Elsevier

Neurocomputing

Volume 73, Issues 4–6, January 2010, Pages 602-612
Neurocomputing

Stator resistance identification based on neural and fuzzy logic principles in an induction motor drive

https://doi.org/10.1016/j.neucom.2009.06.017Get rights and content

Abstract

This paper presents a method for stator resistance identification of an induction motor in an indirect rotor field-oriented control system. This method is based on a simple artificial neural network, in which the rotor time constant is no longer considered to be a constant parameter, but is instead identified using an adaptive model reference system-based procedure. The neural network outputs the estimated rotor speed. The difference between the actual and the estimated rotor speed is used as a signal for either manual or automated fuzzy logic stator resistance identification. Simulations and experiments show the effectiveness of the described approach.

Introduction

An induction motor is the most commonly used electric motor in modern electric drives requiring wide ranges of both speed and power. In addition, induction motor control systems are known to be non-linear control systems because of induction motor parameter variability under different conditions. The heating of motor windings depends on stator and rotor currents leading to variability of stator and rotor resistances. Variable mutual inductance is a consequence of different magnetizing flux levels of the motor. With the development of digital signal processors (DSPs) and power electronics, induction motors can now be used in high-performance variable-speed drives. Some of these drives are based on an indirect rotor field-oriented (IRFO) method. In these cases, the magnetizing level of the induction motor is determined by the rotor flux reference ψr*. In the base speed region, the rotor flux reference is, in most cases, constant [11]. The accuracy and control quality of the IRFO control system is greatly influenced by the value of the rotor resistance Rr used for control. In certain cases, such as overload, the stator current can be much higher than rated, resulting in a significant increase of the rotor resistance. In comparison with a cold motor, it may vary up to 100%.

In the past, several methods have been developed for rotor resistance identification. In the base speed region with a constant magnetizing level of the induction motor, the rotor resistance identification is synonymous with inverse rotor time constant identification. A brief review of the methods for rotor resistance identification is in [17] and includes the following technique classification:

These methods include signal injection in the stator voltage or current with a specific harmonic spectrum. The rotor resistance can be obtained from the spectral analysis of the stator current or stator voltage measurements [2].

The first paper that involved an observer-based technique was Verghese and Sanders [19]. In the previous years, several proposed methods using extended Kalman filters or extended Luenberger observers have been developed to identify the rotor resistance in induction machines [7], [23], many of which include rotor speed estimation. The main drawback of these methods is the fact that the inductances are considered constant.

Model reference adaptive system (MRAS) methods have often been used to identify the rotor resistance due to their simple implementation. The basic idea is that one quantity (vector or scalar) can be calculated in two different ways. One of the two obtained quantities is independent of the rotor resistance. The difference between the two quantities is an error signal, whose existence is assigned to the error in the rotor resistance. The error signal is applied to drive an adaptive mechanism (proportional-integral or integral) which provides correction of the rotor resistance. In most cases, the adaptation process does not work at zero rotor speed and at zero load torque. Ref. [14] shows how the identification procedure can be carried out independent of the stator frequency and the load torque.

There are other methods that are not based on the previous techniques, such as those based on artificial intelligence. Neural networks and fuzzy logic schemes are two examples. In recent years, the use of artificial neural networks (ANNs) in AC drives has been proposed [9], [6], [3], [4]. Advanced control based on artificial intelligence techniques is called intelligent control. Unlike classical control, an intelligent control strategy may not need a mathematical model of the plant.

The most commonly used ANNs are feedforward multilayer networks, where no information is fed back during application. Supervised learning methods, where the neural network is trained to learn the input/output pattern presented to it, are typically used [10]. Most often, a back propagation training procedure is used to adjust the neural network weights. This is generally slow, since the algorithm takes a long time to converge. However, there are the neural networks that consist of an input layer, a layer of input nodes, and one output layer consisting of neurons. This is referred to as a single-layer neural network because the input layer is not a layer of neurons, i.e., no computations occur at the input nodes. This single-layer neural network is called a perceptron, with no hidden layer neurons. Sometimes, this neural network type is known as the two-layer neural network as in [3], [20]. Single-layer neural networks, where the off-line training step is not required, are, therefore, preferable. This type of neural network may learn on-line [15], [16]. Because of this feature, a single-layer ANN is utilized in this paper.

