Listening to the Machine: A Novel Approach to Three-Phase Induction Motor Fault Diagnosis Using Audio Signal Analysis

Bearings are critical components in machinery. An investigation into the feasibility of using audio signals for the automatic classification of healthy and faulty bearings for early detection of bearing faults to prevent catastrophic equipment failures is much needed. The traditional methods often rely upon vibration analysis, which can be expensive and require specialized equipment. This paper aims to develop a reliable and accurate classification system for bearing health diagnoses. A Hybrid CNN model for the task of sound-based audio classification is investigated to analyze the health of ball bearings in Induction motors using the NASA bearing dataset. The main variations between the acoustic signals are discussed in terms of error frequency signatures by examining specific peaks in their spectra. A comparison of existing classifier methods is done to identify the bearing faults at various voltage unbalance levels and load values. In modern conditions, these values and levels are experimented with accuracy rates of over 90%.