Research On The Framework And Method Of Edge-Cloud Collaborative Detection For Blade Surface Damage Of Large Wind Turbine Sets Based On Deep Learning

As a kind of clean energy,the proportion of wind energy in Chinese power structure is increasing year by year.The blade is a vital component of wind turbine(WT)for capturing wind energy.Its surface integrity has a significant impact on safety and generating efficiency of WT.In recent years,some domestic wind farms have begun to use unmanned aerial vehicles(UAVs)to conduct patrol inspection on WT blades.But it still remains that patrol inspectors manually detect the damage after UAVs collect the images of blades.Although the safety of inspectors has been improved,the labor and time costs have not decreased much.In this context,this paper proposes a blade surface damage detection method based on the edge-cloud collaboration architecture and deep learning recognition algorithms for the WT blade inspection based on UAVs.The details are as follows:We divide the blade inspection into the manually operated UAV inspection mode and the fully autonomous UAV inspection mode,and introduce the blade image acquisition process of the two modes.According to the respective advantages of edge computing and cloud computing,a three-layer framework of "cloud-edge-end" for blade damage detection is established.After analyzing the requirements of two UAV inspection modes,we develop the computation offloading strategy considering real-time response and the computation offloading strategy considering precision priority.In order to achieve real-time inspection under the manually operated UAV inspection mode,a lightweight design is carried out for the backbone network of damage detection algorithm.The filter pruning method based on sensitivity analysis is used to compress the model in a controllable way,so that the model can be deployed to the edge nodes near the wind farm and meet the real-time demand.For the side effect that the accuracy of blade damage detection is reduced,knowledge distillation method is adopted to reduce the influence of lightweight methods on the accuracy of the model.In order to meet the requirements of high transmission bandwidth and high precision for the fully autonomous UAV inspection mode,we deploy blade image quality detection algorithms on the edge node of the edge-cloud collaboration architecture to remove the blade images which are too fuzzy or have too small blade area,thus reducing the consumption of useless data on the uploading bandwidth.In the cloud,the detection accuracy of the model is improved by using the backbone network with deeper layers and stronger performance and the model optimization methods based on the Intersection-over-Union(IoU).