Development of a Digital Model of a Gear Rotor System for Fault Diagnosis Using the Finite Element Method and Machine Learning

Abstract

Geared-rotor systems are critical components in mechanical applications, and their performance can be severely affected by faults, such as profile errors, wear, pitting, spalling, flaking and cracks. Profile errors in gear teeth are inevitable in manufacturing and subsequently accumulate during operations. This work aims to predict the status of gear profile deviations based on gear dynamics response using the digital model of an experimental rig setup. The digital model comprises detailed CAD models and has been validated against the expected physical behaviour using commercial finite element analysis software. The different profile deviations are then modelled using gear charts, and the dynamic response is captured through simulations. The various features are then obtained by signal processing, and various ML models are then evaluated to predict the fault/no-fault condition for the gear. The best performance is achieved by an artificial neural network with a prediction accuracy of 97.5%, which concludes a strong influence on the dynamics of the gear rotor system due to profile deviations.