Early performance degradation of ceramic bearings by a twin-driven model

Abstract

Early subsurface cracks make significant changes to the dynamic responses of full ceramic ball bearings, and complex space environment brings challenges to the status monitoring of related devices. This paper proposed a model to predict the early degradation performance under complex working conditions driven by digital twin. LSTM recursive neural network was used to establish the mapping relationship between the stiffness weakening factor which represents the degree of damage and stable degradation evaluation index by multi-feature fusion extracted from vibration signals. Digital twin technology was used to combine the mechanism of physical degradation process with real-time working condition data and historical data to predict the early performance degradation trend of ceramic bearings. The results show that the degradation indexes obtained from the fusion of KF, RMS, and MMFD are more effective in reflecting the early degradation stage of ceramic bearings. The twin-driven early performance degradation evaluation model for ceramic bearings can replace the real degradation trend to a certain extent, which promotes the performance degradation evaluation models of key components using ceramic materials.