【摘要】 针对源域和目标域数据分布差异大及现场故障标签样本难获取而导致诊断精度低的问题，提出了基于多尺度注意力域适应网络(MADAN)的跨域转子故障迁移诊断方法。使用多尺度卷积层直接提取振动信号的深层故障特征；利用空间和通道注意力捕捉重要特征的所属空间位置和通道，并增大其过程传递的权重，进一步增强特征的表达能力；依据域适应方法优化源域与目标域提取特征分布的差异问题，以提高迁移诊断性能。结果表明：相比于其他诊断方法，MADAN方法在跨实验台转子故障迁移任务中具有更高的诊断精度，为跨设备转子故障迁移诊断提供了新的参考。更多还原

【Abstract】 To address the issue of significant differences in data distribution between source domain and target domain, as well as the difficulty in obtaining on-site fault label samples, which lead to low diagnostic accuracy, a cross-domain rotor fault transfer diagnosis method based on multi-scale attention domain adaptation network(MADAN) was proposed. Multiscale convolutional layers were used to directly extract deep fault features of vibration signals. Spatial and channel attention were employed to capture the spatial positions and channels where important features located, and to increase the weights of process transmission, thereby further enhancing the expression ability of the features. Based on the domain adaptation method, the differences in the feature distribution extracted from the source domain and the target domain were optimized to improve the performance of transfer diagnosis. Results show that compared to other diagnostic methods, the MADAN approach achieves higher diagnostic accuracy in cross-rig rotor fault transfer tasks, providing new reference value for cross-device rotor fault transfer diagnosis.更多还原