Classification of Bearing Fault Based on Multi-class Recurrent Neural Network

Aiming at the strong correlation between the current state of the bearing and the state information of the previous moment, this paper proposes to use the multi-class recurrent neural network (RNN) to classify and predict the bearing fault. The bearing vibration and current information are collected...

Full description

Saved in:
Bibliographic Details
Published in2021 International Conference on Artificial Intelligence and Electromechanical Automation (AIEA) pp. 117 - 121
Main Authors Shi-Bo, Li, Yang, Yang, Fan, Li, Gang, Li
Format Conference Proceeding
LanguageEnglish
Published IEEE 01.05.2021
Subjects
bearing
fault identification
Feature extraction
Recurrent neural networks
RNNs
Training
Vibrations
Wavelet analysis
Wavelet domain
wavelet packet
Wavelet packets
Online AccessGet full text

Cover

More Information
Summary:Aiming at the strong correlation between the current state of the bearing and the state information of the previous moment, this paper proposes to use the multi-class recurrent neural network (RNN) to classify and predict the bearing fault. The bearing vibration and current information are collected from three directions of the shaft. The sliding window is used to reduce the noise of the signal, so as to solve the problem of low signal-to-noise ratio of bearing fault signal in actual production. The wavelet packet decomposition and reconstruction technology is used to extract the features, and the bearing signal is converted from the time domain to the frequency domain to construct the features, and the bearing information entropy and wavelet energy are extracted. Three recurrent neural networks (RNNs) are combined to build multi-class RNNs, and the classification and prediction of bearing fault categories were carried out. After training and testing, the model was evaluated by using confusion matrix, and the results verified the effectiveness of the model.
DOI:10.1109/AIEA53260.2021.00032