Enhanced Automated Condition Assessment of Induction Motor Bearings: A Novel Approach Using Matrix Pencil Mean Frequency Signal Processing and Multilayer Perceptron Neural Networks

Abstract

Vibration signal analysis plays a vital role in the condition-based preventive maintenance of induction motor by identifying early signs of motor issues, avoiding costly breakdowns and optimising the motor's maintenance schedule. It provides detailed information very useful for extending the motor's life cycle with proactive, condition-specific maintenance. Furthermore, the vibration signal analysis offers the advantage of identifying the health status of rotating machinery as a whole, as well as its individual components. This paper presents an innovative solution for the automated health assessment of a critical induction motor component: the bearing. Our approach uses the matrix pencil method for signal processing and health signature generation, combined with a multilayer perceptron neural network to detect health conditions from the resulting health signature characteristics. Initially, the matrix pencil is applied to the vibration signal to identify the mean frequency characteristics. This vector provides a holistic view of the signal’s inherent features and transforms its frequency characteristics into a visual spectrum, resulting in improved induction motor bearing fault condition monitoring. Subsequently, the output from the matrix pencil mean frequency analysis is processed by a multilayer perceptron neural classifier, chosen for its low computational cost and high classification accuracy. Experimental validation demonstrates a 100% fault classification rate and automatic identification of defective components. Comprehensive validation further confirms the method’s robustness and feasibility for induction motor bearing fault detection compared to other recently methods.