Online Stator and Rotor Resistance Estimation Scheme Using Artificial ...

IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 54, NO. 1, FEBRUARY 2007

167

Online Stator and Rotor Resistance Estimation Scheme Using Artificial Neural Networks for Vector Controlled Speed Sensorless Induction Motor Drive Baburaj Karanayil, Member, IEEE, Muhammed Fazlur Rahman, Senior Member, IEEE, and Colin Grantham

Abstract—This paper presents a new method of online estimation for the stator and rotor resistances of the induction motor for speed sensorless indirect vector controlled drives, using artificial neural networks. The error between the rotor flux linkages based on a neural network model and a voltage model is back propagated to adjust the weights of the neural network model for the rotor resistance estimation. For the stator resistance estimation, the error between the measured stator current and the estimated stator current using neural network is back propagated to adjust the weights of the neural network. The rotor speed is synthesized from the induction motor state equations. The performance of the stator and rotor resistance estimators and torque and flux responses of the drive, together with these estimators, are investigated with the help of simulations for variations in the stator and rotor resistances from their nominal values. Both resistances are estimated experimentally, using the proposed neural network in a vector controlled induction motor drive. Data on tracking performances of these estimators are presented. With this speed sensorless approach, the rotor resistance estimation was made insensitive to the stator resistance variations both in simulation and experiment. The accuracy of the estimated speed achieved experimentally, without the speed sensor clearly demonstrates the reliable and high-performance operation of the drive. Index Terms—Artificial neural networks (ANNs), induction motor drives, parameter identification, speed sensorlees vector control.

I. INTRODUCTION NDIRECT field oriented vector controlled induction motor drives are widely used in industrial applications for high-performance drive systems. Because indirect field orientation utilizes an inherent slip relation, it is essentially a feedforward scheme, and hence naturally parameter sensitive, particularly to the rotor resistance. The accuracy of the estimated rotor flux is greatly influenced by the value of rotor resistance ( ) used for control. Rotor resistance may vary up to 100% due to rotor heating and recovering this information with a thermal model or a temperature sensor is not desirable. In addition, rotor resistance can change significantly with rotor frequency due to skew/

I

Manuscript received November 20, 2004; revised May 17, 2005. Abstract published on the Internet November 30, 2006. B. Karanayil is with the Power Electronics and Drives Group, University of New South Wales, Sydney NSW 2052, Australia. M. F. Rahman is with the School of Electrical Engineering and Telecommunications, University of New South Wales, Sydney NSW 2052, Australia (e-mail: [email protected]). C. Grantham is with the School of Electrical Engineering and Computer Science, University of New South Wales, Sydney NSW 2052, Australia. Digital Object Identifier 10.1109/TIE.2006.888778

proximity effect in machines with double-cage and deep-bar rotors. A mismatch between the actual rotor flux and the estimated rotor flux which occurs as a result of rotor resistance variation also leads to error between the actual motor torque and the estimated torque and hence to poor dynamic performance. The problem related to rotor resistance adaptation has been reviewed extensively in [1]. Several methods have been reported to minimize the consequences of parameter sensitivity in indirect vector controlled drives. The model reference adaptive control based on torque function reported in [2], could not be used for low frequencies due to the stator resistance voltage drop. Three simple methods of rotor time constant adaptation with reference adaptive control structure was proposed using a torque reference model, reaxis voltage reference active power reference model, and models [3]. Its major drawback was its dependence on stator resistance. The method proposed in [4] was based on the proper selection of coordinate axes, namely, the axis of the rotating frame to be coincident with the stator current vector. However, this method is only valid under the condition of steady–state operation of the motor. The identification algorithm reported in [5] was to minimize the error between the measured stator current trajectory and an estimated current trajectory, the update is done only every 60 s, as the method was largely computational intensive. The methods discussed in [6]–[8] are based on model reference adaptation of either flux or reactive power. The second approach, developed in [9] and [10], was to compensate for rotor resistance variation by adaptive feedback linearization control with unknown rotor resistance. The third identification method is to detect the output signal variation invoked by the artificial injection signal [11]. Also, an extended Kalman filter was used for rotor resistance identification in [12] and [13]. All of these methods assumed that there is no change in the stator resistance during the rotor resistance estimation. Majority of the speed estimation schemes utilize an induction motor model for the speed estimation and require the accurate knowledge of most of the motor parameters [14]. The accurate value of the stator resistance is critical for the correct operation of a speed sensorless drive, since an error between its actual value and the value used in the speed estimator may lead not only to a substantial speed error but also to unstable operation of the drive. As a solution to this problem, many online stator resistance estimation schemes have been proposed [15]–[19]. All of the estimation schemes reported in [15]–[19] have used a model reference adaptive approach with a proportional-integral or integral controllers are used for this purpose.

