322 Journal of Power Electronics, Vol. 6, No.

4, October 2006

JPE 6-4-6

A Novel Technique for Tuning PI-Controllers in Induction

Motor Drive Systems for Electric Vehicle Applications

Ayman Saber Elwer†

†
Dept. Of Power Electronics and Energy Conversion, Electronics Research Institute, Cairo, Egypt

ABSTRACT

In the last decade, the increasing restrictions imposed on the exhaust emissions from internal combustion engines and
traffic limitations have increased the development of electrical propulsion systems for automotive applications. The goal
of electrical and hybrid vehicles is the reduction of global emissions, which in turn leads to a decrease in fuel resource
exploitation. This paper presents a novel approach for control of Induction Motors (IM) using the Particle Swarm
Optimization (PSO) algorithm to optimize the parameters of the Proportional Integral Controller (PI-Controller). The
overall system is simulated under various operating conditions. The use of PSO as an optimization algorithm makes the
drive robust and insensitive to load variation with faster dynamic response and higher accuracy. The system is tested
under variable operating conditions. The simulation results show a positive dynamic response with fast recovery time.

Keywords: Electric Vehicles, Particle Swarm Optimization, Field oriented control, Induction motor

1. Introduction cost, which are all desired for EVs and Hybrid Electric
Vehicles (HEVs). However control of IM is
The major components of an electric vehicle system complicated due to the fact that in obtaining decoupled
are the motor, controller, power source, charger and control of the torque and flux producing components of
drive train. The majority of electric vehicles (EV) the stator phase current, both the magnitude and phase
developed so far are based on dc machines, induction of the stator quantities need to be controlled. In
machines or permanent magnet machines. The addition, there is no direct access to the rotor quantities,
disadvantages of dc machines forced the EV developers such as rotor fluxes and currents [1]. To overcome these
to look into various types of ac machines. The power difficulties, high performance vector control algorithms
density of permanent magnet machines together with have been developed. These algorithms can decouple
the high cost of permanent magnets makes these the stator phase currents by using only the measured
machines less attractive for EV applications. stator current and flux, as well as the rotor speed [2-3].
The main reasons for acceptance of the Induction This drive system has three PI-Controllers which are
motor are its ruggedness, reliability, and inexpensive tuned using PSO instead of traditional tuning methods;
the drive system plays an important role in meeting the
other requirements. It should enable the drive to follow
Manuscript received May 25, 2006; revised August 17, 2006
†
Corresponding Author: elwer@eri.sci.eg any reference speed taking into account the effects of
Tel: +202-3310552, Fax: +202-7292025, ERI, Cairo, Egypt. load impact, saturation and parameter variation. The
 A Novel Technique for Tuning PI-Controllers … 323

MATLAB SIMULINK software package is utilized to foundation for every high performance control scheme
simulate each part of the system under study. The used today.
simulation of the overall system is composed of these
simulated components when they are properly 3. Field Oriented Control
interconnected.
In a separately excited DC machine the axes of the
2. Induction Motor Control Strategies armature and field currents are orthogonal to one
another. This means that the magneto motive forces
Until recently induction motors have been plugged established by the currents in these windings are also
directly into the supply Direct Online (DOL) or orthogonal. If iron saturation is ignored the developed
operated using the relatively crude Open Loop Volts per torque is equal to [5]:
Hertz (V/Hz) control strategy. Both techniques cause
problems with efficiency, reliability and Tem = K a φ(I f )I a (1)
electromagnetic interference. Obviously DOL motors
can only be operated at the supply frequency and are This means that the flux depends solely on the field
therefore incapable of variable speed control. Although winding current. If the flux is fixed then the torque is
the V/Hz strategy can provide speed variation it is varied directly by the armature current. It is for this
unable to provide reliable control under transient reason that DC machines are said to have decoupled or
conditions and can often cause circuit breakers to trip. independent control over torque and flux.
Another problem with open loop control strategies is Unfortunately the operation of induction machines is
that they are only suitable when the motor can be much more complicated. Induction motors are coupled,
operated at steady torque without speed regulation. non linear, multivariable systems whose stator and rotor
The need for more precise control over torque and fields are not held orthogonal to one another. In order to
speed led to the development of closed loop V/Hz achieve decoupled control over the torque and flux
controllers and other algorithms based on the induction producing components of the stator currents, a
motor model. The problem with using the machine technique known as Field Oriented Control is used.
model is that the electromagnetic characteristics are
only valid in steady state so high peak voltages and 4. The Structure of Field Oriented Control
currents still occur under transient conditions. These
motor control strategies were also based on sinusoidal A block diagram for a Field Oriented Controller can be
PWM which is not suited to closed loop PI regulation. seen in the following section. This design uses a more
Replacing the PI regulators with hysteresis controllers robust structure known as indirect FOC, meaning that
can slightly improve performance at the expense of high the rotor angle is not determined directly by measuring
bandwidth noise that it hard to filter from the system. the air gap flux with hall-effect sensors. These sensors
Another problem with early control strategies is that are not particularly suited for use in large industrial
they were unable to consider phase interaction and motors as they can be fragile and sensitive to
could only be operated as synchronous or asynchronous, temperature change [5]. Instead the rotor flux angle is
but not both simultaneously. calculated from a mechanical speed sensor or encoder.
To overcome these limitations a new approach to The key components of the FOC strategy are the Clarke
motor control was required. This was provided in the and Park transform blocks. These transformation and
early 1970s by Hasse and Blaschke who developed the induction motor equations exist in [6]. As can be seen in
theory of Field Oriented Control (FOC) and later in the Figure (1) these map the three phase stator currents onto
mid 1980s by Takahaski and Dependbrock who created a direct and quadrature rotating reference frame that is
Direct Torque Control (DTC) [4]. These strategies are the aligned with the rotor flux. This decouples the torque
324 Journal of Power Electronics, Vol. 6, No. 4, October 2006

