The Application of Deep Learning and Computer Vision Techniques to Magnetic Flux Leakage Data to Accurately Detect and Predict Corrosion Profiles

Abstract Over many years trends in the methods used for analyzing and reporting in-line inspection data have evolved from peak depth in a joint to peak depth for every feature, to clustering and river bottom profiles, and now grid level representations of corrosion defects are becoming a favourable reality. With these advances, comes a key benefit that the understanding of defect profiles has advanced from the well know “boxing” methodologies to detailed, direct representations of pipe defect shape, attributes, and characteristics which in turn opens a clear pathway to overcome the primary challenge of interpreting complex superposition of the magnetic signal responses in areas with complex morphologies and interactions. This paper outlines how advancement in deep learning methodologies with “Big Data” approaches have been applied successfully to magnetic flux leakage data for predictive corrosion sizing at scalable, “life-like” grid level resolutions using computer vison techniques. The proposed methodology utilizes deep learned neural network architectures trained on extensive datasets to accurately characterize defects, thereby providing a more refined and precise analysis compared to conventional methods. This paper will outline some background to the motivations in developing the solution, context of data volumes employed, discuss the challenges of adopting the methodology at scale across an entire pipeline and how cloud computing and advanced data engineering approaches have been applied to provide the scalability needed. The authors will also detail the essential role that vetting of truth data, it’s characterization beyond conventional industry standard approaches for defining defects shapes and types and development of new techniques for data comparison and validation play in ensuring success. Lastly a perspective of how the availability of these advanced algorithms as a comparative view to traditional analysis and interpretation techniques can evolve and provide the means significantly improve the reliability and efficiency of corrosion threat determination in pipelines for the oil and gas industry.