Order Track Variational Stacked Autoencoder Fault Diagnosis Model for Complex Working Conditions

The fault diagnosis effect of stacked autoencoder in complex working conditions is not meeting expectations due to the combined effect of speed fluctuation and noise variation. In order to solve the above problems, a fault diagnosis model of order track variational stacked autoencoder (OTV-SAE) for complex working conditions is proposed. Firstly, a ridge line extraction method based on time-frequency energy is proposed to solve the problem of high dependence of order tracking method on tachometer. Then, the advantages of order spectrum (signal regularity and obvious features) are used to help variational autoencoder to learn noise features and generate samples containing a variety of noise features, which improves the generalization ability of training samples. Experimental results show that OTV-SAE can eliminate the influence of speed fluctuation and improve the robustness of the model to noise, and greatly improve the diagnostic accuracy of the model in complex conditions.