Gear Pitting Fault Detection: Leveraging Anomaly Detection Methods

Monitoring and maintaining the health of gears is crucial for the efficient and safe operation of mechanical systems. Due to harsh operating conditions, gear failures such as wear, pitting, and breakage are common. This study investigates the effectiveness of unsupervised and semi-supervised deep anomaly detection methods for identifying distributed pitting defects in gears using vibration data. In the experimental setup, gear faults of varying severity were created, and vibration data from helical gears were recorded for each level of fault severity. Autoencoders (AE), Variational Autoencoders (VAE), and Deviation Networks (DevNet) have been utilized to detect faulty gears. This study presents the performance of these techniques in predictive maintenance based on the availability of fault data.