Torque Ripple Minimization In PM Synchronous Motors Based On RBF Neural Networks

Permanent-magnet synchronous motor (PMSM) is a typical nonlinear, multi-variable coupled system, and it is also influenced by the nonlinear factors such astorque ripples. The torque ripples will generate speed oscillations and deteriorate theperformance of PMSM. In this paper, after analysing the numerical description of thetorque ripples and other nonlinear factors, a new exact PMSM model is obtainedbased on the traditional PMSM model, which pave the way for torque rippleminimization.It has been developed for neutralizing these pulsating torque components byadopting various estimation and observer schemes. Unfortunately, well-knowndifficulties associated with accurate estimation of instantaneous torque make itdifficult to completely escape the negative effects of motor parameter variations andchanges in motor operating conditions. Radical basis function neural networks (RBFNNs) have the powerful approximating capability, simple structure and outstandinglearning speed in comparison with other sorts of neural networks. Consequently, aRBF neural networks based torque ripple minimization scheme is proposed with thebackground of backstepping control strategy. With its approximation characteristic,online weight-tuning based RBF neural networks estimate the torque nonlinearity andexternal disturbances, and make them satisfy the linear parameter requirement ofbackstepping. At last, backstepping is applied based on the new model of PMSM. Thedistinguished feature of the proposed scheme is to compensate torque ripples bycombining the advantages of RBF neural networks and backstepping.Besides, an AC motor experiment platform is established, which involves a DSPbased controller, a Panasonic servo driver, and a PMSM. The new control strategy isrealized on this experiment system practically. In addition, the simulation andexperiments results are patched to demonstrate the effectiveness of the proposedmethod.