基于LMMD-DANN的无监督风电轴承故障诊断方法OA

The health status of the rolling bearings of wind turbines directly affects the performance of the equipment,but under the condition of variable working conditions,the target domain data is often missing labels,and its diagnostic performance will be greatly reduced.An LMMD-DANN fault diagnosis model is proposed to solve this problem.The model uses a multi-module ensemble architecture that combines a feature extraction network,a local maximum mean difference(LMMD)algorithm,and a domain adversarial neural network(DANN).That is,by introducing the domain adversarial mechanism,an adversarial relationship is established between the feature extractor and the domain classifier,which enhances the feature extraction ability and realizes the feature confusion of cross-domain data.At the same time,the LMMD algorithm is used to focus on local features,which promotes the feature alignment of similar subdomains.The adaptive weighting strategy is introduced to dynamically adjust the proportion of domain classification loss in the loss function.The experimental results on the CWRU and JNU bearing datasets show that the average accuracy of the proposed method in the six variable working conditions reaches 98.88%and 89.1%,respectively,so LMMD-DANN has more advantages in the unsupervised fault diagnosis scenarios of variable working conditions.