Extreme Learning Machines and Particle Swarm Optimization for Induction Motor Faults Detection and Classification
Smail Bazi, M. S. Nait Said · International Review of Electrical Engineering (IREE) · 2015
The Induction motors (IM) are the most common electrical machines used in industrial applications due to their versatility and their reliability. Despite their robustness, they are subjected to many types of faults during their lifetime. This could lead to a sudden damage resulting in the shutting down of the whole production line. To this end, fault monitoring and diagnosis is essential for safe operation and production quality. The aim of this paper is twofold; first we propose a pattern recognition approach for fault detection in IM based on the extreme learning machine (ELM) classifier and the discrete wavelet transform (DWT). Then in a second stage, we propose to optimize the performances of this approach using particle swarm optimization (PSO) by automatically: 1) detecting the most suitable mother wavelet and the decomposition level for modeling the fault signatures and 2) estimating the parameters of ELM (i.e., the regularization as well as the width of the Radial basis function [RBF] kernel). For this purpose, the PSO algorithm uses the cross-validation accuracy as a fitness function for guiding the search process. Experimental results on real and simulated stator and rotor faults are reported and discussed.