Multi objective Optimization of End milling parameters using Firefly and Particle swarm

algorithm.

Abstract

Al/SiC metal matrix composites are non homogeneous and hence while end milling of these
composites, prediction of vibration amplitude is challenging. Proper selection of input parameter
may suppress vibration amplitude which in turn increases tool life and surface finish. In this
study main objective is to optimize machining and geometrical parameters for minimum
vibration using recently developed firefly algorithm and compare the results by particle swarm
algorithm. End milling operations are carried out on Al/SiC metal matrix composites using L27
orthogonal design and vibration at two positions (spindle and work piece holder) are measured
using twin channel piezoelectric accelerometer. Two polynomial regression equations
established between six input and two output variables. Minimizing these two values are two
objectives of optimization algorithms used in this study namely firefly and PSO and constraints
are minimum and maximum values of each controllable input parameters. Confirmation
experiments are conducted using optimized values exhibit that both the methods are suitable for
machining parameter optimization for minimum vibration in spindle as well as work piece
holder.

Key-words: Optimization-firefly-Particle swarm- End milling –Metal matrix composite-vibration

1. Introduction

End milling operation is versatile and efficient metal cutting operation. Hence it is widely used in
aerospace and automobile industries for producing any complex geometrical shapes. In end
milling process the interaction between cutter and work piece is continuously changing, and
hence subjected to varying impact load and cutting force. This makes milling process prone to
vibration and chatter. Chatter is self exited vibration which derives energy to start and grow from
the interaction between cutting tool and work piece. This type of vibration makes the cutting
system instable and it is undesirable. Chatter leads to poor surface quality, uneven tool wear and
reduced material removal rate. To enhance performance of machining process, controlling
vibration is one of the important factors.
Now a days aluminum silicon carbide metal matrix composites are replacing aluminum alloy in
aircraft and automobile industries due to its superior mechanical properties. Metal matrix
composites are non homogeneous in nature due to the presence of reinforcement. When the
cutting tool hit these small perturbations the relative vibration between cutting tool and work
piece increased and it leads to regenerative chatter. Regenerative chatter is the most common
form of chatter occurs while machining metal matrix composites. Proper selection of input
parameters may reduce chatter vibration. Researchers used design of experiments, ANOVA, grey
analysis etc to analyze, predict optimal input parameters for minimum vibration during end
milling operation. (1-5).In recent years for machining parameter optimization several
evolutionary algorithms were used to optimize the input parameters in order to improve the
quality and efficiency of machining.

