Jul 2026· Structural Health Monitoring· 0 citations· 23 references
TL;DR
A novel joint hierarchical suppression framework is developed, which collaboratively operates on channel and spatial dimensions under domain label supervision to identify and remove domain-specific features in shallow network layers and is embedded into fully connected layers to drive the model to learn residual domain-invariant features, thus greatly boosting generalization performance.
Abstract
Domain generalization (DG)-based fault diagnosis can effectively suppress domain drift interference induced by varying operating conditions of rolling bearings, enabling the model to maintain high accuracy and stable diagnostic performance under unseen working scenarios. However, most existing DG methods passively extract domain-invariant features via domain alignment, feature decoupling or data augmentation. Due to the absence of explicit constraints, their performance is restricted and unstable. Furthermore, they fail to take into account the suppression of spurious features triggering “shortcut learning,” which further degrades the generalization capability of the model. To address these limitations, this paper proposes a novel joint hierarchical suppression framework. Firstly, a cross-dimensional suppression algorithm is developed, which collaboratively operates on channel and spatial dimensions under domain label supervision to identify and remove domain-specific features in shallow network layers. Secondly, a global feature suppression mechanism is embedded into fully connected layers to drive the model to learn residual domain-invariant features, thus greatly boosting generalization performance. Finally, the above modules are integrated into an end-to-end trainable architecture to systematically extract robust domain-invariant representations. The proposed approach is verified on three rolling bearing datasets from the University of Ottawa, Huazhong University of Science and Technology, and Lanzhou University of Technology under unseen working conditions, achieving average diagnostic accuracies of 98.73, 96.32, and 99.22%, respectively. Experimental results verify the effectiveness and superiority of the presented method in identifying and eliminating domain-specific features.
In this study, a hybrid deep learning architecture is proposed for robust vibration-based fault diagnosis in industrial machinery by jointly modeling time-domain, frequency-domain, and temporal dynamics. In the time domain, convolutional layers combined with an Enhanced Gated Attention (EGA) mechanism emphasize informative signal components while suppressing noise. Temporal evolution is modeled using Neural Ordinary Differential Equations (Neural ODEs), enabling smooth and stable continuous-time feature representations. In parallel, a Fourier Neural Operator (FNO) extracts frequency-domain characteristics, augmented with gated attention to focus on fault-related spectral patterns. Long Short-Term Memory (LSTM) layers capture long-range dependencies, while Squeeze-and-Excitation (SE) blocks adaptively recalibrate channel-wise feature responses. A multi-scale attention-based fusion module integrates domain-specific representations and auxiliary features to enhance discrimination under varying operating conditions. The proposed model is evaluated on the SUBFv1.0 dataset through extensive ablation studies and experiments under multiple noise levels, achieving 98.41% accuracy in noise-free conditions and maintaining performance above 91% even at 5 dB SNR. Unlike existing multi-path approaches that combine heterogeneous features in a loosely coupled or discrete manner, the proposed architecture uniquely integrates multi-domain feature learning with bidirectional attention mechanisms and continuous-time temporal dynamics, enabling coherent cross-domain interaction and robust fault characterization.
Canan Taştimur· Information Technology and C...· 0 citations
A novel intelligent fault diagnosis framework, termed PGDS-CLNet, is proposed in this study, which is an end-to-end trainable diagnostic network after standard signal normalization and segmentation.
Fanlong Zhu, Junyu Lai, Peiwen Lu et al.· Advances in Mechanical Engin...· 0 citations
Domain generalization (DG) has been widely applied to rotating machinery fault diagnosis. Under extreme working conditions and coupled influencing factors, existing DG methods are prone to issues such as batch oscillations and early gradient optimization conflicts. To address these issues, this article proposes a novel method named the gradient-driven learnable class-center network (GLC-Net). GLC-Net incorporates two core mechanisms: 1) a globally learnable center for momentum updates. GLC-Net replaces passive batch-wise statistical updates with gradient-driven updates to the center parameters, and combines this with a momentum mechanism to alleviate optimization oscillations and 2) dynamic curriculum learning strategy is introduced. This strategy effectively prevents negative transfer by balancing the learning of discriminative features and domain-invariant features. Experimental results on both bearing and gearbox datasets demonstrate that GLC-Net achieves superior generalization performance and robustness.
