Employment of ANN for Predictive Motor Maintenance and Bearing Fault Detection Using Park’s Vector Analysis

Abstract

Predictive maintenance is an emerging concept that is gaining mainstream popularity in industrial automation. It involves continuous monitoring of the machinery’s health, status and performance in real-time. Hence, it allows the industry to schedule maintenance only when specific conditions are met and before the machinery breaks down. Critical machinery such as three-phase induction motors is being used widely in industrial processes. However, they are subjected to many electrical and mechanical stresses due to their long operating times. Bearing faults account for the highest faults in the induction motor. If the faults go unnoticed, they can result in motor failure, which generates economic losses. Thus, the tracking of motor health becomes ineluctable. One of the most effective ways to understand health and efficiency is through motor current signature analysis (MCSA). The proposed model employed MCSA by plotting Park’s vector analysis graph to predict motor health. This graph would then be fed as the input to an artificial neural network (ANN) for automated classification of the healthy or unhealthy motor. The trained model has managed to acquire an accuracy >99% in the test set, thus triumphing over other predictive maintenance methods. It is designed so that the proposed model’s computational complexity is low enough to run on edge computing devices to implement an efficient industrial automation system.