TL;DR
A robust and noise-resilient bearing fault diagnosis framework that integrates advanced signal processing with hybrid deep learning techniques is presented, demonstrating strong robustness and generalization capability.
To tackle the persistent challenges of low bearing fault diagnosis accuracy—specifically the difficulty of extracting faint fault features under complex, variable operating conditions and strong background noise, as well as the tendency to lose deep temporal dependencies—this paper proposes a novel diagnosis method integrating FFT-VMD feature extraction with a Bi-TCN-Bi-GRU neural network. First, to overcome the heuristic parameter selection and frequency-band aliasing inherent in traditional Variational Mode Decomposition (VMD), the Fast Fourier Transform (FFT) is introduced. By extracting global spectral prior information, the FFT guides the VMD to perform adaptive decomposition and effective denoising of non-stationary vibration signals. Subsequently, a dual-path Bi-TCN-Bi-GRU diagnostic model is constructed. The dilated causal convolution mechanism of the Bidirectional Temporal Convolutional Network (Bi-TCN) is utilized to extract deep local spatial features, while the Bidirectional Gated Recurrent Unit (Bi-GRU) is integrated to deeply mine the forward and backward dynamic evolutionary dependencies within the sequence. Experimental results demonstrate that even under severe background noise with a signal-to-noise ratio (SNR) of -2 dB, the proposed method maintains an exceptional diagnostic accuracy of 98.21% on a self-built bearing dataset and 99.90% on the Southeast University (SEU) dataset. This study provides a highly robust and promising solution for enhancing bearing safety and predictive maintenance in complex industrial environments.
Reliable operation of bearings in complex industrial environments is essential. To mitigate the adverse effect of noise interference on fault-pattern recognition in bearing fault diagnosis, this paper proposes a new denoising diagnostic framework, termed the dictionary-wavelet attention network. The framework integrates a dictionary-based denoising autoencoder (DDAE) and a wavelet time–frequency attention (WTFA) module. In the DDAE, the decoder matrix is parameterized as learnable convolutional atoms with different structural preferences, and structured reconstruction is achieved through atom responses driven by encoder-predicted coefficient maps. The WTFA module highlights the principal fault-related informative bands and their neighboring regions, thereby enhancing the specificity of feature representation. Experiments on the Case Western Reserve University dataset and the University of Ottawa variable-speed bearing dataset show that, under severe Gaussian white noise contamination (SNR = −4 dB), the proposed method achieves diagnostic accuracies of 95.94 ± 0.18% and 90.06 ± 0.25%, respectively. Under severe pink noise contamination (SNR = −4 dB), the corresponding accuracies are 92.95 ± 0.17% and 77.03 ± 0.57%, respectively. These results demonstrate that the proposed method provides stronger noise robustness than mainstream deep-learning diagnostic models.
The proposed MSFormer incorporates a parallel multi-scale Convolutional Neural Network architecture and hierarchical Transformer modules to comprehensively process 1D vibration signals to provide a powerful and precise intelligent solution for mechanical fault diagnosis.
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.
To address the issues of low accuracy and insufficient generalization capabilities in traditional methods for diagnosing bearing faults under variable operating conditions, we propose a vision-temporal bimodal multi-channel feature fusion method for rolling bearing fault diagnosis based on domain generalization (DG). This approach constructs a parallel architecture for extracting bimodal features: on one hand, multiple signal processing techniques are employed to transform raw vibration signals into multi-perspective two-dimensional visual feature maps as visual modality input, while simultaneously employing variational modal decomposition to decompose vibration signals into a series of eigenmode functions constituting the temporal modality input. At the model level, a multi-channel large-kernel convolutional network and a global attention-enhanced bidirectional gated recurrent unit network are designed to extract deep features from the visual and temporal modalities, respectively. Subsequently, feature vectors from each channel are concatenated in the feature dimension, with fault classification performed via a progressive dimensionality reduction classifier. Experiments conducted using bearing datasets from case western reserve university and the University of Paderborn in Germany demonstrate that this method can diagnose bearing failures under cross-conditions—even when trained solely on source-domain data and without exposure to target-domain data during training—and that its DG accuracy outperforms that of existing mainstream advanced methods.