Zhixiang Zhang, Tao Lu· IEEE Sensors Journal· 0 citations
Fault diagnosis of rolling bearings is crucial for operational safety. However, the scarcity of labeled data and significant domain shifts are two key challenges. Existing studies neglect robustness to physical disturbances and the interpretability of diagnostic decisions. To address these issues, this paper proposes a physics-informed cross-domain fault diagnosis framework. First, based on bearing fault modulation mechanisms, multi-domain features are extracted. A physics-regularized feature selection strategy combining Sparse Group Lasso with stability selection and Random Forest is used to retain fault-relevant features. Recognizing that velocity variations are a major source of domain shifts, order analysis is introduced as a physics-prior alignment method. Deep CORAL and pseudo-label self-training are combined for unsupervised knowledge transfer. Finally, multi-level interpretability analysis is embedded to quantify domain shift, track adaptation geometry and explain final decisions. Two case studies demonstrate its robustness, trustworthiness and engineering applicability.
Yi Xie, Ruyang Zheng, Xuepeng Guo et al.· Eksploatacja I Niezawodnosc-...· 0 citations
In recent years, deep learning-based bearing fault diagnosis models have demonstrated excellent diagnostic performance under ideal experimental conditions. However, in practical industrial scenarios, the domain shift problem caused by data distribution differences severely undermines the generalization capability of such methods, which is particularly pronounced in cross-device bearing fault diagnosis tasks. Although existing transfer learning-based cross-device bearing fault diagnosis models have achieved encouraging progress, the following shortcomings persist. On the one hand, in cross-device scenarios, the sensitivity of time-domain impact features and frequency-domain resonance characteristics to differences in device structures is different, and the available samples from the target device are often limited. However, most models rely on a single feature domain for modeling, which makes it difficult for them to effectively capture the complete physical information that encompasses both fault‑impulse priors and device‑specific response characteristics, thereby limiting their stability and generalization capability. On the other hand, most models require substantial parameter adjustments during the transfer stage, making them highly prone to overfitting to the individual physical characteristics of the target device, thus undermining the effective knowledge already learned from the source domain. To alleviate the aforementioned issues, a time-frequency collaborative cross-device bearing fault diagnosis model based on supervised transfer learning with limited data is proposed. First, a time-frequency collaborative modeling paradigm is designed, which can extract multi-view complementary information from two physical dimensions: time-domain impact response and frequency-domain structural resonance. This information reflects both the essential attributes of fault patterns and the device‑specific response characteristics, thereby enhancing the robustness of the model to variations in device structures and operating conditions. Second, the low-rank adaptation mechanism is introduced to perform lightweight fine-tuning on key parameters of the pretrained model. This mechanism can constrain parameter updates within a low‑rank subspace, enabling the model to fit the device-specific physical responses of the target device while preserving the stability of the knowledge structure learned from the source domain, thus effectively reducing training complexity and the risk of overfitting. Finally, cross-device diagnosis scenarios are constructed through three real-world cases to comprehensively evaluate the performance of the proposed model. Experimental results indicate that the proposed model outperforms some current mainstream models in terms of both diagnostic accuracy and generalization performance, sufficiently demonstrating its effectiveness and practicality in cross-device scenarios.
The inconsistency in bearing data distributions under diverse conditions often affects the representations of the faulty data and leads to indistinct decision boundaries and even negative transfer resulted from overlapping class distributions, greatly limiting the accuracy of the diagnosis model. To cope with the challenge, a pseudo-label-guided dual-supervised alignment (PDSA) method is developed for bearing fault diagnosis across diverse operating scenarios in this paper. To address the fixed alignment strategy issue, an adaptive distribution alignment layer is incorporated to ResNet18 to achieve dynamic data distribution alignment under varying condition, To enhance classification performances, a dual-supervised mechanism, comprising shallow-layer supervised contrastive learning is introduced through target domain pseudo-labels in target domain and deep-layer regularization class consistency. Experiments on two publicly available bearing datasets demonstrated this model realizes refined class-level alignment, strengthens fault states representation, and shows notable superiority in both accuracy and robustness.