Gearbox Fault Diagnostics Using Deep Learning with Simulated Data

Abstract

Transmission components are prone to fatigue damage due to high and intermittent loading cycles, that cause premature failure of gearboxes. Recently, several vibration-based diagnostics approaches using Machine Learning (ML) and Deep Learning (DL) algorithms have been proposed to identify gearboxes faults. However, most of them rely on a large amount of training data collection from physical experiments, which is often associated with high costs. This paper offers an ML and DL classification performance comparison of several algorithms to diagnose faults in a gearbox based on realistic simulated vibration data. A dynamic model of a single-stage gearbox was developed to generate data for different health conditions. Generated datasets were fed to ML and DL algorithms and accuracy results were compared. Results revealed the superiority of Convolutional Neural Network compared to other classifiers. This research contributes to the prevention of catastrophic failures in gearboxes by early crack detection and maintenance schedule optimization.