Comparative Study of Some Optimization Techniques

Applied to DC Motor Control

Vikrant Vishal1, Vineet Kumar2, K.P.S. Rana3, Puneet Mishra4

Division of Instrumentation and Control Engineering
Netaji Subhas Institute of Technology
New Delhi, India
1
[email protected], [email protected], [email protected], [email protected]

Abstract- Traditional tuning techniques for classical the simple tuning rules for PID controller [3]. These rules are
Proportional-Integral-Derivative (PID) controller suffer from still used extensively in the industry but have many
many disadvantages like non-customized performance measure drawbacks: Firstly, these rules use insufficient process
and insufficient process information. For the past two decades information and performance measure (quarter amplitude
nature inspired optimization algorithms are efficiently being damping) which leads closed loop systems to poor damping
implemented for tuning of PID controllers. In this paper, four [2, 4]. Secondly, since the performance of the controller is
optimization methods namely Genetic Algorithm (GA), affected by changes in process dynamics due to nonlinearities
Accelerated Particle Swarm Optimization (APSO), Differential and parameters’ variations, these rules lack a customized
Evolution (DE) and Cuckoo Search (CS) are studied and used to
performance index.
optimize the controller gains of a Proportional–Integral (PI)
controller for set point tracking in speed control of a DC motor Nature inspired optimization algorithms like GA [5, 6],
by minimizing Integral Time Absolute Error (ITAE). Hardware PSO [7, 8], DE [9] and CS [10] have provided an alternative
validation of the efficiency of above mentioned optimization to classical tuning approach by eliminating the above
algorithms is studied and presented. The plant under study is a mentioned disadvantages. These algorithms explore larger
DC motor control module (MS15) from M/S LJ CREATE™. M/S search space efficiently to achieve the global optimal solution.
National Instruments (NI) based software and hardware Over the years aforementioned algorithms are successfully
components i.e. LabVIEW™ and its add-ons toolkit and data implemented for PID controller tuning [5, 6, 11, 12] but lesser
acquisition (DAQ) card has been utilized for the closed loop
attention has been paid in the literature on hardware validation
control in real time. The system identification is done in
LabVIEW™ and then offline performance optimization is
of effectiveness of these algorithms, particularly the CS and
carried out in MATLAB™. The tuned gains are further used to APSO. Also, the literature shows relatively lesser work on the
study the run time performances in LabVIEW™ environment. comparative study of the performance of different
This is done because MATLAB™ has very good optimization optimization algorithms in run time to tune the PID controller
tools and on the other hand LABVIEW™ makes the parameters.
measurement very easy. From the results obtained it can be In this paper, a comparative study is carried out on the
clearly inferred that CS algorithm outperformed other
optimization capability of GA, APSO, DE and CS, in order to
algorithms studied in this paper, particularly in disturbance
optimally design a PI controller for DC motor speed control.
rejection.
The optimization part is performed in MATLABTM and testing
Keywordsícontroller tuning; DC motor control; PI controller; of optimization results on hardware is done in LabVIEW™
genetic algorithm; accelerated PSO; differential evolution; cuckoo environment. The hardware implemented results of above
search mentioned algorithms are presented and compared in this
work.
I. INTRODUCTION
The rest of the paper is organized as follows: Section 2
Process control industry has seen many advances in the briefly introduces the different optimization techniques used in
past two decades in terms of the controller design and its the paper. Section 3 presents an overview of PID controller.
implementation methods [1]. In spite of these advances Section 4 describes DC motor speed control module, DAQ
classical PID controller is undoubtedly the most popular card specifications, experimental setup and system
controller in the industry because of its simple structure and identification of the motor system used in this experiment.
robust performance in different operating conditions. It has Section 5 discusses about controller parameters optimization
been reported that 97% controllers in refining, paper and pulp and parameters taken in different algorithms. In last, section 6
industries have PID structure [2]. To implement a PID presents the results of the experiment.
controller effectively, tuning of its parameters plays a vital
role. J. G. Zeigler and N. B. Nichols were the first to present