0278-0046/$25.00 © 2007 IEEE

168


Fig. 1. Parameter identification using neural networks.

Fig. 2. Structure of the neural network system for R estimation.

In recent years, the use of artificial neural networks (ANNs) for identification and control of nonlinear dynamic systems in power electronics and ac drives have been proposed [20]–[22], as they are capable of approximating wide range of nonlinear functions to a high degree of accuracy. In this paper, the capability of a neural network has been deployed to have online estimators to address the situation of similar disturbances in both stator and rotor resistances simultaneously. Section II describes an online estimation of rotor resistance ( ) with multilayer feedforward ANN using online training [23]. Multilayer feedforward neural networks are regarded as universal approximations and have the capability to acquire nonlinear input-output relationships of a system by learning via the backpropagation algorithm [24], [25]. It should be possible that a simple twolayer feedforward neural network trained by the backpropagation technique can be employed in the rotor resistance identification. In this estimator, two models of the state variable estimations can be used; one to provide the actual induction motor output states and the other to give the neural model output states. The total error between the desired and actual state variables may then be back propagated as shown in Fig. 1, to adjust the weights of the neural model, so that the output of this model tracks the actual output. When the training is completed, the weights of the neural network should correspond to the paramestimation algorithm eters in the actual motor. However the requires the knowledge of stator resistance ( ) which may also vary up to 50% during operation. It has been observed that the , leads to significant errors in estimation. It is error in hypothesized in this paper that the problem may be overcome to the system using by adding another online estimation for recurrent neural network, discussed in Section III, giving the indirect vector control system, total immunity to both resistance variations. The proposed stator resistance observer was realized with a recurrent neural network trained using the standard backpropagation learning algorithm. The recurrent neural network with feedback loops used in this paper is trained by standard backpropagation algorithm. Such an architecture is known to be a more desirable approach [26] and the implementation reported in this paper confirms this. The rotor and stator resistance estimators described in Sections II and III are investigated by modeling studies using SIMULINK and the results are discussed in Section V. The

proposed resistance estimators are also tested in an experimental set-up for a 1.1 kW squirrel cage induction motor. The results are discussed in detail in Section VI. II. ROTOR RESISTANCE ESTIMATION USING ANNS The basic structure of an adaptive scheme described by Fig. 1 is extended for rotor resistance estimation of an induction motor, as illustrated in Fig. 2. Two independent observers are used to estimate the rotor flux vectors of the induction motor. Equation (1) is based on stator voltages and currents, which is referred as voltage model of the induction motor and (2) is based on stator currents and rotor speed, which is referred as current model of the induction motor (1) (2) The current model (2) can also be written, as in (3) (3) where

,

,

;

, The sample-data model of (3) is shown in (4) (4) ; ; where Equation (4) can also be written as:

.

(5) where

;

;

KARANAYIL et al.: ONLINE SRE AND RRE SCHEME USING ANNS FOR VECTOR CONTROLLED SPEED SENSORLESS INDUCTION MOTOR DRIVE

169

Fig. 3. Two-layered neural network model for rotor flux estimation.

The neural network model represented by (4) is shown in , , and represent the weights of the netFig. 3, where , , and are the three inputs to the network. works and is alIf the network shown in Fig. 3 is used to estimate , and need to be updated. ready known and The standard backpropagation learning rule is then employed and to train the neural network. The weights of the network, are found from training, so as to minimize the cumulative error function (6) and the weight adjustment using generalized delta rule is given by (7) To accelerate the convergence of the error backpropagation learning algorithm, the current weight adjustments is supplemented with a fraction of the most recent weight adjustment, as in (8)

Fig. 4. Relationship between variation of stator current with R .

introduces an error in the estimate of . This variation in estimation. In order to miniprovides the incentive for the mize the error in rotor resistance estimation, resulting from the stator resistance variation, an online stator resistance estimator is integrated, which is discussed in Section III. The effect of the change in stator resistance on the estimation of rotor resistance is explained with Fig. 14 and its effect on the estimated speed is explained with Fig. 17 in Section VI. III. STATOR RESISTANCE ESTIMATION WITH ANNS The voltage and current model equations of the induction motor, (1) and (2) in Section I, can also be written as

