Pipeline Defect Grade Evaluation Based On Metal Magnetic Memory Technology

With the acceleration of the development of all countries in the world,the dependence on fossil energy is increasing.Transportation through pipelines is the ideal way,but various injuries occur as the pipeline goes into service.The metal magnetic memory detection technology can detect the early hidden defects and stress concentration areas in time without affecting the normal operation of the transportation pipeline.However,the detection data is dispersed,which causes the difficulty of quantitative identification of defect grade.Therefore,this paper proposes an improved convolutional neural network model,namely BOA-ICNN model,based on Bayesian optimization algorithm,aiming at the dispersion of magnetic memory detection data,and evaluates the defects in combination with relevant detection standards.To simulate the magnetic signal distribution of pipeline defects with different defect sizes and internal pressures,a model of the relationship between equivalent stress and relative magnetic permeability of ferromagnetic materials under three-dimensional stress state was derived based on the changes in total free energy in magnetic crystals.Finite element magnetomechanical coupling simulation of pipeline defects was conducted using simulation software.According to the simulation results,as the depth increases,the peak value of magnetic field intensity at the defect site shows a trend of first increasing and then decreasing;As the diameter increases,the peak value of magnetic field intensity at the defect shows a non-linear increasing trend;Under different internal pressures,the magnetic field intensity curve at the defect location changes very little with increasing internal pressure.In order to verify the accuracy of the magnetic coupling simulation,L245N pipeline steel is made of pipe specimens,and different circular hole defects are fabricated by using TSC-5M-32 magnetic recorder to obtain the characterization of metal magnetic memory signal under different defect sizes,with the increase of the defect depth,the peak of the defect increases first and then decreases with the trend of nonlinear increase,and verify the accuracy of the simulation results.On this basis,the magnetic field strength tests with different detection methods and directions.Through comparative analysis,it is found that the decay degree of the same defect is different.When the detection direction is different,the magnetic memory signal curve from the north to the south,while the magnetic memory curve from the east to the west shows the trend of decreasing and increasing,which lays the experimental foundation for the grade evaluation of pipeline defects in the next step.Considering the dispersion of magnetic memory detection signal,extract the gradient limit coefficient M,the average gradient value Kavg,gradient maximum Kmax,and the synthetic magnetic memory characteristic parameters,introduce the improved convolutional neural network,using the data for multiple convolution screening of the data,and construct the pipeline defect level evaluation model.In order to reduce the influence of random selection of hyperparameters on the classification accuracy,Bayesian optimization algorithm is used to optimize the hyperparameters and establish an improved convolutional neural network model based on Bayesian optimization algorithm.At the same time,according to the defect depth,the Modified B31G criterion and the GB/T35090-2018 pipeline weak magnetic detection method are combined to optimize the correction coefficient A and the damage degree value G,and then the defect grade is quantitatively evaluated.The verification results show that the accuracy rate is 95%.In order to reduce the noise influence,reduce the magnetic memory signal noise reduction,establish Bayesian optimization improvement Convolutional neural network model,evaluate the pipeline defect data in actual engineering,and verify the excavation.The result shows that the actual damage situation is consistent with the model prediction result,and provides a new idea for the grade evaluation of metal magnetic memory detection technology in actual engineering application.