978-1-4799-2572-8/14/$31.00 2014
c IEEE 1342

Authorized licensed use limited to: Turkish Energy Nuclear and Mineral Research Agency. Downloaded on April 10,2025 at 09:12:12 UTC from IEEE Xplore. Restrictions apply.
 II. PRINCIPLE OF OPTIMIZATION ALGORITHMS gbest: gbest position of swarm
This section presents a brief introduction of GA, APSO, DE pbest: pbest position of particle
and CS optimizing algorithms. In APSO, only the gbest component is used [15], thus the
velocity vector becomes:
Vn+1=Vn+ C1 randn( )+ C2*(gbest,n -currentpositionn) (3)
A. Genetic Algorithm
GA proposed by Goldberg (1989) is a probabilistic search where, randn( ) is drawn from N (0, 1) to replace the second
algorithm based upon the concept of “survival of the fittest” of term. The update in the position is same as (2).
evolution theory. It starts with initial population of likely
solutions of the problem and the application of three operators
C. Differential Evolution
namely; Reproduction, Crossover and Mutation, produce
offsprings which are superior to the parents at the end of each DE was proposed by Storn and Price in 1995 in an attempt
iteration. This operation is repeated for many generations as to solve the chebychev polynomial fitting problem [9, 16]. DE
per the precision requirement to get the optimal solution of the is similar to the structure of GA but requires less computation
problem. The efficiency of the algorithm depends largely on and programming. The main difference is Mutation operation
many factors like population size, mutation rate and crossover [9].
frequency [13]. In spite of the fact that GA suffers from In the mutation operation in DE, new vector is produced
premature and slow convergence, the algorithm has been used by adding the perturbations of two members as a vector to the
efficiently in tuning the PID parameters [5, 6]. third vector. The new vector is then mixed with predefined
parameters in accordance with certain rules to produce test
B. Accelerated Particle Swarm Optimization vector. This is called Crossover. If the test vector of function
Particle Swarm Optimization (PSO) was proposed by is lesser than desired function, then the test vector becomes the
American socio-psychologist James Kennedy and electrical desired vector in next generation. This operation continues
engineer Russell Eberhart in 1995 [7, 8]. PSO is basically a until global optimal solution is achieved.
swarm intelligence based optimization algorithm inspired by
natural phenomenon like Bird Flocking, Animal Herding and D. Cuckoo Search Algorithm
Fish Schooling. In 2010, Xin She Yang and Suash Deb [10] proposed CS
PSO methodology is based upon three simple rules: algorithm via Lévy Flights for optimizing a nonlinear function
Separation, Alignment and Cohesion [14]. In PSO the system using generation of random numbers with symmetric Lévy
is initialized with a population of random solutions. Each distribution obtained by Mantegna’s algorithm. CS algorithm
potential solution is also assigned a randomized velocity, and is based upon the some Cuckoo’s breeds like Ani, Tapera and
the potential solutions, called particles, are then “flown” Guira’s aggressive breeding behavior of laying their own eggs
through hyperspace. Each particle keeps track of its in others birds’ nests.
coordinates in hyperspace associated with the best solution The three fundamental rules for CS as proposed by Yang and
(pbest) it has achieved so far and overall best solution (gbest) Deb are as follows [10]:
[7, 8].
1) Each cuckoo lays one egg at a time, and dumps it in a
The PSO concept consists of, at each time step, changing randomly chosen nest;
the velocity of each particle toward its pbest and gbest. 2) The best nests with high quality of eggs (solutions) will
Acceleration is weighted by a random term. Each particle tries carry over to the next generations;
to modify its current position and velocity according to the 3) The number of available host nests is fixed, and a host
distance between its current position and pbest, and the
can discover an alien egg with a probability pa‫[ א‬0, 1]. In this
distance between its current position and gbest as per (1) and
(2) respectively. case, the host bird can either throw the egg away or abandon
the nest so as to build a completely new nest in a new location.
Vn+1=Vn+ C1 rand1( )*(pbest,n-currentpositionn)+
C2 rand2( )*(gbest,n -currentpositionn) (1) The Lévy flight is performed by following equation for
generating new solutions for ith cuckoo:
Currentposition[n+1] = currentposition[n] + V[n+1] (2)
Xit+1= Xit + ĮْLévy (Ȝ) (4)
Where,
th Where, Į>0 is the step size which should be related to the
Currentposition[n+1]: Position of particle at (n+1) iteration
th scales of the problem of interest. Lévy flights provide random
Currentposition[n]: Position of particle at n iteration
th walks and their random steps are taken from Lévy distribution
V[n+1]: Particle velocity at (n+1) iteration
(5) which has an infinite variance with an infinite mean.
Vn+1: Velocity of particle at (n+1)th iteration
Vn : Velocity of particle at nth iteration Lévy (Ȝ) ~ u > t-Ȝ, 1< Ȝ ”3 (5)
C1: Acceleration factor related to gbest The steps of a cuckoo essentially form a random walk
C2: Acceleration factor related to lbest process which obeys a power-law step length distribution with
rand1( ): Random number between 0 and 1 a heavy tail.
rand2( ): Random number between 0 and 1

