STRUCTURAL HEALTH MONITORING OF WIND TURBINE BLADES USING VIBRATION-BASED DAMAGE DETECTION AND MACHINE LEARNING

Abstrak

Wind turbine blades are subjected to complex, time-varying aerodynamic and gravitational loads over 20-year design lives, making early damage detection essential to prevent catastrophic failure and reduce operation-and-maintenance (O&M) costs, which represent 20–35% of levelised cost of energy (LCOE) for onshore wind. This thesis develops and validates a vibration-based structural health monitoring (SHM) framework for a 45-metre GFRP wind turbine blade, combining high-fidelity finite element (FE) modelling with machine learning classification of damage scenarios. A full-scale FE model of the blade (ANSYS Mechanical, 180,000 shell elements) was validated against experimental modal analysis (EMA) data from a 1:10 scale laboratory specimen; first five natural frequencies matched within 3.2%. Damage scenarios (delamination at 25% and 60% span, trailing-edge debonding, leading-edge erosion) were simulated by selectively reducing element stiffness. Modal strain energy (MSE) damage indicators were extracted from the FE database of 1800 damage cases. A Random Forest (RF) classifier trained on MSE features achieved 96.8% damage detection accuracy and 91.3% location accuracy (5-class) in cross-validation. Deployment on a 5-node wireless sensor network (MEMS accelerometers, 1000 Hz sampling) on the physical specimen confirmed real-world detection of 40 mm × 40 mm artificial delaminations with 93% accuracy, validating the computational framework for field implementation.