Defect Detection And Classification Of Offshore Wind Turbine Blades

The blade is an important part of a wind turbine,and it is also the most vulnerable part.Lightning,sand and dust,irregular loading,etc.will cause damage to the blades.If an early small crack is found,it can be repaired within a few hours;If they are not found in time,they can easily break the blades and cause serious consequences.Compared with onshore wind turbine equipment,offshore wind turbine equipment has long service life,and saving transportation costs.Therefore,computer vision is used to detect and classify the defects of offshore wind turbine blades to provide maintenance workers with clear and accurate results and avoid economic lossesFoggy weather often occurs in offshore areas.Fog interference will reduce the image quality of the wind turbine blades taken by drones and affect the subsequent defect extraction.Therefore,this article uses inverse wavelet transform and wavelet transform to replace the original U-Net up and down sampling,and multi-scale convolution and residual network blocks are used to modify the network.The results show that the image recovered by the improved network compared with other contrast algorithms can retain more detailed information on the basis of complete defogging,and avoid image color distortion.Traditional unsupervised feature extraction methods cannot accurately extract the defect features of wind turbine blades.Using a weakly supervised semantic segmentation method based on convolutional neural networks,Design a reasonable network structure,using spatial continuity,feature similarity and scribble information as the loss function,split out the wind turbine blade part.After a comparative analysis with other unsupervised segmentation algorithms,the results show that the selected weakly-supervised semantic segmentation method can more accurately extract the defect parts.Finally,the idea of transfer learning and the VGG16 network are used to extract the feature data of wind turbine blade defects.Aiming at the problem of classification of defect feature data,first,perform PCA whitening processing on the extracted feature data of wind turbine blade defects to reduce the computational complexity of subsequent classification.Secondly,on the basis of the particle swarm optimization algorithm,the speed and position update equation weights and learning factors are improved,and compared with other optimization algorithms based on PSO,The results show that the improved PSO optimization algorithm has excellent performance in searching for the minimum value on multi-extreme and single-extreme functions.Finally,the improved PSO algorithm and other optimization algorithms are selected to combine with K-means to classify the characteristic data of wind turbine blade defects.The results show that the improved PSO optimization algorithm combined with K-means has better robustness and classification accuracy.