Abstract: Aiming at the problem of the difficulty in distilling the fault feature of the shearer cutting reducer section and the poor robustness of a diagnosis model,a fault diagnosis method for shearer cutting unit reducer based on convolutional neural network was proposed.First,several vibration spectrum kurtosis of health,fatigue pitting, mild wear,severe wear,tooth surface cracks,broken teeth of gear reducer under variable working conditions were extracted by the Kurtogram algorithm. Second, the Kurtosis of vibration spectrum after pixel normalization and gray scale processing were input into the convolutional neural network.The adaptive learning algorithm was used to train the model, and the output of the full connection layer was considered fault features.Finally, combining with the structural principle of the cutting part reducer of a reducer, the Tensor Flow platform was used to program the proposed model, a fault diagnosis test bench was built.Then the recognition rate and feature visualization were used to evaluate the model.The results show that the fault recognition rate of the model is above 99%.The feature visualization under t-SNE algorithm shows that this method solves the complexity of feature extraction in the traditional method and the coupling problem between different fault features. This method provides a new idea for intelligent fault diagnosis of shearer cutting unit reducer.
Key words:  gear reducer; fault diagnosis; convolutional neural network(CNN); Kurtogram algorithm; Kurtosis of vibration spectrum; cutting department of shearer

BAO Cong-wang, JIANG Wei, LIU Yong-zhi, et al. Fault diagnosis method of shearer cutting unit reducer based on CNN［J］.Journal of Mechanical & Electrical Engineering, 2021,38(10):1317-1323,1331.