You are on page 1of 7

ARTICLE IN PRESS

Engineering Applications of Artificial Intelligence 19 (2006) 127–133


www.elsevier.com/locate/engappai

Genetic algorithm-based multi-objective optimization of cutting


parameters in turning processes
Ramón Quiza Sardiñas, Marcelino Rivas Santana, Eleno Alfonso Brindis
Department of Mechanical Engineering, University of Matanzas, Autopista a Varadero km 3 1/2, Matanzas 44740, Cuba
Received 24 March 2004; received in revised form 4 June 2005; accepted 15 June 2005

Abstract

Determination of optimal cutting parameters is one of the most important elements in any process planning of metal parts. This
paper presents a multi-objective optimization technique, based on genetic algorithms, to optimize the cutting parameters in turning
processes: cutting depth, feed and speed. Two conflicting objectives, tool life and operation time, are simultaneously optimized. The
proposed model uses a microgenetic algorithm in order to obtain the non-dominated points and build the Pareto front graph. An
application sample is developed and its results are analysed for several different production conditions. This paper also remarks the
advantages of multi-objective optimization approach over the single-objective one.
r 2005 Elsevier Ltd. All rights reserved.

Keywords: Turning processes; Multi-objective optimization; Genetic algorithms.

1. Introduction mathematical and numerical optimization techniques


(Sönmez et al., 1999).
The selection of optimal cutting parameters, like the In any optimization procedure, it is a crucial aspect to
number of passes, depth of cut for each pass, feed and identify the output of chief importance, the so-called
speed, is a very important issue for every machining optimization objective or optimization criterion. In
processes. In workshop practice, cutting parameters are manufacturing processes, the most commonly used
selected from machining databases or specialized hand- optimization criterion is specific cost, which has been
books, but the range given in this sources are actually used by many authors, from the beginning of the
starting values, and are not the optimal values (Dereli researches in this branch (Taylor, 1907) to some of the
et al., 2001). most recent works (Liang et al., 2001; Wang et al., 2002;
Optimization of cutting parameters is usually a Saravanan et al., 2003; Cus and Balic, 2003; Amiolem-
difficult work (Kumar and Kumar, 2000), where the hen and Ibhadode, 2004).
following aspects are required: knowledge of machining; Sometimes, other criteria like machining time (Chua
empirical equations relating the tool life, forces, power, et al., 1991), material removal rate (Ko and Kim, 1998;
surface finish, etc., to develop realistic constrains; Chien and Tsai, 2003) or tool life (Molinari and Nouari,
specification of machine tool capabilities; development 2002) have been used too. However, these single-
of an effective optimization criterion; and knowledge of objective approaches have a limited value to fix the
optimal cutting conditions, due to the complex nature
Corresponding author. Tel.: +53+45+261432; of the machining processes, where several different
and contradictory objectives must be simultaneously
fax: +53 45 253101
E-mail address: quiza@umcc.cu (R. Quiza Sardiñas). optimized.

0952-1976/$ - see front matter r 2005 Elsevier Ltd. All rights reserved.
doi:10.1016/j.engappai.2005.06.007
ARTICLE IN PRESS
128 R. Quiza Sardiñas et al. / Engineering Applications of Artificial Intelligence 19 (2006) 127–133

Nomenclature T tool life (min)


v cutting speed (m min1)
a depth of cut (mm) V volume of the removed material (mm3)
CT, CF coefficients for tool life and cutting force x decision variables vector
equations ZT, ZL+O tool cost; labour and overhead cost ($)
D dynamic population size zT cost per tool edge ($)
f feed rate (mm/rev) zL, zO specific labour and overhead cost ($/min)
FC, FMAX cutting force and maximum allowed force a, a0 exponents for cutting speed in tool life and
(N) cutting force equations
g(x) constraint function b, b0 exponents for feed rate in tool life and cutting
K number of epoch in each evolutionary period force equations
M material removal rate (mm3 min) g, g0 exponents for depth of cut in tool life and
P, PMOT power required for machining and motor cutting force equations
power (kW) Gi unfeasibility index for the ith individual
Qi dynamic population for the ith epoch Z machine tool transmission efficiency
QEST static population x used tool life
QPAR Paretian population t production rate (time per part) (min)
R, RMAX surface roughness and maximum allowed tS set-up time (min)
roughness (mm) tTC tool change time (min)
rE tool nose radius (mm) t0 time during which the tool do not cut (min)