Several investigations were carried out to optimize machining parameters using different
evolutionary algorithms on different machines with single or multi objectives such as minimize
roughness, tool wear, cutting force , vibration ,machining time etc or maximize metal removal
rate ,cost etc. The Literature survey related to the work of other researchers is given below.
Z.G.Wang (6) optimized milling parameters such as feed rate and speed using GA and hybrid of
GA and SA (GSA) algorithm with the objective of minimizing production time of a component
and concluded that GSA gave better results compared with GA and GP.Saravanan (7) conduct
experiments on turning machine to produce finished product from cylindrical block with cutting
speed, feed and depth of cut as input variables to be optimized. Objective function used is to
minimize production cost and constraints are cutting force, power and tool tip temperature. The
author used six different optimization techniques such as GA,SA, AC, TS,MA and PSO and
conclude that the performance of PSO good compared with other methods.Yanming (8) conduct
experiments on milling machine to optimize machining parameters with the objective of
minimizing machining time and maximizing production rate using modified genetic algorithm
(MGA) and proved that the convergence rate is faster than that of GA.Uros (9) developed cutting
force model using an adaptive neuro-fuzzy inference system based on design of experiments and
then using TLBO algorithm and optimized cutting parameters with an objective of maximizing
cutting force surface area in milling with different constraints and verified the effectiveness of
this algorithm. Palanisamy (10) employed artificial neural network technique to predict the
chatter vibration while milling AISI 1020 steel. The occurrence of chatter vibration for a
particular combination of machining parameter was predicted using a stability lobe diagram and
dynamic stability were analyzed using Nyquist criterion. Palanisamy (11) conducted end milling
operations on mild steel using HSS cutter with the objective of minimizing machining time with
tool life, surface roughness, vibration, cutting force as constraints to optimize the machining
parameter. The author used GA and concludes that the optimized parameters obtained improved
surface finish and efficiency of machining. Babur (12) conduct experiments on Inconel718 to
optimize end milling parameters for minimum surface roughness by integrating neural network
and GA and concluded that GA when coupled with neural network give better results. Recently
firefly algorithm is used by scientists for optimizing complex non linear problem.Bharathi(13)
optimized electrical discharge machining parameters such as current and pulse on time on
hardened die steel using firefly algorithm and studied the effect on surface roughness and
machining time and conclude that FA is more suitable for machining parameter
optimization.Rajkamal (14) applied FA to improve performance of electrical discharge
machining(EDM) and abrasive water jet machining(AWJM).The author conclude that in EDM
metal removal rate , finish increased and AWJM kerf and roughness decreased by using FA
compared with other methods. Dilip kumar (15) applied FA to design shell and tube heat
exchanger with the objective of reducing total cost and compared the results with other non
traditional methods such as GA, PSO, ABC, BBO and CSA and proved that present method is
the most effective method of design optimization of shell and tube heat exchanger. Basker (16)
optimize the machining parameter in milling operation with the objective of maximizing the
profit using different non conventional methods such as GA, CACO, TS and PSO and suggest to
use PSO for optimizing milling parameters.Norfadzlan (17) reviewed the application of PSO for
machining parameter optimization from 2007-2011 and conclude that most machining processes
considered in PSO are Multipass turning followed by end milling. The author also concludes that
the most machining performance measured was different costs followed by surface roughness
and machining time. From the above literature it was concluded that most of the researchers used
GA and PSO to optimize machining parameters of conventional and non conventional machines.
Most of the research work was conducted to improve surface quality, machining time and cost.
Researchers mostly used conventional optimization methods to minimize vibration in end
milling operation. End milling is commonly used machining process and the end mill cuter
geometry is also complex in nature. While end milling non homogeneous materials such as metal
matrix composite the chatter vibration is unpredictable and difficult to control. Some efficient
technique should be implemented to optimize both machining and geometrical parameters to
reduce vibration in both axial and feed direction.

In this study machining and geometrical parameters of end milling operation are optimized
with multi objective of reducing acceleration amplitude in both feed and axial direction. In this
Al/SiC metal matrix composite is prepared using stir casting route. Experiments are conducted
0n prepared composite material using CNC vertical machining centre with high speed end mill
cutter. L 27 Tauguchi orthogonal design is used to conduct experiments with. Helix angle (α),
Nose radius (R), Rake angle (γ), Cutting speed (N), Feed (F), Depth of cut (Y) as input
independent variable and acceleration amplitude measured at two positions (Spindle and work
piece holder) using piezoelectric accelerometer as output performance measure. Two polynomial
regression equations are generated using Minitab software relating input and output variables.
Optimization is done by using two most powerful nontraditional optimization techniques FA and
PSO.The multiple objectives for both the optimization methods are the generated two
polynomial equations. Constraints are selected minimum and maximum values of machining and
geometrical parameters.

2. Experimental Procedure

2.1. Wok piece and tool material.

Work piece material used for experimental work is Al 356+5% SiC metal matrix composite. The
required work piece material is prepared by most commonly used stir casting method as in this
method the distribution of SiC particles are more uniform compared with other methods. The
Prepared composite material is sized into 50mm x 50mm x 50mm cube. Tool material selected
for experimental work is 12mm diameter high speed steel end mill cutter with four flutes.

2.2. Experimental design.

27 experiments are conducted on CNC vertical machining center with six input parameters at
three levels based on Taghuchi orthogonal design. The range and the level of machining
parameters are selected based on literature as well as by conducting pilot experiments. The six
input independent variables selected in this problem are helix angle, nose radius, and rake angle,
cutting speed, feed and depth of cut. The two performance measures, acceleration amplitude in
spindle (channel-I ) and in work piece holder (channel-II) are measured for all 27 experiments.
The level and the range of input parameters are shown in table1.

