Research On Classification And Fitness-for-Service Of Pipeline Girth Weld Defects Based On DCNN

Long-distance oil and gas pipelines have developed rapidly,the total length of which has been over 150000 kilometers.The accidents have occurred frequently in recent years due to the girth weld defects.It is particularly important for pipeline safety to detect and repair girth weld defects in time.MFL in-line inspection,which can obtain all girth welds information,is a commonly used non-destructive testing method for in-service long oil and gas pipelines.However,the MFL signals at girth welds are complex.Therefore,it is difficult to classify and quantify the defects,using traditional manual analysis method based on the shallow morphology of the signal.In this article,the advantages of Deep Convolutional Neural Network(DCNN)in image processing are utilized to study the classification and quantification of pipeline girth weld defects.An image data set,whose samples are MFL images of girth welds,with the results of radiographic testing after actual excavation as labels,is established to be used as research base.At the same time,a quantitative method for additional loads at girth welds based on Analytic Hierarchy Process(AHP)is proposed,which revises the fitness-for-service method based on pipelines internal pressure only.All these provide guarantees for in-service pipelines safety.The research contents are as follows:(1)A classification method on MFL grayscale images of pipeline girth welds based on improved VGG16 is proposed.The VGG16 model structure is improved.An image data set,whose samples are MFL grayscale images of girth welds,with the results of radiographic testing after actual excavation as labels,is established.The image samples are divided into three types: qualified samples,sub qualified samples,and unqualified samples according to the girth weld defects.Images are optimized using median filtering method.The improved VGG16 model is trained by this database.The classification test results indicate that classification accuracy of the trained model is73%,and the recall rate for qualified samples is 93%.The qualified girth welds can be effectively screened out with this method.(2)A generation method on MFL image of pipeline girth welds based on improved DCGAN is proposed.DCGAN model is improved.The image data set,whose samples are MFL signal images of girth welds,with the results of radiographic testing after actual excavation as labels,is established.The image samples,with circular and strip defect labels,are generated with the improved DCGAN.The generated images have high quality and similarity with the original ones.This method effectively expands and enhances the MFL signal image data set,at the same time,improves the problem of imbalanced sample proportions between two types.It lays the foundation for subsequent research on MFL signal image classification.(3)A classification method on MFL signal images of pipeline girth welds based on improved Res Net-50 is proposed.Res Net-50 model is improved.The improved Res Net-50 model is trained by the expanded and enhanced data set containing MFL signal images for circular and strip defects.The classification test results indicate that classification accuracy of the trained model is greater than or equal to 83%,and the recall rate for circular defects is greater than or equal to 97%.This method breaks through the limitations of manually analyzing MFL signals at girth welds,and can effectively identify and classify the types of girth weld defects.(4)The fitness-for-service method of strip defect at girth weld based on DTL,AHP and FAD is proposed.An image data set,whose samples are MFL signal images of girth welds with strip defect,with the defect length as labels,is established.DTL is conducted on this data set with VGG16 model trained on other data set to predict defect length.The additional load of the girth welds is quantified by AHP,by which the FAD method is revised.These researches above have achieved a more accurate fitness-for-service for girth weld defects in non-excavation situations.It can provide more accurate decision-making support for girth weld safety management.