Cross-domain health status assessment of three-phase inverters using improved DANN

Information and large number of fault labels are required to achieve intelligent health status assessment of three-phase inverters. However, the current signals of inverters cannot be sufficiently collected since open-circuit faults (OCFs) occur briefly, which makes it difficult to determine the OCF mode of the various power switches. A transfer learning model that effectively uses a small amount of sample data to achieve domain adaptation is proposed to address this problem. First, collected fault-sensitive signals are subjected to a continuous wavelet transform (CWT) to obtain two-dimensional image data with more abundant fault feature information. Second, the source domain and target domain features are projected into the same feature space through a domain adversarial neural network (DANN) to achieve multi-domain feature extraction and adaptation. Then, in the feature extraction module of the DANN, the deep residual network (Resnet) structure is used to replace the typical convolutional neural network (CNN) structure. Finally, an intelligent diagnosis network is used to identify the health status of the inverter samples under variable conditions. Experimental results show that the proposed model can accurately and effectively realize the cross-domain health assessment of three-phase inverters in the case of small samples. The accuracy of the proposed model is better than that of other classical transfer learning models.