Table 1. Parameters and its levels

levels
Units
Parameter 1 2 3
Helix angle (α) 0
40 45 50
Nose radius (R) mm 0.4 0.8 1.2
Rake angle (γ) 0
8 12 16
Cutting speed (N) m/min 30 60 90
Feed rate (F) mm/rev 0.03 0.04 0.05
Depth of cut (Y) mm 0.5 1 1.5

2.3. Measurement of vibration

End milling operation is subjected to two types of chatter primary and regenerative chatter.
Former ocuurs due to friction between tool and the work piece which is easily controlled. The
regenerative chatter is the common type of self exited vibration which is uncontrollable. In this
the cutting tool vibration generates wavy surface which is regenerated on the work piece. This
regenerative chatter is particularly more in the case of machining non homogeneous work piece
materials for example composites.

In this work vibration in terms of acceleration amplitude is measured at two places using twin
channel piezoelectric accelerometer. One terminal is connected to spindle holder (channel-I)
another terminal is connected to work piece holder (channel-II).The data acquisition system used
is FFT analyzer which uses lab view software. From the captured images peak values are noted
down and are tabulated in table2 .steps involved in experimental procedure is explained in fig1.
 (a) (b)
Figure 1. Experimental set up – (a) accelerometer (b) Data acquisition system to acquire
vibration signals
Output performance
Input process and Geometrical parameters measures

Nose Cutting Acceleration Acceleration

Exp.No
amplitude- amplitude-
radius speed Feed
Helix Rake Depth of Channel-I Channel-II
angle(0) (mm) angle(0 ) (m/min) (mm/rev) cut (mm) (m/sec2) (m/sec2)

1 40 0.4 8 30 0.03 0.5 4.0666 0.29166

2 40 0.4 8 30 0.04 1 3.7066 0.41166

3 40 0.4 8 30 0.05 1.5 3.9900 4.8166

4 40 0.8 12 60 0.03 0.5 9.1333 5.0333

5 40 0.8 12 60 0.04 1 9.4000 0.71666

6 40 0.8 12 60 0.05 1.5 10.0000 1.9666

7 40 1.2 16 90 0.03 0.5 11.3533 0.4566

8 40 1.2 16 90 0.04 1 13.3500 0.71333

9 40 1.2 16 90 0.05 1.5 16.5000 2.0333

10 45 0.4 12 90 0.03 1 11.3330 0.54

 11 45 0.4 12 90 0.04 1.5 11.7500 0.45833

12 45 0.4 12 90 0.05 0.5 10.4666 0.49833

13 45 0.8 16 30 0.03 1 5.3833 0.31666

14 45 0.8 16 30 0.04 1.5 5.0666 0.3266

15 45 0.8 16 30 0.05 0.5 9.4500 0.25166

16 45 1.2 8 60 0.03 1 10.1500 0.39266

17 45 1.2 8 60 0.04 1.5 10.8666 0.5786

18 45 1.2 8 60 0.05 0.5 10.3500 0.45833

19 50 0.4 16 60 0.03 1.5 12.0330 0.54366

20 50 0.4 16 60 0.04 0.5 10.5500 0.3566

21 50 0.4 16 60 0.05 1 11.6660 0.485

22 50 0.8 8 90 0.03 1.5 22.3000 14.033

23 50 0.8 8 90 0.04 0.5 11.3000 0.5433

24 50 0.8 8 90 0.05 1 21.5000 4.4833

25 50 1.2 12 30 0.03 1.5 5.8333 0.3766

26 50 1.2 12 30 0.04 0.5 4.2066 0.2

27 50 1.2 12 30 0.05 1 6.1500 0.37

2.4. Formulation of Regression equation

Second order polynomial regression equation relating six input variables such as helix angle,
nose radius, rake angle, cutting speed, feed and depth of cut. and two output performance
measures as vibration at channel-I and channel-II are generated using MINITAB software.
Minimize the value of these two equations are multi objectives in the present problem. These two
equations are shown in equation (1) and (2)

Minimize acceleration amplitude Channel-I

Ay = 3.7664- 4.3414 α + 7.8747R- 4.8883 γ + 3.0420N -

10.1687F + 8.2221Y+ 0.9670 α2 -1.1525 R2 + 2.0576 γ2 -

0.5884N2+0.9883F2-0.6333Y2-0.5930αF+1.4974αY+1.9301RF-

1.3070RY+1.4471γ F-2.6980 γY+1.9337NF (1)

Minimize acceleration amplitude Channel-I

A x = 5.8242- 5.7263 α +9.4228 R-2.8026 γ -2.0589 N-

4.7793 F+2.3363 Y+1.6772 α2 -1.2978 R2+ 0.6207 γ2+0.5588 N2+

1.2166 F2+0.6025 Y2- 0.6133 αF +0.7597 αY + 0.7179 RF -

0.9121 RY +0.9904 γ F -1.9007 γY + 0.3674 NF (2)

2.5 Constraints

The maximum and minimum levels of six input independent variables are taken as constraints
for the optimization problem. Helix angle (α), Nose radius (R), Rake angle (γ), Cutting speed
(N), Feed (F), Depth of cut (Y).The above end milling parameters are optimized with following
six constraints.

