A vision-based multi-point vibration monitoring method for large wind turbine blades under complex environments

Vision-based vibration monitoring provides a non-contact solution for online monitoring of offshore wind turbine blades. However, existing methods remain sensitive to complex marine environments, resulting in unstable vibration measurements. To address this challenge, a Vibration-Aware Illumination-Structure Adaptive Detector and an Across-Frame Tracking Module (VISAD–AFTM) is proposed for marker-free multi-point blade vibration monitoring using conventional surveillance videos. VISA comprises three branches: an illumination-adaptive branch for suppressing illumination-induced disturbances, a structural detail enhancement branch for enhancing blade contours and local details, and a structural stability constraint branch for improving cross-frame blade-feature stability. VISA is embedded into the YOLOv11 backbone through a residual plug-and-play strategy to construct the illumination-adaptive blade detector. Subsequently, AFTM is designed for blade multi-point vibration extraction. To reduce cross-frame association errors and mismatches, an Identity-Preserving Embedding Calibration (IPEC) mechanism is constructed. Through IPEC, AFTM couples target identity representation with motion-continuity association, improving cross-frame tracking consistency and stability. The method is validated using nighttime backlit video from an edgewise fatigue test of a 108 m full-scale wind turbine blade. The extracted multi-point vibration responses show good agreement with blade strain measurements in both time and frequency domains. The results demonstrate the reliability of the method for practical blade vibration monitoring.