S.Jaseena

Proceedings of the Second National Conference on Control and Power Engineering (CAPECON 2011) 02March2011

Department of Electrical and Electronics Engineering, Hindusthan College of Engineering and Technology,Coimbatore-641 032,India

181

MRAS Sensorless Vector Control Of Induction Motor Using New Sliding-

Mode and Neural Network Flux Observer

1Jaseena.S,

2 M.Dhivya

* 1 PG Student ,*

2 Research Scholar

Department Of Electrical and Electronics Engineering

Anna University Of Technology Coimbatore.

ABSTRACT-In this paper two novel

adaptation schemes are proposed to replace

the classical PI controller used in model

reference adaptive speed-estimation schemes

that are based on rotor flux. The first proposed

adaptation scheme is based on sliding-mode

control and the adaptation law is derived

using Lyapunov Theory to ensure estimation

stability, as well as fast error dynamics. The

other adaptation mechanism is based on

neural network .This new technique uses an

artificial neural network (NN) as a rotor flux

observer to replace the conventional voltage

model. Here a comprehensive study of the

MRAS, Sliding Mode Controller and Neural

Network has been formulated. The learning

algorithm for proposed scheme can be

planning to implemented using a Back

Propagation Algorithm which has been

successfully for a wide variety of applications.

KEYWORDS—Neural network, induction

motor, model reference adaptive control,

sliding-mode control.

I. INTRODUCTION

SEVERAL strategies have been proposed for

rotor speed estimation in sensorless induction

motor drives [1]. Among these techniques,

model reference adaptive systems (MRASs)

schemes are the most common strategies

employed due to their relative simplicity and

low computational effort [1], [2]. Rotor flux,

back electromotive force (EMF), and reactive

power techniques are popular MRAS

strategies that have received a lot of attention.

The back EMF scheme may have stability

problems at low stator frequency and show

low noise immunity, but avoids pure

integration. The reactive power method is

characterized by its robustness against stator

resistance variation while avoiding pure

integration, but suffers from instability [2], [3].

Therefore, rotor flux MRAS, first proposed by

Schauder is the most popular MRAS strategy,

and a lot of effort has been focused on

improving the performance of this scheme.

Generally, the main problems associated with

the low-speed operation of model-based

sensorless drives are related to machine

parameter sensitivity, stator voltage and

current acquisition, inverter nonlinearity, and

flux pure-integration problems [1], [5].

The model reference adaptive system (MRAS)

approach uses two models. The model that

does not involve the quantity to be estimated is

considered as the reference model. The model

that has the quantity to be estimated involved

is considered as the adaptive model (or

adjustable model). The output of the adaptive

model is compared with that of the reference

model, and the difference is used to drive a

suitable adaption mechanism whose output is

the quantity to be estimated. The adaptive

mechanism should be designed to assure the

stability of the control system. Different

approaches have been developed using

MRAS, such as rotor-flux-linkage estimation

based MRAS and back-EMF-based MRAS.

PI controllers are widely used in industrial

control systems applications. They have a

simple structure and can offer a satisfactory

performance over a wide range of operation.

Therefore, the majority of adaptation schemes

described in the literature for MRAS speed

observers employ a simple fixed-gain linear PI

controller to generate the estimated rotor

speed. However, due to the continuous

variation in the machine parameters and the

operating conditions, in addition to the

nonlinearities present in the inverter, fixed-

gain PI controllers may not be able to provide

the required performance. Adaptive control

techniques, such as gain scheduling, where the

PI gains vary with the operating conditions,

are often used to improve the controller

performance. Not much attention has been

devoted to study other types of adaptation



182

mechanisms used to minimize the speed

tuning signal to obtain the estimated speed.

