PSI - Issue 80

Nima Rezazadeh et al. / Procedia Structural Integrity 80 (2026) 411–417 Author name / Structural Integrity Procedia 00 (2019) 000 – 000

416

6

(c)

(d)

Fig. 4. (a) Paderborn feature and prototype distribution with source and target prototypes, (b) Paderborn prototype trajectories across training, (c) CWRU feature and prototype distribution with prototypes; (d) CWRU prototype trajectories across training to final alignment

For Paderborn, the feature-prototype scatter (Fig. 4-a) shows 4 clear clusters that most class centroids overlap tightly across domains, while one remains noticeably shifted, indicating residual misalignment. Its prototype trajectories (Fig. 4-b) converge smoothly without oscillation, with three classes fully aligned and one still lagging, suggesting targeted refinement. For CWRU, the scatter (Fig. 4-c) displays nearly perfect co-location of source and target points and centroids, and the trajectories (Fig. 4-d) move steadily to common endpoints, i.e., uniformly robust domain transfer. 5. Conclusion The proposed prototype attention domain adaptation framework, named Proto-ADDA-X, addresses the problem of domain shift in bearing fault diagnosis. It integrates interpretable feature engineering, dynamic prototype updating, and attention weighting to deliver accurate and transparent fault classification. The framework achieved perfect classification accuracy in unsupervised conditions on both the Paderborn and CWRU datasets. Visualizations of prototype behavior and feature space alignment support the model ’s practical value for real -world diagnostics in rotating machinery. This method offers an effective and lightweight solution for explainable fault diagnosis across different operating environments. References Asutkar S, Tallur S. Deep transfer learning strategy for efficient domain generalisation in machine fault diagnosis. Sci Rep 2023;13. https://doi.org/10.1038/s41598-023-33887-5. Ding Y, Jia M, Zhuang J, Cao Y, Zhao X, Lee C-G. Deep imbalanced domain adaptation for transfer learning fault diagnosis of bearings under multiple working conditions. Reliab Eng Syst Saf 2023;230:108890. https://doi.org/10.1016/j.ress.2022.108890. Herwig N, Borghesani P. Explaining deep neural networks processing raw diagnostic signals. Mech Syst Signal Process 2023;200:110584. https://doi.org/10.1016/j.ymssp.2023.110584. Kim K, Kim YS. Vibration spectrogram analysis for bearing fault diagnosis based on grad-cam for feature selection and statistical approach. Journal of Mechanical Science and Technology 2024;38:5885–98. https://doi.org/10.1007/s12206-024-1010-3. Lessmeier C, Kimotho JK, Zimmer D, Sextro W. Condition Monitoring of Bearing Damage in Electromechanical Drive Systems by Using Motor Current Signals of Electric Motors: A Benchmark Data Set for Data-Driven Classification. PHM Society European Conference 2016;3. https://doi.org/10.36001/phme.2016.v3i1.1577. Li G, Wu J, Deng C, Chen Z. Parallel multi-fusion convolutional neural networks based fault diagnosis of rotating machinery under noisy environments. ISA Trans 2022;128:545–55. https://doi.org/10.1016/j.isatra.2021.10.023. Loparo KA. Case Western Reserve University Bearing Data Center. Bearings Vibration Data Sets, Case Western Reserve University 2012;http. Magadán L, Ruiz-Cárcel C, Granda JC, Suárez FJ, Starr A. Explainable and interpretable bearing fault classification and diagnosis under limited data. Advanced Engineering Informatics 2024;62:102909. https://doi.org/10.1016/j.aei.2024.102909.