Balancing Pipeline Profitability and Failure in Hydrogen Pipelines Using Machine Learning-Derived Operational Rules (C2026-00321)

Francois Ayello, Narasi Sridhar, Ali Mosleh, Theresa Stewart

Hydrogen pipelines face a unique trade-off between energy delivery efficiency and pipeline material integrity. On one hand, pressure must be increased to compensate for hydrogen’s low energy density and ensure sufficient throughput (i.e., economic return). On the other hand, elevated pressure increases the risk of hydrogen embrittlement and material failure (i.e., pipeline rupture). In this work, we leverage an existing predictive model built from validated fracture mechanics principles, gas flow calculations, and equations of state. Rather than requiring operators to run this complex model for each use case, we use the model to simulate a wide range of scenarios and generate a comprehensive synthetic dataset that captures the risk-profitability trade-off. From this dataset, we apply interpretable machine learning methods (i.e., Bayesian rule lists and decision sets) to extract a set of decision rules that define safe and economically optimal operating pressures under various conditions. These rules can be directly used in the field, enabling pipeline operators to make informed decisions quickly and without technical overhead. This approach bridges physics-based modeling and operational simplicity, offering a scalable framework for managing hydrogen or hydrogen-blend pipeline networks.