基于二维灰度图和结构相似生成对抗网络的小样本轴承故障诊断  

Fault Diagnosis for Small Sample Bearings Based on 2D-Structural Similarity Generative Adversarial Networks

作　　者：岳健 刘冰 江国乾[2] YUE Jian;LIU Bing;JIANG Guoqian(Beijing Goldwind Smart Energy Technology Co.,Ltd.,Beijing 101102,China;Institute of Electrical Engineering,Yanshan University,Qinhuangdao 066099,China)

机构地区：[1]北京金风慧能技术有限公司,北京101102 [2]燕山大学电气工程学院,河北秦皇岛066099

出　　处：《轴承》2025年第3期88-96,共9页Bearing

基　　金：国家自然科学基金资助项目(62273299);北京市博士后工作经费资助项目(2015ZZ-13)。

摘　　要：针对轴承故障诊断领域中标记故障样本稀缺,且主流方法在进行数据扩充时存在生成数据质量不高的问题,提出一种基于二维灰度图和结构相似生成对抗网络(SSGAN)的小样本轴承故障诊断方法。首先,将一维振动信号转换为二维灰度图输入SSGAN中进行对抗训练;其次,结合真实故障样本对生成样本进行结构相似性(SSIM)分析,剔除差异性较大的生成样本,得到辅助训练样本;最后,将扩充后的训练样本输入深度卷积神经网络(DCNN)中进行故障诊断。试验结果表明,所提方法在小样本轴承数据集下的故障识别准确率达到了99.10%,与其他故障诊断方法相比具有明显的优越性。Aimed at scarcity of labeled fault samples in the field of fault diagnosis for bearings and the poor quality of data generated by mainstream methods during data expansion,a fault diagnosis method for small sample bearings is proposed based on 2D-structural similarity generative adversarial networks(SSGAN).Firstly,the 1D vibration signal is converted into 2D grayscale image and input it into SSGAN for adversarial training.Secondly,based on real fault samples,the structural similarity(SSIM) analysis of generated samples is carried out,and the generated samples with significant differences are eliminated to obtain the auxiliary training samples.Finally,the expanded training samples are input into deep convolutional neural networks(DCNN) for fault diagnosis.The experimental results show that the proposed method achieves a fault recognition accuracy of 99.10% under a small sample bearing dataset,which is significantly superior to other fault diagnosis methods.

关 键 词：滚动轴承 故障诊断 小样本 灰度图 生成对抗网络 

分 类 号：TH133.33[机械工程—机械制造及自动化] TH165.3