Arman Aryanzadeh

As hydrogen emerges as a critical pillar of the global decarbonization agenda, enabling its safe, cost-effective, and scalable transportation is essential. One strategic approach is the repurposing of existing oil and gas infrastructure—particularly midstream pipeline networks and subsurface storage reservoirs—for hydrogen transmission. This reuse strategy offers significant capital cost savings and deployment speed; however, it introduces substantial technical risks due to hydrogen’s unique properties. Its low molecular weight, high diffusivity, and propensity to cause material embrittlement increase the probability of undetected leakage, structural degradation, and cascading failure events. These limitations necessitate advanced monitoring architectures beyond conventional integrity management systems.

This research proposes an end-to-end framework that combines repurposed infrastructure with intelligent, AI-based leak detection and predictive monitoring systems. The approach comprises four main stages:

1. Sensor Data Integration: Deployment of a multi-modal sensor array (acoustic emission sensors, distributed fiber-optic sensors, pressure and flow meters, and electrochemical H₂ sensors) along pipeline routes and at critical stress points (e.g., valves, weld zones, storage wells). These generate high-resolution spatial-temporal data on strain, vibration, pressure transients, and gas composition.


2. Data Preprocessing and Fusion: Application of signal denoising techniques, time-series synchronization, and data fusion methods to integrate heterogeneous sensor outputs. Feature engineering extracts meaningful physical patterns such as micro-leak signatures, abrupt pressure drops, or harmonic changes indicative of crack propagation.


3. AI Model Architecture: Design of a hybrid machine learning pipeline consisting of:

• Supervised classifiers (e.g., gradient boosting, CNN-LSTM) trained on labeled fault events and benchmark datasets to identify and classify leak types.


• Unsupervised models (e.g., autoencoders, isolation forests) to detect early-stage anomalies in non-linear patterns where no labeled failures exist.


• Physics-informed neural networks (PINNs) integrated with hydrogen flow simulations and fracture mechanics to enhance model interpretability and generalizability under unseen conditions.

    4. Risk Scoring and Decision Support: A real-time risk dashboard correlates leak probability indices, location certainty,            and severity levels. Predictive maintenance schedules are generated by coupling AI output with a digital twin of the              infrastructure, reducing both false alarms and unplanned downtime.

Informed by insights from recent advancements in underground hydrogen storage—including wettability dynamics, brine-rock-H₂ interaction, and caprock sealing failure—this framework ensures that geomechanical and geochemical risks are accounted for during model training and operational calibration.

By combining retrofitted legacy systems with next-generation intelligent diagnostics, this research delivers a scalable blueprint for hydrogen transport that meets stringent safety, environmental, and economic performance metrics. It establishes a practical foundation for integrating hydrogen into national energy grids while ensuring integrity across critical assets during the energy transition.