Few-shot bearing fault diagnosis using GAVMD-PWVD time-frequency image based on meta-transfer learning

被引:6
作者
Wei, Pengying [1 ,2 ]
Liu, Mingliang [1 ,2 ]
Wang, Xiaohang [1 ,2 ]
机构
[1] Heilongjiang Univ, Dept Automat, Harbin 150080, Peoples R China
[2] Key Lab Informat Fus Estimat & Detect, Harbin, Heilongjiang, Peoples R China
关键词
Rolling bearing; Fault diagnosis; Time-frequency image; Few-shot learning; Meta-learning; Transfer learning; Relation network;
D O I
10.1007/s40430-023-04202-0
中图分类号
TH [机械、仪表工业];
学科分类号
0802 ;
摘要
Rolling bearings are crucial components in rotating machinery and often operate under high speeds and heavy loads for extended periods of time. If a bearing fails, it can disrupt the normal functioning of the machinery and lead to economic losses and even casualties. As a result, diagnosing faults in rolling bearings is critical and urgent. Currently, traditional fault diagnosis methods and deep learning-based methods are used for rolling bearing fault diagnosis. However, traditional methods require knowledge of signal processing techniques and selecting fault features through artificial algorithms. On the other hand, deep learning-based methods require a large number of labeled samples, but fault samples are often limited in practice. Additionally, there can be a problem of insufficient generalization ability when bearing working conditions change, which limits the application of deep learning in bearing fault diagnosis. To address this issue, a novel method is proposed in this paper that involves few-shot transfer learning and meta-learning. The method consists of four stages: using genetic algorithm to determine penalty factor and modal numbers adaptively in variational modal decomposition (GAVMD), combining correlation coefficient to eliminate useless modes, obtaining the instantaneous frequency characteristics of useful modes through Pseudo Wigner-Ville Distribution (PWVD), and using GAVMD with PWVD to obtain time-frequency images of the vibration signals of the rotating bearing. Finally, an improved relational network with deep coding ability and attention mechanism (AM) is constructed based on meta-transfer-learning and original relational network (MTLRN-AM). The experiments in this paper are based on the benchmark dataset of bearing fault diagnosis, and the results show that the proposed method has better multi-task learning ability in meta-learning and better classification performance in few-shot scenarios for bearing fault diagnosis. The average recognition rate reached 96.53% and 98% in 10-way 1-shot and 10-way 5-shot, respectively.
引用
收藏
页数:16
相关论文
共 45 条
[1]   The spectral kurtosis: application to the vibratory surveillance and diagnostics of rotating machines [J].
Antoni, J ;
Randall, RB .
MECHANICAL SYSTEMS AND SIGNAL PROCESSING, 2006, 20 (02) :308-331
[2]   Few-Shot Learning approach for plant disease classification using images taken in the field [J].
Argueso, David ;
Picon, Artzai ;
Irusta, Unai ;
Medela, Alfonso ;
San-Emeterio, Miguel G. ;
Bereciartua, Arantza ;
Alvarez-Gila, Aitor .
COMPUTERS AND ELECTRONICS IN AGRICULTURE, 2020, 175
[3]  
Arjovsky M., 2017, Towards principled methods for training generative adversarial networks
[4]  
[边杰 Bian Jie], 2017, [推进技术, Journal of Propulsion Technology], V38, P1618
[5]  
Bowles C, 2018, Arxiv, DOI arXiv:1810.10863
[6]   Parallel Multi-Block ADMM with o(1 / k) Convergence [J].
Deng, Wei ;
Lai, Ming-Jun ;
Peng, Zhimin ;
Yin, Wotao .
JOURNAL OF SCIENTIFIC COMPUTING, 2017, 71 (02) :712-736
[7]   Variational Mode Decomposition [J].
Dragomiretskiy, Konstantin ;
Zosso, Dominique .
IEEE TRANSACTIONS ON SIGNAL PROCESSING, 2014, 62 (03) :531-544
[8]  
Finn C, 2017, PR MACH LEARN RES, V70
[9]   Residual Life Predictions in the Absence of Prior Degradation Knowledge [J].
Gebraeel, Nagi ;
Elwany, Alaa ;
Pan, Jing .
IEEE TRANSACTIONS ON RELIABILITY, 2009, 58 (01) :106-117
[10]   Rolling bearing fault detection of electric motor using time domain and frequency domain features extraction and ANFIS [J].
Helmi, Hamed ;
Forouzantabar, Ahmad .
IET ELECTRIC POWER APPLICATIONS, 2019, 13 (05) :662-669