Jul 2026· Advances in Engineering Technology Research· Vol 17, pp. 329· 0 citations
TL;DR
A transferability-aware weighted fusion strategy integrates multi-source classification results and a shared feature extraction network integrating multi-scale convolutions, LSTM, and ECA mechanism is constructed to extract domain-invariant features.
Abstract
To improve the fault diagnosis accuracy of marine propulsion shaft bearings under variable working conditions and with limited samples, a Multi-scale Deep Multi-source Subdomain Adaptation Network (MDMSAN) is proposed. First, DCGAN is employed to augment source domain samples. A shared feature extraction network integrating multi-scale convolutions, LSTM, and ECA mechanism is constructed to extract domain-invariant features. Private feature extractors are designed for each source-target domain pair, utilizing LMMD for fine-grained subdomain alignment. A transferability-aware weighted fusion strategy integrates multi-source classification results. Experiments on CWRU and PT500 datasets demonstrate that MDMSAN achieves average accuracies of 99.92% and 99.73% under variable working conditions, and maintains 98.85% accuracy with only 1/16 of the target samples, significantly outperforming existing methods.
Experimental validation on the Harbin Institute of Technology aero-engine inter-shaft bearing dataset shows that the proposed model achieves 97% diagnostic accuracy under extreme noise conditions (SNR = -5 dB).
Yang Wang, Boliang Zhang· Scientific Reports· 0 citations
A multi-scale linear attention multi-source subdomain adaptation network (MLAMSAN) that integrates the multi-scale linear attention (MLA) that can achieve fault diagnosis under cross operating conditions through subdomain feature alignments that exhibits the superior diagnostic performance and the strong generalization ability.
Zheng Han, Yuqi Fan, Yaping Wang et al.· Engineering Research Express· 0 citations
Slewing bearings in low-speed, heavy-load equipment generate weak and heterogeneous fault signatures that are difficult to characterize using a single sensor. This study develops a compact dual-branch feature-fusion framework that jointly exploits six-channel vibration and one-channel acoustic-emission (AE) signals. To prevent source-record leakage, complete raw recording groups are assigned to training, validation, and test subsets before segmentation; non-overlapping 1024-point windows are then generated, and channel normalization is fitted using training groups only. Five matched random-seed runs are performed, with the best checkpoint selected exclusively by validation Macro-F1. The fusion model achieves mean test accuracy of 99.10% and Macro-F1 of 0.9910, compared with 97.38%/0.9738 for vibration-only and 98.03%/0.9802 for AE-only. The improvement over vibration-only is statistically significant (p = 0.022 for both Accuracy and Macro-F1), whereas the improvement over AE-only is numerical but does not reach the 0.05 significance level (p = 0.063). The fusion model also obtains the lowest mean Davies-Bouldin index (0.963). These results support compact vibration-AE fusion as an effective diagnostic baseline while also defining its statistical and deployment limitations.
Multi-sensor data fusion is widely recognized as a key enabler for reliable aero-engine condition monitoring under complex operating conditions. However, this paradigm inherently increases system complexity. This paper conducts a systematic empirical study to evaluate the practical efficacy and boundaries of deep feature mining using only a single channel. We propose a Local-Global State Space Model Network (LG-SSMNet), which uses multi-scale 1D CNNs to fit local transient pulse responses and employs a Selective State Space Model (Mamba) to construct a long-term periodic integrator, achieving feature decoupling within a single channel. To address the discretization divergence issue of continuous vibration signals, an empirical stabilization configuration involving dual normalization and stratified learning rates is introduced. Deep quantitative analysis on the HIT aero-engine bearing dataset shows that the four acceleration channels achieve stable accuracies over 99.6% (approaching 100%), while displacement channels show significant divergence. The underperforming channel (Sensor 1) reached only 91.31% due to a critical missed detection rate of outer race faults (4.3% misclassified as healthy). Through joint empirical evidence using ROC-AUC, F1-scores, and qualitative feature visualization, we demonstrate that with reasonably selected acceleration sensors, single-channel deep mining can serve as a practical alternative to multi-channel fusion.
Lei Zhang, Z. Ren· 2026 IEEE International Conf...· 0 citations
Rolling bearings operate under varying working conditions, posing a significant challenge to achieving high-accuracy bearing fault diagnosis. To enhance the fault diagnosis performance of rolling bearings under cross-domain working conditions and noisy environments, this study proposes a similarity-based multiscale cluster region transfer network (SMCRTN), which integrates three core modules: similarity-based processing (SBP), multiscale concatenation U-Net (MCU-Net), and dynamic cluster region transfer (DCRT). Specifically, the SBP module selects source domain samples using entropy score and anomaly score, and optimizes target domain samples through statistical projection. Meanwhile, the MCU-Net incorporates Hilbert-transformed inputs, gated convolution (gated-conv) blocks, and standard convolution blocks to extract multiscale domain-invariant features via dynamic weight adjustment. Furthermore, the DCRT module achieves cross-domain alignment by leveraging cluster-based region transfer and minimizing dynamic entropy-weighted loss. To verify the feasibility and effectiveness of the proposed SMCRTN, comprehensive experiments are performed on the public CWRU dataset and the proprietary PT dataset. Experimental results under various transfer tasks and noise levels demonstrate that the SMCRTN outperforms other intelligent models in terms of diagnostic accuracy and transferability.
Xianfeng Li, Jiantao Shi, Chuang Chen et al.· IEEE Transactions on Reliabi...· 0 citations
To address the challenge of fault diagnosis from high-speed train wheelset bearings towards complex industrial variable speed conditions, we propose a Multi-modal Adaptive Signal Diagnosis Network (MASDN). Firstly, a hybrid architecture combining multi-scale convolution with Informer Encoders is designed to capture time-varying speed features, which dynamically guide vibration feature extraction through adaptive convolutional kernel weighting. Subsequently, combining parallel multi-scale convolutional module enhanced by channel attention mechanisms for robust vibration feature extraction across varying operational conditions. Meanwhile, an adaptive structure incorporating multi-domain constraint loss functions is designed for optimal signal enhancement and noise suppression. On this basis, a speed-guided feature fusion strategy, which automatically prioritizes the most condition-relevant features, is introduced, employing depthwise separable convolution (DSConv) to achieve cross-scale integration of multi-modal features. Experiments indicate that MASDN effectively improves diagnostic accuracy and robustness in scenarios simulating industrial variable speed conditions with ablation studies further validating the effectiveness of each component in the model.
Jingwen Liu, Zengqiang Ma· Measurement science and tech...· 0 citations
