Aiming at the problem that weak fault vibration signals of rotating machinery bearings are easily covered by strong noise and difficult to be recognized,an improved chaotic particle swarm optimization support vector machine fault diagnosis method was proposed.The signal was decomposed by using a local mean decomposition (LMD) algorithm to obtain the product function (PF) component,and the energy normalization process was performed to obtain the time-frequency domain feature set;the spatial dimension of the time-frequency domain feature set was reduced through iterative Laplacian score (ILS);the permutation entropy of the PF component was used as the fitness value of the chaotic particle swarm,and new crossover and mutation strategies were added to establish a new cross-mutation chaotic particle swarm optimization method;the kernel function and penalty factor of the support vector machine were optimized through the improved particle swarm algorithm,and the new classification model was applied to the fault diagnosis of wind turbine generator bearings.The results show that the recognition accuracy of this fault classification model is higher than other classification models.