Torque ripple minimization in switched reluctance motors using fuzzy-neural network inverse learning control

The purpose of this paper is the development of fuzzy-neural network (FNN) inverse learning control algorithms for torque-ripple minimization of SRMs. The approach consists of two FNN modules, which spare the same weight values. The learning FNN module is used to adjust the weight values on-line based on observations of the SRMs' (T-i-/spl theta/) input-output relationship in order to form an approximate dynamic inverse model i(T, /spl theta/) of SRMs. The controlling FNN module is used to predict the SRMs phase current waveforms required to follow a desired torque command. Detailed simulation results show good response characteristics for a four-phase SRM.

I. Introduction

Switched reluctance motors (SRMs) are attractive for industrial applications because of their structural simplicity and low cost, ruggedness, and reliability in harsh environments, and the capability to cover a wide speed range. This motor is intentionally operated in deep magnetic saturation to maximize its torque-to-mass ratio. Thus, all the pertinent characteristics of the machine model, i.e., flux-linkage, inductance, phase torque, etc., are highly nonlinear functions of both rotor position and phase current. The ultimate outcome of all these nonlinearities is that high torque ripples are generated when the SRM is operated under conventional rectangular pulse excitation scheme.