A Machine Learning Based Methodology for Fault Diagnosis in Rotating Machine

Over the course of the past sixty years, a significant amount of effort has been put into investigating the challenge of pinpointing the reasons why spinning machinery breaks down. The intricacy of rotating machinery design and analysis has resulted in an increased focus in recent times on the troubleshooting of faulty rotating machinery. Within the scope of the current research, we investigated a number of potential issues that could occur within these rotating components. Our goal was to identify the feature-classifier combination that would provide the most accurate fault prediction using machine learning. Two distinct phases of the experiment were carried out: in the first, the shaft, rotor, and bearing were the only components taken into account; in the second, the gear component was included. In order to determine which component of a spinning machine was malfunctioning, a microphone and accelerometer were utilised to take recordings of the machine's noise and vibration levels. When it was applied to vibration signals, SVM was able to obtain a maximum efficiency of classification of 99.52% for a total of 12 fault classes. When it came to the categorization of audio signals, SVM scored exceptionally well for issues including 12 classes, but it fared poorly for problems requiring 24 classes, achieving a classification effectiveness of less than 52% in those cases. The use of a sound-based fault diagnosis for a significant number of components or fault classes suggests that an SVM is not an appropriate technique.