Enhancing Pipeline Integrity with Probabilistic Digital Twins and Bayesian Machine Learning (C2026-00407)

Guanlan Liu, Chris Kagarise, Muntathir Al Jumaah, Faisal M. Al-Mutahhar, Ayman Alabdullatif, Lay Seong Teh, Ellen Faust, Alexander Brust, Francois Ayello,

Ensuring the integrity of transmission pipelines is essential for the safe and reliable transport of energy. With the increasing digitalization of the energy sector, digital twins have emerged as powerful tools for representing physical assets through dynamic, data-driven models. These virtual replicas improve understanding of degradation mechanisms, support predictive maintenance, and enable risk-informed decision-making.
This study introduces a probabilistic digital twin framework built using Bayesian networks to model internal pipeline threats, including uniform corrosion, localized corrosion, and erosion. The Bayesian approach offers two key advantages: (1) the ability to develop and update models even with limited or incomplete field data, and (2) explicit representation and propagation of uncertainty across interconnected risk factors.
Through a case study, we demonstrate how the model quantifies pipe wall loss by capturing the dependencies among operating conditions, material properties, and degradation mechanisms. The resulting system provides pipeline operators with a transparent and adaptive tool for assessing risk, guiding inspection priorities, and optimizing maintenance strategies in complex and uncertain environments.