Search

• The RF model is 98 % accurate in predicting the severity of fatigue in hydrogen gas. • Loading conditions and hydrogen partial pressure are the most important features. • Material strength slightly influences the magnitude of hydrogen-enhanced fatigue. • ML models can reliably estimate the probability of failure of hydrogen pipelines. • ML can facilitate the inspection and maintenance planning for hydrogen pipelines. Hydrogen is an environmentally friendly fuel that can facilitate the upcoming energy transition. The development of an extensive infrastructure for hydrogen transport and storage is crucial. However, the mechanical properties of structural materials are significantly degraded in H 2 environments, leading to early component failures. Pipelines are designed following defect-tolerant principles and are subjected to periodic pressure fluctuations. Hence, these systems are potentially prone to fatigue degradation, often accelerated in pressurized hydrogen gas. Inspection and maintenance activities are crucial to guarantee the integrity and fitness for service of this infrastructure. This study predicts the severity of hydrogen-enhanced fatigue in low-alloy steels commonly employed for H 2 transport and storage equipment. Three machine-learning algorithms, i.e., Linear Model, Deep Neural Network, and Random Forest, are used to categorize the severity of the fatigue degradation. The models are critically compared, and the best-performing algorithm is trained to predict the Fatigue Acceleration Factor. This approach shows good prediction capability and can estimate the fatigue crack propagation in low-alloy steels. These results allow for estimating the probability of failure of hydrogen pipelines, thus facilitating the inspection and maintenance planning. [ABSTRACT FROM AUTHOR]