Rotating machine bearing health prognosis using a data driven approach based on KS-density and BiLSTM

Abstract

Rolling element bearings are vital components within rotating machinery, making them a central focus of maintenance in the prognostics and health management sector. This involves closely monitoring their condition to accurately predict the remaining useful life, increasing reliability while minimizing unexpected breakdowns, thereby enabling cost savings through planned maintenance, and enhancing operational stability and security. To achieve this goal, it is necessary to build an online intelligent system for degradation monitoring and failure prognosis by the construction of a robust health indicator and making quantitative measure for bearing degradation. In this paper, an efficient and reliable approach is proposed to estimate the remaining useful life of bearing. A new prediction method is presented by the combination of kernel smoothing density (KS-density) and bidirectional long short-term memory (BiLSTM). Firstly, KS-density smoothens the preliminarily estimated probability distribution function using machinery degradation data. Secondly, the obtained KS-density is used in feed deep learning technique based on BiLSTM models. On this basis, the variation of the signal distribution models between the current faulty state and the normal conditions state is quantified for bearing health assessment. The effective recognition of bearing degradation by the proposed Weibull-based health index is demonstrated through experimental validations utilizing run-to-failure datasets, provided by the centre for intelligent maintenance systems. The comparison with the literature's review show that the prediction results of the proposed approach are more accurate.