Some multi-objective approaches have been reported 2. Cutting process model


in cutting parameters optimization (Lee and Tarng,
2000; Zuperl and Cus, 2003; Cus and Balic, 2003), but 2.1. Decision variables
mainly they use a priori techniques, where the decision
maker combines the different objectives into a scalar In the constructed optimization problem, three
cost function. This actually makes the multi-objective decision variables are considered: cutting speed v, feed
problem, single-objective prior to optimization (Van f, and cutting depth, a. These really are the cutting
Veldhuizen and Lamont, 2000). parameters of the process.
On the other hand, in the a posteriori techniques, the
decision maker is presented with a set of non-dominated 2.2. Objective functions
optimal candidate solutions and chooses from that set.
These solutions are optimal in the wide sense that no Two different, mutually conflicting objectives are
other solution in the search space are superior to them optimized in this model. The first objective is the
when all optimization objectives are simultaneously production rate, t, measured as the entire time required
considered (Abbass et al., 2001). They are also known as to carry out the process:
Pareto-optimal solutions.
Comparing citations by technique, in the last years, V tTC 
t ¼ tS þ 1þ þ t0 , (1)
evidences the popularity of a posteriori techniques (Van M T
Veldhuizen and Lamont, 2000). In dealing with multi- where tS, tTC and t0 are the set-up time, the tool change
objective optimization problems, classical optimization time, and the time during which the tool does not cut
methods (weighted sum methods, goal programming, respectively; V is the volume of the removed metal; T the
min–max methods, etc.) are not efficient, because they tool life, and M the material removal rate. The tool life
cannot find multiple solutions in a single run, thereby is related with the cutting parameters by the Taylor’s
requiring them to be applied as many times as the extended law
number of desired Pareto-optimal solutions.
On the contrary, studies on evolutionary algorithms T ¼ C T v a f b ag (2)
have shown that these methods can be efficiently used to
eliminate most of the above-mentioned difficulties of and the material removal rate can be computed by the
classical methods (Soodamani and Liu, 2000). In this expression
paper, a multi-objective optimization method, based on M ¼ 1000vfa. (3)
a posteriori techniques and using genetic algorithms, is
proposed to obtain the optimal parameters in turning The second objective is the used tool life, x,
processes. considered as the part of the whole tool life which is
ARTICLE IN PRESS
R. Quiza Sardiñas et al. / Engineering Applications of Artificial Intelligence 19 (2006) 127–133 129

