Large Eddy Simulations of a Hydrogen-Air Explosion in an Obstructed Chamber Using Adaptive Mesh Refinement

Following the growing use of hydrogen in the industry, gas explosions have become a critical safety issue. Computational Fluid Dynamic (CFD) and in particular the Large Eddy Simulation (LES) approach have already shown their great potential to reproduce such scenarios with high fidelity. However, the computational cost of this approach is an obvious limiting factor, since fine grid resolutions are often required in the whole computational domain to ensure a correct numerical resolution of the deflagration front all along its propagation. In this context, Adaptive Mesh Refinement (AMR) is of great interest to reduce the computational cost, as it allows to dynamically refine the mesh throughout the explosion scenario, only in regions where Quantities of Interest (QoI) are detected. This study aims to demonstrate the strong potential of AMR for the LES of explosions. The target scenario is a hydrogen-air explosion in the GraVent explosion channel [1]. Using the massively parallel Navier- Stokes compressible solver AVBP, a reference simulation is first obtained on a uniform and static unstructured mesh. The comparison with the experiments shows a good agreement in terms of absolute flame front speed, overpressure and flow visualisation. Then, an AMR simulation is performed targeting the same resolution as the reference simulation only in regions where QoI are detected, i.e., inside the reaction zones and vortical structures. Results show that the accuracy of the reference simulation is recovered with AMR for only 12% of its computational cost.