An Open-Set Diagnosis Method for Compound Faults of Rolling Bearing Based on TFMST and Difficulty-Aware Simulator

Open-set fault diagnosis (OSFD) requires identifying unknown faults efficiently while correctly classifying known faults. However, traditional OSFD methods rarely consider the difficulty levels of the unknown faults. When the compound fault occurs in real world, the characteristic coupling will increase the difficulty of feature extraction and it is easily confused with other fault types. Therefore, the unknown compound faults might have higher difficulty than single faults. To address the issue of open-set recognition (OSR) of compound faults, a new OSFD method for rolling bearing is proposed based on time-frequency multisqueezing transform (TFMST) and difficult-aware simulator (DIAS). First, TFMST is utilized to process the raw vibration signals to extract more discriminative state information by considering both time and frequency directions simultaneously. Then, the transformed time-frequency (TF) images are employed to aid DIAS to construct the OSFD model, in which DIAS can effectively handle the unknown compound faults by simulating the open set with fake samples of diverse difficulty levels. Finally, the performance of the proposed method is evaluated by the use of four kinds of evaluation metrics. The experimental results demonstrate that compared with the comparison methods, the proposed approach achieves better performance in the OSFD tasks of only single faults as well as tasks including compound faults.