It is difficult to identify the vibration damage of wind turbine mechanical components under complex working conditions

and the fault signal characteristics depend on human experience extraction. The original signal is processed by the Complete Ensemble Empirical Mode Decomposition with Adaptive Noise ( CEEMDAN ) method

and the characteristic signal is screened by the Pearson coefficient. Based on the chaos theory

the phase space reconstruction method is used to obtain a more pure signal

and the reconstructed signal is input into the Convolutional Neural Network ( CNN ) to realize fault feature extraction. The results show that compared with the existing methods

the diagnosis model has higher recognition accuracy for different types of bearing faults under different signal-to-noise ratios

and has good generalization and robustness

which can provide reference for the health monitoring and management of wind turbine mechanical components.