Research On X-ray Automatic Flaw Detection Method For Aluminum Conductor Composite Core

Aluminum Conductor Composite Core(ACCC)wire is a new type of conductor for overhead transmission lines.It has many advantages such as light weight,low line loss and large current capacity.So it has good application prospects in power transmission.But its’bending resistance is poor,and it is easy to be damaged due to operation errors when laying the circuit,and which also lays hidden dangers for power transmission.In order to ensure the safety of circuit transmission,a large number of ACCC conductors need to be inspected using non-destructive testing technology,but the inspection process will bring a lot of work,which will easily lead the inspection personnel into a tired state and cause manual false detection.The automatic defect detection technology can save labor cost and provide effective early warning of the risk during the power transmission process.Therefore,the research on the automatic defect detection of ACCC wires is a very meaningful research direction in the industrial world.In this thesis,based on the X-ray scanning technology,for the defects of the ACCC wire X-ray image,the defect contrast is not high,it is not easy to identify,the morphological fluctuation range is large,and there are many interference factors.The automatic defect detection method based on the convolutional neural network is studied.Detection scheme based on target detection and image classification.Experiments show that the recall rate of detection results is as high as 96.17%,and the main research work of this article is carried out from the following three aspects:1.Slightly defective samples are the difficulty of defect detection and belong to difficult samples.In order to solve the problem of shortage of difficult samples in ACCC defect maps,we combine the characteristics of sample imaging maps to design an automatic extraction and embedding expansion scheme for difficult samples to expand.It has been verified by subsequent experiments that this extension method has a significant improvement effect on the detection of minor defects in ACCC X-ray images.2.In this thesis,we study the defect detection network scheme based on the target detection network,and use the Retinanet network with strong real-time performance to detect the defects,and analyze the detection of different defect shapes.This method has obvious effect on the detection of obvious defects and slight creases,but the effect of split detection is poor.3.In this thesis,we study the defect detection network scheme based on classification network,we fix the defect range through preprocessing and other operations,and then detect the defects in combination with the sliding window.This part plays a role in replacing the regression box generation process.Compared with the Retinanet defect detection network,this method has better detection effect on split samples,more real-time performance,and guarantees a higher defect recall rate.