基于属性强度关联性矩阵的零样本滚动轴承故障诊断  

Zero-shot rolling bearing fault diagnosis based on attribute intensity correlation matrix

作　　者：苑茹 马萍[2] 张宏立[2] 王聪 王瑾春 李家声 YUAN Ru;MA Ping;ZHANG Hongli;WANG Cong;WANG Jinchun;LI Jiasheng(School of Electrical Engineering,Xinjiang University,Urumqi 830017,China;School of Intelligence Science and Technology,Xinjiang University,Urumqi 830017,China)

机构地区：[1]新疆大学电气工程学院,乌鲁木齐830017 [2]新疆大学智能科学与技术学院,乌鲁木齐830017

出　　处：《振动与冲击》2025年第2期302-311,共10页Journal of Vibration and Shock

基　　金：新疆维吾尔自治区自然科学基金(2022D01C367,2022D01C89);“天山英才”培养计划(2023TSYCCX0037;2023TSYCQNTJ0020)。

摘　　要：针对传统有监督学习模型难以辨别滚动轴承未知类故障的问题,提出一种基于属性强度关联性矩阵的零样本滚动轴承故障诊断模型。首先,利用样本之间属性强度关系对数据库中故障样本进行细粒度描述,建立各故障样本与故障类别间的关联关系;其次,引入自适应深度可分离残差网络提取故障属性相关的特征信息;最后,根据属性细粒度描述和特征信息,使用属性学习模块预测未知类故障的属性,通过计算其与属性矩阵的欧氏距离,实现零样本轴承故障的诊断。试验结果表明,相较于其他模型,该模型在识别未知滚动轴承故障类别方面取得了优异的性能,平均诊断准确率达到90.45%,验证了该模型的有效性与优越性,为实际生产提供了有益的应用价值。For the issue of traditional supervised models struggling to identify unknown types of rolling bearing faults,a zero-shot rolling bearing fault diagnosis model was proposed based on an attribute intensity correlation matrix.In the method,a granular description of database fault samples was provided based on the intensity of relationships among attributes,thus associative links between each fault sample and its respective fault category were established.Subsequently,an adaptive deep separable residual network was employed to extract feature information pertinent to fault attributes.Ultimately,in accordance with the granular attribute descriptions and extracted feature information,attribute learning modules predict the attribute vector of unknown faults.The diagnosis of zero-shot faults was realized by computing the Euclidean distance relative to the attribute matrix.The experimental results demonstrate that the proposed method significantly outperforms other approaches in identifying unknown rolling bearing fault types,achieving an average diagnostic accuracy of 90.45%,which validates the model’s effectiveness and superiority,offering significant practical value to industrial production.

关 键 词：滚动轴承故障诊断 零样本学习 属性强度关联性矩阵 特征提取 属性学习 

分 类 号：TH133.33[机械工程—机械制造及自动化]