Skip to content
Log in

Similar papers

Open access Jul 2026

Noise-Robust Multi-Domain Feature Integration for Bearing 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.

Canan Taştimur · 0 citations
Open access Aug 2026

DWAN: a dictionary-wavelet attention network for noise-robust bearing fault diagnosis

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.

Y. chen, Jie Hu, Jie Ren · 0 citations
Open access Aug 2026

Adaptive Noise-Aware Bearing Fault Diagnosis via FFT Windowing and Wavelet-Based SNR-Guided LSTM Model Selection with Real-Time FPGA Implementation

This paper presents an adaptive noise-aware bearing fault diagnosis framework that integrates Fast Fourier Transform (FFT)-windowed feature extraction, wavelet-based Signal-to-Noise Ratio (SNR) estimation, and Long Short-Term Memory (LSTM) classification to maintain high diagnostic accuracy under both clean and severely noisy operating conditions. The core novelty of the proposed framework lies in its adaptive model selection mechanism, which automatically selects the most appropriate LSTM classifier according to the estimated SNR, thereby improving diagnostic robustness across different noise environments. Experiments were conducted on two benchmark datasets, the Case Western Reserve University (CWRU) dataset and the Huazhong University of Science and Technology (HUST) bearing dataset, to evaluate the generalization capability of the proposed approach. Two preprocessing pipelines were examined: time-domain normalization before FFT and frequency-domain normalization after FFT. Vibration signals were segmented without overlap to ensure unbiased evaluation. The results demonstrate that both the choice of window function and the normalization strategy significantly influence classification accuracy and robustness. Under noise-free conditions, several window types achieved accuracies above 99%, with triangular and Hamming windows providing the best performance. The combination of triangular windowing and time-domain normalization achieved the highest accuracy of 99.69%. Furthermore, time-domain normalization combined with triangular windowing exhibited superior stability and noise resistance compared with frequency-domain normalization. Under noisy conditions, noise-augmented training was found to be essential for achieving robust generalization. Models trained with moderate noise levels (8–12 dB) provided the best trade-off between accuracy and robustness, whereas excessive noise during training degraded performance. To accommodate varying noise environments, a lightweight wavelet-based SNR estimator was used to categorize operating conditions into low-, medium-, and high-SNR regions and select the corresponding LSTM classifier. The proposed framework was successfully implemented on a ZedBoard FPGA (Field-Programmable Gate Array) development board using a System-on-Chip (SoC) architecture. Experimental results show that, with a sampling frequency of 48 kHz and a processing window of 2048 samples, the proposed system updates the diagnostic result every 42.7 ms, demonstrating its suitability for real-time industrial condition monitoring and intelligent predictive maintenance applications.

Salim Hamouda, Yassine Amirat, Samir Hamdani et al. · 0 citations
Open access Aug 2026

Bearing Fault Diagnosis Based on a Spectral-Guided Adaptive Multi-Scale Convolutional Network

To address the degradation of cross-condition diagnostic performance caused by feature-scale drift in rolling bearing vibration signals under variable operating conditions, this paper proposes a spectral-guided adaptive multi-scale convolutional neural network (SAMACNN). First, PSD sequences and time-frequency features are introduced as dual-stream inputs. While the time-frequency main branch extracts local information, the Spectral Transformer spectral bypass branch captures long-range dependencies in harmonic structures. Second, dynamic gating weights are generated for the multi-scale convolutional branches, enabling sample-conditioned multi-scale feature selection and fusion and alleviating scale mismatch caused by fixed receptive fields and static fusion. Finally, data collected from two bearing fault simulation test rigs are used to verify the effectiveness and superiority of the proposed algorithm. The experimental results show that the proposed SAMACNN method achieves average accuracies of 97.16% and 97.56% on the two datasets, respectively, outperforming the ablation variants and demonstrating strong robustness and generalization capability in complex variable-condition measurement environments.

DaXin Li, Wang Hong, Hai Xue et al. · 0 citations
Preprint Aug 2026

A Time-Frequency Dual-Domain Multi-Scale Convolutional Neural Network for Bearing Fault Diagnosis under Strong Noise

To address the degradation of bearing fault diagnosis accuracy under strong noise, this paper proposes a time-frequency dual-domain multi-scale convolutional neural network. The time-domain branch employs three parallel convolutional kernels to capture multi-scale impulse features, while the frequency-domain branch applies the Fast Fourier Transform to extract noise-robust spectral structure information. Features from both branches are fused for fault classification, yielding a compact model of 110,122 parameters. Experiments on the CWRU bearing dataset across seven signal-to-noise ratio levels demonstrate that the proposed method achieves 99.75% accuracy under clean conditions and maintains 92.50% at -4 dB SNR, representing a 7.25 percentage-point improvement over the single-domain baseline with monotonically increasing gains under stronger noise. Ablation experiments validate the independent performance contributions of the time-domain multi-scale branch and the frequency-domain branch. Comparative experiments against WDCNN, DRSN-CW, MCNN, and 1D-LeNet confirm the superiority of the proposed method under strong noise conditions.

Yanxi Ding, Tingyue Jia · 0 citations