A Probabilistic Framework for Remaining Useful Life Prediction of Bearings
Teng Wang, Zheng Liu, Nezih Mrad
Abstract
This article presents a probabilistic framework for remaining useful life (RUL) prediction of bearings. This framework comprises two phases: 1) early fault detection, i.e., modeling the vibration signal as a series of graphs and then conducting graph spectrum analysis to detect the topological change of graph for early fault detection and 2) RUL prediction, i.e., adopting a probabilistic model for bearing RUL prediction given the detected early fault. Specifically, the least square method and noninformation distribution are employed to set the prior knowledge within this model. Meanwhile, a state-of-the-art Markov chain Monte Carlo method, No-U-Turn sampler, is investigated in posterior sampling for predicting RUL and outputting uncertainty. As a result, this framework can work without the training data from other bearings. Besides, it is an unsupervised framework that makes operators free from manual parameters setting and costly tuning runs. In addition to theoretical derivation, the proposed framework is validated using both synthetic data and real-world data and compared with the representative methods in this field. Excellent results show the effectiveness of this probabilistic framework in the RUL prediction of bearings.