Analysis of Residual Error in Closed Loop Operation of Brushless DC (BLDC) Motor using Novel Neural Network Approach

Abstract

Abstract In existing technique for speed regulation of Brushless DC (BLDC) motor, great performance is achieved but only for fixed speed reference, during dynamic change in reference speed the existing techniques fail due to presence of non-linear complex parameter in the system. The proposed novel and simple Multi-Layer Perceptron (MLP) technique is capable of regulating the speed of a BLDC motor, according to dynamically changing reference paths, by minimizing residual error. The MLP, which is a compound of a deep learning artificial neural network, comprises two layers; one which receives speed error data from the traditional PID controller, and another which predicts the output data. Computations are carried out through Gradient Descent for arbitrary hidden layers, in order to reduce the speed error. As the computed output approaches the target, the optimal data for speed regulation is produced. This proposed MLP technique and Advanced Particle Swarm Optimization (APSO) algorithms have been simulated, using ramp, step, and sinusoidal reference speeds under the same platforms.