4892
0.4854
0.5025
0.4898
0.5141
0.4989
0.5104
0.5072
0.5218
0.4897
0.5099
0.5025
0.4893
0.5226
0.5043
0.5225
0.5025
0.4968
0.5096
0.4884
0.4994
0.5096
0.4892
0.5219
0.5147
0.5029
0.5149
0.5216
0.5157
0.4976
0.5105
0.5212
0.4984
0.4889
0.5226
0.5129
0.5150
0.5096
0.4893
0.4972
0.5178
0.5006
0.5026
0.5012
0.5223

7.629
7.524
5.855
6.209
7.547
6.543
6.438
6.238
5.483
5.768
6.133
6.336
5.593
5.907
7.157
6.829
7.422
6.526
7.616
6.769
7.113
7.394
5.613
7.458
7.187
6.189
7.435
7.173
5.968
5.826
6.735
6.948
6.443
7.501
7.246
5.801
7.010
6.447
6.066
7.602
7.477
5.831
6.383
5.979
6.099
7.582
6.705
7.366
7.328
5.485

0.1887
0.1864
0.1524
0.1599
0.1869
0.1665
0.1643
0.1606
0.1441
0.1504
0.1583
0.1628
0.1487
0.1536
0.1797
0.1729
0.1843
0.1662
0.1887
0.1714
0.1791
0.1838
0.1469
0.1849
0.1802
0.1595
0.1845
0.1800
0.1549
0.1517
0.1707
0.1754
0.1653
0.1861
0.1813
0.1511
0.1770
0.1648
0.1568
0.1884
0.1854
0.1518
0.1637
0.1549
0.1575
0.1877
0.1700
0.1833
0.1824
0.1423

with almost the same value of MRR. Experiment

number 11 of Table 1 leads to MRR of 6.741 g/min
and TW of 0.149 mm for speed of 900 RPM, feed of 45
mm/min of and depth cut of 0.5 mm. In the optimal set
for S. no. 7, Table 5 for speed of 900.798 RPM, feed of
45.5389 mm/min and depth of cut of 0.4854 mm the
MRR is 6.438 g/min and TW is 0.1643mm. The above
discussion signies that there is conformity between

values obtained from the optimisation technique with

the experimental values for relatively same parameter
settings.
The scanning electron microscopy (SEM) photographs of the tool inserts that correspond to the best
values of MRR and TW are shown in Figures 19 and
20. Those values of MRR and TW correspond to the
extreme positions (point 1 and point 2) of the Pareto
optimal set shown in Figure 18. As can be observed
from the photographs, the variation of tool wear and
resultant eect on tool edge is apparent with respect to
the dierent optimal sets of input variables.
11.

Conclusions and future work

Material removal rate and tool wear are important

machining performance measures which directly inuence the productivity and accuracy of the process. In
this paper, for the rst time evolutionary based
approaches were used to both model and optimise
the CNC end milling process. Experimental data based
on CCD were used to develop empirical models for
MRR and TW in terms of prominent machining
parameters such as spindle speed, feed and depth of
cut, with an ecient evolutionary algorithm namely,
GP. Genetic programming is a domain independent
methodology which does not assume any a priori
functional form of the solution and hence it can
accurately model the complex relationships of the
process. The high R2 value of the models and the
normal probability plots prove the eectiveness of the
GP approach to establish substantially valid models.
NSGA-II has been used to simultaneously optimise the
conicting objectives of material removal rate and tool
wear. Accordingly the Pareto-optimal solution set is
generated and presented. This enables the manufacturing engineer to select a particular optimal set of input
variables according to production requirements. The
optimum values are essential for the automation of the
process and implementation of a computer integrated
manufacturing system.
While the present study is sucient to demonstrate
the potential of the approach, for extending its
capability towards real industrial implementation,
future studies may also concentrate to include other
responses such as surface nish, cutting forces and
chatter for multi-objective optimisation. Furthermore,
the models developed may include additional input
parameters as encountered in industrial processes such
as radial depth cut, geometry of cutting edge, diameter
of tool etc. In such a case, the major problem is often
one of determining the relevant input variables from a
highly correlated data set. Also, it would be interesting
to extend the proposed approach to other cutting
conditions and workpiece/tool combinations as the

International Journal of Computer Integrated Manufacturing

1083

Downloaded by [Kondayya D] at 02:21 04 April 2013

Figure 19. SEM photograph of cutting tool insert corresponding to point 1 of Pareto-optimal front. Machining conditions:
x1 931.433 RPM, x2 50.734 mm/min, x3 0.5012 mm, MRR 7.629 g/min, TW 0.1887 mm.

