Abstract: As for the problem of fault diagnosis of nonlinear system in non-Gaussian noises, a new method based on the unscented particle filter(UPF) is proposed, concerning of the shortcoming of degeneracy and estimation precision of generic particle filter. Firstly, normal/abnormal UPF models are established separately, and the calculation method of likelihood probability density function and log likelihood ratio are deducted. Then, the fault detection and diagnosis rule are given, which can forecast both the happening time and type of the fault. At last, some experiments of nonlinear actuator loop of helicopter are carried out, which can demonstrate the validity and superiority of the proposed method.

Key words: Fault diagnosis, Log likelihood ratio, Non-Gaussian, Nonlinear, Unscented particle filter