and flux. Producing components of the stator currents controllers. Loosely defined, adaptive PI-controllers
allows the induction motor to be controlled in much the avoid time-consuming manual tuning by providing
same way as a separately excited DC machine. optimal PI-controller settings automatically as the
Three PI regulators are used to set the output system dynamics or operating points change [7]. There
reference voltages. The first PI regulator compares the are two methods of tuning the PI-Controller; they are
speed set point with the measured mechanical speed of the conventional Ziegler-Nichols method and the
the rotor and produces the stator current quadrature axis Intelligence methods such as the PSO method.
reference, isqref. The stator current direct axis reference
isdref is usually kept constant at the value required to 6. Ziegler-Nichols Method for Tuning the
produce the nominal rotor flux. To operate the motor PI-Controller
above its nominal speed a technique known as Field
Weakening is used to reduce the rotor flux. The Ziegler and Nichols developed PID tuning methods in
reference currents are compared with the measured the early 1940s based on open loop tests (less known
stator currents. The error is used by the PI regulators to than the Cohen-Coon formulas) and also based on a
generate the output stator voltages in the direct and closed loop test, which may be their most widely known
quadrature axes. These are transformed back onto the α achievement.
and β axes using the inverse Park transformer to allow The closed loop method prescribes the following
the output voltage to be generated directly using procedure:
SVPWM. Step 1: Disable any D and I action of the controller
(pure P-controller).
Step 2: Make a set point step test and observe the
PSO
iqsref response.
Vqsref
+ +
Wref PI1 Step 3: Repeat the SP test with increased / decreased
- - PI2
Inv. Inverter controller gain until a stable oscillation is achieved. This
PI3 Par
idsref _ Vdsref
gain is called the "ultimate gain" Ku.
iqs
θ Step 4: Read the oscillation period Pu.
ids
ia ib ic Step 5: Calculate the parameters according to the
Current
Model
Clarke following formulas:
Park
PI: Proportional gain = 0.45 * Ku, integral time =Pu / 1.2
ωm I.M
PID: Proportional gain = 0.6 * Ku, integral time =Pu / 2,
derivative time = Tu / 8
Fig. 1 Field oriented control of Induction Motor This is shown in figure (2)

5. Methods of Tuning the PI-Controller

PI-Controllers have been applied to control almost

any process in current use, from aerospace to motion
control, from slow to fast systems. Alongside this
success, however the problem of tuning PI-controllers
has remained an active research area. Furthermore, with
changes in system dynamics and variations in operating
points PI-Controllers should be returned on a regular
basis. This has triggered extensive research on the
possibilities and potential of the so-called adaptive PI- Fig. 2 Ziegler-Nichols for tuning PI-Controller
 A Novel Technique for Tuning PI-Controllers … 325

