A particle filtering-based approach for remaining useful life predication of rolling element bearings

Rolling element bearings are one of the most widely used components in rotating machinery. However, they are also the components which frequently suffer from damage. Remaining useful life (RUL) prediction of rolling element bearings has received considerable attention, since it can avoid failure risks, and ensure availability, reliability and security. Model-based methods are commonly used in RUL prediction because of their high accuracy in long-time prediction. In model-based methods, a degradation indicator which describes the whole degradation process of bearings, however, is very critical but difficult to be extracted. A model function, used to predict the evolution trend and the RUL of bearings, is difficult to develop as well. In this paper, a particle filtering (PF)-based approach is developed to predict the RUL of rolling element bearings. In this approach, two modules are included, i.e. indicator calculation module and PF-based prediction module. In the first module, a new degradation indicator is calculated based on correlation matrix clustering and weight algorithm. This indicator fuses different characteristics of multiple features, includes more fault information and therefore has a better prediction tendency. In the second module, a PF-based approach is proposed to predict the RUL of bearings. Different from the traditional PF-based approach, a new algorithm of parameter initialization is introduced to calculate the initial parameters of the state space model. Experimental data of rolling element bearings are used to demonstrate the effectiveness of this approach. For comparison, another RUL prediction approach based on adaptive neuro-fuzzy inference system (ANFIS) is also utilized to process the experimental data. The result shows that the proposed approach can effectively calculate the appropriate degradation indicator, initialize the model parameters and perform better in RUL prediction than the ANFIS-based approach for rolling element bearings.