Fault diagnosis of planetary gearboxes based on LSTM neural network and fault feature enhancement

Abstract

To address the shortcomings of Support Vector Machine, deep learning and other artificial intelligence algorithms in the application of gearbox fault diagnosis, an intelligent fault diagnosis method of planetary gearboxes based on the long short-term memory (LSTM) neural network and the fault feature enhancement was proposed. In the proposed method, a sliding window was used to intercept the vibration signals of different local faults of the planetary gearboxes at first. Then, each of the intercepted signals was transformed through the Fast Fourier Transform and the frequency band rich in fault features was selected to enhance the fault features. The data from previous step were used as input to train the LSTM neural network. Finally, the trained LSTM neural network model was used to intelligently extract the fault features in the selected frequency band and achieve identification as well as diagnosis of different local faults of planetary gearboxes. The experimental results show that the proposed method can effectively diagnose different local faults of the planetary gearboxes with better fault recognition accuracy of the network model.