A Weakly Supervised Framework for Anomaly Detection in Hydrogen Blend Transport Networks Using High-Fidelity Simulation Data

The increasing adoption of hydrogen as an energy carrier requires advanced monitoring solutions for transport infrastructures, where intelligent sensing and data-driven analysis can play a key role in improving safety and operational efficiency. However, anomaly detection in hydrogen transport networks remains challenging due to the limited availability of operational data and the complexity of transient behaviors associated with these systems. This work investigates a weakly-supervised anomaly detection framework for hydrogen transport networks based on high-fidelity simulation and data-driven analysis. The proposed methodology combines temporal deep learning architectures and unsupervised representation learning models with an operational threshold calibration strategy based on the trade-off between false positives and false negatives. The proposed framework is validated using a high-fidelity simulation environment that reproduces normal and anomalous operating conditions, including leaks, compressor malfunctions, and delayed activation events. The framework is evaluated through comparative experiments involving different anomaly detection architectures, robustness analysis under measurement noise, and leave-one-topology-out generalization tests. Results demonstrate that the proposed approach can effectively identify abnormal behaviors while maintaining robustness against degraded signal quality and previously unseen operating configurations. The obtained results highlight the effectiveness of the proposed methodology as a framework for developing and validating intelligent monitoring strategies for hydrogen transport infrastructures.