The high-end machineries often suffer from a scarcity of labeled fault samples under alternating load conditions. In addition, class transitions often hinder diagnosis performance. To address these challenges, the evaluate-match-constrain transfer network (EMC-TNet) is proposed, which includes three components: stochastic confident estimation (SCE), enhanced weighted matching (EWM), and interactive refined constraint (IRC). The SCE enhances high-confidence estimations by designing stochastic auxiliary classifiers, achieving the division of known–unknown samples under swarm intelligence. The EWM suppresses private classes of source samples and achieves the effective supervision of target samples by generating enhanced weighted labels through the sample classifier. In addition, it improves discrimination between known–unknown samples by assigning weights via auxiliary classifiers. The IRC refines the classifier discrimination threshold adopting the Shannon entropy weighting strategy and the consistency decision mechanism from dual classifiers, achieving high-precision recognition of known–unknown samples. The effectiveness of the EMC-TNet has been verified on different mechanical testbeds.

Facial expression recognition faces challenges where labeled significant features in datasets are mixed with unlabeled redundant ones. In this paper, we introduce Cross Similarity Attention (CSA) to mine richer intrinsic information from image pairs, overcoming a limitation when the Scaled Dot-Product Attention of ViT is directly applied to calculate the similarity between two different images. Based on CSA, we simultaneously minimize intra-class differences and maximize inter-class differences at the fine-grained feature level through interactions among multiple branches. Contrastive residual distillation is utilized to transfer the information learned in the cross module back to the base network. We ingeniously design a four-branch centrally symmetric network, named Quadruplet Cross Similarity (QCS), which alleviates gradient conflicts arising from the cross module and achieves balanced and stable training. It can adaptively extract discriminative features while isolating redundant ones. The cross-attention modules exist during training, and only one base branch is retained during inference, resulting in no increase in inference time. Extensive experiments show that our proposed method achieves state-of-the-art performance on several FER datasets.

Enclosure circulation of Gas Insulated Switchgear (GIS) and Very Fast Transient Overvoltage (VFTO) are harmful to many respects of power system, including system's security, stability and economic. The effective inhibition of GIS enclosure circulation and VFTO has attracted wide attention. In this paper, GIS enclosure circulation and transient ground potential has been modeled and analyzed. Firstly, this paper discusses the GIS enclosure circulation VFTO, Transient Grounding Potential Rise (TPGR) and the harm, as well as its domestic and foreign researching status. Then, a simplified model has been established, according to which, the steady-state operation of GIS enclosure and connected network circulation theoretical calculation have been investigated according to the GIS shell and grounding grid circulation. The software ATP/EMTP is utilized in the simulation, which takes gas insulated equipment shell connection mode, shell grounding layout and grounding and the grounding grid structure into account. According to the results of theoretical calculation and simulation, the comprehensive optimization of the enclosure circulation and transient potential can be effectively improved and some prevention measures also can be proposed from the deeper understanding of grounding line number and distribution of circulation inhibition. Theoretical results can provide significant reference for the power grid design, operation and protection.