DSP-Based Fuzzy Neural Network PI/PD-Like Fuzzy Controller for Motion Controls and Drives

Abstract

In this paper, an on-line trained fuzzy neural-network PI/PD controller is developed and implemented for speed trajectory tracking of a brushless drive system. The fuzzy neural network (FNN) structure is basically composed of two parallel fuzzy-neural PI/PD-like fuzzy controllers. Each of the fuzzy-neural PI/PD controllers is a four layer control network. Extended Kalman Filter (EKF) is used to adaptively train each FNN parameters on-line. The on-line learning mechanism modifies the weights and the membership functions of the parallel FNN PI/PD-like fuzzy controllers to adaptively control the rotor speed of the drive system. Thus, the proposed architecture-based EKF presents an alternative to control schemes employed so far. The entire system is designed and implemented in the laboratory using a hardware setup. The real-time laboratory implementation is based on a dSPACE DS1104 DSP and MATLAB/Simulink environment. Experimental results have shown that the proposed controller adaptively and robustly responds to a wide range of operating conditions.