Session: 15-03-01: Machine Learning and Deep Learning in Safety, Reliability, and Maintenance I

Paper Number: 172520

Integrated Simulation and Data-Driven Approaches for Hydrogen Leak Safety in Indoor and Outdoor Environments 

The National Renewable Energy Laboratory (NREL) is advancing hydrogen safety and system-level risk assessment through high-fidelity modeling of hydrogen dispersion in both outdoor and indoor environments. These efforts are centered on the implementation of digital twin (DT) frameworks, combined with data-driven inverse modeling techniques, sensor network optimization, and leak localization capabilities across hydrogen infrastructure. A facility-specific digital twin has been developed for the electrolyzer-based hydrogen production system at NREL’s Advanced Research on Integrated Energy Systems (ARIES) platform. This DT integrates site geometry, component layout, leak initiation points, and controlled release parameters with meteorological inputs including wind speed, azimuthal direction, and vertical boundary layer profiles. Using steady-state Reynolds-Averaged Navier-Stokes (RANS)-based computational fluid dynamics (CFD) models with appropriate turbulence closure schemes, the DT simulates hydrogen dispersion under representative environmental conditions.

Simulation results demonstrate that dispersion characteristics vary significantly with ambient wind speed. At high wind velocities, inertial forces dominate, leading to elongated plumes with greater horizontal spread. Under low wind conditions, buoyancy becomes the primary driver, resulting in vertical rise and limited lateral dispersion. These regime shifts impact the spatial distribution of hydrogen concentrations and subsequently affect the effectiveness of fixed-location sensor networks. The digital twin facilitates plume visualization, risk envelope estimation, and data generation for concentration iso-surfaces. These outputs enable scenario-based optimization of sensor placement based on predefined detection thresholds (e.g., >1% vol H₂), supporting more effective alarm triggering and control logic in safety systems.

For indoor applications, a parallel study was conducted in a closed laboratory environment at NREL, employing helium as a surrogate for hydrogen to mimic dispersion behavior under controlled leak scenarios. The experimental campaign was designed to investigate hydrogen dispersion behavior under a range of credible leak scenarios, guided by NFPA standards for hydrogen safety. Controlled gas releases were performed using various orifice diameters and upstream pressures representative of potential failure modes in laboratory settings. These scenarios aimed to replicate realistic leak conditions in terms of flow rate, directionality, and release dynamics. Experiments were conducted under two ventilation modes: passive (HVAC off) and active (HVAC on). In the absence of ventilation, results consistently showed poor air circulation, leading to gas stratification and delayed sensor response. Under active HVAC conditions, airflow patterns; particularly recirculation zones and air curtains near intake vents, significantly influenced dispersion behavior and sensor effectiveness. In certain configurations, the released gas failed to reach detection thresholds due to localized containment by ventilation-induced barriers. CFD simulations replicating these scenarios were used to validate a virtual sensor grid model and quantify detection times, concentration profiles, and spatial coverage. This integrated experimental and modeling approach enabled the evaluation of leak detectability and sensor placement strategies under diverse and realistic operating conditions.

To enhance these capabilities, machine learning (ML) algorithms are being implemented for inverse modeling, specifically for real-time leak source localization based on sparse sensor time-series data. Initial implementations using regression and neural network-based models show promising accuracy and robustness across simulated leak scenarios.

By fusing high-resolution CFD modeling, surrogate gas experimentation, and machine learning, NREL’s platform provides a comprehensive toolset for hydrogen dispersion analysis. These capabilities support safety design, validation of leak detection strategies, and real-time response planning for hydrogen facilities operating under a wide range of environmental and structural conditions. This integrated framework lays the groundwork for dynamic, adaptive hydrogen safety systems capable of supporting the growing deployment of hydrogen technologies in both laboratory and industrial settings.

Presenting Author: Munjal Shah National Renewable Energy Laboratory (NREL)

Presenting Author Biography: Munjal Shah is a Thermal Energy Systems Group researcher at the National Renewable Energy Laboratory (NREL) since 2023. His expertise includes computational fluid dynamics (CFD), finite element modeling, and machine learning. His research includes thermal and mechanical modeling for particle-based concentrated solar power (CSP) receivers, aiming to accelerate industry decarbonization. He also works on the development of thermal energy systems for industrial process heat (IPH) applications and Long-duration energy storage technologies (LDES). Munjal actively participates in CSP and thermal energy storage (TES) research, leveraging high-performance computing for advanced fluid and thermal modeling. He also leads projects focused on dispersion modeling of hydrogen for development and deployment of hydrogen sensor safety technologies for hydrogen storage facilities.

Authors:

Munjal Shah National Renewable Energy Laboratory (NREL)
David Peaslee National Renewable Energy Laboratory (NREL)
Umang Patel National Renewable Energy Laboratory (NREL)
Ian Palin National Renewable Energy Laboratory (NREL)
John Drumm National Renewable Energy Laboratory (NREL)
James Stewart National Renewable Energy Laboratory (NREL)
Kevin Hartmann National Renewable Energy Laboratory (NREL)
Zhiwen Ma National Renewable Energy Laboratory (NREL)
William Buttner National Renewable Energy Laboratory (NREL)