Damage Detection in Gearbox Applications Using Machine Learning

This technical report investigates the application of machine learning for early damage detection in gearbox systems, specifically focusing on Prognostics and Health Management (PHM). The study compares two machine learning approaches—encoding and loss—utilized for detecting damage in a single-stage spur gearbox. Test bench experiments were conducted under various operating conditions, including different damage sizes, speeds, and torques, to evaluate the effectiveness of these approaches. The results indicate that both methods can detect minor damage, with the loss approach providing clearer recognition of different damage stages. The report outlines the significance of early damage detection in enhancing the health management of gearboxes, which is crucial for preventing costly downtime and extending the remaining useful life (RUL) of the equipment. The findings contribute valuable insights for the development of robust damage detection algorithms, essential for improving the reliability of gearbox applications in industrial settings.