Machine learning based integrity decision management of pipeline corrosion clusters

Pipeline interacting corrosion defects usually occur in a cluster in such a way that the failure pressure is not controlled by a single defect. These metal loss defects can impair the service life of the pipeline that could lead to loss of containments and potential harm to the environment and facilities. Therefore, pipeline operators use deterministic and probabilistic integrity assessment to examine these corrosion defects to plan inspection, undertake repairs, or replacement of pipeline sections to prevent such incidents. However, the large amount of interacting metal loss defects captured by in-line inspection tools are generally assessed by conservative physics-based formulations, which are mostly centered on the composite single defect approach to determine the burst pressures. The need to accurately predict the failure pressure of the large amount of clustered corrosion defects in the pipeline requires a computationally efficient machine learning approach that can accommodate variability of the input data effectively. Hence, for this research, categorical machine learning models are trained, validated, and tested using published experimental burst pressure of corrosion cluster with the same defect depth for a test sample. The paper presents this approach, where the predicted pipeline failure pressure of the corrosion clusters captured by real in-line inspection are assessed by generated artificial neural networks and non-linear regression models that provides a total mean deviation percent of 2.52% and 9.36% respectively better than the current models in the literature. This approach provides a pathway for effective decisions by pipeline operators, in reducing pipeline operating and maintenance costs safely.