In this paper two novel adaptation schemes are

proposed to replace the classical PI controller

used in model reference adaptive speed-

estimation schemes that are based on rotor

flux. The first proposed adaptation scheme is

based on sliding-mode control and the

adaptation law is derived using Lyapunov

Theory to ensure estimation stability, as well

as fast error dynamics. The other adaptation

mechanism is based on neural network .This

new technique uses an artificial neural

network (NN) as a rotor flux observer to

replace the conventional voltage model. Both

open and closed- loop sensorless operations

for the new schemes are investigated and

compared with the conventional MRAS speed

observer. Neural networks (NNs) have been

introduced as universal function

approximators to represent the functions with

the weighted sums of nonlinear terms.

Multilayer feedforward NNs have shown a

great capability to model complex nonlinear

dynamic systems. NNs were also combined

with MRAS for online stator and rotor

resistance estimation based on stator current

and rotor flux. The estimated speed represents

one of the NN weights updated online using a

back propagation algorithm.

II. LITERATURE REVIEW

A.ROTOR FLUX MRAS SPEED

OBSERVER

The classical rotor flux MRAS speed observer

shown in Fig.1 consists mainly of a reference

model, an adaptive model, and an adaptation

scheme that generates the estimated speed.

The reference model, usually expressed by the

voltage model, represents the stator equation.

It generates the reference value of the rotor

flux components in the stationary reference

frame from the monitored stator voltage and

current components. The reference rotor flux

components obtained from the reference

model are given by

{ }sDsDsssD

m

rrd piLiRv

L

Lp σψ −−=

(2.1)

{ }sQsQsssQ

m

rrq piLiRv

L

Lp σψ −−=

(2.2)

The adaptive model, usually represented by

the current model, describes the rotor equation

where the rotor flux components are expressed

in terms of stator current components and the

rotor speed. The rotor flux components

obtained from the adaptive model are given by

rqrrd

r

sD

r

mrd

Ti

T

Lp ψωψψ ˆˆˆ

1ˆ −−= (2.3)

rdrrq

r

sQ

r

m

rqT

iT

Lp ψωψψ ˆˆˆ

1ˆ +−= (2.4)

Finally, the adaptation scheme generates the

value of the estimated speed to be used in such

a way so as to minimize the error between the

reference and estimated fluxes. In the classical

rotor flux MRAS scheme, this is performed by

defining a speed tuning signal ωε to be

minimized by a PI controller, which generates

the estimated speed that is fed back to the

adaptive model. The expressions for the speed

tuning signal and the estimated speed can be

given as

rqrdrdrq ψψψψεωˆˆ −= (2.5)

( ) ωεωp

kk i

pr +=ˆ (2.6)

Fig.1 Classical rotor flux MRAS speed

observer

B. VECTOR CONTROL

By splitting the stator current into two

orthogonal components, one in the direction of

flux linkage, representing magnetizing current

or flux component of current, and other

perpendicular to the flux linkage, representing

the torque component of current, and then by



183

varying both components independently, the

induction motor can be treated as a separately

excited DC motor. The implementation of

vector control requires information regarding

the magnitude and position of the flux vector.

Depending upon the method of acquisition of

flux information, the vector control or field

oriented control method can be termed as:

direct or indirect. In the direct method the

position of the flux to which orientation is

desired is strictly measured with the help of

sensors, or estimated from the machine

terminal variables such as speed and stator

current/voltage signals.

The measured or estimated flux is used in the

feedback loop, thus the machine parameters

have minimal effect on the overall drive

performance. Also direct field orientation

method have its inherent problem at low speed

where the voltage drops due to resistances are

dominant, and pure integration is difficult to

achieve. The indirect vector control was

originally proposed in, eliminates the direct

measurement or computation of rotor flux

from the machine terminal variables, but

controls its instantaneous flux position by

summing the rotor position signal with a

commanded slip position signal (also known

as slip frequency control or feed forward

control scheme). The direction of rotor

position need an accurate rotor speed

information and the commanded slip position

is calculated from the model of the induction

motor, that again involves machine parameters

which may vary with temperature, frequency

and magnetic saturation.