7. Design of the FOC Using PSO best fitness in the local neighborhood, designated g, and
the P vector of current particle are combined to adjust
The PSO was originally designed by Kennedy and the velocity along each dimension. That velocity is then
Eberhart [8]. PSO begins with a swarm of particles as the used to compute a new position for the particle as follow
[10]:
initial population. Each particle has a position and a
velocity. The position of the particle encodes the Vj,i = wVj,i +C1 r1(Pj − X j ) +C2r2 (Pgj − X j ) (2)
solution of the problem. The velocity of the particle
represents the value added to the position of the particle Xj:=Xj+Vj (3)
to find its position in the next generation. The algorithm
updates the position and velocity of all particles in each 8. Implementation of PSO
generation, until the algorithm finds an optimum. The
velocity of all particles is initially zero and is updated The PSO program is simple and takes few lines in the
according to the best local position (best fitness) the program. Reducing the time of implementation in the
particle has come across in its lifetime (all generations whole program, the steps of the PSO program is
so far) and the best position any particle in the whole described as follows:
swarm has ever come across. The original PSO Step 1 Generation of initial conditions of each agent
formulae define each particle as a potential solution to a Initial searching points (Xio) and velocities (Vio) of each
problem in D-dimensional space, with particle j agent are usually generated randomly within the
represented as: allowable range. The current searching point is set to
Xj= ( Xj1 ,Xj2 ,………..,XjD ) pbest for each agent. The best-evaluated value of pbest
Also, each particle maintains a memory of its previous is set to gbest and the agent number with the best value
best position as : is stored.
Pj =( Pj1 , Pj2 , ………., PjD ) Step 2 Evaluation of searching point of each agent
Velocity along each dimension is represented as: The objective function value is calculated for each
Vj =( Vj1 , Vj2 , ……..., VjD) agent. If the value is better than the current pbest of the
PI-Controller is a good controller in the field of machine agent, the pbest value is replaced by the current value. If
control, but the problem is the mathematical model of the best value of the pbest is better than the current
the plant must be known in order to solve the overall gbest, gbest is replaced by the best value and the agent
system. Several methods are introduced to tune the PI- number with the best value is stored.
Controller. Our proposed method is using PSO to Step 3 Modification of each searching point
optimize PI-controller parameters; the PSO algorithm is The current searching point of each agent is changing
used on-line to update the PI parameters as shown in using (2) and (3).
figure (1). In this figure there are three PI-controllers Step 4 Checking the exit condition
(PI1, PI2 and PI3) with six constants (Kp1 and Ki1, The current iteration number reaches the predetermined
Kp2 and Ki2, Kp3 and Ki3) that are tuned using PSO. maximum iteration number and then is exited.
Each variable from the six variables has three vectors as Otherwise, the process goes to step 2.
previous (position vector, previous best position vector The PI-controllers shown in figure (1) are tuned using
and velocity vector) vectors assumed of length 10 both Ziegler-Nichols and PSO methods and the results
elements [9]. A higher value of [ w ] favors global are shown through figure (3) to figure (5).
search while a lower value implies local search. We In figure (3) the system is tested under variations in
linearly decrease the value of w over generations to speed reference and constant load and both conventional
favor global search in initial generations and local and PSO methods are used to tune PI-controller
search in the later generations. parameters. In figure (4) a standard driving cycle which
At each iteration, the P vector of the particle with the contains both acceleration and deceleration is applied as
326 Journal of Power Electronics, Vol. 6, No. 4, October 2006

a reference speed with a constant load. Also PI is tuned

using both conventional and PSO methods; finally in
figure (5) the system is tested under a variation in load
and a constant reference speed.

(d) Electromagnetic torque response with Z-N

(a) Speed response

(e) Three-Phase currents with (Z-N)

Fig. 3 transient response of the drive system under

variation in speed reference (100 N.m) with
both (Z-N) and PSO methods

(b) Electromagnetic torque response with PSO

(c) Three-Phase currents with PSO (a) Speed response

 A Novel Technique for Tuning PI-Controllers … 327

(b) Three-Phase currents with PSO (b) Electromagnetic torque response with PSO

(c) Three-Phase currents with Z-N (c) Three-Phase currents with Z-N

Fig. 4 transient response of the drive system under Ramp

acceleration and Deceleration in speed reference
(100 N.m) with both (Z-N) and PSO methods

(d) Three-Phase currents with PSO

Fig. 5 transient response of the drive system under

sudden application of step change in speed
reference and load change from 100-200 N.m
(a) Speed response with both (Z-N) and PSO methods
328 Journal of Power Electronics, Vol. 6, No. 4, October 2006

