《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 of RV gearboxes based on wavelet convolution with attention mechanism

Published:2023-08-14 author:RUAN Qiang, LIU Tao, WANG Zhen-ya, et al. Browse: 1130 Check PDF documents

Fault diagnosis of RV gearboxes based on wavelet convolution

with attention mechanism

RUAN Qiang1,2, LIU Tao1,2, WANG Zhen-ya1,2, ZHANG Bo1,2

(1.Faculty of Mechanical and Electrical Engineering, Kunming University of Science and Technology, Kunming 650500, China;

2.Engineering Research Center for Intelligent Maintenance of Advanced Equipment of Yunnan Province, Kunming 650500, China)

Abstract: Under variable load condition, it is difficult to extract the features of RV reducer gearbox, and its fault mode is also difficult to identify. To solve these problems, a fault diagnosis method (model) of gearbox under variable working conditions based on Laplace wavelet convolution network (LWNet) and attention mechanism (ATT) was proposed. Firstly, based on the principle of Laplace wavelet transform, a wavelet convolutional layer (Laplace Net) was designed instead of the first layer of traditional convolutional neural network to perform adaptive feature extraction on the input signal to obtain more obvious impact features. Then, the attention mechanism was introduced into the model convolutional network to enhance the fault information weight, and SoftMax was used as the classifier for fault diagnosis. In addition, to improve the stability of the model, each convolutional layer was followed by a batch normalization (BN)layer to normalize the features, and Dropout (0.5) was used to prevent overfitting. Finally, the method based on LWNet and ATT was validated using a dataset of RV gearboxes. The research results show that the method based on LWNet and ATT can adaptively locate the impact information of fault signals. Under the condition of constant load, the average diagnostic accuracy is as high as 99.92%. Comparing with the classical deep learning model and recent methods, the accuracy is improved by 3.25%~12.26%. This method has higher diagnostic efficiency and accuracy. Under the condition of variable load, the average accuracy of this method can reach 98.09%, which solves the limitation of feature extraction of vibration signals under variable working conditions.

Key words: transmission; Laplace wavelet convolutional network(LWNet); wavelet transform (WT); attention mechanism (ATT); convolutional neural network(CNN); fault feature extraction

Advanced Search

Online Office System

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