αmin ≤ α ≤ αmax (3)

Rmin ≤ R ≤ Rmax (4)

γ min ≤ γ ≤ γmax (5)

Nmin ≤ N ≤ Nmax (6)

Fmin ≤ F ≤ Fmax (7)

Ymin ≤ Y≤ Ymax (8)

3. Methodology

3.1. Firefly algorithm

The Bio inspired optimization techniques have been successfully employed for the optimization
of different science and engineering problems in the real life as well as in industries due to its
robustness and ability to handle complex problems in an efficient manner. Compared with other
conventional algorithms, metaheristic algorithms give good solutions with less computations and
easy to implement for verity of problems by suitably selecting random numbers.

Fire fly algorithm (FA) is recent swarm intelligence based algorithm developed by Yang (18) in
2008 and is meta heuristic nature-inspired algorithm. Fireflies are one of the families of insects
which are nature’s gift whose behavior differs from other creature. Fireflies emit cold light
flashes with unique pattern from their bodies. This process of producing light from living things
due to chemical reaction is called bioluminescence. Two functions of such flashes are to
communicate with matting partners and to attract potential prey. Fireflies are unisex so that one
firefly will be attracted by other fireflies regardless of their sex.Becasuse of absorption of light
by air, the intensity of light decreases as distance increases. This phenomenon of light intensity
with respect to distance is associated with the objective function to be moved towards better
solution and finally the best solution. This algorithm has been applied in almost all areas of
optimization including continuous, combinational and multi objective optimization in any noisy
environments. In recent years FA is used by the researchers in various fields of engineering such
as industrial optimization,robotics,image processing etc.A comprehensive review of firefly
algorithm by Fister (19) concluded that FA is simple, flexible and versatile and can solve wide
range of real world problems in an efficient way and the convergence rate is very fast compared
with other metaheuristic algorithms such as PSO and GA.In this present study FA is effectively
utilized to optimize machining and geometrical parameters for multiple objectives as, to reduce
vibration amplitude in axial and feed direction while end milling of prepared MMC..Steps
involved in the firefly algorithm are summarized as follows.

1. Firefly position is initialized randomly within the limits of variables of the problem.

2. Controlling factors of the FA algorithm such as randomness factor (α), Randomness reduction
factor (β), absorption coefficient (ᵧ) are defined.

3. Objective function with constrained are defined.

4. Calculating intensity of light (objective function) for all fireflies and select best firefly which
is having high intensity of light and update the position of fireflies.

5. Sort out and rank the fireflies based on intensity.

6. If the intensity of firefly for current iteration is better than previous result, replace it otherwise
keep the previous solution.

7. Update the result till the iterations reach the maximum limit of generation.

8. The intensity of firefly obtained at the end of last iteration is the optimum best solution for the
given problem.

Application of Firefly algorithm for the optimization of end milling parameters. The following
control parameters are selected based on trial run of the algorithm.

Number of Itrations(i) 100

Nmber of fireflies (j) 50

Randomness factor (α) 2

Absorption coefficient (γ) 0.01

Initial attractiveness(β0) 0.1

Initialize the input parameters randomly within the limits are done by the following equations

α = αmin +( αmax - αmin ) rand( ) (9)

R = Rmin +( Rmax - Rmin ) rand( ) (10)

γ = γmin +( γ max - γ min ) rand( ) (11)

N = Nmin +( Nmax - Nmin ) rand( ) (12)

F =Fmin +( Fmax - Fmin ) rand( ) (13)

Y = Ymin +( Ymax - Ymin ) rand( ) (14)

 th
Distance between any two fireflies i and j with positions Xi and Xj of the k component can be
defined as Cartesian distance rij given by the following equation

6
rij = 2
 X
k 1
ik  Xjk  (15)

The attractiveness of firefly  depends on intensity of light (objective function) which decreases as

distance increases. The function  is calculated using the following equation

 =  0e-γr2 (16)

Where  0 is the attractiveness at distance r=0 and is assumed as 1 and γ is absorption coefficient and is

taken as 0 .01.

