Detecting Leading Edge Erosion Of A Wind Turbine Blade Using Deep Learning And Computer Vision Techniques

Abstract

This research presents an integrated methodology for automating the process of detecting leading edge erosion of the wind turbine blade and determining erosion dimensions using advanced artificial intelligence and computer vision techniques. The research also analyzes the effectiveness of the resulting values ​​in evaluating the blade's aerodynamic performance using computational fluid dynamics (CFD). In this research a Mask R-CNN model was developed based on a dataset of eroded blades images. The model was used to automatically detect the location and type of the erosion, instead of using traditional methods that require stopping the turbine and having workers climb up to inspect it. A 3D computer model of the eroded area was then developed to determine the erosion dimensions. This was achieved by applying 3D reconstruction technique to the eroded blade images detected using the Mask R-CNN model developed in this research. The eroded blade was then modeled using the 3D model and the actual erosion dimensions, and a numerical simulation was conducted using CFD to study the effect of error in calculating the erosion dimensions on the blade's lift and drag coefficients. The results demonstrated the effectiveness of the Mask R-CNN model in detecting erosion, with a mean average precision of 84.6%. The 3D model also demonstrated high accuracy in calculating erosion dimensions, with an average error of 0.4 mm, and the maximum relative error in calculating the lift and drag coefficients using CFD was 0.45% and 2.38%, respectively.