《机电工程》杂志社科技期刊核心期刊论文投稿机电一体化机械、仪表电工自动化、计算机浙江大学主办

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

Fault diagnosis model of convolutional neural network based on maximum mean difference

Published:2024-03-26 author:BAO Congwang, CHE Shouquan, LIU Yongzhi, et al. Browse: 933 Check PDF documents

Fault diagnosis model of convolutional neural network based on

maximum mean difference

BAO Congwang, CHE Shouquan, LIU Yongzhi, CHEN Jun, ZHANG Caihong

(School of Mining and Mechanical Engineering, Liupanshui Normal University, Liupanshui 553000, China)

Abstract: In view of the difficulty of bearing fault data collection in engineering scenarios and the low accuracy and stability of bearing fault diagnosis under small samples, a fault diagnosis method of rolling bearing under small samples,namely convolutional neural network (CNN) diagnosis model (method) based on maximum mean discrepancy (MMD) was proposed. Firstly, the fault simulation signal was obtained according to the bearing fault mechanism, and the confrontation training model between the simulation signal and a few real samples was constructed based on the generative adversarial network, the pseudo-domain samples were obtained, and the training data set was expanded. Secondly, the cross-entropy loss and MMD were used as the optimization criteria of the CNN, the scaling factor was introduced, the network was dynamically optimized, and the scaling factor of 0.05 was selected as the optimal network structure parameter according to the test results, and the training model of fault diagnosis was constructed. Finally, the training set of the model was composed of the pseudo-domain samples with 1 024 data points and the real samples, which was normalized and input into the constructed network model. With MMD as constraint, convolution and pooling operations were carried out to achieve feature extraction, and the model was optimized by back propagation to achieve iterative updating of the diagnostic model parameters. The experimental results show that the proposed method can significantly improve the accuracy of bearing fault diagnosis and recognition under small samples. When the number of samples is only 16, the recognition rate can reach more than 95%, which proves that the method can still obtain a high fault recognition rate in bearing fault diagnosis under small samples.

Key words: rolling bearing; fault diagnosis; small sample size; generative adversarial network; convolutional neural network(CNN); maximum mean discrepancy(MMD); crossentropy loss

Advanced Search

Online Office System

Chinese Core Periodicals
Chinese Sci-tech Core Periodicals
SA, INSPEC Indexed
CSA: T Indexed
UPD:Indexed