Founded in 1971 >
Chinese Sci-tech Core Periodicals >
British Science Abstracts (SA, INSPEC) Indexed Journals >
United States, Cambridge Scientific Abstract: Technology (CSA: T) Indexed Journals >
United States, Ulrich's Periodicals Directory(UPD)Indexed Journals >
United States, Cambridge Scientific Abstract: Natural Science (CSA: NS) Indexed Journals >
Poland ,Index of Copernicus(IC) Indexed Journals >
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
Abstract: In the diagnosis of variable working condition gear faults, traditional deep learning models have low diagnostic accuracy and poor generality due to the problems of diverse gear operating conditions, few fault sample data, significant differences in data distribution, and imbalanced fault data. To address these problems, a variable working condition gear fault diagnosis (VWFD)method(modal) based on meta-learning techniques was proposed. Firstly, the original vibration signals of the gear were resampled using overlapping sampling to increase the number of fault samples. Secondly, the resampled fault data was transformed into time-frequency feature maps through short-time Fourier transform (STFT), making the data form more suitable for the model's input to extract more complete fault features. Then, the Inception modules were introduced into the prototype network based on meta-learning technology to improve its feature expression ability and obtain more comprehensive fault feature information. Finally, various fault metric prototypes were established through the optimized prototype network, and fault classification was performed using a metric classifier to achieve gear fault diagnosis under variable working conditions. A series of comparative experiments were designed to verify the rationality of the proposed model structure and the introduction position and quantity of Inception modules,and the experimental results were analyzed. The research results show that the VWFD method can achieve higher diagnostic accuracy than other fault diagnosis methods. For example, in the 5-way 5-shot experiment group with the same load and different speed types of variables working conditions, the average diagnostic accuracy of VWFD is as high as 91.26%, while the diagnostic accuracy of support vector machine (SVM), convolutional neural networks (CNN), and prototypical nets (PN) are only 74.48%, 87.22%, and 89.56%, respectively.
Key words: variable working condition gear fault diagnosis (VWFD); overlapping sampling technology; meta-learning technique; prototype network; short-time Fourier transform (STFT); Inception module