【摘要】 针对滚动轴承故障样本不足和特征信息获取不全面导致故障诊断准确率低的问题，提出了一种基于时频特征融合和关系网络的少样本故障诊断方法。该方法结合元学习的训练策略，首先设计了一个特征提取模块，用于获取滚动轴承振动信号的时频域信息并进行融合，以此加强获取特征的全面性；其次使用关系网络的度量模块计算支持样本和查询样本的相似得分，最终实现故障诊断。实验结果表明，在CWRU数据集的跨工况场景下，本方法展现出了优异的性能，故障诊断准确率最高可达99.82%,并有效验证了特征提取模块的有效性，显著提升了滚动轴承故障诊断的准确性和可靠性。更多还原

【Abstract】 Aiming at the low fault diagnosis accuracy caused by insufficient fault samples of rolling bearings and incomplete feature information acquisition, this paper proposes a few-shot fault diagnosis method based on time-frequency feature fusion and relation networks. This method combines a meta-learning training strategy. Firstly, a feature extraction module is designed to obtain and fuse the time-frequency domain information of rolling bearing vibration signals, which enhances the comprehensiveness of the extracted features. Secondly, a metric module of the relation network is used to calculate the similarity scores between support samples and query samples, ultimately achieving fault diagnosis. Experimental results demonstrate that in cross-working condition scenarios of the CWRU dataset, this method exhibits outstanding performance, with a maximum fault diagnosis accuracy of 99.82%. Additionally, it effectively verifies the validity of the feature extraction module, significantly improving the accuracy and reliability of rolling bearing fault diagnosis.更多还原