Abstract
Sustainable operation of pipeline networks for oil and gas transportation requires diagnostics capable of both detection and characterization of pipeline defects in particular corrosion defects. Current defect analysis techniques can identify and characterize the geometric features of metal loss defects or defect clusters such as peak depth, length, and width with limited accuracy. Probabilistic data driven models have also shown an ability to predict error bounds for individual defect characteristics as opposed to overall defect tolerance. The prediction accuracy of the health of a pipeline with metal loss defects such as corrosion can be improved with additional detail in the corrosion surface profile as this affects the burst pressure. This will enable operators to apply more accurate corrosion growth models and simulations that can forecast the reduction in pipeline capacity and facilitate more targeted diagnostic and mitigation plans. To this end, a data-driven workflow is proposed to automate the detection, classification, and surface prediction of external corrosion defects. This combines experimental MFL data and validated MFL simulations and leverages both image-based and physics-informed machine learning methodologies.
