Computational Fluid Dynamics (CFD) modelling of rollover and application to Liquefied Natural Gas storage

The interest to study and predict Liquefied Natural Gas (LNG) rollovers dates back to the 60s following La Spezia incident, which highlighted the need to take extra care when storing LNG with different densities together. However, LNG rollovers - which are complex physical phenomena during which a stratified LNG is suddenly and rapidly mixed while releasing large amounts of vapour - can be hazardous. In order to limit any resulting overpressure and subsequent releases of vapour into the atmosphere, rollovers have to be better understood. Currently, 0D lumped-parameter models are used in the LNG industry, but by restricting rollovers as the consequence of the average density equalisation of two adjacent layers, it leaves little room to tank design and filling procedure improvements. The novelty of this work is the development of an LNG rollover predictive code, rolloverFoam, based on Computational Fluid Dynamics and capable of providing spatial representations - in 2D and 3D - of the phenomenon while predicting rollover occurences reliably. RolloverFoam, which is based on the Navier-Stokes equations, integrates the effects of buoyancy via the Boussinesq approximation. Besides, unlike previous studies, the effects of turbulence are also accounted for in the code, and among the existing RANS methods, k-e models incorporating the inuence of buoyancy were used. Given the lack of data publicly available for LNG rollovers, rolloverFoam was at first applied to simulate small-scale experiments with mixtures of Freon. Later, it was applied to a medium-scale LNG rollover experiment and the well-known La Spezia incident. The numerical results, obtained both for Freon and LNG, have shown very encouraging agreements with experimental data while providing some insights into the physics underpinning the phenomenon thanks to 2D and 3D spatial representations. Besides, an advantage of rolloverFoam is its adaptability to more complex geometries compared to traditional lumped-parameter models currently used in the LNG industry. Hence, the method was finally applied to simulate a stationary medium-scale FLNG Moss type tank