Adaptive stochastic-filter-based failure prediction model for complex repairable systems under uncertainty conditions

Dynamical reliability assessment and failure prediction are effective tools for ensuring the efficiency, availability, and safety of repairable systems. To achieve better assessment performance, accurate modeling failure recurrence data are the core of prediction approaches. However, because of the uncertainties from the environmental conditions and repair activities, the failure counting model is usually not well established. To solve this problem, in this paper, we propose an adaptive recursive-filter-based dynamical failure prediction approach for complex repairable systems. First, based on the framework of the state space model, a fusion model that fuses Brownian motion into a nonhomogeneous Poisson process is proposed to characterize failure process under multiple uncertainty conditions. Then, an adaptive statistical inference method based on a Bayesian recursive filter and the EM algorithm is derived to update the model parameters and estimate the initial states adaptively. To verify the effectiveness of the proposed approach, a real gas pipeline compressors reliability prediction problem was implemented.