2014 IEEE International Advance Computing Conference (IACC) 1343

Authorized licensed use limited to: Turkish Energy Nuclear and Mineral Research Agency. Downloaded on April 10,2025 at 09:12:12 UTC from IEEE Xplore. Restrictions apply.
 III. OVERVIEW OF PID CONTROLLER

The block diagram of closed loop control system is shown

in Fig. 1. The parallel version of PID controller is described
by:
t
de ( t ) (6)
u ( t ) = K p e ( t ) + K i ³ e (τ ) d τ + K d + us
0
dt

Where, u(t) is the controller output, e(t) is the error, us is the

bias, Kp is the proportional gain, Ki is the integral gain and Kd
Fig. 2. Schematic Diagram of DC Motor Control Module [17]
is the derivative gain.
3) Tachogenerator Output:
The controller output is therefore sum of three terms:
Proportional action, which corresponds to proportional +5V(approximately) at maximum speed counter
control; Integral action, which gives a control action clockwise
proportional to the time integral of error. This ensures that the -5V (approximately) at maximum speed clockwise
steady state error becomes zero. The derivative action is 4) Potentiometer Output :
proportional to the time derivative of the error and allows Analog voltage varying between +5V and -5V for 1
prediction of the future error. complete revolution of output shaft
In this paper, PI controller is used for DC motor speed control. 5) Eddy Current Brake:
A 2 position Eddy Current brake is fitted to the
module to provide disturbance to the motor
IV. SYSTEM DESCRIPTION 6) Motor Drive Input :
+5V for maximum speed counter clockwise
This section discusses the DC motor control module, DAQ -5V for maximum speed clockwise
card, experimental setup and the system identification of DC B. DAQ Card Specifications
motor.
The DAQ card used in this experiment is NI PCI-6221 and
A. DC Motor Control Module has the following specifications:
The MS15 DC Motor Control Module shown in Fig. 2 1) Input Voltage range: ±10V
enables the user to perform closed loop positional and speed 2) Output Voltage range: ±10V
control of DC motor. The speed and direction of the motor can
3) Sampling Rate: 250kS/s
be controlled by either an analog signal or a pulse width
modulated digital signal. 4) Input Channels: 8 Differential, 16 Single ended.
5) Output Channels: 2
The DC Motor Control Module has following specifications
[17]: C. Experimental Setup
1) Motor Speed: The scheme and snapshot of experimental setup is shown
Upto 2500 RPM in either direction in Fig. 3 and Fig. 4 respectively. It consists of MS15 DC
2) Power Supplies: Motor Module connected to Intel Core i2, 1.99 GHz HP PC
+5V at approximately 400mA, with Windows 7 operating system via NI PCI-6221 DAQ
card. The main application program is developed on
+12V at approximately 0.5 to 0.9A,
LabVIEW™ and the offline tuning of PID parameters utilizing
-12V at approximately 0.5 to 0.9A
above mentioned optimization algorithms has been done on
MATLAB™ on the PC.
The run time data of speed of the motor for setpoint tracking
and disturbance rejection is collected at the sampling rate of
100 samples/second from the DC Motor Control Module.