(8) where is the training coefficient, and positive momentum constant and

(12)

is a user-selected (13) Using the discrete form of (12)

Similarly, the changes in

can be determined as follows: (9)

(14) where

; ;

can be calculated from either The rotor resistance from (10) or from (11) (10) (11) The rotor resistance estimator (RRE) described in this section has used the rotor fluxes , derived from the voltage model of the induction motor. These rotor fluxes are deof the motor [see (1)]. Modpendent on the stator resistance eling results in Section V clearly shows that maximum possible

;

. , , and are calculated from the motor The weights parameters, motor speed , and the sampling interval . To examine the effect of stator resistance variation in the amplitude of stator current, modeling studies were carried out with a ramp change in stator resistance with the drive operated with RFOC. The stator current profile is shown in Fig. 4. Initially, the stator resistance is increased from its nominal value of 6.03 –8.0 and the stator current now drops, as indicated by the is deforward arrow in Fig. 4. Later the stator resistance creased from 8 to 6.03 and the stator current now increases

170

Fig. 5. on (14).


D axis stator current estimation using recurrent neural network based

Fig. 6. on (18).

Q axis stator current estimation using recurrent neural network based

to a different value, as indicated by the reverse arrow. The relationship between stator current and stator resistance is nonlinear which could be easily mapped using a neural network. Equation (14) can be represented by a recurrent neural network, as shown in Fig. 5. The standard backpropagation learning rule is then employed to train the network. The weight is the result of training so as to minimize the cumulative error function (15) The weight adjustment for

is given by (16)

To accelerate the convergence of the error backpropagation learning algorithm, the current weight adjustments is supplemented with a fraction of the most recent weight adjustment, as in (17) (17) where is the training coefficient, is a user-selected positive momentum constant. Similarly, using the discrete form of (13)

(18) Equation (18) can be represented by a neural network, as shown in Fig. 6. can be calculated from (19) The stator resistance

(19) The stator resistance of an induction motor can be, thus, estimated from the stator current using the neural network system as indicated in Fig. 7. Having estimated the two most critical parameters and , the induction motor speed can be easily estimated with the state

Fig. 7.

R

estimation using ANN.

equations, with very good accuracy. The speed estimation algorithm is explained in Section IV and their associated errors are explained together with the experimental results in Section VI. IV. SPEED ESTIMATION The methods of obtaining the rotor and stator resistances are already explained in Sections III and IV. The rotor speed can be synthesized from the induction motor state equations, and can be written as

(20) The flux linkages , are dependent on as dicdepends tated by (1). Also, the estimated stator resistance on the estimated rotor resistance , as shown in (19). Equation (20), thus, depends significantly on the variation of motor parameters, however, as the two critical parameters for flux esand are modified by adaptation techniques, the timation speed estimation was found to be accurate as demonstrated by the experimental results discussed in Section VI.


171

TABLE I INDUCTION MOTOR PARAMETERS

Fig. 8. Block diagram of the indirect vector controlled induction motor drive with online stator and rotor resistance tracking.

V. ANALYSIS OF RESULTS A. Modeling Results The block diagram of a rotor flux oriented induction motor drive, together with both stator and rotor resistance identifications, is shown in Fig. 8. The induction motor is controlled with a rotor flux oriented vector controller, as shown in this figure. The voltage model fluxes are estimated from the measured stator voltages and currents using a programmable cascaded low-pass filter (PCLPF) [27]. The stator voltages are PWM voltages and are filtered with hardware filters and only the sinusoidal voltages are taken into the PCLPF flux estimators. The rotor resistance is estimated using ANNs by the RRE block, as described in Section II. The stator resistance is estimated by the stator resistance estimator (SRE) block described in Section III. The estimation depends on , as given by (12)–(14). The rotor flux linkages , used in estimating the stator resistance is already compensated by the RRE, and hence the estimation error in is avoided. The use of ANNs in identification algorithms in Sections II and III are verified by simulations with the help of SIMULINK. In order to investigate the performance of the drive for parameter variations in rotor resistance , a series of simulations were conducted by introducing error between the actual value and the value used in the controller. Similarly, another series of simulations were conducted by introducing error between the actual stator resistance and the one used in the controller . All these investigations were conducted for the drive running at 1000 rev/minute and with a constant load torque of 7.4 Nm. The parameters of the motor used for modeling studies are in Table I. 1) With RRE and SRE Off: Initially, a 40% error was introduced between and , and and simultaneously at 1.5 s, after switching off both the RRE and SRE blocks in Fig. 8.