Figure 20. SEM photograph of cutting tool insert corresponding to point 2 of Pareto-optimal front. Machining conditions:
x1 935.714 RPM, x2 36.642 mm/min, x3 0.5223 mm, MRR 5.485 g/min, TW 0.1423 mm.

modelling and optimisation algorithms are of generic

type and the approach is systematic.
References
Angeline, P.J., 1995. Evolution revolution: an introduction
to the special track on genetic and evolutionary
programming. IEEE Expert, 10 (3), 610.
Ashour, A.F., Alvarez, L.F., and Toropov, V.V., 2003.
Empirical modelling of shear strength of RC deep beams
by genetic programming. Computers and Structures, 81,
331338.
Azlan Mohd, Z., Habibollah, H., and Saan, S., 2011.
Integration of simulated annealing and genetic algorithm
to estimate optimal solutions for minimising surface
roughness in end milling Ti-6AL-4V. International
Journal of Computer Integrated Manufacturing, 24 (6),
574592.
Bala Murugan, G., Biswanath, M., and Sukamal, G., 2009.
Optimisation of machining parameters for hard machining: grey relational theory approach and ANOVA.
International Journal of Advanced Manufacturing Technology, 45, 10681086.
Baskar, N., et al., 2005. Optimization of machining
parameters for milling operations using non-conventional methods. International Journal of Advanced Manufacturing Technology, 25, 10781088.
Baykasoglu, A., 2008. Gene expression programming based
meta-modelling approach to production line design.
International Journal of Computer Integrated Manufacturing, 21 (6), 657665.

Benyamin, G., and Daniel, R.L., 2002. Automated nonlinear

model predictive control using genetic programming.
Computers and Chemical Engineering, 26, 631640.
Brezocnik, M., Kovacic, M. and Ficko, M., 2004. Prediction
of surface roughness with genetic programming.
Journal of Materials Processing Technology, 157158,
2836.
Choudhury, I.A. and El-baradie, M.A., 1996, Analysis of
model building techniques for the development of
machinability database systems. International Journal of
Production Research, 34 (5), 12611277.
Deb, K., et al., 2002. A fast and elitist multiobjective genetic
algorithm: NSGA-II. IEEE Transactions on Evolutionary
Computation, 6 (3), 187192.
Ding, T.C., et al., 2011. Optimization of cutting parameters
for desirable surface roughness in end-milling hardened
AISI H13 steel under a certain metal removal rate.
Advanced Materials Research, 188, 307313.
Dolinsky, J.U., Jenkinson, I.D. and Colquhoun, G.J., 2007.
Application of genetic programming to the calibration of industrial robots. Computers in Industry, 58,
255264.
El-Mounayri, H., Kishawy, H., and Tandon, V., 2002.
Optimized CNC end-milling: a practical approach.
International Journal of Computer Integrated Manufacturing, 15 (5), 453470.
Fonseca, C.M. and Fleming, P.J., 1993. Genetic algorithms
for multiobjective optimization: formulation, discussion,
and generalization. In: Proceedings of the 5th international conference on genetic algorithms, San Mateo, CA:
Morgan Kaufmann, 416423.

