1

Tuning of Optimum PID Controller Parameter Using Particle

Swarm Optimization Algorithm Approach

Wan Azhar Wan Yusoff

Nafrizuan Mat Yahya
Azlyna Senawi
Fakulti Kejuruteraan Mekanikal,
Universiti Malaysia Pahang,
Karung Berkunci 12, 25000 Kuantan, Pahang.

Abstract: In this paper, an artificial intelligence method, optimal gains of the PID such as by Cohen and Coon in
Particle Swarm Optimization (PSO) algorithm is presented for 1953, Åström and Hägglund in 1984 or by Zhuang and
determining the optimal proportional-integral-derivative (PID) Atherton in 1993 [1].
controller parameters of a typical servo motion system. This To obtain the optimal parameter tuning, it is highly
paper demonstrates in detail on how to employ the PSO method desirable to increase the capabilities of PID controllers by
to search efficiently the optimal PID controller parameters of a
typical servo motion system. In order to assist estimating the
adding new features. Most in common, artificial
performance of the proposed PSO-PID controller, a new time- intelligence (AI) techniques have been employed to
domain performance criterion function has been used. The improve the controller performances for a wide range of
proposed approach yields better solution in term of rise time, plants while retaining the basic characteristics. AI
settling time, maximum overshoot and steady state error techniques such as artificial neural network, fuzzy system
condition of the system. Compared to conventional Ziegler – and neural-fuzzy logic have been widely applied in order
Nichols method, the proposed method was indeed more efficient to get proper tuning of PID controller parameters [6-10].
and robust in improving the step response of a typical servo Recently, a new evolutionary technique, Particle
motion system. Swarm Optimization (PSO) was first introduced in 1995
by Kennedy and Eberhart for unconstrained continuous
optimization problems [11-12]. Its development was
INTRODUCTION based on observations of the social behavior of animals
such as bird flocking, fish schooling and swarm theory.
Even in a decade where advanced control algorithms The PSO is initialized with a population of random
mostly based on some kind of optimization procedure solutions. The PSO has some attractive characteristics
have achieved a high degree of maturity, Proportional where it has memory and therefore, knowledge of good
Integral Derivative (PID) controllers are still widely used solutions is retained by all particles. There exist
in industrial applications even though many new control constructive cooperation between particles where
techniques have been proposed [1-2]. The reason is that it particles in the swarm share information between them.
has a simple structure which is easy to be understood by The theoretical framework of PSO is very simple and
the engineers, and under practical conditions, it has been PSO is easy to be coded and implemented using computer
performing more reliably compared to more advanced and [11]. In fact, the PSO technique can generate a high
complex controllers [3-4]. The main propose of designing quality solution within shorter calculation time and stable
a PID controller is to determine the three gains and they convergence characteristics than other stochastic methods
are proportional gain (kp), integral gain (ki) and derivative [13]. Thus, this technique has gained much attention and
gain (kd) of the controller [2]. However, the three wide applications in various fields recently [14-20].
adjustable PID controller parameters should be tuned The rest of the paper is organized as follows. In topic
appropriately [1]. follows, a brief discussion about PID controller, basic PID
Over the years, several heuristic methods have been servo motion system and performance estimation of PID
developed for the tuning of PID controllers. The first controller are presented. Next, the PSO method and its
method used the classical tuning rules proposed by implementation into the PSO-PID controller are viewed in
Ziegler and Nichols [5]. Generally, it is always hard to detail. Further, the simulation results are presented in
determine optimal or almost optimal PID parameters with table form and discussed. Finally, the discussion of the
the Ziegler-Nichols method in many industrial plants [5]. results followed by conclusion of the research is provided.
Other than original works done by Ziegler and Nichols, a
great number of methods have been proposed to obtain
 2
PID CONTROLLER [5, 21] therefore the following approximation as shown in (5) has
been made.
The PID controller is used to improve the dynamic
response as well as to reduce or eliminate the steady-state G (s )≈ 1 (5)
error. The derivative controller adds a finite zero to the
open-loop plant transfer function and improves the The servomotor is modeled as a lump inertia, J, a
transient response. The integral controller adds a pole at viscous damping term, b, and a torque constant, Kτ. The
the origin, thus increasing system type by one and lump inertia term is comprised of both the servomotor and
reducing the steady state-error due to a step function to load inertia. It is also assumed that the load is rigidly
zero. coupled such that the torsional rigidity moves the natural
The continuous form of a PID controller, with input mechanical resonance point well out beyond the servo
e(⋅) and output u pid (⋅) , is generally given as : controller’s bandwidth. This assumption allows us to
model the total system inertia as the sum of the motor and
⎡ 1 d ⎤ load inertia for the frequencies that will be control.
u pid (t ) = k p ⎢ e (t ) + ∫ e (τ ) d τ e (t )⎥
t

