Multi-sensor signal fusion for a compound fault diagnosis method with strong generalization and noise-tolerant performance

Abstract

Abstract The gearbox is one of the main components of rotating machinery, for which the complex and changeable working environment leads to frequent failures. To solve the problems of the low diagnostic accuracy of compound faults, weak generalization ability, and difficult diagnosis in a noisy environment, a new method is proposed based on a convolutional neural network. First, the fusion of vibration signals collected by multiple sensors forms a one-dimensional sequence as the input of the network. Second, the random destruction of input and minimal batch normalization mechanisms are introduced to improve the noise tolerance and generalization ability of the model. Finally, the accuracy is estimated by ten-fold cross-validation, which ensures the reliability of the diagnostic results when there are few fault samples. The resulting method is used in compound fault diagnosis under different working conditions. The results show that the method has an accuracy rate of more than 99.5% for gearbox compound-fault classification, strong generalization, and good noise-tolerant performance.