To get ideal decoupling, the controller should

track the machine parameters and for this

various adaptation methods have been

proposed. However it has been reported that

the controller performance is adequate within

normal operating temperatures for most of the

high performance applications, and the

parameter adaptations methods may be

essential only in the case of critical

applications. In contrast to direct method the

indirect method controls the flux in an open

loop manner. Field orientation scheme can be

implemented with reference to any of the three

flux vectors: stator flux, air gap flux and rotor

flux. It has been shown that out of the three the

orientation with respect to rotor flux alone

gives a natural decoupling between flux and

torque, fast torque response and better

stability. Hence in this work orientation along

rotor flux is considered.

C. FLUX OBSERVER AND SPEED

ESTIMATION

There are many techniques involved in

implementing different types of field oriented

control. Most of the methods require precise

estimation of either the rotor position or speed.

The speed sensors increase the cost and size of

the drive, lower the system reliability, and also

require special attention to measure noise. The

estimation of rotor flux by integration of the

open loop machine voltages arise difficulties

at low speed. Finally, although the indirect

field orientation is simple and preferred, its

performance is highly dependent on accurate

knowledge of the machine parameters.

Research in induction motor control has been

focused to remedy the above problems. Much

work has been reported in decreasing the

sensitivity of the control system to the

parameter variation and estimating, rather than

measuring the rotor flux and speed from the

terminal voltages and currents. This eliminates

the flux or speed sensor, there by achieving

sensorless control. Many speed estimation

algorithms and speed sensorless control

schemes have been developed during the past

few years. One of the major problems with the

terminal quantities-based flux observers

designed in the past is their sensitivity to the

machine parameters, specifically, to rotor

resistance for the current model observer and

to stator resistance in case of the voltage

model flux observer. To overcome these

problems various control techniques have been

tried to improve the rotor flux estimation.

III. PROPOSED MODEL

A. SLIDING MODE

CONTROLLER

The sliding mode control is especially

appropriate for the tracking control of motors,



184

robot manipulators whose mechanical load

change over a wide range. SM control (SMC)

is a variable structure control with high

frequency discontinuous control action that

switches between several functions depending

on the system states. It can be one of the most

effective and robust control strategies, in

addition to its capability to cope with bounded

disturbance as well as model imprecision that

makes it suitable for robust nonlinear control

of induction motor drives. The principle of

SMC is to define a switching control law to

drive the nonlinear state trajectory onto a

switching surface and maintain this trajectory

sliding on this surface for all subsequent time.

The control law is defined based on Lyapunov

theory to guarantee the motion of the state

trajectory toward the sliding surface. This is

done by choosing a hitting control gain to

maintain the derivative of Lyapunov function

as always a negative definite. Advantages of

sliding mode controllers are that it is

computationally simple compared adaptive

controllers with parameter estimation and also

robust to parameter variations. The

disadvantage of sliding mode control is sudden

and large change of control variables during

the process which leads to high stress for the

system to be controlled. It also leads to

chattering of the system states.

B. SLIDING MODE MRAS SPEED

OBSERVER

The classical SM strategy applied for control

applications is modified here to fit with the

speed-estimation problem. Hence, a novel SM

rotor flux MRAS (MRAS-SM) is developed to

replace the conventional constant gain PI

controller. A new speed-estimation adaptation

law for the SM scheme is derived based on

Lyapunov theory to ensure stability and fast

error dynamics. Defining the speed tuning

signal (2.5) and choosing a sliding surface s as

0, >+= ∫ kdtks ωω εε

(3.1)

such that the error dynamics at the sliding

surface s = 0 will be forced to exponentially

decay to zero. When the system reaches the

sliding surface, this gives

(3.2)

and the error dynamics can be described by

ωω εε k−=& (3.3)

The SMC law can be found using Lyapunov

theory and defining the Lyapunov function

candidate as

2

2

1sv = (3.4)