Ti t
+ Td
dt
(1) The actual motor position, θ(s) is usually measured
⎣ ⎦
either by an encoder or resolver coupled directly to the
where kp is the proportional gain, Ti is integral time motor shaft. Again, the underlying assumption is that the
constant and Td is the derivative time constant. We can feedback device is rigidly mounted such that its
also rewrite as mechanical resonant frequencies can be safely ignored.
d External shaft torque disturbances, Td are added to the
u pid (t ) = k p e (t ) + k i ∫ e (τ ) dτ e (t )
t

0
+ kd
dt
(2) torque generated by the motor’s current to give the torque
available to accelerate the total inertia, J.
where ki = kp / Ti is the integral gain and kd = kpTd is the Around the servo drive and motor block is the servo
derivative gain. controller that closes the position loop. A basic servo
In simple form, the PID controller transfer function is controller generally contains both a trajectory generator
ki
and PID controller. The trajectory generator typically
C (s ) = k p + + k d s (3) provides only position set-point commands labeled in Fig.
s
1 as θ*(s). The PID controller operates on the position
error and outputs a torque command that is sometimes
scaled by an estimate of the motor’s torque constant, K̂ t .
BASIC PID SERVO MOTION SYSTEM [22] If the motor’s torque constant is not known, the PID gains
are simply re-scaled accordingly. Due to the exact value
The basic components of a typical servo motion of the motor’s torque constant is generally not known, the
system are depicted in Fig. 1. According to this figure, symbol “^” is used to indicate it is an estimated value in
the servo drive closed a current loop and is modeled controller. In general, equation (6) holds with sufficient
simply as a linear transfer function G(s). In their most accuracy so that the output of the servo controller (usually
basic form, servo drives receive a voltage command that +/- 10 Volts) will command the correct amount of current
represents a desired motor current. Motor shaft torque, T for a desired torque.
is related to motor current, I by the torque constant, Kτ as
show in (4). Kˆ t ≈ K t (6)

T ≈ Kτ I (4) There are three gains to adjust in the PID kp, ki and kd.
These gains all act on the position error defined as (7).
The output of the PID controller is a torque signal.

error (t) = θ*(t) - θ(t) (7)

PERFORMANCE ESTIMATION OF PID

CONTROLLER [5, 23]

