Attention-based State Decoupler of Vibration Transmission in Multi-working Conditions of Rolling Bearings

Abstract

Bearing fault diagnosis in multi-working conditions is one of the biggest challenges all the time. The paper proposed an attentional-based state decoupler of vibration transmission for rolling bearing fault diagnosis in multi-working conditions. The decoupler consists of three parts which are the vibration encoder, state-decoupling module and classification module. A vibration transmission graph of the smallest core-bearing rotating machinery model is embedded in the decoupling module. The decoupler is trained under one certain condition and tested under the others. The test results show that the proposed decoupler is capable of disentangling the vibration states and recognizing the bearing defects in condition of various loads and frequencies, with higher interpretability.