Research On Fast Classification And Recognition Technology Of Wind Turbine Blade Damage Image

A study on the rapid classification and recognition of surface damage of wind turbine blade has been carried out combing the facing problem of such detection.The major contents include the design of the framework of rapid classification and recognition for blade surface damage image with variable scale,the procedure and application deployment of corresponding data,and the search path planning method for the global image of blade surface damage.The study contents further extend to the establishment of data set of convolutional neural network for classification and recognition as well as the training and testing of blade damage images of wind turbine blade through network model,which leads to the obtainment of network model that can be applied to the classification and recognition of such damage.Such model can achieve the basic requirements of the application in terms of the classification and recognition effect of the wind turbine blade damage.Following is the further illustration of the specific works done.Firstly,according to the structural features of wind turbine and the fast development of unmanned aerial vehicle(UAV)field,a rapid classification and recognition method for the blade surface damage with variable scale by taking UAV as the image acquisition platform has been put forward.In a global scale,UAV can remotely take the image of blade surface and obtain its global image.Based on the global-scale image,a global coordinate system of blade surface can be built up.The suspected damage on the surface can be displayed by pre-processing the image of blade surface and further be demarcated the position coordinates in the global coordinate system.Based on all known suspected damage position coordinates on the blade,all the positions were traversed and the sequence and path were optimized,enabling the UAV to traverse and search all positions with suspected damage on the blade surface according to the planned path.In terms of local scale,UAV can come close to the wind turbine blade and optimize the searching path for the planned suspected damage positions and further search and capture the local image of the suspected damage positions one by one.After such images being captured,the images should be preliminarily screened so as to eliminate the interfering images including lightning arrester flash contact and eddy current generator.Then,three convolutional neural network models have been adopted to train the classification and recognition of the surface damage on the wind turbine blade.Through the comparison analysis,the network model VGG-16,the one with the lowest accuracy of classification and recognition for such damage image was ruled out.As a result,Goog Le Net and Res Net-50 training model structure have been adopted,with an accuracy of 73.68% an74.24%,respectively,to rapidly classify and recognize the damage image on the surface of wind turbine surface.The corresponding application deployment of two network models mentioned above can improve the efficiency of classification and recognition of such surface damage,shorten the routine inspection time and lower down the maintenance cost,which will be of significant meaning for the safe operation of the wind turbine generator system.