Controller design attempts to minimize the system

error produced by certain anticipated inputs. The system
Fig. 1: Basic P.I.D Servo Control Topology [22] error is defined as the difference between the desired
response of the system and its actual response.
For the purposes of this discussion, the transfer Performance criteria are mainly based on measures of the
function of the current regulator or really the torque system error. Basically, PID controller design method
regulator can be approximated as unity for the relatively using criterion as tabulate in TABLE I.
lower motion frequencies that are interested in and
 3
TABLE I changing the velocity of each particle toward its pbest and
PERFORMANCE ESTIMATION OF PID gbest locations at each time step. As example, the jth
CONTROLLER [23] particle is represented as xj = (xj,1 , xj,2 , . . . ,xj,g) in the g-
dimensional space. The best previous position of the jth
Name of Criterion Formula
particle is recorded and represented as pbestj = (pbestj,1 ,
∞
e (t ) dt
Integral of the Absolute
Error (IAE)
IAE = ∫ pbestj,2 , . . . , pbestj,g). The index of best particle among
0 all particles in the group is represented by the gbestg. The
∞
ISE = ∫ e(t ) dt
Integral of Square Error 2 rate of the position change (velocity) for particle j is
(ISE) 0 represented as vj = (vj,1, vj,2, . . . , vj,g). The modified
Integral of Time- ∞ velocity and position of each particle can be calculated
ITSE = ∫ te(t ) dt
2
weighted Square Error 0
using the current velocity and distance from pbestj,g to
(ITSE) gbestg as shown in the following formulas:
Integral of Time- ∞
weighted Absolute ITAE = ∫ t e(t ) dt v (jt, +g1) = w • v (jt, )g + c1 ∗ rand ( ) ∗ ( pbest j , g − x (jt, )g )
0
Error (ITAE) + c 2 ∗ rand ( ) ∗ (gbest g −x (t )
j,g ) (9)
A disadvantage of the IAE and ISE criteria is that its
minimization can result in a response with relatively small x (jt, +g1) = x (jt, )g + v (jt, +g1) (10)
overshoot but a long settling time because IAE and ISE
performance criterion weights all errors equally j = 1, 2, …, n
independent of time. Although the ITSE performance g = 1, 2, …, m
criterion weights errors with time, the derivation
processes of the analytical formula are complex and time where
consuming. n number of particles in a group;
In this paper, the performance criterion in the time m number of members in a particle;
domain was used as proposed by [5] for evaluating the t (t ) pointer of iterations(generations);
PID controller. A step of good control parameters kp, ki v j,g velocity of particle j at iteration t,
and kd can yield a good step response that will result in v gmin ≤ v (j t, )g ≤ v gmax ;
performance criteria minimization in the time domain. w inertia weight factor;
These performance criteria in the time domain include the c1, c2 acceleration constant;
overshoot Mp, rise time, tr, settling time, ts and steady rand( ) random number between 0 and 1;
state error Ess. Therefore, a new performance criterion x (jt, )g current position of particle j at iteration t;
W(K) is defined as [5]: pbestj pbest of particle j;
gbest gbest of the group
W (K ) = (1 − e (− β )
) • (M p + E ss )
(8) max
+ e − β
• (t s − t r ) The parameter v g determined the resolution, or
fitness, with which regions were searched between the
where K is [kp, ki , kd] and β is the weighting factor. present position and the target position. If v g
max
is too
max
The performance criterion W(K) can satisfy the high, particles might fly past good solutions but if v g is
designer requirements using the weighting factor, β value.
too low, particles may not explore sufficiently beyond
β can set larger than 0.7 to reduce the overshoot and
steady-state error or smaller than 0.7 to reduce the rise local solutions.
time and settling time. The constant C1 and C2 represent the weighting of the
stochastic acceleration terms that pull each particle
toward pbest and gbest. C1 and C2 were often set to be 2.0
OVERVIEW OF PARTICLE SWARM according to past experience. This because low values
allow particle to fly far from the target region before
OPTIMIZATION [5, 11, 13, 24]
being tugged back while high values result in abrupt
The particle swarm optimization (PSO) has been movement toward or past target regions.
Generally, the inertia weight w is set according to
introduced by Kennedy and Eberhart in 1995. PSO is
derived from the social-psychological theory, and has equation (11) below. Suitable selection of w provides a
been found to be robust in complex systems. Each particle balance between global and local explorations, thus
requiring less iteration on average to find a sufficiently
is treated as a valueless particle in g-dimensional search
space, and keeps track of its coordinates in the problem optimal solution.
space associated with the best solution (evaluating value) w max − w min
w = × iter (11)
and this value is called pbest. The overall best value and iter max
its location obtained so far by any particle in the group
where itermax is the maximum number of iterations or
that was tracked by the global version of the particle
swarm optimizer gbest. The PSO concept consists of generations and iter is the current number of iterations.
 4
IMPLEMENTATION OF A PSO-PID CONTROLLER Step 5: Modify the member velocity v of each individual
[5, 13] K according to (13)

