Data-driven approaches are widely employed for diagnosing open-circuit faults in inverters. However, their performance often deteriorates significantly under unknown operating conditions due to reliance on known training domains. To address this, this paper proposes a cross-domain differential attention network (CDDAN). Firstly, a wavelet principal component feature extraction strategy is introduced. This method precisely locates sensitive frequency bands through wavelet transformation, then employs principal component analysis to compress them into compact, discriminative features, effectively suppressing noise and redundancy. Subsequently, the CDDAN module utilizes a differential attention mechanism to conduct parallel comparative learning across multiple known operating conditions. This mechanism explicitly captures feature differences, thereby extracting key fault characteristics insensitive to operating condition variations. To further enhance the model's generalization capability, a multi-scale differential loss function was designed. This function synergistically optimizes multiple objectives, simultaneously improving diagnostic accuracy and cross-operating-condition adaptability. Experimental results demonstrate that this method can effectively diagnose all 22 open-circuit faults under unknown operating conditions, outperforming existing mainstream approaches.

Rotating machinery is a key component in modern industry, and its operating condition directly affects equipment safety and production reliability. However, discrepancies among different machines cause source–target distribution shifts, while fault annotation for target machines is costly, limiting the performance of deep learning-based diagnosis under cross-machine scenarios with limited labels. To address these issues, this paper proposes a multi-scale time–frequency semantic consistency model based on self-supervised transferable representation learning, termed MTFSC. First, augmented waveform views and multi-scale frequency-domain views are constructed from unlabeled source-domain vibration signals for self-supervised pre-training without source labels. Then, a time-domain impulse-aware feature extractor and a time–frequency decoupled spectral feature extractor are designed to enhance local impulsive responses and emphasize fault-sensitive time–frequency patterns. Furthermore, a semantic-aware soft contrastive loss is developed to mine potential semantic neighbors from multi-scale frequency-domain structural similarity, reducing false-negative effects in conventional hard-label contrastive learning. Finally, the pre-trained time-domain extractor is transferred to the target machine and fine-tuned with limited labeled samples. Experimental results show that MTFSC outperforms comparison methods under different labeled sample ratios and achieves an average accuracy of 97.5% across four cross-machine diagnostic tasks.

Fault diagnosis in industrial environments is often constrained by the scarcity of labeled data, variability of operating conditions and cross-device deployment. Moreover, relying solely on vibration signals results in limited diagnostic accuracy and robustness for motors under complex working conditions. To address these challenges, this paper proposes a novel self-supervised motor fault diagnosis framework based on vibration–current multimodal contrastive learning, in which current features are innovatively incorporated into the contrastive learning paradigm. Specifically, a vibration–current multimodal contrastive learning is first developed to enhance discriminative representation learning while improving cross-modal feature consistency, thereby enabling deeper semantic interaction and complementary information fusion between different modalities. Subsequently, to address the adaptive feature encoding of multimodal signals, a dynamically weighted multi-scale residual encoder is constructed, which effectively extracts heterogeneous modal features by adaptively adjusting the importance of multi-scale representations. Finally, the effectiveness of the proposed framework is validated through three case studies. Experimental results demonstrate that an average diagnostic accuracy of 97.75% was achieved across all cross-device tasks using only 5% labeled samples. In addition, the proposed model contains 11.89M parameters and requires only 2.58s for fine-tuning, highlighting its robustness, reliability, and strong application potential in label-limited industrial scenarios.

The three-phase voltage source inverter is a key component in the drive system of electric construction machinery, and its reliable operation directly affects production efficiency and operational safety. However, frequent and sudden load variations introduce strong dynamic disturbances, altering the open-circuit fault characteristics of the inverter and hindering real-time fault detection. Traditional diagnostic methods, which typically rely on fixed thresholds or steady-state features, struggle to maintain accuracy under such varying conditions. To address this challenge, this paper proposes a dynamic adaptive open-circuit fault diagnostic method. The proposed approach requires only three-phase current measurements and constructs phase current vectors using even-odd symmetric components, enabling efficient detection of multiple fault types without complex computation. In addition, a distributed adaptive threshold algorithm is developed to automatically adjust thresholds based on the relative variations of the current vector. Experimental results demonstrate that the proposed method can accurately identify open-circuit faults while maintaining strong robustness and generalization capability under varying load conditions.

Rotating machinery plays a critical role in transmission systems, while the scarcity of fault samples and labeled data limits the performance of existing diagnostic methods under few-shot conditions. This paper proposes a few-shot fault diagnosis method for rotating machinery based on a time-frequency dual-stream Mamba architecture and joint metric learning. Raw vibration signals and multi-scale short-time Fourier transform time-frequency views are constructed to characterize transient impacts and frequency-band energy distributions from complementary perspectives. A dual-stream feature extraction network is developed, where the time-domain branch captures impulsive fault patterns, and the time-frequency spatial Mamba and channel attention branches extract spatial dependencies and fault-sensitive responses. Moreover, a hierarchical selective fusion mechanism is introduced to adaptively integrate complementary features across different domains. Finally, a covariance-based joint metric learning module is designed to model class distributions using second-order statistics of support samples and classify query samples through local similarity aggregation. Experimental results show that the proposed method consistently outperforms comparison methods under different rotational speeds, training sample scales, and 8-way 1-shot/5-shot tasks, validating its effectiveness and robustness for few-shot rotating machinery fault diagnosis.