Multiscale Multidisciplinary Machine Learning Modeling for Pipeline Risk Assessment

Abstract Ensuring the safety and reliability of pipeline systems across diverse components, materials and operating conditions necessitates the integration of information from various fields of expertise. Machine learning can be a powerful tool to analyze multiscale and multidisciplinary data, enhancing our ability to make informed decisions and manage pipeline risks effectively. Also, machine learning technology, when applied to pipeline risk assessment, can be used to organize valuable pipeline information, and uncover hidden correlations. However, machine learning faces challenges like overfitting and identifying false correlations without a real causal impact. Understanding the causes of pipeline failures is essential for taking appropriate actions. Since data can be imperfect with measurement error, a probabilistic risk prediction model is necessary to incorporate uncertainty considerations. This publication uses a previously published Bayesian network model for Stress Corrosion Cracking (SCC) pipeline risk assessment to demonstrate how Bayesian networks can be used to aggregate data at multiscale and from multidisciplinary factors. The model is reviewed with field data to share knowledge with pipeline engineers using machine learning for pipeline risk assessment.