The PID controller using the PSO algorithm was v (jt, +g1) = w • v (jt, )g
developed to improve the step transient response of (
+ c 1 * rand ( ) * pbest j,g − k (j t, g) )
typical servo motion system. It was also called the PSO- (13)
PID controller. The PSO algorithm was mainly utilized to + c2 * rand ( ) * (gbest g −k (t )
j,g )
determine three optimal controller parameters kp, ki, and j = 1 , 2 K n, g = 1, 2K 3
kd, such that the controlled system could obtain a good
step response output.
( t +1 )
If v j , g > V g , then v (jt,+g1) = Vgmax
max
In this paper, to apply the PSO method for searching Step 6:
the controller parameter, we use the “individual” to
replace the “particle” and the “population” to define the If v (jt, +g1) < V gmax, then v (jt,+g1) = Vgmax
“group”. The three controller parameters kp, ki, and kd,
composed an individual K by K ≡ [kp, ki, kd]; hence there Step 7: Modify the member position of each
are three members in an individual. These members are individual K according to (14)
assigned as real values. If there are n individuals in a
population, then the dimension of a population is n x 3. A k (j t, g+1) = k (j t, g) + v (jt, +g1) (14)
set of good control parameters kp, ki, and kd, can achieve a
good step response and result in minimization of such that k gmin ≤ k (j t, g+1 ) ≤ k gmax
performance criteria in the time domain including the min
where k g and k g
max
settling time (ts) , rise time (tr), maximum overshoot (Mp) represent the lower and upper
and steady state error (Ess). In the same time, we defined bounds, respectively, of member g of the individual K.
the evaluation value, f as in (12) which is reciprocal of the For example, when g is 1, the lower and upper bounds of
min max
performance criterion W(K) as in (8). the kp controller parameter are k g and k g respectively.
1
f = (12) Step 8: If the number of iterations reaches the maximum
W (K )
then, go to Step 9. Otherwise, go to Step 2.
It employs the smaller W(K) the value of individual K, the
higher its evaluation function. Step 9: The individual that generates the latest gbest is an
In order to limit the evaluation value of each optimal controller parameter.
individual of the population within a reasonable range, the
Routh-Hurwitz criterion must be utilized to test the
closed-loop system stability before evaluating the SIMULATION EXAMPLES RESULTS
evaluation value of an individual. The feasible individual
and small value of W(K) if the individual satisfied the The simulation of the Typical Servo Motion System
Routh-Hurwitz criterion stability test applied to the without PID controller, PSO-PID controller and Ziegler
characteristic equation of the system. and Nichols-PID Controller were implemented by
The searching procedures of the proposed PSO-PID MATLAB Version 7.2 and executed on the Pentium 4
controller were shown as follows [5, 24, 25]: 2.66GHz personal computer with 1GB RAM.

Step 1: Specify the lower and upper bounds of the three Typical Servo Motion System without PID Controller
controller parameters and initialize randomly the
individuals of the population including searching The block diagram of the Typical Servo Motion System
points, velocities, pbests and gbest. without PID is shown in Fig. 2 below. Result of this step
response of the Typical Servo Motion System without
Step 2: For each initial individual K of the population, PID controller has shown in following Fig. 3. To simulate
employ the Routh-Hurwitz criterion to test the this case, we found that performance criteria of the system
closed-loop system stability and calculate the in the time domain as the TABLE II below.
values of the four performance criteria in the
time domain, namely Mp, Ess, tr and ts.

Step 3: Calculate the evaluation value of each individual

in the population using the evaluation value, f
given by (12).

Step 4: Compare evaluation value of each individual with

its pbest. The best evaluation value among the
pbest is denoted as gbest. Fig. 2: Block Diagram of the Typical Servo Motion
System without PID Controller
 5

Fig. 4: Block Diagram of the Typical Servo Motion

System with PSO-PID Controller

Fig. 3: Step response of the Typical Servo Motion System

without PID controller

TABLE II
PERFORMANCE CRITERIA OF TYPICAL SERVO
MOTION SYSTEM WITHOUT PID CONTROLLER

Performance Criteria Value

Settling time (ts) 3.91 s
Rise time (tr) 0.0070 s
Fig. 5: Step response of the Typical Servo Motion System
Maximum overshoot (Mp) 97.7578 % with PSO-PID controller with β =1.0
Steady state error (Ess) 0

Typical Servo Motion System With PSO-PID Controller

According to the trial, the following PSO parameters are

used for verifying the performance of the PSO-PID
controller in searching the PID controller parameters:
• the member of each individual is kp, ki and kd;
• population size = 25;
• inertia weight factor w is set by (11), where wmax =
0.9 and wmin = 0.4; Fig. 6: Step response of the Typical Servo Motion System
• Range of three controller parameter: with PSO-PID controller with β = 1.5
o kp : minimum value = 0 ,
maximum value = 1.50;
o ki : minimum value = 0 , maximum value = TABLE III
1.00; BEST SOLUTION OF TYPICAL SERVO MOTION
o kd: minimum value = 0 , maximum value = SYSTEM WITH PSO-PID CONTROLLER WITH THE
1.00; DIFFERENT VALUES OF Β
• The limit of change in velocity : β 1.0 1.5
o V kmax
p
= k pmax / 2 Number of 25 25
o V k = k imax / 2
max
Iteration
i
o V kmax = k dmax / 2 kp 0.0092 0.0099
d
• Acceleration constant C1 and C2 = 2 ki 0.0015 0.0017
kd 0.0095 0.0100
The block diagram of the Typical Servo Motion System
tr 0.0120 0.0110
with PSO-PID controller is shown in Fig. 4 below.
Results of this step response of the Typical Servo Motion ts 0.0220 0.0200
System with PSO-PID controller with β = 1.0 and β = 1.5 Mp 0.1115 0.1084
has shown in following Fig. 5 and Fig. 6 respectively. The Ess 0 0
simulation results that showed the best solution were Evaluation Value 13.3876 11.6415
summarized in the Table III below.
 6
As can be seen, through about 25 iterations or
generations, the PSO method can prompt convergence
and obtain good evaluation value, thus achieve better
performance criterion that are rise time, settling time,
percentage of overshoot and steady state error condition.
These results show supremacy of the PSO-PID controller
that can search optimal PID controller parameter quickly
and efficiently.