According to Lyapunov theory, if the function

v& is negative definite, this will ensure that the

state trajectory will be driven and attracted

toward the sliding surface s, and once reached,

it will remain sliding on it until the origin is

reached asymptotically .The time derivative of

the Lyapunov function can be calculated as

( )ωω εε ksssv +⇔= &&& (3.5)

By solving the above equations finally gets

0),(ˆ2

1>+

+= MssignM

f

kfr

ωεω (3.6)

Equation (3.6) represents the switching law of

the SM controller and could be written in

general form as

seqr uu +=ω̂ (3.7)

where equ is the equivalent control that defines

the control action that keeps the state

trajectory on the sliding surface, su is the

switching control that depends on the sign of

the switching surface, and M is the hitting

control gain that makes a negative definite.

The block diagram of the novel MRAS

observer employing SM adaptation

mechanism (MRAS-SM) is shown in Fig.2

0=+= ωω εε ks &&



185

Fig. 2 MRAS-SM speed observer.

IV. NEURAL NETWORK

MRAS FLUX OBSERVER

Recently, the re-emerging artificial neural

networks (ANN) techniques have been widely

applied in the field of system identification

and control. In addition, they have been used

in some power electronic applications, such as

inverter current regulation, dc motor control,

flux estimation, and observer based control of

IM. Evidently, neural network techniques are

showing promise as a competitive method of

signal processing for power electronics

applications when they have the advantages of

extremely fast parallel computation, immunity

from input harmonic ripple, and fault tolerance

characteristics due to distributed network

intelligence.

Fig.3 Proposed NN-MRAS speed observer.

Neural network is well known for its learning

ability and approximation to any arbitrarily

continuous function. In this research, an

adaptive neural network controller is

developed for IM speed control which

constitutes a dynamic mapping. The learning

algorithm applied is the back propagation

which has been trained and replaced the

controller, it can be retained online for plant

parameter variation to make it adaptive. The

feed forward neural network is usually trained

by back propagation training scheme. With the

network initially untrained and its weights

selected at random, so an output signal is

obtained for a given input pattern. The actual

output is compared with the desired output and

the weights are adjusted by the supervised

back propagation training algorithm until the

errors become acceptably small. The fig.3

shows the proposed NN-MRAS speed

observer.

V SIMULATION RESULTS

A. VECTOR CONTROL OF AN

INDUCTION MOTOR Speed error is used to calculate the torque

reference and current is used to calculate the

flux reference. With the help of the torque

error and flux error reference Isabc is

calculated. The reference Isabc is compared

with the actual Isabc and trigger is given to the

inverter.From the figure we can see that speed

is inversely proportional to torque.



186

Fig.4 Vector control of an induction motor

using PI controller.

Fig .5 Performance measures of induction

motor Torque,Rotor speed and Stator

currents using PI controller.

B. SLIDING MODE

CONTROLLER

SM control (SMC) is a variable structure

control with high frequency discontinuous

control action that switches between several

functions depending on the system states. It

can be one of the most effective and robust

control strategies, in addition to its capability

to cope with bounded disturbance as well as

model imprecision that makes it suitable for

robust nonlinear control of induction motor

drives. The principle of SMC is to define a

switching control law to drive the nonlinear

state trajectory onto a switching surface and

maintain this trajectory sliding on this surface

for all subsequent time. The control law is

defined based on Lyapunov theory to

guarantee the motion of the state trajectory

toward the sliding surface. By using the

equations of sliding surface we can implement

the sliding mode controller.

Fig.6 Performance measure of induction

motor using sliding mode controller.

V. CONCLUSION In this paper two novel nonlinear adaptation

schemes are proposed to replace the fixed gain

PI controller,which is conventionally used for

rotor flux MRAS observer. One of these

schemes is based on SM theory where a novel

speed estimation adaptation law is derived

based on Lyapunov theory to ensure

estimation stability with fast error dynamics.

