Aiming at the problem that the traditional bearing fault diagnosis method relies heavily on prior knowledge and expert experience in feature extraction in the face of strong noise and non-stationary signal recognition,a CWT-CNN-based centrifugal pump bearing fault identification method was proposed combining traditional signal processing methods with deep learning algorithms.The continuous wavelet transform (CWT) was used to transform the original 1D vibration signal into a 2D time-frequency map with richer fault feature information,and the 2D time-frequency map was then input to the convolution layer to complete the automatic feature extraction,finally fault identification was completed on the SoftMax layer.After the verification of the public bearing data set of Western Reserve University and the centrifugal pump vibration bearing collection experimental platform built in the laboratory,the fault identification accuracy of this method can reach more than 90%.