The steady-state values of the torque and rotor flux linkages are shown in Fig. 9(a). The estimated torque is 4% lower than the actual motor torque, as shown in Fig. 9(a) (ii). The rotor flux linkage in the motor increases by 21% compared with its estimated value, when the error in rotor resistance is introduced, as shown in Fig. 9(a) (iv). 2) With RRE On and SRE Off: Later, simulations were repeated after switching only the RRE block on with the SRE block switched off, for the same errors introduced in Fig. 9(a). estimated in this case is higher than the by 1.7%, The as shown in Fig. 9(b)(i). The estimated torque is 1.35% higher than the real motor torque, as shown in Fig. 9(b) (ii), but the estimated rotor flux linkage is 1.5% lower than the actual rotor flux linkage, as indicated in Fig. 9(b) (iii). 3) With RRE On and SRE On: Finally, simulations were carried out with both the RRE and SRE blocks switched on. The dynamic torque and rotor flux linkage are shown for both of the cases in Fig. 9(b). The error reported in the previous paragraph, between estimated and real torques and rotor flux linkages have largely disappeared in this case. The estimated rotor resistance has tracked the real rotor resistance of the motor very well, as the estimation error now drops to 0.3%, as in Fig. 9(b) (i). However, there is a small but insignificant error of 4.4%, as shown Fig. 9(b) (iv), for the estimated stator resistance with respect to the real stator resistance. Fig. 9(a) and (b) describe the possible steady-state errors encountered in a situation where a step change in resistance is applied, only for the purpose of investigation and verification of the technique. However, the practical variation in resistances is very slow, and a corresponding analysis is also carried out, and the results are indicated in Fig. 10. The simulations are done in three steps. At first, the drive system is analyzed after introand , and and keeping both ducing error between RRE and SRE turned off. Repeated simulations were also carried out, with RRE on and SRE off. The estimated in this case by 1.1%, as shown in Fig. 10(i). The estiis higher than the mated torque is 1.3% higher than the real motor torque, as shown in Fig. 10(ii), but the estimated rotor flux linkage is 1.5% lower than the actual rotor flux linkage, as indicated in Fig. 10(iii). Finally, both rotor and stator resistance estimators are investigated with both RRE and SRE switched on. The estimated rotor resistance has tracked the real rotor resistance of the motor very well, as the error now drops to 0.3%, as in Fig. 10(i). However there was a small but insignificant error of 5%, as shown with respect to Fig. 10(iv), for the estimated stator resistance the stator resistance , but its effect on the rotor flux oriented

172


Fig. 10. Performance of the induction motor drive with a ramp change in stator and rotor resistances with and without RRE and SRE.

Fig. 11. Experimental setup for the resistance identification in induction motor drive.

B. Experimental Results

Fig. 9. Performance of the drive with and without rotor and stator resistance compensations. (a) 40% step change in R and R , R R uncompensated. (b) 40% step change in R and R , R R compensated.

&

&

control is negligible, as the errors between torques and rotor flux linkages are virtually eliminated.

In order to verify the proposed stator and rotor resistance estimation algorithms, a rotor flux oriented induction motor drive was implemented, as shown in Fig. 11, in the laboratory. The experimental setup is built around a dSPACE DS1104 controller board residing in PC. An IGBT inverter with a switching frequency of 5 kHz is used for driving the induction motor. Hand coded C programs with the real-time reference library functions are used to develop the control programs. The current and flux sampling interval controllers were implemented with 100 and the speed controller with 500 . The proposed rotor resistance estimation block used 1000 sampling time and the stator resistance estimation block used 100 . An encoder with 5000 pulses per revolution is used for position and speed feedbacks. A permanent magnet DC motor coupled to the induction motor


173

^ in experiment. Fig. 14. Estimated stator resistance R Fig. 12. Estimated resistances R and R in the experiment.

