【摘要】 为解决小样本变工况轴承故障诊断中故障诊断模型参数多且泛化性能弱、故障诊断率低、诊断速度慢的问题，提出了将高效通道注意力(Efficient Channel Attention, ECA)机制与元迁移学习(Meta Transfer Learning, MTL)相结合的在线故障诊断方法。首先，将不同工况的原始振动信号转化为二维灰度图像，采用改进后的残差网络作为特征提取器进行特征提取。在不提升模型复杂度的情况下，增强了模型对重要特征的关注度，增强了模型的特征提取能力。之后，将提取到的特征与现场数据结合进行元训练，获得训练参数。最后，在元测试阶段，利用不同工况的元学习任务对模型进行微调，实现在线变工况轴承故障诊断。对比实验验证了本文方法的有效性和泛化能力。更多还原

【Abstract】 An online fault diagnosis method combining efficient channel attention(ECA) mechanism and meta transfer learning(MTL) is proposed to solve the problems of many model parameters, weak generalization performance, low fault diagnosis rate and slow diagnosis speed in small sample variable bearing fault diagnosis.Firstly, the original vibration signals of different working conditions are transformed into two-dimensional grey-scale images, and the improved ResNet(Residual Network) is used as a feature extractor for feature extraction.Without enhancing the complexity of the model, the focus of the model on important features is enhanced and the feature extraction capability of the model is enhanced.Afterwards, the extracted features are combined with field data for meta training to obtain training parameters.Finally, in the meta-testing stage, the model is fine-tuned using meta-learning tasks for different operating conditions to achieve online variable operating condition bearing fault diagnosis.Comparative experiments validate the effectiveness and generalization capability of the method.更多还原