Rolling bearing fault diagnosis method based on SSA-WDCNN

Abstract

The invention relates to a rolling bearing fault diagnosis method based on SSA-WDCNN, and the method comprises the following steps: S1, carrying out the singular spectrum analysis of a noise-containing rolling bearing vibration signal, and obtaining a reconstructed vibration signal; S2, inputting the reconstructed vibration signal into a fault diagnosis model for training; and S3, sending an original signal to be diagnosed into the trained fault diagnosis model, and performing rolling bearing fault diagnosis. Compared with the prior art, the singular spectrum analysis method is used for processing the noise-containing rolling bearing vibration signal, the bearing vibration signal is diagnosed based on the convolutional neural network with the first convolutional layer as the wide convolutional layer, the extraction effect and accuracy of the signal feature are improved, and the diagnosis accuracy and the diagnosis efficiency of the rolling bearing fault are further improved.

Description

technical field [0001] The invention relates to the field of fault diagnosis, in particular to a rolling bearing fault diagnosis method based on SSA-WDCNN. Background technique [0002] Rolling bearings are one of the key components in rotating machinery, and their health status affects the normal operation of the entire rotating machinery system. Timely detection of bearing failures can effectively reduce losses in production. At present, the fault diagnosis of rolling bearings mostly studies its vibration signals. However, the fault feature components in the signal, especially the early weak fault features, are easily overwhelmed by noise and other irrelevant signal components, resulting in equipment failures that cannot be detected in time. Therefore, it is particularly important for mechanical fault diagnosis to use the signal decomposition method to decompose the vibration signal into a series of sub-signal components with clear physical meaning, and then extract the fa...

AI Technical Summary

Problems solved by technology

[0003] In the prior art, the feature extraction of the fault signal is realized by using an autoencoder to denoise the noisy signal first, and then input the signal into a one-dimensional convolutional neural network (One-Dimension CNN, 1-DCNN) model, but Two models need to be trained and only for test signals with a signal-to-noise ratio of less than -1dB

In addition, wavelet packet decomposition is used to eliminate part of the interference of the original signal, and then Empirical mode decomposition (EMD) is used to extract fault features to realize bispectrum demodulation of the entire signal, but the final bispectrum resolution rate is not high

In addition, the improved integrated empirical mode decomposition (ensemble empirical mode decomposition, EEMD) of EMD is also applied in signal noise reduction. In the prior art, the EEMD method and the block threshold strategy are combined to denoise the vibration signal. Relatively good experimental results have been achieved, but there is a large deviation between the reconstructed signal and the clean signal, and the scope of use is limited to signals with small signal-to-noise ratios

In the current related research, the effect of noise reduction and signal feature extraction on large noise has not been ideal, resulting in some fault signals are prone to diagnostic deviations under the interference of large noise environments

Images