Digital Twin-Assisted Bearing Fault Diagnosis Using Multi-Fidelity Incremental Learning

Digital twin (DT)-assisted intelligent fault diagnosis has been a hot topic, which can support personalized monitoring of critical machinery. A central challenge is that diagnostic models using deep learning (DL) suffer from the problem of catastrophic forgetting if personalized faults occur in dynamic environments. To deal with this issue, this article presents a multi-fidelity incremental learning (MFIL) for the continuous diagnosis of key faults in rolling bearings. An effective bearing DT model of the shaft system is constructed to generate sufficient low-fidelity (LF) simulation data. With the assistance of multi-fidelity information fusion and feature boosting, the MFIL method is proposed to mitigate the catastrophic interference induced by new datastreams including limited high-fidelity (HF) physical data. A real-world diagnosis problem in rolling bearings consisting of incremental faulty types is utilized to evaluate the proposed method. The proposed MFIL obtains the average accuracy of 96.87% using only 15 HF data and 100 LF data. The result demonstrates that the proposed MFIL can effectively diagnose different health conditions over time while guaranteeing computationally efficient.