1 NEURO-FUZZY COMPENSATION STRATEGY TO MINIMISE TORQUE RIPPLE IN SWITCHED RELUCTANCE MOTOR DRIVES L. Henriques*, L. Rolim*, W. Suemitsu*, P.J. Costa Branco** and J. A. Dente** *UFRJ/COPPE-Elétrica CP 68504 - CEP 21945-970 RJ Brasil, walter@dee.ufrj.br **Mechatronics Lab., Instituto Superior Técnico, Lisboa, 1049-01 Portugal, pbranco@alfa.ist.utl.pt Abstract: Simple power electronics and fault tolerance are advantages of SRM drives. However, excessive torque ripple has limited their application. This paper presents a novel method of controlling the motor currents to minimise the torque ripple based on a neuro- fuzzy compensator. In the proposed controller, a compensating signal is added to the output of a PI controller, in a current-regulated speed control-loop. The compensating signal is learned prior to normal operation, in a self-commissioning run, but the neuro-fuzzy methodology is also suitable for on-line self-learning implementation, for continuous improvement of the compensating signal. Copyright  2000 IFAC Keywords: Switched reluctance motors, Learning control, Compensation, Electrical machines, Fuzzy control, Actuators, Fuzzy systems, Speed control 1. INTRODUCTION In the last years, some works (Belfore and Arkadan, 1997; Bolognani and Zigliotto, 1996) have been proposed on modelling and control of SR motors using soft-computing techniques. Artificial intelligence-based fuzzy, and/or neural, and/or genetic controllers have demonstrated a number of advantages over conventional ones and, more significantly, helping to incorporate "some intelligence" into them (Costa Branco and Dente, 1998). Fuzzy control of a SR drive has been implemented with success by Rodrigues et al. (1997), and has shown to be effective for speed control in applications where some degree of torque ripple is tolerated, as is the case in many industrial applications. Nevertheless, in servo control applications or when smooth control is required at low speeds, elimination of torque ripple becomes the main issue for an acceptable control strategy of the SRM. In this case, even using a fuzzy PI-like control, is not satisfactory because the controller response, which is used as reference signal for current control, gives rise to sustained torque pulsations. Furthermore, the magnitude of these pulsations changes with the motor speed and with its load. Results (Husain and Ehsani, 1996; Lovatt and Stephenson, 1994; Wallace and Taylor, 1992) have been published, which use many different compensating strategies. Some works (Lovatt and Stephenson, 1994) use the inverse of the static torque-current-rotor position relationship that are tabulated previously and stored in memory. However, these methods are quite laborious and mainly sensitive to parameter variations. A novel compensation method is proposed in this paper, which is based upon a self-tuning neuro-fuzzy compensator with adaptation capacity to motor parameters change. The compensating signal is added to the output of a PI controller. Therefore, the compensator has the objective of to learn how modulate the currents to reduce the torque ripple. 2. THE PROBLEM: TORQUE RIPPLE With only a PI-speed controller, it is not possible to obtain a ripple-free motor speed at any speed range because it would require a ripple-free output torque for this purpose. If it is supposed that the speed is constant and equal to the reference one in steady state, then the PI controller's signal (i.e. the reference current) would be constant. Figure 1 shows, however, that a constant current reference produces a pulsating torque, rendering the ripple-free speed control unfeasible. The results shown correspond to the current-regulated, full-load operation of a 750W SR motor, at rated speed (1800rpm). At high speeds, torque pulsation occurs at higher frequencies, thus causing less speed ripple due to the natural filtering provided by the mechanical load inertia. Fig. 1. Torque ripple from constant current reference.