June 11- 14, 2013 Jeju, Korea

Modelling dynamic liquid-gas systems: Extensions to the volume-of-uid solver

Johan A. Heyns 1 , Oliver F. Oxtoby1 and Ridhwaan Suliman1
1

Aeronautic Systems, Council for Scientic and Industrial Research PO Box 395, Pretoria 0001, South Africa
Abstract

This study presents the extension of the volume-of-uid solver, interFoam, for improved accuracy and efciency when modelling dynamic liquid-gas systems. Examples of these include the transportation of liquids, such as in the case of fuel carried onboard air- and spacecraft or liquid natural gas on tankers. As part of the development three extensions are considered: Firstly, a revised surface capturing formulation is proposed; secondly, a new weakly compressible volume-of-uid formulation is presented; and lastly, a piecewise-linear interpolation of the pressure derivative is implemented. For the evaluation of this solver, a number of test cases with dynamic liquid-gas ows are considered where the results are compared with experimental measurements as well as numerical predictions from the current interFoam solver .

1. INTRODUCTION
Various industries benet from the accurate modelling of multi-uid liquid-gas ow. Examples include maritime and naval engineering which require the accurate prediction of wave impact loads on xed and oating structures. Furthermore, with the transportation of liquids it is necessary to accurately describe the two-uid ow as it greatly inuences sloshing induced impact pressures measured on container walls. In commercial aircraft, for example, fuel can constitute a signicant portion of take off weight and may be subjected to highly dynamic loading conditions. On liqueed natural gas (LNG) carriers stringent restrictions are placed on tank lling levels due to potentially large sloshing induced loads. With the aim of modelling immiscible liquid-gas ow accurately and efciently, Computational Fluid Dynamics (CFD) is increasingly employed as it has in recent years developed to a point where it can provide a cost effective alternative. Most existing liquid-gas free-surface models treat both gas and liquid as incompressible [14], neglecting the affect of density changes due to pressure variations. However, when considering high density ratio liquid-gas systems under low Mach number ow conditions, it may be required to extend the volume-of-uid method to account for variations in the gas density. Various experimental studies [57] suggest that under certain conditions the system may be subjected to large variations in pressure where the compressibility of the gas may effect the predicted pressure loads. By means of a non-dimensional

Corresponding author: Johan A. Heyns (jheyns@csir.co.za)

analysis, a weakly compressible formulation for low Mach numbers may be derived [8]. It is found that by accounting for variations in gas density, some high frequency pressure oscillations, observed in experimental results, may be captured. Furthermore, it allows for the evaluation of absolute pressures and demonstrates some improvement in computational efciency. With the aim of increased accuracy when modelling the evolution of the free-surface interface, especially when considering dynamic or violent ow conditions, the interface capturing formulation is extended. The proposed formulation combines a second-order accurate higher-resolution interpolation scheme [4] with an articial compressive term [9]. This allows for the preservation of a sharp interface, while maintaining the integrity of the interface shape [10]. When evaluating dynamic sloshing, it is found that the formulation provides a closer approximation of the peak pressures and load frequencies. If the density of the liquid and gas varies signicantly, small changes in the pressure gradient over the interface will induce inaccurate acceleration of the gas [3]. Consequently, this results in the formation of spurious oscillations in the velocity eld over the interface. The formation of spurious velocities are particularly evident on cut cell Cartesian meshes, typically generated by snappyHexMesh. To mitigate this behaviour, the pressure derivative is interpolated using a piece-wise linear approximation [11]. For high density ratio systems, it is found that this approach may result in a notable reduction in the magnitude of the said spurious velocities. To evaluate the proposed solver for modelling liquid-gas systems, various partially lled tank congurations are considered. Numerical results from these analyses are compared with experimental measurements as well as numerical results from the interFoam solver. It is found that through the implementation of the proposed extensions an improvement in accuracy and computational efciency when modelling liquid-gas systems may be achieved.

2. NUMERICAL MODEL
Using the Eulerian volume-of-uid (VOF) approach a control volume partially lled with liquid and gas can be considered, where the volume fraction occupied by the liquid and the gas are respectively denoted and 1 . For a given cell an averaged velocity, u, and pressure, p, can be assumed if a homogeneous ow model is employed. The governing equations describing the uid ow are respectively the gas continuity, volume fraction or liquid continuity equation as well as an averaged liquid-gas momentum equation [8] uj (1 ) g = (1) g t xj ( u j ) + =0 t xj ( ui ) ( ui u j ) p ui + + = gi + t xj xi xj xj (2) (3)

where the gas continuity equation (1) contains a temporal term that accounts for variations in gas density using a weakly compressible formulation for low Mach number ows. In the previous equations time is denoted t , g is gravitational acceleration, xi is the Cartesian coordinate in the direction i and the mixture density and dynamic viscosity can be calculated from the volume fractions = l + (1 )g = l + (1 )g where the density and viscosity are denoted and . The subscripts l and g are used to represented the liquid and gas. The volume-of-uid or liquid continuity equation (2) remains unchanged from the incompressible formulation and existing VOF surface capturing schemes [4, 9, 1214] may be employed. With the aim of improving the accuracy when modelling the evolution of the free-surface interface a blended higher

resolution articial compressive (HiRAC) formulation [10] is implemented. The formulation combines a second-order Crank-Nicolson discretisation and a blended higher resolution scheme [4] with an articial compressive term [9]. This allows for the preservation of a sharp interface, while maintaining the integrity of the interface shape. The semi-discrete form of the VOF equation reads n+1 n = t 1 (ui ) 2 xi
n+1

