Structural assessment and machine learning-based fatigue prediction of wind turbine blades

This thesis investigates the structural behaviour and fatigue performance of composite wind turbine blades by integrating full-scale mechanical testing with machine learning (ML) techniques. As wind energy systems grow in importance for delivering sustainable and cost-effective power, accurate prediction blade performance remains a technical challenge due to complex loading, material variability, and environmental effects. The primary objective of this research is to develop a hybrid experimental-ML framework to improve fatigue life prediction and reduce reliance on time-consuming full-scale testing. To achieve this, two full-scale blades - a 13-metre glass fibre-reinforced epoxy blade and a 4.5-metre blade made from woven Twintex - were subjected to static, dynamic, and fatigue testing in accordance with IEC TS 61400-23. Key mechanical parameters such as tip deflection, strain, natural frequencies, and stiffness were experimentally obtained and validated against finite element models developed in ABAQUS. Complementing the physical tests, a suite of supervised ML models - Random Forest (RF), Gradient Boosting (GB), Bagging, Decision Trees (DT), XGBoost, and Gene Expression Programming (GEP) – were trained using experimental and literature-based fatigue data. Input features included fibre volume fraction, maximum/minimum stress, frequency, R-value, and laminate thickness. The best performing models (RF and Bagging) achieved mean R² values above 0.91 during 10-fold cross-validation, while Grading Boostinghad the lowest average MAE (0.73 log cycles), indicating high absolute accuracy in fatigue life prediction. Feature importance analysis and SHAP (SHapley Additive exPlanations) values revealed that loading frequency is the most influential factor in fatigue progression, followed by stress levels and thickness, aligning with physical fatigue mechanisms. The use of explainable AI techniques enabled the interpretation of model behaviour and enhanced model transparency for practical deployment. A user-friendly Graphical User Interface (GUI) was also developed using Python’s Tkinter framework, allowing non-programmers to predict fatigue life and mechanical performance based on defined inputs. This GUI acts as a bridge between data science models and engineering applications, promoting user accessibility and supporting real-time decision-making in blade design, testing, and maintenance. The thesis contributes significantly to the field by offering a validated, scalable, and interpretable approach to structural performance prediction, reducing the dependency on full-scale testing and enabling faster design iterations. It aligns closely with the goals of structural health monitoring (SHM), predictive maintenance, and digital twin strategies for wind energy systems. The hybrid framework developed here serves not only to improve current design and assessment methodologies but also sets a foundation for future research into physics-informed ML, real-time SHM integration, and model generalisation to a broader class of composite structures. The framework demonstrates the potential of combining experimental mechanics with explainable AI to accelerate innovation in sustainable energy technologies.

Publisher

University of Galway

Rights

CC BY-NC-ND

cbn

Collections

University of Galway Theses (Research Masters Theses)