consumed in the process constraints, some mechanisms based on the improve-


V ments proposed by Kurpati et al. (2002) were added,
x¼ 100%. (4) which are based on the following principles:
MT
Usually, for a particular cutting process, tS, tTC, t0  In a given population, feasible solutions are preferred
and V can be considered constant values, so that the over infeasible solutions, i.e. feasible solutions should
objectives are functions of T and M, and, consequently, have a better rank than the infeasible ones.
they depend upon the decision variables v, f and a.  The amount of infeasibility (or the extent of
constraint violation) is an important piece of
2.3. Constrains information and should not be ignored while hand-
ling constraints.
There are some constraints which affect the selection  The number of violated constraints is also an
of the optimal cutting conditions and will be taken into important piece of information and should be taken
account. In the first place, they are the allowed values into consideration while handling them.
for the cutting parameters, given by the tool maker, and
restricted by the bottom and top permissible limits: A relevant aspect of the Coello and Toscano’s micro-
aMIN papaMAX , (5a) GA is the use of a static population, which does not
change in all the processes. This static population
f MIN pf pf MAX , (5b) allows, in each new evolutionary period, the addition
of a set of ‘‘fresh’’ individuals, using few computational
vMIN pvpvMAX . (5c) resources.
Also there are some constraints related to the machine Fig. 1 shows the block diagram of the proposed genetic
features. The cutting force, FC, must not be greater than algorithm, which will be explained, in the next sections.
a certain maximum value, FCMAX, given by the
strength and stability of the machine and the cutting 3.2. Generation of the initial populations
tool. The cutting force is computed from empirical
expressions in the form: The first step in the proposed GA is the generation of
0 0 0 the individuals for the initial populations. In fact, two
F c ¼ C F v a f b ag , (6) populations must be created: static and dynamic. For
0 0 0
where CF, a , b and g are coefficients experimentally each individual in both populations, values for the
obtained. decision variables are randomly selected from the valid
Another machine-related constraint is the maximum range, taking into account the restrictions given by Eqs.
allowable value for cutting power, P, which must not (5a)–(5b):
surpass the machine motor power, PMOT (considering x1 ¼ rndðaÞ : aMIN papaMAX , (9a)
the friction losses in the transmission):
vF C PMOT Z x2 ¼ rndðf Þ : f MIN pf pf MAX , (9b)
P¼ 4
p . (7)
6  10 100 x3 ¼ rndðvÞ : vMIN pvpvMAX . (9c)
Finally, in the finishing operations, the obtained
The three values of each individual are encoded to
surface roughness, R, must be smaller than the specified
form a code string that represents itself. This code
value, RMAX, given by technological criteria, so that the
string, the so-called ‘‘chromosome’’, is composed by
following equation is satisfied:
binary elements (0 or 1), and has 96 characters (32 for
125f 2 each decision variable).
R¼ pRMAX , (8)
rE
3.3. Evaluation of the individuals
where rE is the tool nose radius.
For each individual in the dynamic population, the
objective functions (production rate and used tool life)
3. Optimization algorithm implementation are evaluated, using the corresponding values of cutting
parameters (decision variables).
3.1. General description Three constraints are also evaluated. In order to
facilitate handling, these constraints were scaled result-
The proposed genetic algorithm is based on the ing in the form
micro-GA, developed by Coello and Toscano (2001).
This is an elitist algorithm with a technique to maintain 100P
g1 ¼  1p0, (10a)
adequate population diversity. In order to handle ZPMOT
ARTICLE IN PRESS
130 R. Quiza Sardiñas et al. / Engineering Applications of Artificial Intelligence 19 (2006) 127–133

3.4. Selection, crossover and mutation

By selection, crossover and mutation it is possible to a


obtain new population from the current one. In this
approach, selection was carried out by tournament. To
create each individual, for the new population, two
couples of candidates to parents must be randomly
selected. In each couple, both candidates are compared,
taking into account the following rules:

 A feasible individual is always better than an


unfeasible one.
 In a couple of feasible individuals, one of them is
better if it dominates the other one. If no individual
dominates the other, both are equally good.
 In a couple of unfeasible individuals, the best is the
one which has the smaller unfeasibility index.

Crossover operator combines the code strings of the


two ‘‘victorious’’ candidates (one for each couple). In
the proposed approach, a two-point crossover is
implemented, because in this way it is less disruptive
than multi-point crossover, and it helps the diversity
preservation better than the single-point one.
Finally, by mutation it is possible to obtain some
random changes in the code string of the new
individuals. This is a technique that helps to introduce
new features in the population. Of course, there is no
guaranty that these new features would be advanta-
geous; therefore, the mutation likelihood should be kept
very low as the high value will destroy good individuals,
and degenerates the genetic algorithm into a random
search method (Cus and Balic, 2003). In the proposed
micro-GA, a mutation likelihood of 104 was selected.

3.5. Update of the elitist population


Fig. 1. Block diagram of the proposed genetic algorithm.

