Multimodal Ground-Based SHM for Wind Turbine Blade Damage Detection and Early Warning

Abstract

Early-stage defect detection in wind turbine blades remains challenging due to limited robustness of singlemodality monitoring under real operating conditions. This paper presents a ground-based multimodal structural health monitoring (SHM) framework that integrates RGB imaging, infrared thermography, acoustic sensing, and vibration analysis. The system performs synchronized data acquisition and feature extraction, followed by adaptive late fusion with reliability-aware weighting. Experimental evaluation using Leave-One-Turbine-Out validation shows that the proposed approach achieves a macro F1-score of 0.90 and reduces the false alarm rate to 0.07. Probability of Detection (POD) analysis indicates a minimum detectable crack size of approximately 12 mm. In addition, multimodal integration improves Remaining Useful Life (RUL) prediction accuracy compared to SCADA-based methods. The results demonstrate that adaptive multimodal fusion enhances detection reliability and enables scalable ground-based monitoring without dependence on UAV inspection.