A Novel Bayesian Update Method for Parameter Reconstruction of Remaining Useful Life Prognostics

Remaining useful life (RUL) prediction is a core component for reliability research and condition-based maintenance (CBM). In the existing parameter-reconstruction method, the degradation trajectory of an in-situ unit is reconstructed by the weighted sum of that of historical units. However, this method requires an optimization problem to be solved for each new measurement, which leads to an excessively consumed time and does not satisfy the requirements of online prognostics and decisionmaking. In this paper, these weights are assumed as a set of probabilities, based on which they can be updated via Bayesian estimation, instead of solving the optimization problem at each observation epoch. To verify the proposed approach, a data set developed by a commercial simulation tool for aircraft turbofan engines is involved. In light of the implement situation of the proposed approach on this data set, the absolute error of the prognostics result and the consumed time for computation are significantly reduced compared with the existing approach.