Robust deep learning-based fault detection of planetary gearbox using enhanced health data map (enHDMap) under domain shift problem

Abstract The conventional deep learning-based fault diagnosis approach faces challenges under the domain shift problem, where model encounters different working conditions from ones it was trained on. This challenge is particularly pronounced in of planetary gearboxes due to complicated vibrations they generate, which can vary significantly based on system characteristics gearbox. To solve this challenge, paper proposes a robust deep-learning-based fault-detection for by utilizing an enhanced health data map (enHDMap). Although there HDMap method that visually expresses vibration signal gearbox according gear meshing position, greatly influenced machine operating conditions. In study, domain-specific features are further removed, while fault-related enhanced. Autoencoder-based residual analysis and digital image-processing techniques employed address domain-shift problem. performance proposed validated significant problem conditions, as demonstrated studying two test rigs with configurations operated stationary non-stationary Validation accuracy measured all 12 possible scenarios. achieved detection accuracy, outperforming prior methods most cases.