Detection and sizing of metal-loss defects in oil and gas pipelines using pattern-adapted wavelets and machine learning

Signals collected from the magnetic scans of metal-loss defects have distinct patterns. Experienced pipeline engineers are able to recognize those patterns in magnetic flux leakage (MFL) scans of pipelines, and use them to characterize defect types (e.g., corrosion, cracks, dents, etc.) and estimate their lengths and depths. This task, however, can be highly cumbersome to a human operator, because of the large amount of data to be analyzed. This paper proposes a solution to automate the analysis of MFL signals. The proposed solution uses pattern-adapted wavelets to detect and estimate the length of metal-loss defects. Once the parts of MFL signals corresponding to metal-loss defects are isolated, artificial neural networks are used to predict their depth. The proposed technique is computationally efficient, achieves high levels of accuracy, and works for a wide range of defect shapes.