Downloaded by [Kondayya D] at 02:21 04 April 2013

1084

D. Kondayya and A. Gopala Krishna

Gree, D.J. and Aldrich, C., 1998. Empirical modelling of

chemical process systems with evolutionary programming. Computers and Chemical Engineering, 22 (78),
9951005.
Horn, J., Nafpliotis, N. and Goldberg, D.E., 1994. A niched
Pareto genetic algorithm for multiobjective optimization.
In: Proceedings of the 1st IEEE conference on evolutionary
computation, IEEE World Congress on Computational
intelligence, IEEE Service Center, Piscataway, New
Jersey. Vol. 1. 8287.
_
Ihsan
Sonmeza, A., et al., 1999. Dynamic optimization of
multipass milling operations via geometric programming.
International Journal of Machine Tools and Manufacture,
39, 297320.
Jain, N.K., Jain, V.K., and Jha, S., 2007. Parametric
optimization of advanced ne-nishing processes. International Journal of Advanced Manufacturing Technology,
34, 11911213.
Kopac, J. and Krajnik, P., 2007. Robust design of ank milling
parameters based on grey-Taguchi method. Journal of
Materials Processing Technology, 191, 400403.
Kovacic, M., et al., 2007. Prediction of the bending capability
of rolled metal sheet by genetic programming. Materials
and Manufacturing Processes, 22 (5), 634640.
Koza, J.R., 1992. Genetic programming: on the programming
of computers by means of natural selection. Cambridge,
MA: MIT Press.
Krain, H.R., Sharman, A.R.C., and Ridgway, K., 2007.
Optimisation of tool life and productivity when end
milling Inconel 718TM. Journal of Materials Processing
Technology, 189,153161.
Lu, H.S., et al., 2009. Grey relational analysis coupled with
principal component analysis for optimization design of
the cutting parameters in high-speed end milling. Journal
of Materials Processing Technology, 209, 38083817.
Nastran, M. and Balic, J., 2002. Prediction of metal wire
behavior using genetic programming. Journal of Materials Processing Technology, 122, 368373.
Oktem, H., 2009. An integrated study of surface roughness
for modeling and optimization of cutting parameters
during end milling operation. International Journal of
Advanced Manufacturing Technology, 43, 852861.
Oktem, H., Tuncay, E., and Fehmi, E., 2006. Prediction of
minimum surface roughness in end milling mold parts
using neural network and genetic algorithm. Materials
and Design, 27, 735744.
Palanisamy, P., Rajendran, I., and Shanmugasundaram, S.,
2008. Prediction of tool wear using regression and ANN
models in end-milling operation. International Journal of
Advanced Manufacturing Technology, 37, 2941.

Riolo, R., Terence, S., and Bill, W., eds., 2007. Genetic
programming theory and practice IV. Ann Arbor: Genetic
and Evolutionary Computation series, Springer.
Samanta, B., 2009. Surface roughness prediction in machining using soft computing. International Journal of
Computer Integrated Manufacturing, 22 (3), 257266.
Sardinas, R.Q., Santana, M.R., and Brindis, E.A., 2006.
Genetic algorithm based multiobjective optimization of
cutting parameters in turning processes. Engineering
Applications of Articial Intelligence, 19, 127133.
Sette, S. and Boullart, L., 2001. Genetic programming:
principles and applications. Engineering Applications of
Articial Intelligence, 14, 727736.
Shunmugam, M.S., Bhaskara Reddy, S.V. and Narendran,
T.T., 2000. Selection of optimal conditions in multi-pass
face-milling using a genetic algorithm. International
Journal of Machine Tools and Manufacture, 40, 401414.
Tandon, V., El-Mounayri, H., and Kishawy, H., 2002. NC
end milling optimization using evolutionary computation. International Journal of Machine Tools and Manufacture, 42, 595605.
Tolouei-rad, M., and Bidhendi, I.M., 1997. On the optimization of machining parameters for milling operations.
International Journal of Machine tools and Manufacture,
37 (1), 116.
Urbanski, J.P., et al., 2000. High speed machining of moulds
and dies for net shape manufacture. Materials and
Design, 21, 395402.
Wang, Z.G., Wong, Y.S., and Rahman, M., 2004. Optimisation of multi-pass milling using genetic algorithm and
genetic simulated annealing. International Journal of
Advanced Manufacturing Technology, 24,727732.
Wang, Z.G., et al., 2006. Multi-objective optimization of
high-speed milling with parallel genetic simulated annealing. International Journal of Advanced Manufacturing
Technology, 31, 209218.
Wong, S.V., and Hamouda, A.M.S., 2003, Machinability
data representation with articial neural network.
Journal of Materials Processing Technology, 138, 538
544.
Yih-fong, T., 2007. A hybrid approach to optimise multiple
performance characteristics of high-speed computerised
numerical control milling tool steels. Materials and
Design, 28, 3646.
Zitzler, E. and Thiele, L., 1998. Multiobjective optimization
using evolutionary algorithms a comparative case
study. In: A. E.Eiben, eds. Proceedings of the 5th international conference on parallel problem solving from
nature. Amsterdam: Springer, 292304.