Investigation of liquefied natural gas (LNG) dispersion using computational fluid dynamics

Liquefied Natural Gas (LNG) is currently playing an important role in the global energy markets. This is evidenced by the growing demand and increased construction of LNG facilities across Europe and United States. One of the challenging problems within the LNG industry is to protect the general public from hazards which could result from accidental spill. A spill of LNG creates a flammable gas cloud which disperses through the atmosphere constituting fire and explosion hazards. The most commonly performed hazard analysis involves verifying compliance with regulations such as NFPA 95A. One method that is currently being used to establish compliance is dispersion modelling using Computational Fluid Dynamics (CFD). However, real terrain dispersion simulation presents unique difficulty due to issues related to complex turbulent phenomena development, particularly in the presence of obstacles such as buildings in the path of the dispersing cloud The present study aims to demonstrate the potential of Large-Eddy Simulation (LES) for CFD simulation of LNG dispersion. For this purpose, ANSYS CFX is used to simulate LNG dispersion based on the Coyote Series Experiment. Turbulence in the flow field was prescribed via Smagnorinsky sub-grid scale model originally developed for atmospheric turbulence. Under the Smagorinsky model, effect of adopting different closure models for Smagorinsky coefficent has been investigated. This include a constant value coefficient, the Wall damping closure (LES-WALE) and the scale-invariance closure (dynamic model). Computational results are reported and compared with experimental data. Also, results were compared with those obtained from Reynolds-Averaged Navier-Stokes (RANS) simulations conducted as part of this study. Results of the simulations demonstrate that LES performs better than RANS in reproducing experimentally observed concentration trends. Of all three LES closure models of smagorinsky coefficient investigated in this study, the constant value approach with a Smagnorinsky constant of 0.1 produced results which compare more favourably with experiment. The poor performance of the LES-WALE and LES dynamic models is thought to have resulted from the models being over-dissipative and under-dissipative respectively in the near-ground region