Did you know that 64% of U.S. energy commodities are transported via pipelines? The Pipeline and Hazardous Materials Safety Administration, which monitors the transport of crude oil and natural gas, is dedicated to ensuring the safe transportation of energy and hazardous materials. It reported 297 incidents in 2023, resulting in 44,000 barrels spilled. To minimize the severity of future incidents, developing effective disaster response and management strategies is necessary.
Algorithms generated by artificial intelligence offer a potential solution given their ability to assist with accurate prediction, response time, and mitigation. While AI-based approaches have shown promise, especially with managing repetitive tasks and automating processes, there remain challenges including assessing risk from aging pipeline infrastructure, accurately classifying incidents based on various factors such as leak severity, financial impact, and response urgency.
In a new study, “AI-Driven Framework for Predicting Oil Pipeline Failure Causes Based on Leak Properties and Financial Impact,” for the Journal of Pipelines Systems Engineering and Practice, authors Tanzina Afrin, Nita Yodo, and Ying Huang seek to overcome these challenges. They secured a data set of oil pipeline incidents from 2010 to 2022 from the PHMSA and utilized various learning algorithms to predict failure causes for oil pipeline incidents based on pipeline properties, leak severity, and financial impacts. Learn more about this research and how authorities can leverage AI-driven analysis to enhance emergency response preparedness and mitigate the impact of oil pipeline incidents at https://doi.org/10.1061/JPSEA2.PSENG-1830. The abstract is below.
Abstract
The current oil pipeline system is at risk for leaks and ruptures due to aging infrastructure, corrosion, and extreme weather. Addressing these vulnerabilities demands a robust disaster preparedness and response strategy that can be adapted to various incident causes. This study evaluates a data-driven framework with artificial intelligence (AI)–based algorithms for classifying failure causes in oil pipelines. For this purpose, a data set summarizing oil pipeline incidents from 2010 to 2022 obtained from the Pipeline and Hazardous Materials Safety Administration, which comprises various incident attributes, including the severity of incidents and estimated financial impacts, was processed for data balancing with a variation of synthetic minority oversampling technique. Further, this study employs extreme gradient boosting, random forest, 𝐾-nearest neighbors, and support vector machine algorithms for multiclassification tasks in predicting the failure cause. The results indicate that majority-class incident causes can be effectively predicted across algorithms, achieving over 90% accuracy for corrosion failures and over 80% for equipment failures. The proposed methods can be integrated into the mitigation stage of a disaster management framework to aid with decision-making in maintenance activities and improve emergency response based on incident urgency.
See how you could apply AI failure predictions to improve pipeline maintenance in the ASCE Library: https://doi.org/10.1061/JPSEA2.PSENG-1830.