(ui ) xi

[uc |i (1 )] xi

(4)

where the articial term, which is only activated in the interface, reduces the smearing of the free-surface through the compressive velocity, uc . The volume fraction face value, f , is interpolated using a blended high-resolution scheme, which switches between a compressive and more diffusive scheme based on the alignment of the free-surface and the mesh. As the density of the liquid and gas varies signicantly, small changes in the pressure gradient over the interface may induce inaccurate acceleration of the gas. This consequently results in the formation of spurious oscillations in the velocity eld over the interface. To overcome this, Panahi et al. [11] discretised the spatial pressure derivative in a piece-wise linear manner. They noted that the velocity eld is continuous throughout the domain, so that for a given cell face 1 p x c
f

(5)

where c is a constant. To compute the interpolated face pressure, the pressure is treated as piece-wise linear and the density as piece-wise constant. Eq. (5) is integrated from node, p, to the cell face, f , to nd an expression for p f in terms of the pressure at node p
f p

1 p dx = x

c dx
p

(6)

which yields p f pp = pL c 2 (7)

where L is the distance between node p and n. In a similar fashion, the face pressure in terms of the pressure at node n is obtained pn p f = n L c 2 (8)

The resulting interpolated pressure at the cell face is found by subtracting Eq. (7) from Eq. (8) pf = p pn + n p p p + n (9)

which is found to provide a fair approximation of the pressure gradient over the free-surface of uids with large density differences.

3. NUMERICAL RESULTS
In the rst test case, a partially lled tank with bafe conguration is modelled. Due to the dynamic or violent sloshing of the liquid, this test case allows for the benchmarking of different surface capturing formulations. The numerical results are also evaluated against pressures obtained from experimental measurements [8, 15]. A schematic of the tank is shown in Figure 1; it is 25 % lled with liquid and

Figure 1: Schematic representation of partially lled tank with bafes Table 1: Material properties of the liquid and gas Liquid (Water) Gas (Air) Density (kg/m3 ) 998 1.21 3 Viscosity (kg/m s) 1.002 10 1.812 106 Acoustic velc (m/s) 343.2 is subjected to sinusoidal lateral excitation with an acceleration of 0.6 g and an excitation frequency of 1.1 Hz. For the analysis a structured mesh with 12000 nodes is employed and viscous no slip boundary conditions are specied on the tank walls. Furthermore, for the analysis the material properties as given in Table 1 are employed and a gravitational acceleration of g = 9.81 m/s2 is assumed. In Figure 2 the pressure measured on the bottom bafe is shown. The numerically predicted results for different surface capturing schemes are compared to experimental measurements. Results from the interFoam solver are denoted MULES [16], CICSAM is the scheme proposed by Ubbink and Issa [4] and HiRAC [10] is the scheme implemented as part of this work. From these results it appears the combination of the second-order temporal discretisation along with articial compressive term results in an improvement when evaluating the peak pressures.
Experimental MULES CICSAM HiRAC

Pressure 0

6 t (s)

10

12

Figure 2: Predicted pressures on the lower bafe of the tank To evaluate the inuence of variations in gas density the weakly compressible solver is compared to a solver assuming the gas is incompressible. In Figure 3 a shorter time period of the analysis is shown during which the top bafe opening is being closed off by liquid. During this time the compressible gas acts as a spring system absorbing and releasing energy. By accounting for variations in gas density the weakly compressible formulation provides a closer approximation of measured pressure. It is furthermore found that the weakly compressible formulation results in a 30 % reduction in computational time. It is

suspected that by accounting for variations in gas density the numerical system is softened, resulting in the improved convergence rates.

Figure 3: Predicted pressures for the weakly compressible solver The next test case considers the extension to unstructured meshes typically employed when modelling complex geometries. For this a cylindrical tank partially lled with liquid at rest is modelled. A cut-cell Cartesian mesh, with renement towards the tank wall, is generated with snappyHexMesh. It is found that in the region where the mesh under go a change in cell size on the liquid-gas interface, it may result in numerical imbalances in the pressure eld. In Figure 4 the cut-cell mesh along with the velocity magnitude is shown, highlighting the problem. In the gure the velocity magnitude along the cross-section of the domain is plotted and compared. It is found that the piece-wise linear interpolation provides a reduction in spurious velocities as it provides a more accurate representation of the pressure eld.

