Research On Defect Quantitative Characterization And Profile Reconstruction In Submarine Pipelines Based On Magnetic Flux Leakage Testing

Magnetic flux leakage(MFL)testing is an important means of submarine oil and gas pipeline inspection.Defect quantitative characterization and profile reconstruction are the most important links in MFL data analysis,which directly affect the safety assessment of submarine pipelines.This dissertation is supported by the National High Technology Research and Development Program of China(No 2011AA090301)and the Special National Key Research and Development Plan(No 2016YFC0802300).This research focused on the key issues of MFL data processing,conducted systematic and in-depth research on girth weld and defect signal recognition,defect quantitative characterization and profile reconstruction,and made important progress.A series of new MFL data processing methods were proposed and integrated into the submarine pipeline MFL inspection system.It is applied in the Shengli Sea area and provides technical guarantee for the assessment,abandonment,reconstruction and safe operation of submarine pipelines.The method of target detection for pipe girth welds is researched,and the girth weld recognition and data segmentation are realized.In the segmented pipeline data,the clustering method is used to extract defect data.Quantitative defect characterization is further performed on the basis of 3D inspection data fusion.Based on the generative adversarial network,the 3D profile reconstruction of defects is realized.The main works are summarized as follows:1.Study on technology and signal processing of MFL testing for submarine pipelinesThe principle of MFL testing,submarine pipeline MFL testing system,MFL signal analysis method,submarine pipeline MFL testing operation process,and data analysis process are studied and analyzed.According to the characteristic that the submarine pipelines are all girth welds,the analysis process is designed.The process of segmenting the pipeline data by using the girth weld signal first and then identifying the defect area of the segmented data is proposed to provide a basis for subsequent research.Analyze the test situation and test equipment requirements of the MFL pig of the submarine pipeline during the development process.Capacity test devices and pull test devices were constructed,which provided necessary experimental data for the subsequent study of the MFL signal of defects.2.Research on girth weld object detection method based on CNNBased on the submarine pipeline components MFL signal analysis and traditional anomaly detection methods,a threshold-based method for pipeline girth weld recognition is proposed.Based on the computer vision target detection model,a multi-layer convolutional neural networks(CNN)designed for the feature extraction of girth welds is proposed,and a pipeline girth weld object detection algorithm based on deep convolutional networks(DCNN)is proposed.Based on the inspection data of submarine pipeline engineering,a girth weld data set was constructed and the method was verified.The results show that it still maintains good accuracy in the presence of noise interference and local data loss.It has certain robustness and realizes the identification of girth weld signals and pipeline segmentation.3.Research on defect region identification method of MFL data based on K-means clusteringAiming at the traditional threshold problem,the defect signal recognition method is studied,and an adaptive threshold defect marking method is proposed to avoid the arbitrary selection of artificial threshold.Furthermore,the clustering method is studied,and a method for identifying the region of the magnetic flux leakage signal defect based on K-means clustering is proposed without threshold.The verification shows that the defect detection rate meets the detection rate requirement.The defect signal area contains the necessary features of the defect,reaching the defect quantitative inversion standard.The method can also provide new ideas for defect detection of other nondestructive testing methods.4.Research on quantitative characterization of defects based on 3D MFL data fusionAiming at the problems of inadequate utilization of MFL signal and the need to improve the accuracy,the current method of MFL data fusion was studied,and a 3D MFL data fusion model was proposed.All three component of MFL signal are input to the network for automatic feature extraction,which maximizes the data utilization rate.The objective loss function of joint regression of defect length,width and depth is designed,and the regression prediction of defect scale is realized using DCNN.A classification and quantization algorithm based on 3D MFL signal is proposed,and the method of categorizing defects first and then quantifying them according to categories is used to further improve the accuracy of defect quantification.Using the method of finite element simulation and pig pulling test,a defect quantization data set is established.The verification shows that the error meets the requirements,has good accuracy and generalization performance,and is superior to traditional methods.Related achievements have formed a set of data analysis system and successfully applied the submarine pipeline MFL inspection project.5.Research on 3D profile reconstruction method based on GANA method for expressing the remaining wall thickness of the pipeline based on image gray scale is proposed.Using the method of finite element simulation and pulling test,the defect profile inversion data set is constructed.Aiming at the imbalance between the speed and accuracy of defect 3D contour inversion,a fully convolutional network is applied to realize defect 2D edge contour and defect 3D contour inversion.In order to solve the problem of inversion contour blurring,a defect 3D profile reconstruction model based on generative adversarial networks(GAN)was proposed.Generate network to perform defect contour inversion,discriminate network to determine whether the contour is true or false,and iteratively train iteratively.The verification shows that the method has high MFL signal defect 3D contour reconstruction accuracy and engineering application capabilities.It provides a new way for three-dimensional rapid reconstruction of defects,which can be used for full pipeline MFL imaging and provides a basis for refined safety assessment of submarine pipelines.