The relationship for movement of firefly from present position to new position for helix angle is
e calculated using the following equation

αnp=(αcp +(αb –αcp ) x  ++(αb –αcp ) x α (rand-0.5) (17)

Similarly new position for other input variables such as R, γ, N, F and Y are calculated. These
values are substituted in the objective function equation (1) and (2) to get the minimum values of
vibration in channel I and channel II respectively.

PSO algorithm

Particle swarm optimization (PSO) is swarm intelligence based nondeterministic optimization

technique. It was developed by James Kennedy and Russell Eberhart in the year 1995 by
observing swarming habits of birds or fishes .It uses evolutionary computation technique similar
to genetic algorithm but there is no evolution terms such as cross over and mutation. In PSO it is
not necessary to encode the parameters hence it is most suitable if more number of parameters
are to be optimized.

PSO algorithm is population based stochastic algorithm. The concept behind the PSO as follows
There are group of birds randomly searching for their foods. Food is available at only one place
only. They have no idea where the food is available but they got idea how for it is from their
place. So the strategy they are adopting is simply following the bird which is nearer to the food.
In optimization each bird is represented by a particle in space and is evaluated by objective
function. Each particle moves with velocity and follows the current optimum particle The
optimization procedure is initialized with a population of random solution and updating the
generation till reaches optimum solution. It can able to maintain multiple potential solutions at a
time. Each potential solution is represented by a particle in space. In all the iterations each
particle is updated by two best values. Among the particles the best solution arrived so far is
called particle best pbest. A best solution arrived among the population so far is called global
best gbest.After arriving best values the position and velocities are updated. The steps, evaluate
fitness of each particle, update best values, update position and velocity are repeated until
termination condition is reached. The steps involved in PSO are summarized as follows.

1. Initialize the population of N particles with random position and velocity vector.

2. Calculate the fitness value of each particle, if the fitness value is better than pbest, set the
current fitness value as new pbest.

2. Choose the particle with best pbest value of all particles as gbest.

3. For each particle calculate velocity and position.

4.Particle velocity on each dimension is fixed to a maximum velocity V max .If the sum of
acceleration would cause the velocity on that dimension to exceed V max the velocity on that
dimension is limited to V max .

5. Terminate if maximum number of iterations reached, otherwise go to step 2.

Application of PSO for the optimization of end milling parameters. The following control
parameters are selected based on trial run of the algorithm.

Number of particles (N) =50

Number of iterations performed (i) 100

Cognitive coefficient (C1) 2

Social coefficient (C2) 2

Inertia coefficient (w) min- 0 max- 1

Initialize the input parameters randomly within the limits are done by the equations 9-14.

The above values are substituted in equation (1) and (2) to find the fitness value.

Calculation of pbest value

Minimum value of acceleration amplitude obtained in channel I and Channel II for current
iteration is considered as the pbest value for that particular solution only.

Calculation of gbest value

Minimum value of acceleration amplitude obtained in channel I and Channel II for entire
iteration executed is considered as the gbest value.

Velocity (V) and position (p) of helix angle (α) is updated using the following equation

V(α) =( w) present(α)+ C1 rand( )(pbest(α) -present (α))+ C2 rand( ) (gbest( α) -present (α) (18)

p(α) = V(α) +present (α). (19)

Similarly velocity and position for other input parameters such as nose radius(R), rake angle (γ),
cutting speed (N), feed (F) and depth of cut(Y) are updated. These values are substituted in the
objective function equation (1) and (2) to get the minimum values of vibration in channel I and
channel II respectively.