This paper deals with an IRFO control system, including inverse rotor time constant identification and stator resistance adjustment. In this paper, the rotor flux space vector is estimated using four different types of induction motor models (so called voltage and current models), and each is described in the stationary reference frame (α, β) or in a synchronously rotating reference frame (d, q). The stationary reference frame is only introduced to identify the inverse rotor time constant as described in Section 3. The rotor field orientation is carried out in the synchronously rotating reference frame. The single-layer ANN, utilized in this paper, is designed in the same reference frame. First, the rotor flux magnitude is estimated in the stationary reference frame by the voltage model (reference model) and by the current model (adaptive model). The error signal of the rotor flux magnitude of the two estimators is applied to drive a PI mechanism which provides correction of the inverse rotor time constant. Compared with the method described in [12], the stator resistance is not a constant parameter, but is identified as described later. Second, the rotor flux space vector is estimated in the d, q reference frame by the voltage model (reference model) and by the single-layer ANN model (adaptive model). The errors between the generated rotor flux components are applied to drive a PI estimator which provides rotor speed estimation. The ANN is simple because it contains only an input layer and an output layer. The ANN contains two weights dependent on the inverse rotor time constant and one weight dependent on the rotor speed. The weights dependent on the inverse rotor time constant are tuned based on the inverse rotor time constant identification as described in this section. The weights dependent on the rotor speed are tuned based upon the estimated rotor speed. Adjustment of these weights is performed in such a way that the error between actual and estimated rotor speed converges quickly to zero. Finally, the accuracy of the inverse rotor time constant identification, as well as the ANN rotor speed estimation, depends on the stator resistance. Because of that, any mismatch between actual and estimated rotor speed will result from inaccurate stator resistance.

In this work, the stator resistance is adapted manually first and then automatically by using a fuzzy logic principle, in order to achieve zero error between the actual and estimated rotor speed. The adaptation is based on a simple rule: if the estimated rotor speed is higher than the actual rotor speed then the identified stator resistance should be decreased, and vice versa. The rule was established experimentally, by observing the impact of the identified stator resistance variation on the speed error. This simple rule is correct only if the inverse rotor time constant and the motor inductances are properly identified. In this paper, the inverse rotor time constant is identified and the magnetizing flux level of the induction motor is constant. Constant mutual inductance is a consequence of the fixed magnetizing flux level in the motor. The stator and rotor leakage inductances are constant in normal regimes of the IRFO control system.

Simulation and experimental results over a wide range of rotor speed are presented.

Section snippets

IRFO control system

Fig. 1 shows the IRFO control system of induction motor, including both inverse rotor time constant identification and stator resistance identification.

An induction motor can be described by the following equations in a synchronously rotating (d, q) reference frame [11]:us=isRs+dψsdt+jωeψs0=irRr+dψrdt+j(ωeωr)ψrJdωrdt=32P2Lm4Lr|ψr×is|RωωrP2tlψs=Lsis+Lmirψr=Lmis+Lrir,where us, is, ir, ψs, ψr denote space vectors of the stator voltage, stator current, rotor current, stator flux, and rotor flux,

Identification of inverse rotor time constant

There are many methods for estimation of the (inverse) rotor time constant. One group of on-line rotor time constant adaptation methods is based on the principles of MRAS. This is the approach with relatively simple implementation requirements.

This paper utilizes the identification of inverse rotor time constant described in [12] with a small modification of (7) in order to eliminate the DC drift as suggested in [22]. Replacing the actual rotor flux space vector ψr with the estimated rotor flux

Stator resistance tuning based on ANN

The MRAS theory, as described in Section 3, is utilized in order to estimate the rotor speed of induction motor. The rotor flux space vector is estimated in the d, q reference frame by the voltage model (reference model) and by the ANN-based model (adaptive model) of the induction motor. Conventionally, the current model is used as the adaptive model because it is the rotor speed-dependent one. The difference between flux space vectors estimated using the two ways is then used in an adaptive

Simulation results

A simulation program of the complete control system in MATLAB-Simulink environment is developed. This program deals with an idealized pulse width modulation (PWM) inverter, i.e. it contains just the first harmonic of the stator voltage. The frequency of the first harmonic ωe is determined by the IRFO control system as shown in Fig. 1.

