Autonomous structural health monitoring of composite wind turbine blades using guided waves and machine learning

This research paper presents a guided wave (GW)-driven framework for structural health monitoring of composite wind turbine blades, leveraging both experimental and numerical data in conjunction with a hybrid machine learning (ML) approach for accurate damage identification and classification. High-fidelity ultrasonic GW signals were collected under controlled laboratory conditions for pristine and damaged blade states, including erosion damage, longitudinal debonding, and transverse debonding. Finite element simulations, incorporating a tri-array of sensors, were further employed to enhance spatial resolution and replicate complex wave-damage interactions. All GW signals were converted into time–frequency representations using scalogram analysis, enabling rich feature encoding of frequency dispersion characteristics for each damage case. These scalogram images were used as input to a two-stage ML classifier based on transfer learning, which first performs binary damage detection, followed by multi-class classification of damage types. The proposed model achieved high classification accuracy across both synthetic and experimental datasets, with statistical confidence intervals confirming the robustness of predictions. This methodology demonstrates the viability of integrating physics-informed data with ML to enable automated, high-resolution health status monitoring of composite blades and supports its scalability for deployment in operational wind energy systems.