A Study on Pipeline Defect Magnetic Flux Leakage Field Simulation and Recognition Based on an Enhanced YOLOv8 Model

With the increase of service time of oil and gas pipeline, the pipeline body is prone to corrosion, dent and other defects. These defects pose significant threats to pipeline safety. Magnetic flux leakage (MFL) internal inspection technology offers advantages of no coupling agent and ease of automation. To regularly detect and identify pipeline defects, MFL inspection has been widely applied for online non-destructive testing. Improving defect identification accuracy based on MFL data is crucial for ensuring pipeline safety. This paper is based on the principles of MFL internal inspection. We reconstruct the MFL internal detector using finite element software. Simulations of MFL signals for defect-containing pipeline segments are conducted, creating a simulation database for pipeline defect MFL fields. Based on the simulation data, cubic spline interpolation is used to expand the dataset and generate pseudocolor maps. This study also improve the detection head of YOLOv8 by adding a parallelized patch-aware attention mechanism. These improvements enhance small target detection capabilities and enable classification of different defect types. Thus, the accuracy of defect identification in MFL detection is improved. The results show that, after training on the simulation dataset, the model achieves over 90% accuracy and over 95% recall. This indicates that the proposed algorithm can accurately identify pipeline defects in MFL data and effectively classify them.