一种用于跨域轴承故障诊断的深度自适应网络EI北大核心CSCD

针对轴承在不同工况条件下的振动数据存在分布差异,导致诊断精度低的问题,提出一种新的深度自适应网络用于跨域条件下的轴承故障诊断。采用傅里叶变换将原始时域振动信号转换为频域信号并通过深度特征提取器提取其分类特征;利用最大均值差异(maximize mean discrepancy,MMD)来进行深度特征的边缘分布对齐;利用Wasserstein度量网络将源域中有标签数据的类别结构与目标域中无标签数据的类别结构进行匹配,即对齐不同域的类别条件分布,使得故障数据在不同域的分布能够更好的对齐,从而提高模型在目标域未标签数据集上的分类准确率。试验利用凯斯西储大学公开的故障轴承数据集进行了两种跨域条件的模型迁移,验证了该网络在不同迁移场景中都具有较高的准确率,且优于其他深度自适应网络。

A deep adaptive network for cross-domain bearing fault diagnosis

Here,aiming at the problem of low diagnosis accuracy caused by distribution difference of bearing vibration data under different working conditions,a new deep adaptive network was proposed for bearing fault diagnosis under cross-domain conditions.Fourier transformation was used to convert the original time-domain vibration signal into frequency-domain one,and its classification features were extracted with the deep feature extractor.The maximum mean discrepancy(MMD)was used to do marginal distribution alignment of deep features.Wasserstein metric network was used to match category structure having labeled data in source domain with that having no labeled data in target domain,i.e.,align category conditional distributions in different domains,make distributions of fault data in different domains being better aligned,and improve the classification accuracy of the model on unlabeled data sets in target domain.Fault bearing data sets published by Case Western Reserve University were used in tests to do model migration under two cross-domain conditions.It was shown that the proposed network has higher accuracy in different migration scenarios,and is superior to other deep adaptive networks.