摘要
由于滚动轴承故障样本获取困难,导致训练样本分布往往呈现极强的不平衡性,严重影响轴承智能故障诊断的准确率。针对滚动轴承训练样本不平衡的问题,提出一种基于约束式自编码器-生成对抗网络(constrained autoencoder-generative adversarial network, CAE-GAN)的故障诊断方法,通过增强故障样本特征以提高诊断模型的精度。首先结合自编码器和生成对抗网络,构建一种基于编码-解码-判别结构的网络模型,以提高生成器捕捉真实样本分布的能力;为进一步提高生成样本的质量,提出一种基于距离约束的方法以限制不同类别样本之间的距离,从而避免生成样本全部来自同一类型。通过滚动轴承故障诊断试验证明了该方法能有效提高生成样本的质量,解决样本不平衡问题,轴承故障诊断准确率较其他方法有明显提高。
Due to difficulty in obtaining fault samples for rolling bearing,the distribution of training samples often exhibits strong unbalance to seriously affect the accuracy of bearing intelligent fault diagnosis.Here,aiming at the problem of unbalanced training samples of rolling bearing,a fault diagnosis method based on constrained autoencoder-generative adversarial network(CAE-GAN)was proposed to enhance features of fault samples,and improve the accuracy of diagnosis model.Firstly,a network model based on encoding-decoding-discrimination structure was constructed by combining AE and GAN to improve the generator’s ability to capture actual sample distribution.To further improve the quality of generated samples,a method based on distance constraint was proposed to limit distances between different types of samples,and thereby avoid all generated samples coming from the same type.Finally,rolling bearing fault diagnosis tests showed that the proposed method can effectively improve the quality of generated samples,solve the problem of samples unbalance,and obviously improve the accuracy of bearing fault diagnosis compared to other methods.
作者
李可
何坚光
宿磊
顾杰斐
包灵昊
薛志钢
LI Ke;HE Jianguang;SU Lei;GU Jiefei;BAO Linghao;XUE Zhigang(Jiangsu Provincial Key Lab of Advanced Food Manufacturing Equipment&Technology,Jiangnan University,Wuxi 214122,China;Wuxi Branch,Jiangsu Special Equipment Safety Supervision Inspection Institute,Wuxi 214071,China)
出处
《振动与冲击》
EI
CSCD
北大核心
2023年第23期65-70,86,共7页
Journal of Vibration and Shock
基金
国家自然科学基金(52175096)。
