The fault characteristics of complex equipment are characterized by uncertainty, nonlinearity and so on. For prevention of failure may cause serious consequences, it is necessary to improve the accuracy of fault prediction. As fault prediction has the characteristics of uncertainty, we design a method of fault prediction, which combines fuzzy mathematics closeness degree with particle filter algorithm to predict fault. The new method uses the membership function to describe the normal system with the normal fuzzy sets and the abnormal system with the abnormal fuzzy sets, and uses particle filter algorithm to calculate predictive value and the membership degree. Then we can calculate the closeness degree of predicted value of the normal membership degree with normal and abnormal fuzzy subset to implement a fault prediction. This method predicts whether the three tank system is faulty by the change of water level of the T2 tank in the three tank system and makes test by the fault of the UH-60 planet gear disc when the crack begins to increase, and we have compared with Particle filter fault prediction based on Dynamic Time Warping match, Fault prediction algorithm based on stochastic perturbation particle filter and FDI method based on particle filter. The feasibility of the proposed method is verified by experiments, which can predict the failure of the system in time.