^ estimation. Fig. 15. Stator currents in R

Fig. 13. Rotor fluxes in R estimation.

is used to load the induction motor. A constant load torque is maintained by using the current control loop in the load circuit. In order to examine the capability of tracking the rotor resistance of the induction motor with the proposed estimator, a temperature rise test was conducted at an ambient temperature of 25 C. The motor was running at a constant speed of 1000 rev/min and with a load torque of 7.4 Nm. The results of esand estimated stator resistance timated rotor resistance collected from the experiment for nearly an hour are shown by the top and bottom traces of Fig. 12. Fig. 13 shows the axis ), the voltage rotor flux linkages of the current model ( ) and the neural model ( ), taken at the end model ( of heat run. All of the flux linkages are in the stationary reference and are updated with a frame. The flux linkages is updated sampling time of 100 , whereas the flux follows the flux linkage only at 1000 . The flux linkage , due to the online training of the neural network. The coefficients used for training are and . To verify the stator resistance estimation, an additional 3.4 per phase was added in series with the induction motor stator, with the motor running at 1000 rev/min with a load torque of

7.4 Nm. The estimated stator resistance together with the actual stator resistance is shown in Fig. 14. The estimated stator resistance converges to 9.4 within less than 200 ms. Fig. 15 shows both the measured axis stator current and the one estimated by folthe neural network model. The neural model output lows the measured values , due to the online training of the network. The neural model current estimate is updated with a sampling time of 100 . The coefficients used for training are and . The modeling results, as described in Fig. 9(b), indicates that the proposed rotor and stator resistance estimators can converge in a short time, as low as 200 ms corresponding to a 40% step change for both stator and rotor resistances simultaneously. In order to compare the stator resistance estimation for simulation and experiment, simulation is repeated with SRE and RRE is applied without blocks in Fig. 8. Then, a step change in a step change in and the results are recorded as the upper trace in Fig. 16. The bottom trace in this figure is the same as the top trace of Fig. 14. The estimation time in modeling is in very close agreement with the one obtained from experiment. C. Speed Estimation Experiments were conducted to verify the performance of the speed sensorless operation of the drive together with the proposed rotor and stator resistance adaptation. To examine the ef-

174


Fig. 18. Experimental results of dynamic speed estimation.

Fig. 16. Effect of stator resistance estimator on steady-state speed estimation— experimental results.

Fig. 19. Experimental results for low-speed estimation.

Fig. 17. Effect of stator resistance estimation on dynamic speed estimation— experimental results.

fect of stator resistance in the estimated speed, an additional 3.4 per phase was added in series with the induction motor stator, with the drive operated with RFOC with the SRE block in Fig. 8 off. The estimated speed dropped by 10 rev/min from 1000 rev/min, as indicated in Fig. 17. Fig. 18 shows the effect of stator resistance estimation during the speed reversal. and the speed Fig. 18(i) shows the plots of estimated speed measured by an incremental encoder, before switching in the additional 3.4 resistance. Fig. 18(ii) shows the results after switching in the additional resistance with the SRE block off. Finally, the SRE block is turned on and the results are recorded, as in Fig. 18(iii). The distortion in the estimated speed in Fig. 18(ii) has totally disappeared in Fig. 18(iii). An additional detailed

Fig. 20. Experimental results of speed estimation during a step load.

view of the speed estimation results of Fig. 18(iii) is plotted separately in Fig. 19 for better readability. To analyze the low-speed performance of the proposed estimations, the experiment was repeated for a speed reversal from to and the speed estimation was found to be satisfactory, as shown in Fig. 20.


Fig. 21. Comparison of stator resistance estimations.

In order to examine the capability of the speed sensorless drive for load transients, a step load of 7.4 Nm was applied to the motor shaft, and the estimated and measured speeds were monitored. Fig. 21 shows these results when the drive was operating at 1000 rev/min. VI. CONCLUSION This paper has presented a new online estimation technique in the presence of variations for for the rotor resistance estimation the speed sensorless induction motor drive. The was found to be totally insensitive to variations, as a result of the stator resistance estimation which is embedded separately. Investigations carried out in this paper have clearly shown for a speed senthat two ANNs can be used in estimating sorless RFOC drive, in the face of significant variations in , which can occur due to motor heating. Both the rotor and stator resistance variations can be successfully estimated using the adaptation capabilities of neural networks. The implementation of these neural network techniques required only a small increase of the computation times. The feasibility and validity of the proposed identification has been proved by the experimental results. The experimental results discussed in this paper have used an SPWM inverter, and hence the line-to-line motor terminal voltages are measured after necessary hardware filtering in the PWM voltages. We are currently investigating the use of an inverter with space vector modulation and DC voltage measurement together with dead time compensation to replace the measurement of filtered PWM voltages to improve the results of the stator and rotor flux linkage estimations for speeds lower than 100 rev/min, thereby enabling us achieve better low-speed results. REFERENCES [1] R. Krishnan and A. S. Bharadwaj, “A review of parameter sensitivity and adaptation in indirect vector controlled induction motor drive systems,” IEEE Trans. Power Electron., vol. 6, no. 4, pp. 623–635, Oct. 1991. [2] R. D. Lorenz and D. B. Lawson, “A simplified approach to continuous on-line tuning of field oriented induction motor drives,” IEEE Trans. Ind. Appl., vol. 26, no. 3, pp. 420–424, May/Jun. 1990.

