Three Dimensional Reconstruction Method of Pipeline Defects Based on CNN-Transformer
This paper studies the method of defect reconstruction in long distance oil and gas pipelines based on MFL signals.Because the inverse solution is ill posed,and the Deep Learning model has strong nonlinear mapping ability and feature extraction ability,the CNN-Transformer hybrid architecture model is developed as a quantitative model to predict the defect size.Additionally,it studies the correction methods for simulated MFL signals to reduce the discrepancy between simulation data and experimental data.It is verified that the error between this simulated axial component data and the peak value of the experimental data decreases by 83.73%on average,while the peak error of the radial component decreases by 28.25%on average,which solves the problem that the training samples of the Deep Learning model are not sufficient.Moreover,the hybrid architecture model trained by the corrected dataset as the training set has better prediction accuracy when predicting the defect size.The Mean Relative Error of the model in predicting the length,width and depth of the defect is reduced by 21.35%,22.58%and 21.55%respectively,which has high accuracy and robustness.
MFL detectionfinite element analysisConvolutional Neural NetworksTransformer modelsignal correction
万方数据
维普