The maintenance of an elitist population (so-called


FC Paretian population) is a common technique to preserve
g2 ¼  1p0, (10b) the fittest individuals. At the end of each evolutionary
F MAX
period, the non-dominated individuals (Paretian solu-
R tions) are selected from the dynamic population and
g3 ¼  1p0. (10c) added to the elitist one.
RMAX
After the addition of new individuals, the elitist
For the unfeasible individuals, i.e. those that violate populations should be filtered in order to eliminate
at least one constraint, the unfeasibility index, repre- dominated and staked individuals.
senting the ‘‘amount’’ of unfeasibility, is computed. For
the ith individual, this index, Gi, can be computed as
P3   3.6. Creation of the new dynamic population
j¼1 max gj ðxi Þ; 0
Gi ¼ 1 PD P3  , (11) At the end of an evolutionary period, and to begin the
D k¼1 j¼1 max gj ðxk Þ; 0
following one, a new dynamic population is created, to
where xi represents the vector of decision variables for carry out the evolution process anew. This new dynamic
the ith individual xi ¼ (x1i, x2i, x3i); and D, the dynamic population is generated by the random selection from
population size. The unfeasibility index allows one to the whole set composed by the individuals of the static
compare two unfeasible individuals. population and the current dynamic population.
ARTICLE IN PRESS
R. Quiza Sardiñas et al. / Engineering Applications of Artificial Intelligence 19 (2006) 127–133 131

This kind of selection enables one to begin the new For the implementation of the genetic algorithm the
evolutionary period with an initial population contain- following parameters were established: static population
ing individuals near the Pareto front and ‘‘fresh’’ size: 500; dynamic population size: 10; maximum
individuals, avoiding the convergence to a limited sector Paretian population: 20; number of epochs: 25; number
of the Pareto front. of evolutionary periods: 100. The proposed algorithm
These evolutionary processes are repeated while the was implemented in C++.
elitists population form a uniform Pareto front. The In Table 1, the obtained Paretian points after the
process is stopped after a maximum number of evolutionary process are shown. As can be seen, the
evolutionary processes is achieved. limiting constraint was the cutting power.
In Fig. 2, the obtained non-dominated points were
plotted to form the Pareto front, which makes the
analysis and decision-making process easier.
4. An application sample
Analysing the above figure, some interesting aspects
may be noted. First, the Pareto frontier has a typical
On a CNC lathe, it will be machined a steel bar by
asymptotical profile, with two clearly defined zones:
means of a P20 carbide tool. The lathe has a 10 kW
motor and a transmission efficiency of 75%. A
1. From point 1 to 5: There is significant decrease in the
maximum cutting force of 5000 N is allowed by the
used tool life (over three times) and a corresponding
part-tool-machine system. Empirical models were ex-
increase in the total operation time. It should be
perimentally obtained for tool life and cutting force,
pointed out that this time increase (about 12%) is
resulting in
more moderate than used tool life decrease.
5:48  109 2. From point 5 to 14: Operation time increases by
T¼ , (12a) about 10% and the used tool life has a small decrease.
v3:46  f 0:696  a0:460
An additional information can be extracted from a
6:56  103  f 0:917  a1:10
FC ¼ . (12b) cost graph, made from the above Pareto graph. Two
v0:286
Selected cutting tools have the following limits in its 10
allowed cutting parameters: 1

aMIN ¼ 0:5 mm; aMAX ¼ 6:0 mm; (13a) 8


2
 [%]

6
f MIN ¼ 0:15 mm=rev; f MAX ¼ 0:55 mm=rev; (13b)
3
4
vMIN ¼ 250 m=min; vMAX ¼ 400 m=min: (13c) 4
5 6 7 8 9 10 11 12 1314
3
In the considered operation 219,912 mm will be 2
removed. Set-up time, tool change time and the time 0.8 0.9 1.0 1.1 1.2
 [min]
during which the tool does not cut have been assumed as
0.15, 0.20 and 0.05 min, respectively. Fig. 2. Pareto front.

