A novel method for identifying sudden degradation changes in remaining useful life prediction for bearing

Abstract

This paper presents a novel approach to enhancing the accuracy of Remaining Useful Life (RUL) predictions for bearings, addressing the limitations of traditional methods that fail to capture sudden changes in bearing health states. Conventional methods often rely on the monotonicity of a single feature, such as root mean square (RMS), and are unable to continuously monitor health changes. To overcome these challenges, a prototypical network is employed to identify sudden changes in bearing health states, referred to as critical points in this paper. By comparing the health states at the current time and the initial time, the critical point can be determined through the results of the prototypical network. Once the critical point is identified, the hyperparameters of the prototypical network are fixed and transformed to enable RUL prediction. Consequently, RUL can be predicted following the critical point. Furthermore, the prototypical network updates the initial time and continuously analyses the vibration signal to determine the next critical point. This process repeats until no further signals are input. Moreover, validation on two public datasets demonstrates the effectiveness of the proposed method in improving RUL prediction accuracy. Results indicate that the proposed method enhances model precision in bearing RUL prediction through critical point detection (CPD). Incorporating CPD offers a novel perspective for RUL prediction in industrial bearing applications.