2 December 2025 in Measurement Science and Technology
Fault diagnosis methods based on deep learning often require a large amount of labeled data to train the network. However, in engineering applications, fault data collection is limited, and deep learning methods lack interpretability, which restricts the effectiveness of fault diagnosis models. This paper introduces a novel Generative Adversarial Network (RCTECAM GAN) to address the problem of class imbalance and the interpretability of neural networks. Firstly, ResCNN-Transformer (RCT) is designed as the generator architecture, which comprehensively captures the features of raw one-dimensional time-series signals from both global and local perspectives, thereby improving the quality of generated samples. Secondly, Eigen-CAM (ECAM) is embedded in the classifier to highlight key regions that significantly contributing to the final diagnostic results, introducing interpretability into RCTECAM GAN. Experimental results demonstrate that the proposed method exhibits excellent performance in handling data class imbalance, achieving a fault diagnosis accuracy of 99.31%, which outperforms existing methods in addressing data class imbalance and enhancing the interpretability of deep learning models. This provides new ideas for solving the problems of class imbalance and interpretability of neural networks in the field of fault diagnosis.
25 November 2024 in Measurement Science and Technology
The mechanism analysis of the sudden change in load inertia caused by the failure of large petrochemical fan blades is unclear, which makes it difficult to diagnose imbalanced samples faults based on data-driven methods and has the problem of poor interpretability. Based on this issue, a fault diagnosis method for imbalanced samples of broken blades in large petrochemical fan under sudden changes in load inertia is proposed. This method firstly establishes a failure mechanism model for large petrochemical fan blades, revealing the physical characteristics between inertia, torque and speed under sudden changes in load inertia. Secondly, based on the failure mechanism model, the fault characteristics of broken blades in petrochemical fan are extracted to solve the problem of difficult feature extraction of fault samples in imbalanced samples. Finally, a data-driven diagnostic model was constructed under the constraint of sudden changes in load inertia to improve the interpretability of fault diagnosis for large petrochemical fan with broken blades. The experiment shows that the proposed method significantly improves the diagnostic accuracy and effectiveness for detecting faults in broken blades of petrochemical fans, achieving a fault diagnosis accuracy of 99.03%.
6 May 2026 in Engineering Research Express
Under the influence of complex operating conditions and strong noise interference, achieving precise and efficient fault diagnosis for rotating machinery remains a significant challenge. Although deep learning-based fault diagnosis methods have achieved high recognition accuracy, they often lack effective explanations for the model's decision-making basis. To address this issue, this paper proposes an interpretable dual-stream heterogeneous information fusion model named Interpretable MS-ResCNN-Transformer (IMS-RCT). First, a cascaded feature extraction architecture based on MS-ResCNN and Transformer is constructed to achieve synergistic modeling of local fault features and global dependencies. Second, a dual-stream heterogeneous information learning strategy is designed, utilizing 1D vibration signals and 2D time-frequency spectrograms as parallel inputs, enabling the model to fully exploit complementary features in both the time and time-frequency domains. Furthermore, an interpretability analysis framework based on Score-Class Activation Mapping (Score-CAM) is introduced to achieve effective attribution and visual explanation of the diagnostic decision basis. Experimental results demonstrate that IMS-RCT exhibits excellent diagnostic performance under various fault scenarios, with a fault recognition accuracy reaching 99.21%, which is significantly superior to multiple comparison methods. Meanwhile, the interpretability analysis validates a strong consistency between the features focused on by the model and typical fault mechanisms, further enhancing the reliability and trustworthiness of the model in practical engineering applications.
1 October 2026 in Measurement Science and Technology
Accurate assessment of fault severity in rolling bearings under strong noise remains a critical challenge for intelligent predictive maintenance. In this study, the diagnostic task is formulated as joint fault type and severity classification, with emphasis on severity-level discrimination. In practical industrial environments, vibration signals are frequently contaminated by complex noise, including Gaussian noise, pink noise, Laplace (impulsive) noise, and combined interference, which severely mask transient fault impulses and distort time–frequency representations. To tackle this problem, we propose a Dual-Stream Gated Cross-Modal Attention Fusion Network (DS-GCMAF). The framework simultaneously processes one-dimensional (1D) raw vibration sequences and two-dimensional (2D) time–frequency representations. A bidirectional cross-modal multi-head attention mechanism is introduced to facilitate effective information interaction and alignment across heterogeneous feature spaces. Meanwhile, a quality-aware adaptive gating strategy is employed to dynamically regulate the contribution of each modality based on its reliability. In low signal-to-noise ratio (SNR) conditions, the less reliable 2D branch is selectively attenuated, and the more robust 1D stream serves as an anchor for cross-modal calibration. Extensive experiments were carried out on the Paderborn University (PU) dataset and HUSTbearing dataset under four complex noise types. On the PU dataset across varying SNR levels (−8 dB to 8 dB), DS-GCMAF achieves 85.80% accuracy at −8 dB, surpassing the strongest baseline by 2.69 percentage points. On the HUSTbearing dataset under −10 dB, the proposed method attains 74.04% accuracy under the most challenging combined noise condition, significantly outperforming existing single-stream and conventional fusion approaches. These results demonstrate the superior robustness and its effectiveness in joint fault-type and severity classification under highly noisy environments.