Fig. 1. Block Diagram of Simple Closed Loop Control System

Fig. 3. DC Motor Speed Control Scheme

1344 2014 IEEE International Advance Computing Conference (IACC)

Authorized licensed use limited to: Turkish Energy Nuclear and Mineral Research Agency. Downloaded on April 10,2025 at 09:12:12 UTC from IEEE Xplore. Restrictions apply.
 Fig. 6. Closed Loop Measured and Simulated Data for DC Motor at Speed
100 RPM.

The operating point chosen for set point tracking and

Fig. 4. Snapshot of DC Motor Control Module disturbance rejection is 100 RPM. The values of PI controller
parameters Kp and Ki calculated by each algorithm are
D. System Identification tabulated in Table 2. The Fitness vs. Number of iterations
For system identification the DC motor is excited with 4.5V in curve for each algorithm is shown in Fig. 7, where fitness
open loop and the transfer function is estimated using System corresponds to ITAE.
Identification Toolkit of LabVIEW™. The transfer function
shows that the DC motor behaves mainly as a first order VI. RESULTS AND DISCUSSION
system having a very small component of second order
The DC motor’s behavior for set point tracking for
dynamics.
different values of Kp and Ki obtained by CS, DE, APSO and
0 . 78026 (7)
y(s) = u (s) GA is shown in Fig. 8. The figure clearly shows that CS has
1 + 0 . 251767 s + 0 . 0040954 s2 the fastest response whereas GA has sluggish. The PI
controller action in this case is shown in Fig. 9.
Where y is DC motor speed scaled in voltage and u is the
excitation voltage. For converting DC motor speed (RPM) into To analyze the controller’s disturbance rejection
corresponding voltage a tachogenerator is used. The capability, a 2 position Eddy Current brake is used. The brake
tachogenerator is assumed to have zero order dynamics with a is applied suddenly, first at position 1 and then at position 2
constant scaling factor of 75.4248 RPM/V. Fig. 5 shows open and the response of DC motor is observed at different
loop measured and simulated data for u = 4.5V. The curve optimized gains as shown in Fig. 10 and Fig. 11 respectively.
clearly shows that the estimated transfer function closely
matches the actual transfer function of the DC motor. The
closed loop simulation and run time curve of DC Motor TABLE I. OPTIMIZATION SPECIFICATIONS
running at operating speed of 100 RPM is shown in Fig. 6. S.NO. SPECIFICATION VALUES
1. No. of Iterations 100
2. Dimention 2
V. CONTROLLER OPTIMIZATION 3. Initial Population 20
4. Step Size (Fixed) 0.01
The performance index taken for optimization is ITAE 5. ODE Solver ODE1
given by:
TABLE II. OPTIMIZED PI PARAMETERS
ITAE = ³ t e ( t ) dt (8)
PI Gains
Algorithms
The optimization specifications for GA, DE, CS and APSO Proportional Gain (Kp) Integral Gain (Ki)
are same as tabulated in Table I. Built-in toolbox is used for DE 4.7516 17.2746
GA 2.4909 11.1350
GA and codes for DE, CS and APSO are developed in
APSO 4.2344 16.7322
MATLAB™. CS 4.7499 17.2810