This paper has presented an entirely new

application of an NN to give an improved

MRAS speed observer scheme suitable for

speed sensorless induction motor drives. A

multilayer feedforward NN estimates the rotor

flux components from present and past

samples of reference stator voltages and

measured currents. The new scheme making

use of the offline trained NN observer as a

reference model in MRAS scheme. Firstly we

have to model Sliding Mode control of an

induction motor using MATLAB SIMULINK.

This work has been extended to MRAS

sensorless speed control of an induction motor

using Neural Network Flux Observer.

VI. REFERENCES [1] Shady M. Gadoue, Damian Giaouris, and

John W. Finch, “MRAS Sensorless Vector

Control of an Induction Motor Using New

Sliding-Mode and Fuzzy-Logic Adaptation

Mechanisms,” IEEE Transactions On Energy

Conversion, VOL. 25, NO. 2, June 2010

[2] Shady M. Gadoue, Damian Giaouris, and

John W. Finch, “Sensorless Control of

Induction Motor Drives at Very Low and

Zero Speeds Using Neural Network Flux

Observers,” IEEE Trans.On Industrial

Electronics, VOL. 56, NO. 8, AUGUST 2009

[3] Y. A. Kwon and D.W. Jin, “A novel

MRAS based speed sensorless control of

induction motor,” in Proc. 25th Annu. Conf.

IEEE Ind. Electron. Soc., 1999, pp. 933–938.

[4] Yemna BENSALEM and Mohamed

Naceur Abdelkrim, ” A Sensorless Neural

Model Reference Adaptive Control for

Induction Motor Drives,” 2009 International

Conference on Signals, Circuits and Systems.



187

[5] S. Maiti, C. Chakraborty, Y. Hori, and M.

C. Ta, “Model reference adaptive controller-

based rotor resistance and speed estimation

techniques for vector controlled induction

motor drive utilizing reactive power,” IEEE

Trans. Ind. Electron., vol. 55, no. 2, pp. 594–

601, Feb. 2008.

[6] J. W. Finch and D. Giaouris, “Controlled

AC electrical drives,” IEEE Trans. Ind.

Electron., vol. 55, no. 1, pp. 1–11, Feb. 2008.

[7] P. Vas, Sensorless Vector and Direct

Torque Control. New York: Oxford Univ.

Press, 1998.

[8] V. Utkin, “Sliding mode control design

principles and applications to electric drives,”

IEEE Trans. Ind. Electron., vol. 40, no. 1, pp.

23–36, Feb. 1993.

[9] M. Comanescu and L. Xu, “Slidingmode

MRAS speed estimators for sensorless vector

control of induction machine,” IEEE Trans.

Ind. Electron., vol. 53, no. 1, pp. 146–153,

Feb. 2006.

[10] P. Vas, Artificial-Intelligence-Based

Electrical Machines and Drives- Application

of Fuzzy, Neural, Fuzzy-Neural and Genetic

Algorithm Based Techniques. New York:

Oxford Univ. Press, 1999.

[11] B. Karanayil, M. Rahman, and C.

Grantham, “Stator and rotor resistance

observers for induction motor drive using

fuzzy logic and artificial neural networks,”

IEEE Trans. Energy Convers., vol. 20, no. 4,

pp. 771–780, Dec. 2005.

[12] M. S. Zaky, M. M. Khater, S. S.

Shokralla, and H. A. Yasin, “Widespeed-

range estimation with online parameter

identification schemes of sensorless induction

motor drives,” IEEE Trans. Ind. Electron., vol.

56, no. 5, pp. 1699–1707, May 2009.

APPENDIX

QsDs ii ,

Stator current components in the

stator frame.

J Rotor inertia.

mL Mutual inductance.

rs LL , Stator and rotor self-inductances.

p Differential operator.

rs RR , Stator and rotor resistances.

rT Rotor time constant.

sQsD vv ,

Stator voltage components in the

stator frame

ωε Speed tuning signal.

σ Leakage coefficient.

rqrd ψψ , Components of the rotor flux

linkage vector.

rω Angular rotor speed.

S.Jaseena

Documents

Transcript of S.Jaseena