Abstract: In order to solve the problem of difficult blade defect diagnosis caused by severe signal under-sampling in tiptiming technology, a defect diagnosis method for rotating blade based on convolution neural network (CNN) was proposed.In the method, the under-sampled tip timing signal was subject to fast Fourier transform directly to obtain the beat signalof blade at actual vibration frequency. Then, the beat signal was directly used as the input signal of the one-dimensionalCNN model, and by optimizing the network structure and parameters, the sensitive features that can characterize the stateof rotating blades were automatically mined and applied for model training, so as to obtain the applicable diagnosis modelfor defect identification of rotating blades. Finally, the feasibility of the proposed method was verified by the vibration testof rotating blades. The results show that the!!CNN-based defect diagnosis method for rotating blades could solve the analysisproblem resulted from the under-sampling of blade tip timing signal and the average diagnosis accuracy rate reaches 89%, which provides a new way for defect diagnosis of rotating blades.