A novel quantitative diagnosis method for rolling bearing faults based on digital twin model

Dual-impulse behaviors of rolling bearings have been widely researched for quantitative diagnosis. However, it is challenging to accurately extract entry and exit moments of the fault from noise-contaminated raw signals. To address this issue, a novel quantitative diagnosis method based on digital twin model is proposed to assess the fault severity from the original signal waveform. Specifically, the quantitative diagnostic criterion for bearing faults is derived to reveal the instantaneous response characteristics of dual-impulse behaviors, and then a digital twin model is constructed to characterize the fault characteristics of the measured signal with noise-free twin signals. Subsequently, a recursive parameter optimization strategy based on cosine similarity (RPOS-CS) is proposed to optimize the twin model in real time, and fault parameters of the optimal signal will be applied to evaluate the fault size of the bearing. Finally, kernel density estimation is employed to perform uncertainty analysis on multiple diagnosis results, thereby realizing interval estimation and significantly enhancing the reliability of diagnosis results. Both simulated and experimental signals are utilized to validate the efficacy of the proposed method, and the further comparative analysis shows that it exhibits high diagnostic accuracy and outstanding reliability.