FAULT CLASSIFICATION AND DETECTION OF HIGH-SPEED ROLLING BEARING USING SOFT COMPUTING TECHNIQUES

Abstract

High-Speed Bearings fault diagnostics is a very critical aspect to increase the dependability and safety of modern industrial systems. Artificial Intelligence (AI) techniques have been gaining traction in academia and industry as an innovative and effective solution for defect detection and classification. As a result, this dissertation report aims to provide a comprehensive overview of the various processes involved in machine fault classification, as well as an introduction to various Machine Learning Algorithms such as k-Nearest Neighbor (k-NN), Naive Bayes, Support Vector Machine (SVM), Artificial Neural Network (ANN), and Decision Tree. The merits, drawbacks, and practical consequences of various AI algorithms are then examined. After that, there's also a comprehensive literature assessment of these AI algorithms in industrial applications. This dissertation work firstly focuses on classification of high speed bearing faults using various Machine Learning Techniques and comparing their performances. The Case Western Reverse University Bearing data is used for training and validation of different algorithms. Vibration response of the bearing are obtained and processed to obtain Time domain, Frequency Domain and Time-Frequency Domain features which are then ranked using Minimum Redundancy Maximum Relevance (MRMR) algorithm and subsequently fed into machine learning models. Secondly, this dissertation focuses on using Particle Swarm Optimization (PSO) technique with Power Spectral Entropy (PSE) as the objective function to optimize the parameters of Variational Mode decomposition (VMD). Fault frequency is than detected and compared with theoretical fault frequencies from most sensitive Intrinsic Mode Function (IMF) obtained from parameter optimized VMD.