Application of metaheuristic- based approach for the diagnosis of electrical machines

Abstract

Fault diagnosis is critical in control theory for improving safety, performance, and reliability in controlled processes. Detecting system errors, component faults, and abnormal operations promptly is essential for diagnosing the source and severity of malfunctions. Various solutions have been proposed for fault diagnosis and residual generation. The widely used squirrel cage three-phase induction motor in industrial applications demands reliable operation due to its robustness, low cost, and standardization. Induction motor failures, particularly inter-turn short circuits in stator windings, can lead to significant downtimes and adverse effects on both people and installations. Additionally, unbalanced supply voltage, which is a common and significant phenomenon, may hinder our ability to differentiate it from the detection of inter-turn short circuits. In this thesis, we have contributed to the field by presenting a comprehensive solution for the detection, classification, and severity estimation of faults in induction machines using artificial intelligence tools. My work includes developing a mathematical model that incorporates Stator Inter Turn Short-Circuit faults across three phases of an induction motor. I implemented a Feed Forward Neural Network (FFNN) approach for fault detection and phase-specific classification, followed by a Multi-Layer Perceptron (MLP) Neural Network approach for detecting and classifying both Stator Inter Turn Short-Circuit and Unbalanced Supply Voltage faults, as well as estimating the severity of short-circuits. I explored various types of Artificial Neural Networks (ANNs) to improve fault detection, classification accuracy, and severity estimation. My thesis is organized into four chapters: an introduction covering induction motor faults and diagnostic techniques, development and validation of induction motor models using MATLAB/Simulink, implementation of FFNN and MLP processes for fault detection and severity estimation, and a comprehensive Fault Identification process utilizing multiple ANNs in a cascaded manner. My skills include mathematical modeling, MATLAB/Simulink proficiency, and expertise in Neural Network development for fault diagnosis and optimization in industrial applications.