Abstract: To solve certain kinds of fault prognostic problems, an algorithm based on particle filter is presented. On its the state estimation stage, the algorithm estimates the posterior distribution of system fault progression model states and parameters based on joint estimation and particle filter. On the state prediction stage, two different approaches are used. One samples the current state posterior distribution iteratively and uses the sampled particles to form the state prior distribution for some future time. The other uses the datadriven prognostic algorithm to predict the system measurements, to convert the problem of predicting prior distribution to the problem of estimating posterior distribution. The algorithm also adopts Gaussian mixture model to approximate the distribution of random variables, to guarantee that the above two approaches can fuse efficiently under one prognostic frame and improve prediction precision and reliability gradually. On the prognostic decision stage, based upon the above calculated state distribution combined with certain fault criteria the distribution of system remaining useful lifetime can then be inferred. Simulation results demonstrate the validity and feasibility of the proposed algorithm.

Key words: fault prognostics, state estimation, joint estimation, particle filter, Gaussian mixture model, distribution of remaining useful lifetime