Design Of Wind Turbine Monitoring And Alarm System Based On Machine Vision

Thanks to the vigorous development in the field of new energy,wind energy,as one of the new energy,has been used more and more widely because of its rich reserves,high economic benefits and broad prospects for commercial development.Wind turbine is an important equipment to convert wind energy into electric energy.Because the wind turbine system is complex and the failure rate is very high,the monitoring and alarm of wind turbine has become particularly important.Gearbox bolts are an important part of wind turbine.The load and vibration amplitude generated by wind turbine at high speed will lead to the loosening of gearbox bolts.Thesis presents a non-contact bolt looseness detection method based on machine vision.Thesis firstly trains and evaluates the target detection network（Res Net34-YOLOv5）,the AP（Average Precision）value on the test set reached 0.993,and realizes the recognition of bolts in images.Then,two images of the monitored bolts in different periods are collected,and then the images are spatially perspective transformed by feature point matching to make them share the same coordinate system.Finally,the two images are registered by the strength-based method.If the bolt rotates during the two inspections,the difference characteristics will appear in the registration error of the two images as the basis of bolt looseness detection.According to the experimental results,the method can accurately detect the looseness of the bolt when the bolt rotates more than 5°.When the wind turbine is operating,some of its parts may have too high temperature,which may lead to fire.Since most wind turbines are deployed in remote areas and there are few people around,they may not be found until a large-scale fire occurs.Therefore,it is necessary to carry out fire monitoring and alarm inside the wind turbine cabin.Thesis combines Swin Transformer with YOLOv5 and proposes a new target detection network（Swin-YOLOv5）,which realizes the recognition of flame and smoke in the image.According to the experimental results,the AP value of Swin-YOLOv5 for detecting the flame category reached 0.881,and the AP value for detecting the smoke category reached 0.923.