Typical Servo Motion System with Ziegler and Nichols- Fig. 7: Determining fo and Ko of Ziegler and Nichols
PID Controller method

In order to emphasize the advantages of the proposed

PSO-PID controller, we also implemented the Ziegler and
Nichols-PID Controller into the Typical Servo Motion
System. The Ziegler and Nichols method basically boils
down to these two steps (fundamental of servo motor):

Step 1: Set ki and kd to zero. Excite the system with a

step command. Slowly increase kp until the shaft
position to oscillate. At this point, record the
values of kp and set ko equal to this value. Record
the oscillation frequency, fo. Fig. 7 shows the Fig. 8: Step response of the Typical Servo Motion System
result of slowly increasing on the proportional with Ziegler and Nichols-PID Controller
term. The system begins to oscillate at
approximately 0.5Hz (fo = 0.5Hz) with ko
approximately 5e-5 Nm/rad TABLE IV
PERFORMANCE CRITERIA OF TYPICAL SERVO
Step 2: Set the final PID gains using equation (15) MOTION SYSTEM WITH ZIEGLER AND NICHOLS-
PID CONTROLLER
k p = 0.6 k 0 Nm/rad ; Performance Criteria Value
Settling time (ts) 4.3980 s
k i = 2 f o k p Nm/(rad.sec) ; (15) Rise time (tr) 0.5160 s
kp Maximum overshoot (Mp) 34.2869 %
kd = Nm/(rad.sec) ;
8 fo Steady state error (Ess) 0

