Introduction

International Standard Serial Number:

ISSN 1001-4551

Sponsor:

Zhejiang University;

Zhejiang Machinery and Electrical Group

Edited by:

Editorial of Journal of Mechanical & Electrical Engineering

Chief Editor:

ZHAO Qun

Vice Chief Editor:

TANG ren-zhong,

LUO Xiang-yang

Tel:

86-571-87041360,87239525

Fax:

86-571-87239571

Add:

No.9 Gaoguannong,Daxue Road,Hangzhou,China

P.C:

310009

E-mail:

meem_contribute@163.com

Bearing fault diagnosis based on two-channel deep learning
Published:2023-09-20 author:SONG Xiao-cheng, CEN Yue-feng, ZHANG Yu-lai, et al. Browse: 1044 Check PDF documents
Bearing fault diagnosis based on two-channel deep learning


SONG Xiao-cheng, CEN Yue-feng, ZHANG Yu-lai, CEN Gang

(School of Information and Electronic Engineering, Zhejiang University of Science and Technology, Hangzhou 310000, China)


Abstract: In recent years, the research on bearing fault diagnosis based on deep learning and vibration analysis has developed rapidly, gradually replacing the traditional manual feature extraction and machine learning based methods. For the problems of the existing deep learning models in bearing fault diagnosis, such as poor generalization ability of a single model and insufficient local feature extraction ability due to mutual exclusion of functions of common hybrid models. A two-channel network algorithm based on convolutional neural network (CNN) and recurrent neural network (RNN) was proposed. Firstly, one-dimensional CNN (1D-CNN) and bidirectional long short-term memory neural network (BiLSTM) were selected as the base model. Then, the pooling layer in the 1D-CNN was replaced by a multi-headed self-attention (MS). Finally, multi-headed self-attention (MS) was introduced in the BiLSTM, and the overall model was referred by MSCNN-MSBiLSTM, and they were verified by the experiments with bearing fault data provided by case western reserve university (CWRU). The research results show that the MSCNN-MSBiLSTM obtains 99% classification accuracy, which yield better experimental results compared with other comparison models. The common hybrid model is able to obtain better stability and generalization performance compared with the single model after the introduction of multi-headed self-attention (MS). It is able to extract deeper features compared with the common hybrid model.

Key words:  recurrent neural network(RNN); deep learning; convolutional neural network(CNN); bidirectional long short-term memory network(BiLSTM); multi-headed selfattention(MS)
Advanced Search
Online Office System
  • Chinese Core Periodicals
  • Chinese Sci-tech Core Periodicals
  • SA, INSPEC Indexed
  • CSA: T Indexed
  • UPD:Indexed


2010 Zhejiang Information Institute of Mechinery Industry

Technical Support：Hangzhou Bory science and technology

You are 1895221 visit this site