An adaptive particle filter was proposed for the gas path component abrupt fault diagnosis of turbo-fan engine characterized by a nonlinear non-Gaussian system. In order to reduce the computational burden and ensure the filtering accuracy

the relation between the filtering accuracy and the sampling number was analyzed. The number of particles was adjusted in the filtering process according to the variance of the state variables. The proposed method could reduce the number of particles in the filtering process and computation time while guaranteed the filtering accuracy. The extended Kalman filter (EKF) was introduced to update the particles and generate the importance probability density function helping to avoid the particle degeneracy to some extent. A series of simulations on a turbo-fan engine indicates that the root mean square error of the improved particle filter for the turbo-fan engine fault diagnosis is reduced by 50% than the conventional particle filter

and the computational burden is also reduced by 30%.