Detection Of Surface Defects Of Wind Turbine Blades Based On Deep Learning

In recent years,wind power has developed rapidly,and fault diagnosis of wind turbines has become more and more important.Due to the low efficiency of manual visual inspection of the surface defects of the wind turbine blades,poor safety,and too much reliance on the subjective consciousness of the inspectors,the installation of various sensors is expensive,and the environmental requirements are high,so deep learning algorithms are used for blade surface defect detection,which is not only safe and efficient,but also more objective and low in cost.This thesis studies the current status of surface defect detection of wind turbine blades and the image recognition algorithm based on deep learning models,builds an experimental simulation platform and builds the original data set of wind turbine blades for defect detection,and based on the Mask-RCNN instance segmentation model to identify and segment the surface defects of the wind turbine blade.In view of the small number of self-built sample data sets,increase the number of training samples through data enhancement technology,and experiments are used to verify the effectiveness of data augmentation technology for deep learning model training in small samples.In addition,in order to eliminate the adverse effect of the image background on defect detection,the blades image is removed based on the adaptive threshold segmentation method,highlighting the blades information,reducing the model’s misrecognition of the background,and designing experiments to verify the effectiveness of the proposed method.In summary,this thesis provides two methods to improve the performance of deep learning recognition models,and provides two ideas for optimizing performance for the application of deep learning models to the detection of surface defects on wind turbine blades,which can be referenced by related researchers.