通过集成域对抗训练和最大平均差异的故障诊断  

Fault diagnosis through integrated domain adversarial training and maximum average difference

作　　者：张宇 张公政 Zhang Yu;Zhang Gongzheng(Hegang Digital Technology Co.,Ltd.,Shijiazhuang 050035,China;Hegang Digital Technology Co.,Ltd.,Shenzhen Branch,Shenzhen 518000,China;Xiongan Saibo Intelligent Technology Co.,Ltd.,Xiongan 070001,China)

机构地区：[1]河钢数字技术股份有限公司,石家庄050035 [2]河钢数字技术股份有限公司深圳分公司,深圳518000 [3]雄安威赛博智能科技有限公司,雄安070001

出　　处：《现代计算机》2024年第20期63-67,共5页Modern Computer

基　　金：2023年工信部工业互联网创新发展工程-基于工业互联网平台的重点行业生产设备智能运维系统项目(ZTZB-23-990-020);河北省省级科技计划资助-重大科技成果转化专项(23280101Z)。

摘　　要：目前,工业物联网已成功地应用于智能制造业。物联网中的大量数据促进了基于深度学习的工业设备健康监测的发展。由于在不同工作条件或设备上收集的机械故障诊断监测数据存在域不匹配,用训练数据训练的模型在实际应用中可能无法发挥作用。因此,研究具有域自适应能力的故障诊断方法至关重要。提出了一种基于改进的域自适应方法的智能故障诊断方法。具体来说,分别使用最大平均差异和域对抗训练关于特征空间距离和域不匹配的两个特征提取器来增强特征表示。由于单独的分类器训练特征提取器,进一步利用集成学习获得最终结果。实验结果表明,该方法在解决域不匹配故障诊断中是有效的,且具有实际应用价值。Currently,the industrial Internet of Things has been successfully applied in the intelligent manufacturing industry.The large amount of data in the Internet of Things has promoted the development of industrial equipment health monitoring based on deep learning.Due to domain mismatch in the mechanical fault diagnosis monitoring data collected under different working con-ditions or equipment,models trained with training data may not be effective in practical applications.Therefore,it is crucial to study fault diagnosis methods with domain adaptive capabilities.An intelligent fault diagnosis method based on an improved do-main adaptive method was proposed in this paper.Specifically,two feature extractors for feature space distance and domain mis-match are trained using maximum average difference and domain adversarial training to enhance feature representation.Due to the separate classifier training feature extractor,further ensemble learning is utilized to obtain the final result.The experimental results show that this method is effective and has practical value in domain mismatch fault diagnosis.

关 键 词：域自适应 域对抗性训练(DAT) 集成学习 故障诊断 最大均值差异(MMD) 

分 类 号：TH17[机械工程—机械制造及自动化] TP277[自动化与计算机技术—检测技术与自动化装置]