Artificial Neural Network and ESPRIT-TLS Combination-Based Approach for Fault Bearing Recognition in Induction Machines

Abstract

Several studies have been conducted in recent decades on the detection of bearing faults to develop methods for real-time monitoring. Faults are found in almost all electromechanical systems and the major cause of damage to machinery, resulting in serious accidents to humans and material environments. While methods such as MCSA (Motor Current Signature Analysis) and PCA (Principal Component Analysis) have been widely used for obtaining and analyzing the machine's stator current and reducing its dimensions and noise, they do not allow for easier detection and parameter estimation when a defect appears, and sometimes help from an expert is needed to explain the signals. In this paper, the use of ANN-GA (Artificial Neural Networks-Genetic Algorithm) is investigated in combination with the best variant of the ESPRIT (Estimation of Signal Parameters via Rotational Invariant Techniques) method applied on stator current signals for efficient real-time bearing fault detection, to avoid any help being required from experts. MATLAB simulations of these variants revealed that the TLS variant highly discriminates the fault. By using this variant to prepare the data in the search of the best ANN model, in combination with genetic algorithms (used for optimizing the search of ANN parameters), two architectures capable of discriminating this bearing fault in real time in time and frequency domain were obtained.