EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Explainable Physics‐Informed Hybrid Modeling for Anomaly Detection and Structural Fatigue Assessment in Wind Turbines Using SCADA Data

Rawnak Omar, Mabrouk Mosbahi, Anoire Ben Jdidia, Fathi Djemal, Antonio Pantano, Tullio Tucciarelli, Hassen Trabelsi, Mohamed Haddar

Shock and Vibration · 2026-01 · 2026 (1)

DOI: 10.1155/vib/3021162 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Shock and Vibration, online erschienen am 24.07.2026. Verfasst von Rawnak Omar und 7 weiteren Autor:innen. Die Arbeit nennt 41 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: Università degli Studi di Palermo.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Explainable artificial intelligence (XAI) is increasingly essential for trustworthy decision‐making in wind energy systems, particularly for anomaly detection and condition monitoring using SCADA data. While recent studies have achieved high accuracy in short‐term power forecasting, they often rely on black‐box models and fail to account for long‐term structural fatigue. This study proposes a physics‐informed and explainable hybrid framework for anomaly detection and fatigue assessment in wind turbines using SCADA data from the Yalova wind farm (Turkey). The proposed approach integrates a random forest–XGBoost stacking ensemble with a Palmgren–Miner‐based fatigue index to capture cumulative structural degradation under variable wind conditions. Hyperparameter optimization is performed using Optuna to enhance model performance. To improve model transparency, both global and local explainability techniques are incorporated, including the SHapley Additive exPlanations (SHAP) for global feature importance analysis and LIME for instance‐level interpretation. The proposed framework achieved very high predictive performance, with an F1‐score of 0.9992, recall of 0.9996, precision of 0.9989, and ROC–AUC of 1.0000. Precision–recall analysis further confirmed strong performance under a class imbalance condition. These results should, however, be interpreted with caution, as the anomaly labels are derived from threshold‐based criteria applied to SCADA variables. An ablation study reveals a trade‐off between predictive accuracy and physical consistency, showing that the integration of fatigue‐related features enhances interpretability and alignment with mechanical degradation processes. SHAP and LIME analyses highlight wind speed and power deviation as primary drivers of anomalies, modulated by fatigue accumulation effects. The proposed framework provides a transparent and physically grounded approach that supports condition monitoring and contributes to more informed maintenance decision‐making in wind energy systems.

Förderung

  • Università degli Studi di Palermo

Zitieren

APA

Omar, R., Mosbahi, M., Jdidia, A. B., Djemal, F., Pantano, A., Tucciarelli, T., Trabelsi, H., & Haddar, M. (2026). Explainable Physics‐Informed Hybrid Modeling for Anomaly Detection and Structural Fatigue Assessment in Wind Turbines Using SCADA Data. Shock and Vibration, 2026(1). https://doi.org/10.1155/vib/3021162

Vancouver

Omar R, Mosbahi M, Jdidia AB, Djemal F, Pantano A, Tucciarelli T, et al. Explainable Physics‐Informed Hybrid Modeling for Anomaly Detection and Structural Fatigue Assessment in Wind Turbines Using SCADA Data. Shock and Vibration. 2026;2026(1). doi:10.1155/vib/3021162

BibTeX
@article{Omar2026,
  author = {Omar, Rawnak and Mosbahi, Mabrouk and Jdidia, Anoire Ben and Djemal, Fathi and Pantano, Antonio and Tucciarelli, Tullio and Trabelsi, Hassen and Haddar, Mohamed},
  title = {Explainable Physics‐Informed Hybrid Modeling for Anomaly Detection and Structural Fatigue Assessment in Wind Turbines Using SCADA Data},
  journal = {Shock and Vibration},
  year = {2026},
  volume = {2026},
  number = {1},
  doi = {10.1155/vib/3021162}
}
Kostenlos starten →