Proceedings of the International Conference on Theory and Applications of Mathematics and Informatics - ICTAMI 2004, Thessaloniki, Greece

COMPUTER SIMULATION OF REAL TIME IDENTIFICATION FOR INDUCTION MOTOR DRIVES

Abstract. The paper was meant to be an unitary approach to state and parameter identification in induction motor drives. The main methods for rotor flux, mechanical parameters and load torque estimation were compared and classified. In the area of state and parameter simultaneous identification a general adaptation mechanism, valid for all the solutions based on a reference model, was introduced. Four of the most used solutions for sensorless drives were discussed and compared. There are two general approaches to state and parameter identification in induction motor drives: the Kalman filter, and the linear state observer (Luenberger) plus an adaptation mechanism. For each problem there are also some particular solutions, based not on a general theory, but on solving various specific equation. Theoretical analyze, numerous simulations and experiments show in every case the superiority of general solutions on those called particular. From these two general solutions, the one based on a linear state observer was chosen, because it is much simpler and demands a considerably smaller computation time than a Kalman filter, while their performance is very close. This observer is completed by an adaptation mechanism (for electrical parameters identification) and possible by a least square algorithm (for mechanical parameters identification). This solution was improved (in simplicity, reliability and computation time reduction) for an industrial implementation. Keywords. Sensorless drives, Kalman filter, linear state observer.

1. Induction Motor Model for Real Time Identification. It is the simplified motor model, used in the control of any FOC drive. In stator reference frame its equations are:

295

Marius Marcu, Ilie Utu, Leon Pana, Orban Maria - Computer simulation of real time identification for induction motor drives

Rotor flux identification is determinant for the performance of any DFOC system, no matter how complex this is. That is why it is the first discussed. Moreover, the results obtained in this part are very important for understanding and comparing different solutions of state and parameter simultaneous identification problem. 2. Linear State Observers (Luenberger Observers). They are given by the general state estimation theory. A full continuos or discrete observer is given by: where the gain matrix L controls the observer pole placement, as it is demanded by stability, dynamics, noise sensitivity conditions. As for the induction motor the A matrix has variable parameters, L has to assure these conditions for any parameters value. Usually this is done trough some relations assuring a proportionality of matrix A and observer poles. Each sampling time the matrix gain for the continuos time observer is thus computed, and then its discrete equivalent is obtained.

296

Marius Marcu, Ilie Utu, Leon Pana, Orban Maria - Computer simulation of real time identification for induction motor drives 3. Kalman Filter. It is a state observer with the same structure but having a gain matrix denoted K, computed each step, so that a quadratic performance criterion is minimized. This criterion is constructed according to covariance matrices of input and measurement noise. For the system:

if we denote covariance matrices of w and v with Q and respectively R, the Kalman Filter is:

where simple subscripts show the time moment, while the doubles are to be read like this: k,k-1: computed at k moment, based on information at k-1 moment. Obviously it is the most complex of all the state estimation algorithms and it demands the greatest computation time (including a matrix inverse operation). 4. Simulation Testing of Induction Motor Rotor Flux Observers. For evaluating their performance, observers described have been simulated in various operating conditions. In all cases the drive was controlled by the analog system, while the digital observers were working in open loop. The following situations have been considered: operation under normal conditions: machine parameters are identical to those used in the algorithms, and voltage, current and speed measurement is not affected by noise. However the measured voltage is according to real PWM and filter characteristics. Machine started and operated at three reference speed levels: small (30 rpm, i.e. a stator frequency of 1 Hz), medium (990 rpm, i.e. a stator frequency of 33 Hz) and high (2010 rpm, i.e. a stator frequency of 67 Hz); motor rotor resistance is different to that used by algorithms. Rotor resistance is assumed 40% higher than the rated value, used by flux observers; 297

Marius Marcu, Ilie Utu, Leon Pana, Orban Maria - Computer simulation of real time identification for induction motor drives measurement white noise of about 10% the rated current, voltage and speed. Noise covariance matrices are known. Simulation has been made, in various operating conditions, for DFOC schemes based on both Luenberger and VI observers. Systems in which rotor flux is estimated by a Luenberger observer have been found superior, especially in what regards the low operation. Fig.1 and 2 presents the results for a starting and a reversal process.

