Research On Electromagnetic Ultrasonic Defect Assessment Method For Carbon Steel Pipeline Based On Deep Learnin

Pipeline is a major tool for oil and gas transportation,and pipeline health assessment is a key task in industry.In recent years,electromagnetic ultrasonic pipeline defect detection technology based on deep learning has gradually emerged.In this paper,we address the problems of current pipeline ultrasonic defect assessment methods and carry out research on ultrasonic defect assessment methods based on deep learning,focusing on data enhancement and defect identification model issues,as follows:(1)Data enhancement methods for defective echo signals are carried out in two paths,direct enhancement and indirect enhancement.The former uses a generative adversarial network approach to perform direct data enhancement analysis on one-dimensional features processed by dimensional transformation and data filtering;the latter uses a bidirectional generative adversarial network approach to perform indirect feature data enhancement analysis on two-dimensional echo signals,using its encoding structure to map potential features to real samples,enabling the generation of new data samples while obtaining more adequate feature information.The two data enhancement methods are comprehensively evaluated by observing the feature distribution and the similarity of the generated data to the original echo signal.(2)A defect evaluation model capable of obtaining potential information is established based on the enhanced feature data.The encoder of the dual generative adversarial network is used to replace the coding structure of the convolutional neural network to form a recognition model of the dual generative convolutional neural network with potential feature data enhancement capability,and the firefly sparrow search algorithm is introduced to optimize the parameters of the recognition model to improve its global convergence capability when the sparrow search falls into local extremes,and then to achieve qualitative discrimination of the type of pipe defects and quantitative prediction of the size.quantitative prediction.(3)Based on the above research content,a pipeline ultrasonic defect evaluation software platform is developed.The functions of reading,signal pre-processing,data enhancement and defect identification based on twisted guided wave pipeline ultrasound signals are realized in one click.The complex data and information are presented to the user in the form of graphics or images,allowing the user to understand and process the information more intuitively.The effectiveness of the defect assessment method proposed in this paper is verified by electromagnetic ultrasonic pipe defect detection experiments.The comparison results of several machine learning and deep learning models show that the proposed method can control the prediction error to about 0.3%,while the error of the traditional signal amplitude quantification method is about 5%.Since the method proposed in this paper can not only qualitatively and quantitatively discriminate the defect types and sizes of pipes,but also realize the size prediction of unknown data,it can be used as an effective discriminative method for pipeline health assessment work in industrial fields.