Table 1
Paretian points obtained as outcomes from the optimization process

No. t (min) x (%) a (mm) f (mm/rev) v (m/min) M (mm3/min) T (min) F C ðNÞ P (kW)

1 0.85 9.22 1.6 0.55 400 346,520 6.9 1125 7.5


2 0.87 6.75 1.7 0.55 358 333,372 9.8 1257 7.5
3 0.91 3.90 1.9 0.55 294 311,495 18.1 1528 7.5
4 0.94 2.77 2.1 0.55 261 298,575 26.6 1727 7.5
5 0.95 2.46 2.1 0.55 250 294,281 30.3 1800 7.5
6 0.96 2.45 2.2 0.53 250 292,619 30.6 1800 7.5
7 0.98 2.39 2.5 0.45 250 284,831 32.3 1800 7.5
8 1.00 2.33 2.9 0.38 250 277,270 34.0 1800 7.5
9 1.02 2.28 3.3 0.32 250 270,135 35.7 1800 7.5
10 1.03 2.24 3.7 0.29 250 265,117 37.0 1800 7.5
11 1.06 2.18 4.3 0.24 250 256,782 39.4 1800 7.5
12 1.08 2.13 4.9 0.20 250 250,600 41.2 1800 7.5
13 1.10 2.08 5.5 0.18 250 244,608 43.2 1800 7.5
14 1.12 2.05 6.0 0.16 250 240,880 44.5 1800 7.5
ARTICLE IN PRESS
132 R. Quiza Sardiñas et al. / Engineering Applications of Artificial Intelligence 19 (2006) 127–133

1.80
Labour and Overhead Cost
1.60
Tool Cost

1.40

1.20

1.00
Cost ($)

0.80

0.60

0.40

0.20

0.00
1 2 3 4 5 6 7 8 9 10 11 12 13 14
Points

Fig. 3. Cost graph.

main costs must be computed: tool cost, ZT, which is significant element to select cutting parameters, ignoring
computed by tool life and operation cost. Consequently, in these
circumstances, cutting parameters corresponding to
Z T ¼ x  zT , (14) point 14 must be selected, warranting the lowest cutting
where zT is the cost for each tool edge; and the labour time, despite its elevated tool consumption and high
and overhead cost, ZL+O, which is computed by cost.
Z LþO ¼ ðzL þ zO Þ  t, (15)
where zL and zO are the specific labour and overhead 5. Conclusions
cost. For this sample, the cost for each tool edge has
been selected as 14.17$ and the labour and overhead As can be remarked in the exposed sample, a
specific cost, as 0.26 and 0.18$/min, respectively. posteriori multi-objective optimization offers greatest
In this graph (Fig. 3) tool cost and labour and amount of information in order to make a decision on
overhead cost, for each Paretian point, were plotted. selecting cutting parameters in turning. By means of
Obviously, from point 1 to 14, a continuous (but not Pareto frontier graphics, several different situations may
uniform) decrease in the tool cost is shown. On the be considered, facilitating the choice of right parameters
contrary, in this direction, an increase in the labour and for any condition.
overhead cost can be noted. Furthermore, in points 5 The proposed micro-GA has shown obtain several,
and 6 the lowest total cost values are achieved. uniformly distributed points, in order to arrange the
Studying the forehand graphics, several decisions may Pareto front, at a reasonably low computational cost.
be made, depending on the real conditions. For instance, Aspects like diversity maintenance and constraints
if the considered workshop is working in ordinary handling have been successfully sorted for the studied
conditions, the most convenient cutting parameters problem.
should correspond to points 5 or 6, which have a low Cost analysis can complement the Pareto front
use of tool life and an acceptable operation time. information, and it helps the decision-making process.
Moreover, in these points the lowest costs are reached. The proposed model must be enlarged to include more
On the contrary, if the workshop has small and constraints, such as cutting surface temperature.
discontinuous productions, that do not cover all the
time, then there is no necessity to use low operation
times. In this case, to obtain a large tool life may be References
more important, whence, cutting parameters matching
Abbass, H.A., Sarker, R., Newton, Ch., 2001. PDE: A Pareto–frontier
to point 1 must be selected. differential evolution approach for multi objective optimization
Finally, in special conditions, such as wartime or problems. Proceedings of the Congress on Evolutionary Computa-
other emergencies, the production volume is the most tion, Vol. 2, Piscataway NJ, pp. 971–978.
ARTICLE IN PRESS
R. Quiza Sardiñas et al. / Engineering Applications of Artificial Intelligence 19 (2006) 127–133 133

