Intelligent non-contact health monitoring of wind turbines via dynamic vision and adaptive kernel SVM

Abstract To address the challenges of complex sensor installation and high maintenance costs in traditional wind turbine condition monitoring, this paper proposes a non-contact monitoring method based on image processing and Adaptive Kernel Support Vector Machine (AK-SVM). A three-layer architecture—remote data acquisition, edge computing, and central monitoring—is established. High-resolution cameras and meteorological sensors capture turbine video and environmental parameters; edge devices perform image preprocessing, frame-difference analysis, and feature-point tracking to estimate turbine rotational speed; the monitoring center employs the AK-SVM model to classify the operational states of turbines. Experimental results demonstrate that the proposed method accurately estimates rotational speed under diverse meteorological conditions. Moreover, the AK-SVM consistently outperforms conventional SVMs with linear, polynomial, RBF, and sigmoid kernels in terms of accuracy, recall, and F1-score, showing enhanced robustness and environmental adaptability. This study provides a low-cost, non-invasive, and intelligent solution for wind farm operation and maintenance.