Fault diagnosis for hybrid dynamic systems with imperfect model based on particle filters

A particle filter is put forward for fault diagnosis of hybrid dynamic system with imperfect models. Firstly, the divergence of the general particle filter for imperfect systems is discussed. Secondly, two kinds of statistics are put forward, i.e. normalization factor and belief of maximal a-posteriori probability state. Finally, threshold logic is presented to detect unknown-faults, and its correctness is proven under some reasonable assumptions.