298

Marius Marcu, Ilie Utu, Leon Pana, Orban Maria - Computer simulation of real time identification for induction motor drives

Fig. 1. Estimated rotor flux magnitude and components on stator q and d (dotted) axis. Luenberger observer, in off-line operation, 0.2 ms sampling time - experimental results. a), b) motor speed of 2010 rpm; c), d) motor speed of 990 rpm; e), f) motor speed of 60 rpm; 299

Marius Marcu, Ilie Utu, Leon Pana, Orban Maria - Computer simulation of real time identification for induction motor drives

Fig.2. Starting and reversal process for a DFOC system based on a Luenberger flux observer. The sampling time is 0.2 ms - experimental results. a) reference and measured speed, starting process; b) stator current components on stator q and d(dotted) axis; c) estimated rotor flux components on stator q and d(dotted) axis; d) stator reference currents, in rotor flux frame - flux controlling current is dotted; e) reference and measured speed, reversal process; f) stator current components on stator q and d(dotted) axis. 300

Marius Marcu, Ilie Utu, Leon Pana, Orban Maria - Computer simulation of real time identification for induction motor drives 5. Rotor Flux and Speed Simultaneous Identification. Sensorless drives. Respecting the most utilized terminology, the solutions are given by: methods based on an adaptation mechanism: ELO - (Extended Luenberger Observer) - the reference model is the induction motor, while the adjustable one is a Luenberger state observer; EKF - (Extended Kalman Filter); ELO Method The adaptation mechanism is obtained from the general one, using the particular expression of each variable, that is:

That means (K >0 is an arbitrary constant):

301

Marius Marcu, Ilie Utu, Leon Pana, Orban Maria - Computer simulation of real time identification for induction motor drives

Fig.3. Structure of a sensorless DFOC, based on ELO speed estimation method. EKF Method

Fig. 4. Structure of a sensorless DFOC, based on EKF speed estimation method. ELO and EKF schemes are clearly the best. These two schemes are different especially on what regards method complexity and computation time, ELO being superior. In this way is repeated the conclusion given by the 302

Marius Marcu, Ilie Utu, Leon Pana, Orban Maria - Computer simulation of real time identification for induction motor drives comparison of Luenberger and Kalman flux observers. Again, and practically from the same reason, the solution based on Luenberger observer is preferred.

Fig. 5. Simulation results for an ELO sensorless drive. a) measured (dotted) and estimated speed; b) same, zoom on time; c) measured and reference (dotted) speed; d) estimated and reference (dotted) speed; e) estimated rotor flux on stationary d (dotted) and q axis; f) measured (dotted) and estimated stator current on stationary q axis; 303

Marius Marcu, Ilie Utu, Leon Pana, Orban Maria - Computer simulation of real time identification for induction motor drives

Fig. 6. Simulation results for an EKF sensorless drive. a) measured (dotted) and estimated speed; b) same, zoom on time; c) estimated rotor flux on stationary d (dotted) and q axis; d) reference stator currents on synchronous frame: flux current dotted; e) rotor flux angle; 304

Marius Marcu, Ilie Utu, Leon Pana, Orban Maria - Computer simulation of real time identification for induction motor drives

References 1. I. Boldea: Parametrii mainilor electrice, Editura Academiei, Bucureti, 1991. 2. C. K. Chui, G. Cheng: Kalman filtering with real time applications, Springer Verlag, 1991. 3. A. Fransua, R. Mgureanu: Maini i acionri electrice; elemente de execuie, Editura Tehnic, Bucureti, 1986. 4. M. Marcu, I. Uu: Sisteme de reglare vectorial ale mainilor asincrone. Editura Universitas, Petroani, 2000.

Authors: Marius Marcu - University of Petroani, Petrosani, Romania Ilie Utu - University of Petroani, Petrosani, Romania Leon Pana - University of Petroani, Petrosani, Romania Maria Orban - University of Petroani, Petrosani, Romania

305