A Two-Stage Data-Driven Approach for Image-Based Wind Turbine Blade Crack Inspections

A two-stage approach for precisely detecting wind turbine blade surface cracks via analyzing blade images captured by unmanned aerial vehicles (UAVs) is proposed in this paper. The proposed approach includes two main detection procedures, the crack location and crack contour detection. In locating cracks, a method for extracting crack windows based on extended Haar-like features is introduced. A parallel sliding window method is developed to scan images and the cascading classifier is developed to classify sliding windows into two classes, crack and noncrack. Based on detected windows containing cracks, a novel clustering algorithm, the parallel Jaya K-means algorithm, is developed to assign each pixel in crack windows into crack and noncrack segments. Crack contours are obtained based on boundaries of crack segments. The effectiveness and efficiency of the proposed crack detection approach are validated by executing it on a personal computer and an embedded device with UAV-taken images collected from a commercial wind farm. Computational results demonstrate that the proposed approach can successfully identify both the blade crack locations and crack contours in UAV-taken images.