Fig. 5. Open Loop Measured and Simulated Data Fig. 7. Fitness vs. Iteration Curve for APSO, CS, DE and GA

2014 IEEE International Advance Computing Conference (IACC) 1345

Authorized licensed use limited to: Turkish Energy Nuclear and Mineral Research Agency. Downloaded on April 10,2025 at 09:12:12 UTC from IEEE Xplore. Restrictions apply.
 rejection at brake position 1 and 2 is shown in Fig. 15 and in
Fig. 16. It is clear from the figures that CS algorithm has the
minimum ITAE in both set point tracking and disturbance
rejection. The run time and simulated error curves for setpoint
tracking are shown in Fig. 17 and Fig. 18 respectively.
Table III shows the percentage improvement in
performance of controller tuned by CS, DE and APSO with
respect to GA in terms of ITAE for set point tracking and
disturbance rejection. Significant reduction in ITAE is
observed in case of CS tuned controller in comparison to
Fig. 8. Set Point Tracking Performance other.
TABLE III. ITAE IMPROVEMENT

CS DE APSO
SETPOINT
3.37 % 0.18 % -0.32 %
TRACKING
DISTURBANCE
16.47 % 15.17 15.00 %
REJECTION *
DISTURBANCE
22.39 % 21.37 % 19.62 %
REJECTION **
* Brake Position 1.
** Brake Position 2.
Fig. 9. Controller Action for Set Point Tracking

Fig. 12. Controller Action for Disturbance Rejection at Brake Position 1.

Fig. 10. Disturbance Rejection at Brake Position 1.

Fig. 11. Disturbance Rejection at Brake Position 2. Fig. 13. Controller Action for Disturbance Rejection at Brake position 2.

It is observed that at brake position 1, there is a decrease in

motor speed by 20 RPM and at brake position 2, the speed
decreases by 40 RPM.
The PI controller action for disturbance rejection when
Eddy Current brake is at position 1 and at position 2 is shown
in Fig. 12 and Fig. 13 respectively. Since PI controller rejected
the disturbance in a very small time interval, Figs. 10-13 are
shown only for 1 second to distinguish the performance of
controller tuned by studied algorithms.
The value of ITAE calculated at different optimized gains
for set point tracking is shown in Fig. 14 and for disturbance Fig. 14. ITAE for Set Point Tracking

1346 2014 IEEE International Advance Computing Conference (IACC)

Authorized licensed use limited to: Turkish Energy Nuclear and Mineral Research Agency. Downloaded on April 10,2025 at 09:12:12 UTC from IEEE Xplore. Restrictions apply.
 VIII. REFERENCES

