Remaining useful life prediction via a hybrid DBN-KF-based method: A case of subsea Christmas tree valves

Abstract

Remaining useful life (RUL) is the critical goal of fault prediction, which provides theoretical support for subsequent maintenance decisions of a system. The state measurement of industrial equipment is often accompanied by a large amount of random noise. In addition, the parameters of the degradation are often random. This kind of uncertainty makes carrying out RUL prediction difficult. To this end, a novel hybrid model-data-driven RUL prediction method based on a fusion of Kalman filter (KF) and dynamic Bayesian network (DBN) is proposed in this paper. The hybrid DBN-KF-based method provides a more comprehensive evaluation and improves accuracy compared with the traditional KF method. By enhancing the performance of observation values through DBN, the optimal estimation of the system state is implemented. Estimation error and observation error are fully considered. In addition, the uncertainty distribution of degradation parameters and environmental parameters is integrated into the state estimation model. RUL is determined based on the system state calculated by the proposed method and the failure threshold obtained by the system characteristics. Numerical simulation is carried out for a subsea Christmas tree valves to demonstrate the advantages of the proposed RUL prediction method.