175

[3] T. Rowan, R. Kerkman, and D. Leggate, “A simple on-line adaptation for indirect field orientation of an induction machine,” IEEE Trans. Ind. Appl., vol. 37, pp. 720–727, Jul./Aug. 1991. [4] C. C. Chan and H. Wang, “An effective method of rotor resistance identification for high-performance induction motor vector control,” IEEE Trans. Ind. Electron., vol. 37, pp. 477–482, Dec. 1990. [5] J. Holtz and T. Thimm, “Identification of the machine parameters in a vector-controlled induction motor drive,” IEEE Trans. Ind. Appl., vol. 27, pp. 1111–1118, Nov./Dec. 1991. [6] H. Sugimoto and S. Tamai, “Secondary resistance identification of an induction motor applied model reference adaptive system and its characteristics,” in Proc. Conf. Rec. IEEE-IAS Annu. Meeting, 1985, pp. 613–620. [7] L. Zhen and L. Xu, “A mutual MRAS identification scheme for position sensorless field orientation control of induction motors,” in Proc. Conf. Rec. IEEE-IAS Annu. Meeting, 1995, pp. 159–165. [8] T. Nouguchi, S. Kondo, and I. Takahashi, “Field-oriented control of an induction motor with robust on-line tuning of its parameters,” IEEE Tran. Ind. Appl., vol. 33, pp. 35–42, Jan./Feb. 1997. [9] R. Marino, S. Peresada, and P. Valigi, “Adaptive input-output linearizing control of induction motors,” IEEE Trans. Autom. Control, vol. 38, pp. 208–221, Feb. 1993. [10] R. Marino, S. Peresada, and P. Tomei, “Global adaptive output feedback control of induction motors with uncertain rotor resistance,” IEEE Trans. Autom. Contr., vol. 44, pp. 967–983, May 1999. [11] T. Matsuo and T. A. Lipo, “A rotor parameter identification scheme for vector controlled induction motor drives,” IEEE Trans. Ind. Appilcat., vol. 21, pp. 624–632, May/Jun. 1985. [12] L. C. Zai and T. A. Lipo, “An extended Kalman filter approach in rotor time constant measurement in PWM induction motor drives,” in Proc. Conf. Rec. IEEE-IAS Annu. Meeting, 1987, pp. 177–183. [13] D. Atkinson, P. Acarnley, and J. Finch, “Observers for induction motor drives,” IEEE Trans. Ind. Electron., vol. 27, pp. 177–183, Dec. 1991. [14] K. Rajashekara, A. Kawamura, and K. Matsus, Eds., Sensorless Control of AC Motor Drives. Piscataway, NJ: IEEE Press, 1996. [15] I. H. Ha and S. H. Lee, “An online identification method for both stator and rotor resistances of induction motors without rotational transducers,” IEEE Trans. Ind. Elctron., vol. 47, no. 4, pp. 842–853, Aug. 2000. [16] K. Akatsu and A. Kawamura, “Sensorless very low-speed and zero speed estimations with online rotor resistance estimation of induction motor without signal injection,” IEEE Trans. Ind. Appl., vol. 36, pp. 764–771, May/Jun. 2000. [17] G. Guidi and H. Umida, “A novel stator resistance estimation method for speed-sensorless induction motor drives,” IEEE Trans. Ind. Appl., vol. 36, pp. 1619–1627, Nov./Dec. 2000. [18] R. J. Kerkman, B. J. Seibel, T. M. Rowan, and D. Schlegel, “A new flux and stator resistance identifier for AC drive systems,” in Proc. Conf. Rec. IEEE-IAS Annu. Meeting, 1995, pp. 310–318. [19] T. G. Habetler, F. Profumo, G. Griva, M. Pastorelli, and A. Bettini, “Stator resistance tuning in a stator-flux field-oriented drive using an instantaneous hybrid flux estimator,” IEEE Trans. Power Electron., vol. 13, pp. 125–133, Jan. 1998. [20] S. K. Mondal, J. O. P. Pinto, and B. K. Bose, “A neural network based space-vector PWM controller for a three voltage-fed inverter induction motor drive,” IEEE Trans. Ind. Appl., vol. 38, no. 3, pp. 660–669, May 2002. [21] B. Burton, R. G. Harley, G. Diana, and J. L. Rodgerson, “Implementation of a neural network to adaptively identify and control VSI-fed induction motor stator currents,” IEEE Trans. Ind. Appl., vol. 34, no. 3, pp. 580–588, May-Jun. 1998. [22] L. Ben-Brahim, S. Tadakuma, and A. Akdag, “Speed control of induction motor without rotational transducers,” IEEE Trans. Ind. Appl., vol. 35, no. 4, pp. 844–850, Jul./Aug. 1999. [23] B. Karanayil, M. F. Rahman, and C. Grantham, “Rotor resistance identification using artificial neural networks for an indirect vector controlled induction motor drive,” in Proc. 27th Annu. Conf. Industrial Electron. Soc., IECON2001, Denver, CO, Nov.-Dec. 29–2, , vol. 2, pp. 1315–1320. [24] K. Funahashi, “On the approximate realization of continuous mappings by neural networks,” Neural Netw., vol. 2, pp. 183–192, 1989. [25] K. Hornik, M. Stinchcombe, and H. White, “Multilayer feedforward networks are universal approximations,” Neural Netw., vol. 2, pp. 359–366, 1989. [26] D. T. Pham and X. Liu, Neural Networks for Identification, Prediction and Control. New York: Springer-Verlag, 1995. [27] B. K. Bose, Modern Power Electronics and AC Drives. Englewood Cliffs, NJ: Prentice-Hall, 2002.