[1] K. J. Astrom, “Process Control-Past, Present and Future,” Control

Systems Magazine, IEEE, Vol. 5, No. 3, pp. 3-10, August, 1985.
[2] K. J. Astrom and T. Hagglund, “Revisiting the Zeigler-Nichols
Step Response Method for PID Control,” Journal of Process
Control, Vol. 14, pp. 635-650, 2004.
[3] J. G. Zeigler and N. B. Nichols, “Optimum Settings for Automatic
Controllers,” Trans. ASME, Vol. 64, pp.759-768, 1942.
Fig. 15. ITAE for Disturbance Rejection at Brake Position 1. [4] T. Haggling and K. J. Astrom, “Revisiting the Zeigler-Nichols
Tuning Rules for PI Control-Part II The Frequency Response
Method,” Asian Journal of Control, Vol.6, No.4, pp. 469-482,
December 2004.
[5] C. Ou and W. Lin, “Comparison between PSO and GA for
Parameters Optimization of PID Controller,” Proceedings of 2006
IEEE International Conference on Mechatronics and Automation,
pp. 2471-2475, June 25-28, 2006, Luoyang, Henan.
[6] H. Zhang, Y. Cai and Y. Chen, “Parameter Optimization of PID
Controllers Based on Genetic Algorithm,” Proceedings of
International Conference on Health Networking, Digital
Ecosystems and Technologies (EDT), Vol. 1, pp. 47-49, April 17-
Fig. 16. ITAE for Disturbance Rejection at Brake Position 2.
18, 2010, Shenzhen.
[7] R. Eberhart and J. Kennedy, “A New Optimizer Using Particle
VII. CONCLUSION Swarm Theory,” Proceedings of 6th International Symposium on
Micro Machine and Human Science, pp. 39-43, October 4-6, 1995
Nagoya, Japan.
In this study, some optimization techniques i.e. DE, GA, [8] J. Kennedy and R. Eberhart, “Particle Swarm Optimization,”
APSO and CS algorithms have been investigated for the tuning Proceedings of IEEE International Conference on Neural
Networks, Vol. 4, pp. 1942-1948, Nov 27-Dec 1, 1995, Perth, WA.
of PI controller, for set point tracking applied to a DC Motor
Speed Control. The performance comparison was carried out [9] L. G. Yang and L. M. Guang, “The Summary of Differential
Evolution Algorithm and its Improvements,” Proceedings of 3rd
for setpoint tracking and disturbance rejection. Though, International Conference on Advance Computer Theory and
comparable performance is observed in case of set point Engineering (ICACTE), Vol. 3, pp. 153-156, August 20-22, 2010,
tracking but in the case of disturbance rejection the Chengdu.
performance of CS tuned controller is found to be superior to [10] X. S. Yang and S. Deb, “Engineering Optimization By Cuckoo
other algorithms studied in this paper. This paper has Search,” International Journal of Mathematical Modelling and
considered the offline tuning of the controller gains and as a Numerical Optimization, Vol. 1, No. 4, pp. 330-343, December 23,
2010.
future scope of this work online tuning can be considered.
[11] M. M. R. A Milani, T. Cavdar and V. F. Aghjehkand, “Particle
Swarm Optimization-Based Determination of Zeigler-Nichols
Parameters for PID Controller of Brushless DC Motors,”
International Symposium on Innovations in Intelligent Systems and
Applications (INISTA), pp. 1-5, July 2-4, 2012, Trabzon.
[12] I. Chiha, J. Ghabi and N. Liouane, “Tuning PID Controller With
Multi-Objective Differential Evolution,” 5th International
Symposium on Communications Control and Signal Processing
(ISCCSP), pp. 1-4, May 2-4, 2012, Rome, Italy.
[13] A. Kumar and S. Chakarverty, “Design Optimization using
Genetic Algorithm and Cuckoo Search,” Proceedings of IEEE
International Conference on Electro/Information Technology
(EIT), pp.1-5, May 15-17, 2011, Mankato, MN.
Fig. 17. Run Time Error [14] C. W. Reynolds, “Flocks, Herds and Schools-A Distributed
Behavioral Model,” Computer Graphics, Vol. 21, No. 4, pp. 25-34,
1987.
[15] X. S. Yang, S. Deb and S. Fong, “Accelerated Particle Swarm
Optimization and Support Vector Machine for Business
Optimization and Applications,” Networked Digital Technologies
(NDT2011), Communications in Computer and Information
Science, Vol. 136, pp. 53-66, March 23, 2011.
[16] R. Dong, “Differential Evolution verses Particle Swarm
Optimization for PID Controller Design,” Proceedings of 5th
International Conference on Natural Computation (ICNC), Vol. 3,
pp. 236-240, August 14-16, 2009, Tianjin.
[17] LJ Technical Systems. DC Motor Control Module User Manual.
Fig. 18. Simulation Error LJ Group, Holtsville, NY, USA, 1st edition.

2014 IEEE International Advance Computing Conference (IACC) 1347

Authorized licensed use limited to: Turkish Energy Nuclear and Mineral Research Agency. Downloaded on April 10,2025 at 09:12:12 UTC from IEEE Xplore. Restrictions apply.