Machine Learning-Based Dipole Modeling of 3-D Magnetic Flux Leakage Caused by Outer-Surface Linear Cracks for Pipeline In-Line Inspection

Conventional magnetic dipole models (MDMs) fail to properly model the 3-D magnetic flux leakage (MFL) caused by outer-surface cracks for pipeline in-line inspection (ILI), due to the difficulty in quantifying the permeability distortion around a crack. In this article, we propose the machine learning-based MDM (ML-MDM), as a hybrid physics-informed and data-driven model (HPDM), which can fast model the 3-D MFL caused by outer-surface linear cracks of pipelines. The machine learning algorithms based on data-driven models are used to construct a virtual crack to evaluate the nonlinear effect of permeability distortion in the residual pipe wall on the MFL, and the physical theory of MDM combines the real crack with the virtual crack to rapidly compute the 3-D MFL. The effectiveness of the proposed method is verified by the 3-D finite element method (FEM) simulations and the pipeline inspection gauge (PIG) ILI experiments. The results show that ML-MDM has the similar MFL modeling accuracy as FEM, but much less time cost than FEM.