9. Conclusions gbest : Global best value

Ia : Armature current
The overall system is simulated and tested when Ka : Torque constant
subjected to various operating conditions which are Ki : Integral Constant of PI-Controller
suitable for EV. The motor is controlled for both speed Kp : Proportional Constant of PI-Controller
and currents using PI-controllers which are tuned using pbest : personal best value
a PSO algorithm to optimize the PI parameters (Kp and Pgi : Global best value
Ki) for each controller instead of the traditional Z-N Pj : a personal best position in search space
method. r1 and r2 : are random numbers between 0 and 1
The system is first tested for a step change in the speed Tem : Electromagnetic torque of the motor
of the motor under constant load. As shown in figure Vi : current velocity
(3) the speed response of the tuned system using PSO is Vk : current velocity
faster and has lower overshot than applying the Z-N Vk+1 : modified velocity
method. The system is tested using a speed reference W : the inertia weight
that includes acceleration and deceleration (more Xj : current position in search space
suitable than for EV) under constant load and the Xk : current position in search space
response is also is faster and has lower overshot with Xk+1 : modified searching point
PSO. Finally the system is tested under a change in the φ (I f ) : Flux
load and a step change in the speed reference. From the
response of the drive system shown in figure (5) it is
Acknowledgment
obvious that the response of the system with PSO is
faster, has lower overshot and has lower steady state
Author gratefully thanks the staff of Journal of Power
error. From these results, the PSO succeeds in tuning
Electronics for their effort in order to support science,
on-line the PI-controller more efficiently than the
through this journal
traditional method, and shows a more dynamic
response.

References
APPENDIX (A)

Motor Parameters [1] Vas, Peter, Sensorless Vector and Direct Torque Control"
New York: Oxford University Press, 1998.
Power=37Kw
[2] R.W. De Doncker and D.W. Novotny, " The Universal
2P=2
Field Oriented Controller" IEEE Trans. On I.A., vol.30,
J=1.662 Kg.m2
Jan/Feb 1994, pp.92-100.
Β=0.1 N.m.s [3] T.G. Habetler, F.Projumo, M.Pastorelli, "Direct Torque
Lm=34.7mH Control of Induction Machine Over a Wide Speed
Lls=0.8 mH Range" IEEE-IAS Conf. Rec., 1991, pp.600-606.
[4] J. Holtz, “The Representation of AC Machine
L−lr = 0.8 mH
Dynamics by Complex Signal Flow Graphs,” IEEE Trans.
Rs=0.087 Ω Ind. Elect., Vol 42, No. 3, June 1995, pp. 263 – 271.
Rr- =0.228Ω [5] H Tajima, Y Hori, “Speed Sensorless Field Orientation
Control of the Induction Machine,” IEEE Trans. Industry
APPENDIX (B) Appl., Vol 29, No. 1, 1993 pp 175 – 180.
[6] S. E Lysherski, Electromechanical Systems, Electric
LIST OF SYMBOLS Machines and Applied Mechatronics, CRC Press, Boca
C1, C2 : Acceleration Coefficient Raton, 2000.
 A Novel Technique for Tuning PI-Controllers … 329

[7] A.Besharati, W.Lo,K.M Tsang " Self-Tuning of PID

Controller Using Newton-Raphson Search Method "
IEEE, Trans. Om I.E., vol.44, No.5,pp.717-725.
[8] R.C.Eberhart, J.Kennedy " A New Optimizer Using
Particle Swarm Theory" Proc. Of sixth International
Symposium On Micro Machine and Human Science
(Nagoya-Japan), IEEE Service center 1995, pp. 39-43.
[9] A.S.Elwer, S.Wahsh, O.Khalil and A.Nour eldin "
Intelligent Fuzzy Controller Using Particle Swarm
Optimization For Speed Control of PMSM " IECON
2003 conf. Proc. Roanoke, Virginia, USA, 2-6 Nov.
2003, pp.1762-1766.
[10] Y.H.Shi, and R.C.Eberhart " Parameter Selection in
Particle Swarm Optimization " The 7th, annual Conf., on
Evolutionary Programming, San Diego, USA, 1998.

Ayman Saber Elwer was born in

Kaliobya, Egypt, in 1972. He received the
B.S. in electrical engineering from
Zagazig University, Egypt 1995, with
hounr Degree, M.S. degree in electrical
engineering from Ain Shams University,
Egypt, in 2001, the Ph.D. degree in in electrical engineering
from Cairo University, Egypt, in 2005 he has been with the
Department of Power Electronics and Energy Conversion,
Electronics Research Institute, where he is currently an Ass.
Professor. His research interests are in Artificial Intelligence,
advanced control of electrical machines, and power
electronics.