A Diagnosis Scheme of Gearbox Faults Based on Machine Learning and Motor Current Analysis

Abstract

Condition monitoring of gearbox elements is a crucial task for manufacturers in order to guarantee machines availability, reliability and labor safety. Thus, motor current-based maintenance presents many advantages over traditional vibration-based maintenance, as it is non-invasive, inexpensive, and widely applicable since the majority of today’s machines are driven by induction motors. Therefore, several studies have been realized recently in order to develop efficient condition monitoring programs based on motor current analysis. In this paper, a diagnostic method of gearbox faults based on motor current analysis is developed using supervised machine learning techniques. A method is proposed to remove the effect of the load level on the classification efficiency by using the sum of the phase currents instead of the single-phase currents. A dimensionality reduction flowchart based on the singular value decomposition (SVD) algorithm is proposed in this study in order to remove the operating speed effect on the diagnostic accuracy. Two robust health indicators independent of the operating speed and load are constructed and injected as inputs of varying machine-learning models in order to classify the different health states of the gearbox. The developed health indicators showed a good accuracy in diagnosing gears and bearings faults.