Due to their short duration, concealability, and random occurrence, intermittent faults have become the most dangerous hazard in electronic circuit systems. However, existing intelligent diagnostic methods often struggle to provide substantial effectiveness for such objects. This article proposes a scheme to enhance the output signal of electronic circuits to make the features more intuitive and facilitate fault feature mining, and designs a scheme to mine tiny fault features from global signals. First, the circuit’s output signal undergoes an S-transform to obtain its time-frequency domain characteristics. Next, a sequence and excitation networks attention module is employed to allocate weights to different channels. Finally, the aforementioned output is used as input to the Swin transformer framework to thoroughly explore fault features. Three electronic circuits are used as experimental circuits to test the proposed method. The experiment shows that the proposed diagnostic method is fast and has an accuracy of over 97%. Therefore, the effectiveness of the designed strategy that includes multiple attention mechanisms in mining intermittent fault features in electronic circuit systems has been demonstrated.

Learning the original data distribution and generating new samples has proven to be an effective approach in addressing the issue of data imbalance. This paper combines the strengths of generative adversarial networks and variational autoencoder, proposing a novel data augmentation method named variational autoencoding generative adversarial networks with self-attention. Specifically, an encoding-decoding process is introduced during the generative adversarial process to provide distribution information for the generator’s sampling space, thereby accelerating the model’s convergence speed and simultaneously improving the quality of generated samples. Additionally, a self-attention module is incorporated into the discriminator to capture global information from the input data, guiding the generator. During the training process, overlapping sampling and feature-layer matching are employed. Comparative experiments with other advanced algorithms on both public and engineering datasets with multiple imbalanced cases demonstrate that the proposed method can generate high-quality samples, effectively enhance original imbalanced data, and exhibit strong generalization and robustness.

Big data-based air conditioning fault diagnosis research has developed rapidly in recent years, but in actual engineering, the fault sample size of air conditioning systems is much smaller than the normal sample size, and the resulting sample imbalance problem makes conventional data-driven diagnostic methods based on low accuracy and poor stability. In order to solve the problem of unbalanced fault diagnosis of air-conditioning chillers, this paper proposes an integrated learning-based diagnostic model, which achieves diagnosis by combining multiple base models and by majority voting. The method uses four classification models, namely, random forest model, decision tree model, k nearest neighbor model, and isomorphic integration model, as base classifiers, and synthesizes the four base classifiers into a heterogeneous integration algorithmic model (IMV) through integrated learning, and performs diagnostic detection of seven types of typical faults of chiller units using the majority voting method of integrated learning. The effectiveness of the proposed algorithm is verified on the RP-1043 dataset, and the experimental results show that the accuracy of the heterogeneous integrated algorithm model (IMV) can reach 96.87%, which is a significant improvement compared with the accuracy of the other four base classifier models (81.04%–96.25%). Therefore, the integrated learning model has some application prospects in fault diagnosis when targeting unbalanced datasets.

Cross-domain fault diagnosis has garnered significant attention in rolling bearing health monitoring. However, most existing studies focus on fault identification under constant operating conditions and thus struggle to address the data distribution discrepancies caused by variable speed and load in practice. Thus, this research introduces a class-domain alignment with domain importance perception method (CDADIPM) for fault diagnosis across varying operating conditions. CDADIPM integrates conditional maximum mean discrepancy (CMMD) with adversarial learning to extract both shared and class-specific features across domains, enhancing cross-domain feature representation. Additionally, a domain importance perception mechanism is introduced to dynamically assess and assign transfer weights to multiple source domains based on distribution differences within each category, ensuring more accurate and efficient knowledge transfer. To further prevent the exclusion of valuable source domain knowledge due to large distribution differences, a weight constraint mechanism is designed to ensure the effective participation of all source domains. Experimental results demonstrated that CDADIPM achieved accuracies exceeding 97% across all tasks, surpassing the performance of the latest multi-source domain adaptation methods. These results highlight its superior engineering applicability and broader potential for practical deployment.

Multi-source domain adaptation methods have garnered considerable attention in the field of fault diagnosis due to their ability to integrate diverse knowledge from different domains. However, traditional approaches often overlook the heterogeneity among data domains and the imbalance in their respective contributions. Typically, these methods rely on static transfer strategies to directly align all source and target domain data, which limits the potential of multi-source cross-domain diagnostic learning. To address this issue, a dynamic-static transfer-aware dual attention network (DSTDAN) is proposed in this study, specifically designed for fault diagnosis of core components in compressors. First, a dynamic transfer strategy is developed, in which online adaptive parameter updates are employed to finely align feature distributions across domains. Second, to account for intrinsic domain-specific differences, a static transfer strategy is introduced to enable robust cross-domain knowledge extraction, complementing the dynamic strategy and jointly forming a dynamic-static matching module. Finally, a dual-classifier adversarial training mechanism enhanced with an attention mechanism is constructed. Through multi-task collaborative learning, domain contributions to the diagnostic task are automatically weighted, effectively mitigating negative transfer and improving overall diagnostic performance. Extensive experiments conducted on rolling bearing dataset demonstrate that DSTDAN significantly outperforms existing mainstream transfer approaches.