Amiolemhen, E., Ibhadode, A.O.A., 2004. Application of genetic Lee, B.Y., Tarng, Y.S., 2000. Cutting-parameter selection for
algorithms—determination of the optimal machining parameters in maximizing production rate or minimizing production cost in
the conversion of a cylindrical bar stock into a continuous finished multistage turning operations. Journal of Materials Processing
profile. International Journal of Machine Tools and Manufacture Technology 105 (1–2), 61–66.
44 (12–13), 1403–1412. Liang, M., Mgwatu, M., Zuo, M., 2001. Integration of cutting
Chien, W.T., Tsai, C.S., 2003. The investigation on the prediction of parameter selection and tool adjustment decisions for multipass
tool wear and the determination of optimum cutting conditions in turning. International Journal of Advanced Manufacturing Tech-
machining 17-4PH stainless steel. Journal of Material Processing nology 17, 861–869.
Technology 140 (1–3), 340–345. Molinari, A., Nouari, M., 2002. Modeling of tool wear by diffusion in
Chua, M.S., Loh, H.T., Wong, Y.S., Rahman, M., 1991. Optimization metal cutting. Wear 525, 135–149.
of cutting conditions for multi-pass turning operations using Saravanan, R., Asokan, P., Vijayakumar, K., 2003. Machining
sequential quadratic programming. Journal of Materials Proces- parameters optimisation for turning cylindrical stock into a
sing Technology 28 (1–2), 253–262. continuous finished profile using genetic algorithm (GA) and
Coello, C.A., Toscano, G., 2001. A micro-genetic algorithm for simulated annealing (SA). International Journal of Advanced
multiobjective optimization. In: Zitzler, E. (Ed.), First International Manufacturing Technology 21, 1–9.
Conference on Evolutionary Multi-Criterion Optimization, Lecture Sönmez, A.I., Baykasoglu, A., Dereli, T., Filiz, I.H., 1999. Dynamic
Notes in Computer Science. Springer, New York, pp. 126–140. optimization of multipass milling operation via geometric pro-
Cus, F., Balic, J., 2003. Optimization of cutting process by GA gramming. International Journal of Machine Tools & Manufactur-
approach. Robotics and Computer Integrated Manufacturing 19, ing 39, 297–320.
113–121. Soodamani, R., Liu, Z.Q., 2000. GA-based learning for a model-based
Dereli, D., Filiz, I.H., Bayakosoglu, A., 2001. Optimizing cutting object recognition system. International Journal of Approximate
parameters in process planning of prismatic parts by using genetic Reasoning 23, 85–109.
algorithms. International Journal of Production Research 39 (15), Taylor, F.W., 1907. On the art of cutting metals. Transactions of the
3303–3328. ASME 28, 310–350.
Ko, T.J., Kim, H.S., 1998. Autonomous cutting parameter regulation Van Veldhuizen, D.A., Lamont, G.B., 2000. Multiobjective evolu-
using adaptive modeling and genetic algorithms. Precision En- tionary algorithms: analizing the state-of-the-art. Evolutionary
gineering 22 (4), 243–251. Computation 8, 125–147.
Kumar, R., Kumar, V., 2000. Optimum selection of machining Wang, J., Kuriyagawa, T., Wei, X.P., Gou, G.M., 2002. Optimization
conditions in abrasive flow machining using neural networks. of cutting conditions using a deterministic approach. International
Journal of Material Processing Technology 108, 62–67. Journal of Machine Tools & Manufacture 42, 1023–1033.
Kurpati, A., Azarm, S., Wu, J., 2002. Constraint handling improve- Zuperl, U., Cus, F., 2003. Optimization of cutting conditions during
ment for multiobjective genetic algorithms. Structural and Multi- cutting by using neural networks. Robotics and Computer
disciplinary Optimization 23, 204–213. Integrated Manufacturing 19, 189–199.

You might also like