(a) Cut-cell Cartesian mesh

(b) Velocity magnitudes

Figure 4: Evaluation of pressure interpolation on a cut-cell Cartesian mesh The last test case evaluates the damping ratio of a cylindrical tank with a single ring bafe, comparing the numerical predictions to experimental results as well as empirical correlations for linear wave motion [17]. The damping ratio, which represents a decay in the slosh wave amplitude, is typically a function of the submergence height, h, tank radius, r and bafe geometry. A schematic representation of the cylindrical tank is shown in Figure 5 and the material properties as given in Table 1 are used. Figure 6a compares the numerical predictions to experimental measurements for different submergence

Figure 5: Schematic representation of open cylindrical tank heights and wave amplitudes. The data points represent various experimental ndings and the solid lines are calculated using the empirical correlations suggested by ONeill [18]. It is found that the numerical results provide a reasonable approximation of the damping ratio. In Figure 6b the results of the solver with piece-wise linear interpolation of the pressure is compared to the interFoam solver, showing that through the more accurate representation of the pressure eld a better approximation of the damping ratio is achieved. As expected, due to the relatively small linear wave amplitude of these analyses, the results appear to be insensitive to the surface capturing formulation.

(a) Comparison with experimental

(b) Comparison with interFoam

Figure 6: Damping ratios for a cylindrical tank with a bafe

4. CONCLUSION
In conclusion, this study presents extensions to the volume-of-uid solver, interFoam. A new surface capturing formulation is implemented and the the solver is extended to account for variations in gas density by means of a weakly compressible formulation for low Mach number ows. Furthermore, a new piece-wise linear interpolation of the pressure derivative is implemented. From the results presented, it appears these extensions may provide some improvement in both accuracy and computational efciency when modelling liquid-gas systems.

REFERENCES
[1] T Wacawczyk and T Koronowiczy. Modeling of the wave breaking with CICSAM and HRIC high-resolution schemes. In Proceedings of the European Conference on Computational Fluid Dynamics ECCOMAS CFD, 2006.

[2] Dongming Liu and Pengzhi Lin. Three-dimensional liquid sloshing in a tank with bafes. Ocean engineering, 36(2):202212, 2009. [3] Rainald L ohner, Chi Yang, and Eugenio O nate. On the simulation of ows with violent free surface motion. Computer Methods in Applied Mechanics and Engineering, 195(41):55975620, 2006. [4] O Ubbink and RI Issa. A method for capturing sharp uid interfaces on arbitrary meshes. Journal of Computational Physics, 153(1):2650, 1999. [5] GN Bullock, C Obhrai, DH Peregrine, and Henrik Bredmose. Violent breaking wave impacts. part 1: Results from large-scale regular wave tests on vertical and sloping walls. Coastal Engineering, 54(8):602617, 2007. [6] C Lugni, M Brocchini, and OM Faltinsen. Wave impact loads: The role of the ip-through. Physics of uids, 18:122101, 2006. [7] OM Faltinsen, OF Rognebakke, and AN Timokha. Classication of three-dimensional nonlinear sloshing in a square-base tank with nite depth. Journal of uids and structures, 20(1):81103, 2005. [8] Johan A Heyns, Arnaud G Malan, Thomas M Harms, and Oliver F Oxtoby. A weakly compressible free-surface ow solver for liquid-gas systems using the volume-of-uid approach. Journal of Computational Physics, 2013. [9] H Jasak and HG Weller. Interface tracking capabilities of the inter-gamma differencing scheme. Unpublished, February, 42, 1995. [10] JA Heyns, AG Malan, TM Harms, and OF Oxtoby. Development of a compressive surface capturing formulation for modelling free-surface ow by using the volume-of-uid approach. International Journal for Numerical Methods in Fluids, 2012. [11] Roozbeh Panahi, Ebrahim Jahanbakhsh, and Mohammad S Seif. Final report on an experimental investigation of sloshing. Applied ocean research, 28(3):171181, 2006. [12] M Darwish and F Moukalled. Convective schemes for capturing interfaces of free-surface ows on unstructured grids. Numerical Heat Transfer, Part B: Fundamentals, 49(1):1942, 2006. [13] S Muzaferija, M Peric, P Sames, and T Schellin. A two-uid Navier-Stokes solver to simulate water entry. In Proc 22nd Symposium on Naval Hydrodynamics, pages 277289, 1998. [14] Bruno Lafaurie, Carlo Nardone, Ruben Scardovelli, St ephane Zaleski, and Gianluigi Zanetti. Modelling merging and fragmentation in multiphase ows with surfer. Journal of Computational Physics, 113(1):134147, 1994. [15] BEELAB test reference BAE808: Laboratory simulation of sloshing in part-lled tanks, 2008. [16] Henrik Rusche. Computational uid dynamics of dispersed two-phase ows at high phase fractions. PhD thesis, Imperial College London (University of London), 2003. [17] F. T. Dodge. The new dynamic behaviour of liquids in moving containers. Southwest Research Inst., 2000. [18] J.P. ONeill. Development of a vof-fractional step solver for oating body motion simulation, 1960.