Simulation results are obtained when the stator resistance is manually tuned (Section 5.1) and when the stator resistance is identified by fuzzy logic principles (

Experimental results

In order to compare the theory with a test case, a control algorithm was executed on the dSpace DS1104 microcontroller board. The other parts of the experimental setup of the field-oriented control system were an induction motor rated at 1.5 kW, a dynamometer test bench rated at 5 kW, a PWM inverter with IGBT transistors (SKM100GB123D, made by Semikron), an encoder with 1800 pulses and the PC computer. The frequency of space vector PWM modulation was 2.5 kHz. The dead time of the frequency

Conclusion

A study of the IRFO control system of an induction motor including deviations in the stator resistance has been carried out. MRAS-based identification of the inverse rotor time constant at a rated magnetizing level was included in the observed system. The identified inverse rotor time constant is an input parameter for a single-layer ANN. The ANN outputs components of the rotor flux space vector in the synchronously rotating reference frame and the estimated rotor speed. Here, the ANN presents

Dinko Vukadinovic was born in Banja Luka, Bosnia and Herzegovina, in 1973. He received his B.E. degree from the University of Split, his M.E. degree from the University of Zagreb and his Ph.D. degree from the University of Split, Croatia, in 1997, 2002 and 2005, respectively, all in electrical engineering. He became a junior researcher in the University of Split, Faculty of Electrical Engineering, Mechanical Engineering and Naval Architecture, Department of Electric Power Engineering, in 1998.

References (23)

  • L. Baghli et al.

    Signal analysis and identification for induction motor sensorless control

    Control Engineering Practice

    (2006)
  • P. Roncero-Sanchez et al.

    Rotor-resistance estimation for induction machines with indirect-field orientation

    Control Engineering Practice

    (2007)
  • A.G. Aissaoui et al.

    A fuzzy logic controller for synchronous machine

    Journal of Electrical Engineering

    (2007)
  • L. Ben-Brahim et al.

    Speed control of induction motor without rotational transducers

    IEEE Transaction on Industry Application

    (1999)
  • B.K. Bose, Intelligent control and estimation in power electronics and drives, in: IEEE International Electric Machines...
  • B.K. Bose

    Modern Power Electronics and AC Drives

    (2002)
  • B. Burton et al.

    Implementation of a neural network to adaptively identify and control VSI-fed induction motor stator currents

    IEEE Transaction on Industry Application

    (1998)
  • T. Du et al.

    Design and application of extended observers for joint state and parameter estimation in high-performance ac drives

    IEE Proceedings—Electric Power Applications

    (1995)
  • Z. Ibrahim et al.

    A comparative analysis of fuzzy logic and pi speed control in high-performance AC drives using experimental approach

    IEEE Transaction on Industry Application

    (2002)
  • S.K. Mondal et al.

    A neural network based space-vector PWM controller for a three voltage-fed inverter induction motor drive

    IEEE Transaction on Industry Application

    (2002)
  • H.T. Nguyen et al.

    A First Course in Fuzzy and Neural Control

    (2003)
  • Cited by (20)

    View all citing articles on Scopus

    Dinko Vukadinovic was born in Banja Luka, Bosnia and Herzegovina, in 1973. He received his B.E. degree from the University of Split, his M.E. degree from the University of Zagreb and his Ph.D. degree from the University of Split, Croatia, in 1997, 2002 and 2005, respectively, all in electrical engineering. He became a junior researcher in the University of Split, Faculty of Electrical Engineering, Mechanical Engineering and Naval Architecture, Department of Electric Power Engineering, in 1998. In 2006 he became an Assistant Professor in the University of Split. His research interests include induction motor control systems, digital signal processors and artificial neural networks. Dr. Vukadinovic is a member of the KES International.

    Mateo Basic was born in Split, Croatia, in 1982. He received his B.E. degree from the University of Split, Croatia, in 2006, in electrical engineering. He became an assistant in the University of Split, Faculty of Electrical Engineering, Mechanical Engineering and Naval Architecture, Department of Electric Power Engineering, in 2008. His research interests include induction motor control systems.

    Ljubomir Kulisic was born in 1948 in Zadar, Croatia. He received his B.E. degree from the University of Split and his M.E. degree from the University of Zagreb, Croatia, in 1974 and 1995, respectively, both in electrical engineering. He became an assistant in the University of Split, Faculty of Electrical Engineering, Mechanical Engineering and Naval Architecture, Department of Electric Power Engineering, in 1974. Since 2003 he has been a Senior Lecturer. His research interests are power electronic converters, simulation of power semiconductor devices and control of high performance drives. Mr. Kulisic is a member of the Croatian Society for Communications, Measurement, Electronic and Automation (KoREMA) and the Institute of Electrical and Electronic Engineers (IEEE).

    View full text