Using the values got in Step 1 into the equation (15),

the optimum PID gains according Ziegler and Nichols DISCUSSION AND CONCLUSION
method are then:
This paper presents a novel design method for
-4
kp = 3.0e Nm/rad determining the PID controller parameters using the PSO
ki = 3.0e-4 Nm/(rad.sec) method. The proposed method integrates the PSO
kd= 7.40e-5 Nm/(rad.sec) algorithm with the new time-domain performance
criterion into a PSO-PID controller. Through the
Fig. 8 shows the result of Typical Servo Motion System simulation of a typical servo motion system, the results
with Ziegler and Nichols-PID Controller. Table IV show that the proposed controller can perform an efficient
summarize the performance criteria of the system in the search to obtain optimal PID controller parameter that
time domain. achieve better performance criterion that are rise time,
settling time, percentage of overshoot and steady state
error condition.
This PSO-PID controller has been compared Ziegler-
Nichols PID method to verify it being more superior .The
comparison from rise time show that the PSO-PID
achieves less time that is 0.01sec compared to Ziegler-
Nichols PID that need 0.52sec. The system using Ziegler-
Nichols method takes about 4.4sec to finally settle
between 2% of the final value making it very difficult to
incorporate into any high performance motion control
 7
application. In contrast, the PSO method gives a quicker problem. Proceedings of the 2nd International Conference
settling time that it takes only 0.02sec to settle. The PSO on Machine Learning and Cybernetics, 1583-1585.
method instantaneously reduces the maximum overshoot [16] Ahmed, T. (2004). Adaptive particle swarm optimizer for
of the system to only 0.1% compared to the Ziegler- dynamic environments. M. Sc. thesis, The University of
Texas at Arlington, United States
Nichols method that recorded 34.3% of maximum [17] Baumgartner, U., Magele, Ch. and Renhart, W. (2004).
overshoot. However, there is no steady state error for both Pareto optimality and particle swarm optimization. IEEE
methods. Transactions on Magnetics, Vol. 40( 2), 1172-1175.
Therefore, it is clear from the results that the proposed [18] Ning, L., Fei, L., Debao, S. and Chang, H. (2004). Particle
PSO method has more robust stability and efficiency and swarm optimization for constrained layout optimization.
can solve the searching and tuning problems of PID Proceedings of the 5th World Congress on Intelligent
controller parameters more easily and quickly than the Control and Automation, 2214 -2218.
Ziegler-Nichols method. [19] Nunez, J.J.J. (2004). Particle swarm optimization
applications in power system engineering. M.Sc. thesis,
University of Puerto Rico, Mayagues Campus, Puerto Rico.
[20] Asokan, P., Baskar, N., Babu, K., Prabhaharan, G. and
REFERENCES Saravanan, R. (2005). Optimization of surface grinding
operations using particle swarm optimization technique.
[1] Pedret, C., Vilanova, R., Moreno, R. and Serra, I. (2002). A Journal of Manufacturing Science and Engineering, Vol.
refinement procedure for PID controller tuning. Computer 127, 885-892.
and Chemical Engineering, Vol.26, 903-908. [21] Chang, W.-D. and Yan, J.-J. (2005). Adaptive robust PID
[2] Chang, W.-D., Hwang, R.-C. and Hsieh, J. –G. (2003). A controller design based on a sliding mode for uncertain
multivariable on-line adaptive PID controller using auto- chaotic systems. Chaos, Solitons and Fractals, Vol. 26,
tuning neurons. Engineering Applications of Artificial 167-175.
Intelligence, Vol.16, 57-63. [22] Kaiser, D., Fundamentals of Servo Motion Control.
[3] Tan, K.K., Huang, S. and Ferdous, R. (2002). Robust self- http://www.parkermotion.com/whitepages/ServoFundamen
tuning PID controller for nonlinear systems. Journal of tals.pdf
Process Control, Vol.12, 753-761. [23] Wang, X., Wang, Y., Zhou, H. and Hui, X. (2006). PSO-
[4] Valério, D. and Costa, J. S. (2006). Tuning of fractional PID: A novel controller for AQM Routers. Proceedings of
PID controllers with Ziegler-Nichols-type rules. Signal IEEE and IFIP International Conference on Wireless and
Processing, Vol. 86, 2771-2784. Optical Communication Network (IEEE/IFIP WOCN
[5] Gaing, Z.L. (2004). A Particle Swarm Optimization 2006), 1-5.
approach for optimum design of PID controller in AVR [24] Yoshida, H., Fukuyama, Y., Kawata, K., Takayama, S. and
system. IEEE Transactions on Energy Conversion, Nakanishi, Y. (2001). A Particle Swarm Optimization for
Vol.19(2), 384-391. reactive power and voltage control considering voltage
[6] Chen, M. and Linkens, D.A. (1998). A hybrid neuro-fuzzy security assessment. IEEE Transactions on Power Systems,
PID controller. Fuzzy Sets and Systems, Vol.99, 27-36. Vol.15(4), 1232-1239.
[7] Chu, S.-Y. and Teng, C.-C. (1999). Tuning of PID [25] Liu, Y., Zhang, J. and Wang, S. (2004). Optimization
controllers based on gain and phase margin specifications design based on PSO algorithm for PID controller.
using fuzzy neural network. Fuzzy Sets and Systems, Vol. Proceedings of 5th World Congress on Intelligent Control
101, 21-30. and Automation, Vol. 3, 2419-2422.
[8] Lan, L. H. (2006). Stability analysis for a class of Takagi-
Sugeno fuzzy control systems with PID controllers.
International Journal of Approximate Reasoning, In Press.
[9] Kim, S.-M. and Han, W. Y. (2006). Induction motor servo
drive using robust PID-like neuro-fuzzy controller. Control
Engineering Practice, Vol.14, 481-487.
[10] D’Emilia, G., Marra, A. and Natale, E. (2006). Use of
neural networks for quick and accurate auto-tuning of PID
controller. Robotic and Computer-Integrated
Manufacturing, In Press.
[11] Kennedy, J. and Eberhart, R.C. (1995). Particle swarm
optimization. Proceedings of the International Conference
on Neural Networks, 1942-1948.
[12] Fan, H. (2002). A modification to particle swarm
optimization algorithm. Engineering Computations, Vol.
19, 970-989.
[13] Kao, C. –C, Chuang, C.-W and Fung, R. –F (2006). The
self-tuning PID control in a slider-crank mechanism system
by applying particle swarm optimization approach.
Mechatronics, Vol.16, 513-522.
[14] Tayal, M. (2003). Particle swarm optimization for
mechanical design. M. Sc. thesis, The University of Texas
at Arlington, United States.
[15] Wang, K.-P., Huang, L., Zhou, C. –G. and Pang, W.
(2003). Particle swarm optimization for traveling salesman