Fault prognostic algorithm based on dual estimation and particle filter

Supposing that system fault evolution process is expressed by a state space model with unknown slow time-variant parameters, then the fault prognostics can be formulated as a problem of estimating model states for some future time while knowing all the information about system till time step now. An algorithm based on dual estimation and particle filter was presented. This algorithm includes two major stages: on the state estimation stage, it used an iterative procedure to estimate the posterior distributions of system states and parameters alternatively based on two parallel connected particle filters; on the state prediction stage, the algorithm sampled former estimated posterior distributions iteratively and used the sampled particles to form the prior distributions of system states for some future time. Based upon above calculated results and combined with certain fault criterions, the time to failure could then be inferred by computing the probability of system failure. Comparing with the joint estimation approach, the simulation result demonstrates the validity and feasibility of proposed algorithm.