Presentations

Abstract

Leading-edge erosion is a critical issue for the wind energy industry, causing damage to wind turbine rotor blades that results in reduced aerodynamic efficiency and an annual energy production (AEP) loss of up to 3.7%. While fast, non-contact inspections using thermal and visual cameras from the ground provide a way to monitor these damages without requiring downtime, interpreting the data from these inspections is labor-intensive and requires considerable expertise. Artificial intelligence (AI), particularly image segmentation techniques like convolutional neural networks (CNNs), offers a promising approach to automate the analysis of inspection data. However, the effectiveness of AI models depends heavily on the availability of large, high-quality datasets that capture the wide range of conditions and damage types found in real-world scenarios. The KI-VISIR project focused on creating such a dataset to support the development of AI-based inspection technologies. By utilizing existing thermal and visual inspection techniques, the project collected over 2200 high-resolution visual images and 1200 thermograms from 30 operational wind turbines under various conditions. The resulting dataset, which will be made publicly available prior to the conference, serves as a valuable reference for training AI algorithms, enabling more accurate and efficient identification of rotor blade damage. By making this dataset freely accessible, the KI-VISIR project aims to facilitate the advancement of AI-based methods for wind turbine inspections, reducing the manual workload involved in data interpretation and improving the overall efficiency of maintenance processes across the wind energy sector.