Fault Detection Of Bolt Missing Pins In Transmission Line Based On Improved SSD Network

Bolts are the most numerous connecting components in transmission lines and play a very important role.Due to the long-term exposure to the harsh field environment and suffering from wind and rain,the faults such as missing pins,nut looseness and bolt corrosion are very likely to occur.If faults cannot be eliminated in time,they will inevitably affect the normal operation of the transmission line,and even cause serious consequences of falling off key components,resulting in regional power outages and production stagnation,affecting economic development.Therefore,it is very necessary to conduct regular inspections of transmission lines.It is necessary to quickly and accurately detect bolt faults in inspection pictures collected by the UAV,and provide corresponding solutions to eliminate potential safety hazards.Missing pins is the most common fault in bolts of transmission line.Finding and dealing with bolt missing pins faults in a timely manner can not only ensure the safety and stability of the transmission line,but also provide a reference for the subsequent detection of other types of bolt faults.It is of great significance to apply artificial intelligence technology and combine the relevant knowledge of transmission lines to study an intelligent detection method for bolt missing pins faults in transmission line.When using the traditional method to detect the bolt faults of the transmission line,the inspection image is firstly subjected to image segmentation and other preprocessing operations,and then the target detection is carried out according to the geometric shape and texture features of the bolt targets.However,the detection accuracy and detection efficiency are poor,a lot of manpower and time are consumed,and the massive image data collected by the UAV cannot be processed in time.In recent years,deep learning technology has been widely used in the field of image target detection,and many achievements have been achieved.Using deep learning technology,the characteristics of bolt faults targets in inspection pictures can be automatically learned,and batch processing of images collected by UAV can be realized.It has positive research significance and application value to detect whether there is a bolt fault in the transmission line.Collecting pictures by the UAV and using deep learning technology to process pictures in batches can save a lot of manpower and time,and do not require staff to work at high altitudes,reducing safety risks.Most importantly,it can greatly improve the inspection efficiency of transmission lines,improve the speed and the accuracy of fault detection.Based on the previous research and analysis,combined with the characteristics of bolt missing pins faults in transmission lines,this paper aims at the problem that the bolt target in the bolt missing pins faults detection data set accounts for too small in the picture and the number of missing pins bolt targets in a single picture is small.Based on the traditional image data enhancement method s,a small target random paste data enhancement method is designed.Through statistical analysis,according to the length-width ratio of the bolt target,the parameters of the anchor box length-width ratio in the network model are reasonably adjusted and optimized.In view of the problem of insufficient feature extraction and utilization of the original network model,on the basis of the original SSD network model structure,two feature map fusions are added,which can effectively utilize the feature information in the network model and enhance the feature information of the each layer’s feature maps.It is helpful to detect the bolt missing pins faults.In terms of feature fusion,this paper designs a inter-level cross feature pyramid for the first feature fusion.The fusion method of inter-level cross connection is used to fuse the feature information of high-level and low-level feature maps,which can enhance the detailed information of high-level feature maps and low-level feature maps.At the same time,an adaptive spatial feature fusion mechanism is introduced to perform secondary weighted feature fusion on the feature map to further enhance the feature information of the output feature map.The experimental results show that the improved network model in this paper and the bolt missing pins faults detection method proposed in this paper have the highest detection accuracy.On the general data set PASCAL VOC2007 test set,the m AP value reaches 79.8%;In the fault detection data set,the AP value of normal bolts reaches 87.93%,and the AP value of missing pins bolts reaches 89.15%.In the follow-up research,this paper will further reduce the model complexity and parameter quantity by simplifying the network model and optimizing the algorithm design,and improve the detection speed of bolt missing pins faults.At the same time,based on this research,the deep learning technology will be extended and applied to other types of bolt faults such as nut looseness and bolt corrosion.