TY  - EJOU
AU  - Hwang, Jiwon 
AU  - Lee, Sangjun 

TI  - Deep Learning-Based Spatiotemporal Surrogate Framework for Reconstructing Transient 3D Hydrogen Explosion Overpressure Fields
T2  - Computer Modeling in Engineering \& Sciences

PY  - 
VL  - 
IS  - 
SN  - 1526-1506

AB  - Computational Fluid Dynamics (CFD) has become the benchmark approach for analyzing hydrogen explosions because it accurately resolves transient shock-wave propagation and complex blast-wave interactions. However, high-fidelity CFD simulations remain computationally expensive for repeated safety evaluation and large-scale parametric studies. To address this limitation, this study proposes a deep learning-based spatiotemporal surrogate framework for reconstructing transient three-dimensional (<math id="mml-ieqn-1"><mn>3</mn><mrow><mi mathvariant="normal">D</mi></mrow></math>) hydrogen explosion overpressure fields. High-resolution CFD datasets were generated using the OpenFOAM-based <monospace>radXiFoam</monospace> solver by systematically varying blast-wall height and setback distance, producing approximately 20.4 million spatiotemporal pressure samples from 1859 monitoring locations. Three representative sequential deep learning architectures, namely Long Short-Term Memory (LSTM), Transformer, and Temporal Convolutional Network (TCN), were comparatively evaluated using an identical training and evaluation protocol. Among the investigated models, TCN achieved the highest reconstruction accuracy for an unseen interpolation test configuration, yielding a Root Mean Square Error (RMSE) of 0.001113 and a coefficient of determination (<math id="mml-ieqn-2"><msup><mi>R</mi><mrow><mn>2</mn></mrow></msup></math>) of 0.9778. The proposed framework successfully reconstructed the spatiotemporal evolution of hydrogen explosion overpressure fields, including blast-wave propagation, reflection, diffraction, and attenuation, while substantially reducing computational cost compared with conventional CFD simulations. Overall, the proposed surrogate framework provides a computationally efficient surrogate modeling approach for transient three-dimensional overpressure field reconstruction, supporting hydrogen explosion hazard assessment and protective barrier evaluation.
KW  - Hydrogen explosion; surrogate modeling; temporal convolutional network; spatiotemporal overpressure reconstruction; computational fluid dynamics; deep learning

DO  - 10.32604/cmes.2026.086869