Image Analysis for Wind Turbine Blade Structures and Materials

The increasing size of wind turbine blades makes quality control of the blades increasingly challenging and costly. At the same time, tighter design tolerances require manufacturers of blades and blade materials to ensure that structural and material
properties follow the specifications closely. In this thesis, I use computer-based image analysis to address the challenges of extracting the structural and material properties needed for quality control of wind turbine blades. The goal is to present methods suitable for quality control solutions that can assist human evaluation or automate quality control of wind turbine blades.
For the evaluation of blade structures, I propose a method based on graph cut optimization for segmentation and surface detection in both 2D and 3D ultrasound images. To make this method scale to large 3D datasets, a new type of graph structure for multi-label segmentation has been developed along with new high-performance parallel and serial versions of state-of-the-art graph cut algorithms. The research contributions are generic and applicable to a range of different optimization and computer vision tasks. Furthermore, I discuss some of the challenges of analyzing ultrasound images of wind turbine blades and why the presented method is suitable for this purpose. For the evaluation of blade materials, I present an approach for estimating fiber orientations in fiber-reinforced composites using structure tensor analysis. The method uses Gaussian kernels and analytical eigendecomposition that makes it tolerant to noise, resolution invariant, and fast. The implementation uses vector operations for even faster computations on modern hardware. Then, I demonstrate the use of structure tensor analysis for characterizing fiber orientations in unidirectional fiberreinforced composites commonly used in wind turbine blades. Finally, I discuss some of the challenges and things to consider when dealing with orientation information in 3D. The work presented in this thesis allows important structural properties to be extracted from large 3D images which form the basis for automated quantitative evaluation of wind turbine blades and fiber-reinforced composites.