176


Baburaj Karanayil (M’86) received the B.Tech. degree in electrical engineering from Calicut University, Calicut, India, in 1984 and the M.Tech. degree in electrical engineering from the Indian Institute of Technology, Bombay, in 1986. From 1986 to 1994, he worked in the power electronics industry manufacturing both uninterruptible power supplies and variable speed drives in India, in design, systems engineering, and quality control. Since 1995, he has been working as a Professional Officer with the Power Electronics and Drives Group, University of New South Wales, Sydney, Australia. His primary technical interests are in power electronics and its applications to power supplies and in modern control techniques for electrical drives. Dr. Karanayil is a Member of the IEEE Industry Applications, Industrial Electronics and Power Electronics Societies.

Muhammed Fazlur Rahman (M’79–SM’96) graduated in electrical engineering from the Bangladesh University of Engineering and Technology, Dhaka, Bangladesh, in 1972, and received the M.S. and Ph.D. degrees from the Institute of Science and Technology, University of Manchester, Manchester, U.K., in 1975 and 1978, respectively. He was subsequently a Systems Design Engineer at the General Electric Company (U.K,) for two years before joining the National University of Singapore in 1980. In 1988, he joined the University of New South Wales, Sydney, Australia, as a Senior Lecturer. He is at present an Associate Professor in the School of Electrical Engineering and Telecommunications. His research interests are in power electronics, motor control, and motion control systems. Dr. Rahman is an active member of the IEEE of Power Electronics, Industry Applications and Industrial Electronics Societies.

Colin Grantham received the B.Sc. and Ph.D. degrees from the University of Newcastle-upon-Tyne, U.K., in 1969 and 1972, respectively. He then joined the British Approvals Service for Electrical Equipment in Flammable Atmospheres (BASEEFA), where he helped introduce new types of hazardous atmosphere protection in the U.K. In 1975, he moved to the Military Vehicles and Engineering Establishment (MVEE), where he stayed until 1981 when he took up a post as Lecturer in the School of Electrical Engineering and Computer Science, University of New South Wales, Australia. He is at present an Associate Professor in the School of Electrical Engineering and Telecommunications, University of New South Wales. His research interests cover the fields of electrical machines and drive systems, electrical safety, and electrical equipment for hazardous atmospheres.