Data Fusion of Complementary Axial and Circumferential Magnetic Flux Leakage Inline Inspections and Effects on Safe Remaining Life

Abstract Measurements from magnetic flux leakage (MFL) in-line inspection tools include two types of uncertainty: random uncertainty arising from essential variables such as velocity and magnetization level; and systemic uncertainty from the measurement principle. The accuracy of the measurement principle is subject to the angle between its magnetic field and the metal loss anomaly. To gain a comprehensive assessment of anomalies, pipelines can be inspected with two MFL techniques with perpendicular magnetic fields, i.e., axial MFL (MFL-A) and circumferential MFL (MFL-C). Currently, the inspection data from each MFL tool is analyzed separately, and two inspection reports are generated respectively. Comparing the two inspection reports, the conflicting results due to the change in angle become obvious and expert opinion is relied upon to make the most appropriate assessment. Fusing the original complementary data sets together overcomes the subjectivity, enables one unique outcome, and greatly reduces the uncertainty. To take advantage of the complementary information of two MFL inspections, ROSEN’s novel machine learning-based data fusion model is applied to the aligned MFL data sets to generate a high-resolution 3D metal loss depth map. The validity of the proposed model is verified using the field data from six operational pipelines of varying diameters. The performance regarding depth prediction and burst pressure calculation is assessed using API STD 1163 Level 3 validation methods. The depth comparison of the derived 3D metal loss depth maps versus laser scans shows strong correlation in both morphology and depth. The safe remaining life of metal loss anomalies in the studied pipelines have been calculated to demonstrate differences due to the new levels of measurement certainty. In addition, as 3D depth maps are now generated, the results differ from traditional box profiles for burst pressure calculations such as Remaining Strength (RSTRENG) and Plausible Profiles (Psqr). The consistent performance on all six pipelines proves the generalization of the data fusion model.