Result and discussion

Firefly optimization results

End milling operations are carried out on CNC vertical machining centre on Al/SiC metal matrix
composites using high speed steel end mill cutter. L27 Taghuchi orthogonal design is used to
conduct experiments with six input variables and vibration at channel-I is measured as output
performance measure. In this study recently developed firefly algorithm is used to optimize input
variables for minimum vibration. Polynomial regression equation, equation (1) is derived
between six input controllable variable and output performance measure (vibration at channel-I)
Minimize the value of equation (1) with constraints given in equation (3-8) is the objective for
this optimization.MATLAB software is used to do the optimization. The firefly convergence
graph for vibration in channel-I is shown in Figure2. From Figure 2 it is observed that it
converged at 36th iteration and the corresponding minimum value of acceleration is 2.5630
m/sec2.The corresponding optimum input parameters are helix angle (α) 46.030 , nose radius (R)
0.41mm, rake angle (γ) 8.20, cutting speed (N) 32m/min, feed (F) 0.0488mm/rev, and depth of
cut (Y) 0.512mm.

Polynomial regression equation, equation (2) is derived between six input controllable variable
and output performance measure (vibration at channel-II) Minimize the value of equation (2)
with constraints given in equation (3-8) is the objective for this optimization. MATLAB software
is used to to do the optimization. The firefly convergence graph for vibration in channel-II is
st
shown in Figure3. From Figure 3 it is observed that it is converged at 41 iteration and the
corresponding minimum value of acceleration is 0.0785m/sec2.The corresponding optimum input
parameters are helix angle (α) 41.10 , nose radius (R) 0.412mm, rake angle (γ)15.90, cutting
speed (N) 45.4m/min, feed (F) 0.0312mm/rev, and depth of cut (Y) 1.42mm.

Figure2.Firefly convergence graph for vibration at Channel-I

 Figure3.Firefly convergence graph for vibration at Channel-II

PSO optimization results

The same problem optimized by firefly algorithm is optimized by PSO algorithm in order to
compare the result and performance of both the algorithms. Minimize the value of equation (1)
with constraints given in equation (3-8) is the objective for this optimization also.MATLAB
software is used to do the optimization. The PSO convergence graph for vibration in channel-I is
shown in Figure4. From Figure 4 it is observed that it converged at 18th iteration and the
corresponding minimum value of acceleration is 2.5052m/sec2.The corresponding optimum input
parameters are helix angle (α) 470 , nose radius (R) 0.434mm, rake angle (γ) 8.40, cutting speed
(N) 33m/min, feed (F) 0.049mm/rev, and depth of cut (Y) 0.524mm.

Minimize the value of equation (2) with constraints given in equation (3-8) is the objective for
this optimization also. MATLAB software is used to to do the optimization. The PSO
convergence graph for vibration in channel-II is shown in Figure5. From Figure5. it is observed
nd
that it is converged at 26 iteration and the corresponding minimum value of acceleration is
0.0788m/sec2.The corresponding optimum input parameters are helix angle (α) 41.540, nose
radius (R), 0.415mm, rake angle (γ) 15.860, cutting speed (N) 43.68m/min, feed (F)
0.0342mm/rev, and depth of cut (Y) 1.42mm.
 Figure4.PSO convergence graph for vibration at Channel-I

Figure5.PSO convergence graph for vibration at Channel-II

Comparison of optimization results of FA and PSO

In our study the work piece material used is composite which is non homogeneous. While end
milling such type of materials the regenerative chatter vibration is continuous and non linear. To
optimize such type of problems the performance and convergence of PSO is better compared
with FA. The reasons might be PSO generating completely different random number for entire
iterations compared with FA. The wrong selection of randomness factor in FA also (α) diverts
the problem and trapped into several local optima. FA does not memorize the previous better
solution and so each firefly move regardless of previous better situation randomly. Therefore
convergence is slow compared with PSO. This is proved in our study by comparing the figure
2,3,4 and 5
Conclusion

Following conclusions are arrived from the above multi objective optimization study carried out
using bio inspired optimization methods.

 Both firefly and PSO are suitable for multi objective optimization of milling parameters
 Convergence rate is better for PSO to minimize vibration at channel –I and channel-II
compared with firefly algorithm. It indicates PSO is most suitable for optimizing milling
parameters effectively for complex problems with noisy atmosphere.
 In FA by carefully selecting randomness factor α better results can be obtained.
 Minimum value of vibration obtained by both the optimization methods is almost similar.
 From the above results obtained it is concluded that both the methods are suitable for
multi objective optimization of milling parameters for reduced vibration. By means of
optimization vibration is reduced considerably compared with experimental results.