摘要
为解决火电机组灵活运行导致的数据分布差异问题,本文提出一种多域特征自适应对齐方法用于旋转机械跨工况故障诊断,核心思想是通过对齐多个源域与目标域以提取域不变信息,从而促进诊断知识的迁移,提高对目标任务的诊断性能。首先,采用深度残差网络连接多个卷积单元提取单源域子特征,并通过最小化域适配损失使得每个源域与目标域的特征空间对齐,然后,依次建立单源域分类器并通过对输出结果的自适应加权,实现目标域故障类型的准确判断。实验结果表明,本文所提方法能有效适配多源域与目标域的信息,具有更好的跨工况泛化能力和诊断性能。
To solve the problem of data distribution differences caused by the flexible operation of thermal power units,a multi-domain feature adaptive alignment method is proposed for cross-service fault diagnosis of rotating machinery.The core idea is to extract domain-invariant information by aligning multiple sources and target domains,thus facilitating the migration of diagnostic knowledge and improving the diagnostic performance for the target task.First,a deep residual network connects multiple convolutional units to extract single source domain sub-features.The feature space of each source domain is aligned with the target domain by minimizing the domain adaptation loss.Then,a single-source domain classifier is built in turn,and the output results are adaptively weighted to determine the fault type in the target domain accurately.The experimental results show that the proposed method can effectively adapt the information of multiple sources and target domains and has better generalization ability and diagnostic performance across working conditions.
作者
袁电洪
徐翔
周阳
赵敏
邓艾东
许猛
Yuan Dianhong;Xu Xiang;Zhou Yang;Zhao Min;Deng Aidong;Xu Meng(China Energy Suqian Power Generation Co.,Ltd,Suqian 223803,China;National Engineering Research Center of Power Generation Control and Safety,Southeast University,Nanjing 210096,China)
出处
《信息化研究》
2023年第1期54-60,共7页
INFORMATIZATION RESEARCH
基金
国家重点研发计划(No.2022YFB4100403)
关键词
轴承
领域自适应
故障诊断
无监督学习
自适应对齐
bearing
domain adaption
fault diagnosis
unsupervised learning
adaptive alignment
作者简介
袁电洪(1974—),男,高级工程师,主要研究方向为燃煤电厂智慧建设和生产管理等;通信作者:赵敏(1997—),女,硕士生,主要研究方向为轴承故障诊断和状态评估等。
