基于改进加权域对抗网络和混合注意力机制的水电机组滚动轴承故障诊断方法
Fault Diagnosis for Rolling Bearing of Hydroelectric Unit Based on Improved Weighting Domain Adversarial Network and Hybrid Attention Mechanism
胡志平 1王焕河 1田凡 1陈万凯 2许颜贺3
作者信息
- 1. 湖北白莲河抽水蓄能有限公司,湖北 黄冈 438600
- 2. 重庆川仪软件有限公司,重庆 400700
- 3. 数字流域科学与技术湖北省重点实验室,华中科技大学,湖北 武汉 430074
- 折叠
摘要
针对水电机组滚动轴承可采集故障样本匮乏、难以完成高效准确故障诊断目标的问题,提出一种基于改进加权域对抗网络(improved weighting domain adversarial network,IWDAN)和混合注意力机制(hybrid attention mechanism,HAM)的机组轴承故障诊断方法.首先,利用小波变换(wavelet transform,WT)将轴承一维振动信号转换为二维时频图,实现更高维度的信号表征;其次,利用IWDAN对源域时频图进行自适应加权,提取更为有效的域间共享特征;最后,将所提取特征作为HAM的输入,有效抑制冗余信息干扰,显著提升诊断效率与精度.通过机组轴承诊断实例分析,验证所提IWDAN-HAM方法具有更加优越的性能,可为机组维护策略的制定提供可靠数据基础.
Abstract
Aiming at the problem of insufficient fault samples and unsatisfied diagnosis accuracy,a fault diagnosis method for rolling bearing of hydroelectric unit was proposed in this paper based on improved weighting domain adversarial network(IWDAN)and hybrid attention mechanism(HAM).First,the one-dimensional vibration signal of bearing was transformed into a two-dimensional time-frequency spectrum by the wavelet transform(WT)method,which contributes to characterize the signal in the higher dimensions.Subsequently,in order to ex-tract more efficient shared features between different domains,the time-frequency spectrum in source domain was weighted adaptively using the developed IWDAN model.Finally,the extracted features were served as the inputs of HAM method to effectively suppress the interfer-ence of redundant information and further improve the diagnosis efficiency and accuracy.Based on the case analysis of unit bearing diagno-sis,the superior diagnosis performance of the developed IWDAN-HAM method was validated convincingly,and the corresponding results were helpful to provide reliable data foundation for the formulation of unit maintenance strategies.
关键词
水电机组/滚动轴承/故障诊断/域对抗网络/注意力机制Key words
hydroelectric unit/rolling bearing/fault diagnosis/domain adversarial network/attention mechanism引用本文复制引用
出版年
2025
国家科技期刊平台
NSTL
维普
万方数据