Predicting failure pressure of corroded gas pipelines: A data-driven approach using machine learning

Abstract

This paper presents a novel methodology utilizing machine learning (ML) techniques to accurately forecast the failure pressure of corroded pipelines, which are pivotal for oil, gas, and petroleum product transport. In contrast to prior approaches, this study overcomes their limitations by integrating physically significant factors tied to corroded pipeline failure mechanisms and delivering interpretable models. Robust ML models are constructed using dependable experimental data from pertinent literature sources, ensuring research rigor. To bolster model interpretability, the SHAP method is deployed, facilitating a comprehensive grasp of each feature's impact on predictions. The performance and efficacy of the developed ML models are rigorously assessed using real-world pipeline incidents reported by the PHMSA in the US, further validating their practical utility. The findings not only affirm the models' precision and dependability but also yield valuable insights, underscoring their importance. This study makes a substantial contribution to the enhancement of pipeline integrity management by enabling precise failure pressure predictions in corroded pipelines. The insights gleaned from this research hold practical implications, enabling well-informed choices in pipeline design, integrity management, and risk assessment. Thus, it stands as a valuable resource for pipeline operators aiming to optimize resource allocation and bolster overall pipeline safety and efficiency.