A Physics-Informed Machine Learning Framework for Structural Damage Detection: Integrating Modal Analysis with SHAP-Based Feature Importance
Journal of Studies in Civil Engineering · 03.06.2026 · 3 (1) · 74-92
DOI: 10.53898/jsce2026315 ↗Kurz gefasst
Fachartikel in der Zeitschrift Journal of Studies in Civil Engineering, online erschienen am 03.06.2026. Verfasst von Bibek Pokhrel und 5 weiteren Autor:innen. Die Arbeit nennt 44 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
Structural health monitoring systems increasingly rely on machine learning, yet these models often function as black boxes that obscure the physical meaning of their predictions. This study addresses this gap by developing a novel Frequency Importance Weighted Damage Index (FIWDI) that bridges physics-based modal analysis with interpretable machine learning through SHAP value integration. The framework was validated using the Z-24 Bridge benchmark dataset, where we extracted modal frequencies from 45 test configurations spanning healthy and progressively damaged states. A CatBoost classifier achieved perfect classification accuracy (100%), with SHAP analysis revealing that the fourth vibration mode contributed 34.54% to damage detection—an unexpected finding that challenges conventional assumptions about lower-mode dominance. The FIWDI demonstrated superior sensitivity compared to traditional frequency-based indices. Statistical validation confirmed robustness across environmental variations. This work demonstrates that interpretable machine learning can enhance rather than replace physics-based understanding in structural engineering applications.
Zitieren
APA
Pokhrel, B., Awasthi, B., Bohora, A., Kumari, D., Kunwar, P., & Paudel, B. (2026). A Physics-Informed Machine Learning Framework for Structural Damage Detection: Integrating Modal Analysis with SHAP-Based Feature Importance. Journal of Studies in Civil Engineering, 3(1), 74-92. https://doi.org/10.53898/jsce2026315
Vancouver
Pokhrel B, Awasthi B, Bohora A, Kumari D, Kunwar P, Paudel B. A Physics-Informed Machine Learning Framework for Structural Damage Detection: Integrating Modal Analysis with SHAP-Based Feature Importance. Journal of Studies in Civil Engineering. 2026;3(1):74-92. doi:10.53898/jsce2026315
BibTeX
@article{Pokhrel2026,
author = {Pokhrel, Bibek and Awasthi, Bighyan and Bohora, Aayushma and Kumari, Damnee and Kunwar, Prabesh and Paudel, Brindal},
title = {A Physics-Informed Machine Learning Framework for Structural Damage Detection: Integrating Modal Analysis with SHAP-Based Feature Importance},
journal = {Journal of Studies in Civil Engineering},
year = {2026},
volume = {3},
number = {1},
pages = {74--92},
doi = {10.53898/jsce2026315}
}
EUVIMED