【摘要】 无监督域自适应已成为多工况下轴承故障诊断的一种重要方法。然而，现有多源无监督域自适应方法往往忽略不同视角信号对于跨域故障诊断的贡献，不足以全面表达轴承的故障特征。此外，这些方法的不同源域对同一目标域的预测结果存在差异。为此，提出一种融合时频特征的多源无监督域自适应(time-frequency features fused multi-source unsupervised domain adaptation, TFFMUDA)轴承故障诊断方法。该方法以时域和频域信号为输入，通过特征耦合机制实现两种故障特征的互补，并利用分类器对齐策略增强了不同源域对于同一目标域的诊断一致性。通过实际轴承故障案例的试验结果表明，所提方法相较于现有无监督域自适应轴承故障诊断方法能获得更清晰的故障类决策边界并具有更好的目标域诊断精度。更多还原

【Abstract】 Unsupervised domain adaptation becomes an important method for bearing fault diagnosis under multiple operating conditions. However, existing multi-source unsupervised domain adaptive methods often ignore contributions of signals from different perspectives to cross-domain fault diagnosis, and they are insufficient to fully express fault characteristics of bearings. In addition, there are differences in their prediction results of the same target domain due to these methods having different source domains. Here, a time-frequency features fused multi-source unsupervised domain adaptation(TFFMUDA) bearing fault diagnosis method was proposed. This method could take time-domain and frequency-domain signals as inputs, complement mutually the two types of fault features with feature-coupling mechanism, and enhance the diagnostic consistency of different source domains for the same target domain using classifier alignment strategy. The experimental results of actual bearing fault cases showed that the proposed method can obtain clearer fault class decision boundaries and better target domain diagnostic accuracy compared to existing unsupervised domain adaptive bearing fault diagnosis methods.更多还原