Eddy Signal Defect Detection Algorithm Based On Modified SSD

With the rapid development of world economy,the demand for oil and gas keeps rising.Oil and gas pipelines are essential to national economy.Transportation of these resources is an important part for the utilizing of them.With the advantages of high efficiency and low cost,pipeline transportation has become an indispensable part of oil and gas transportation.Pipeline eddy current inspection is an important part of pipeline defect detection,for it can identify internal defects from external defects,which MFL detection cannot,so eddy current inspection is of great significance.This paper studies and analyzes eddy current data collected from metal pipelines,and mainly completes the following aspects of work:designing an adaptive threshold eddy current signal defect detection algorithm that integrates eddy current and magnetic flux leakage signal,which performs well on low-noise pipelines;designing a method of converting eddy current signal into pseudo-color image;designing an improved defect detection algorithm based on SSD network,which performs well on both low and high noise pipelines.First,a pipeline defect detection algorithm based on adaptive threshold,magnetic flux leakage and eddy current signal is designed.Analyzing the disadvantages of fixed threshold algorithm according to the characteristics of eddy current signals,introduce an algorithm based on adaptive threshold for eddy current signal analysis.After the limitation of the method only considering eddy current signal that external defects may be identified as internal defects is proven through experiments,magnetic flux leakage signals are taken into the modified algorithm as an auxiliary examination method,the modified algorithm deals well with the problem above,and it’s efficient for low-noise pipeline detection.Secondly,the limitations of traditional threshold method in high-noise pipeline detection is proven After experimental comparison,SSD is determined to be the basis of the new detection algorithm.Due to the complex environment of in use submarine pipeline,there is more interference in the detection process,and collected eddy current signals are seriously influenced by noise,traditional threshold method is difficult to competent for defect detection task,so after comparing three commonly used target detection algorithm based on deep learning,SSD network with relatively faster detection speed and higher accuracy is selected as the foundation of the new detection algorithm.Thirdly,a method is designed to transform the original eddy current signal into pseudocolor image to complete data preparation of SSD network.Since the SSD network needs to take pictures as input,a method is designed to convert the eddy current signal into pseudo-color image.The resulting image has high resolution and relatively clear defect characteristics,which is suitable for future labeling and network training.Fourth,a modified SSD network is designed for high-noise pipeline detection.Multiple sets of data were used to test original SSD network for pipeline detection.After analyzing the results,targeted improvements were made for its weak detection ability of small targets.A receptive field module and feature fusion mechanism are added to the network,and experiments show that the modified algorithm has high detection accuracy for both low and high noise pipelines.