Grid Resolution for the Simulation of Sloshing using CFD

Bernhard Godderidge , Mingyi Tan , Chris Earl & Stephen Turnock

Introduction
Sloshing occurs when a tank is partially lled with a liquid and subjected to an external excitation force [1]. Ships with large ballast tanks and liquid bulk cargo carriers, such as very large crude carriers (VLCCs), are at risk of exposure to sloshing loads during their operational life [2]. The inclusion of structural members within the tanks dampens the sloshing liquid sufciently in all but the most severe cases. However, this approach is not used for Liqueed Natural Gas (LNG) carriers and the accurate calculation of the sloshing loads is an essential element of the LNG tank design process [3, 4]. The increase in global demand for LNG has resulted in a new generation of LNG tankers with a capacity in excess of 250,000 m3 , compared to 140,000 m3 today. A prerequisite for the safe operation of these LNG tankers is an accurate calculation of the sloshing loads experienced by the containment system [5, 6]. The work of Abramson [7] summarizes the methods available in modern sloshing analysis, and Ibrahim [8] gives an up-to-date survey of analytical and computational sloshing modeling techniques. A more general modeling technique is the solution of the Navier-Stokes equations using Computational Fluid Dynamics (CFD). Some recent examples of CFD sloshing simulation include Hadzic et al. [9], Aliabadi et al. [10], Standing et al. [11], Kim et al. [12], Rhee [13] and El Moctar [14]. Sloshing ows are treated as a transient problem in CFD. While the number of sloshing oscillations can vary, a large number of time steps, usually O(102 ) to O(103 ) per oscillation are required. Design optimization or the use of a numerical wave tank to gather statistical sloshing pressure data [15] requires long simulation times or multiple runs. Parameters such as time step size, grid spacing and model choice directly inuence the complexity and computational cost of a CFD model.

Hinatsu [16] by Godderidge et al. [17, 18]. The effect of grid spacing when capturing impact pressure caused by an enclosed air bubble is investigated. It is found that local ow features are best suited to indicate that the ow is sufciently well resolved. These ndings are further investigated using larger, geometrically similar sloshing tanks. The initial grid geometry is used to simulate the scaled sloshing ow at two and four times the initial grid size. Then, the grid is rened to give the same mesh spacing as in the rst problem.

Sloshing Problem
Sloshing in rectangular container, induced by pure sway motion, is investigated in the present study. Figure 1 shows the tank dimensions, locations of pressure monitor points and axis system orientation. The CFD model was validated using the experimental steady state sloshing pressures given by Hinatsu [16]. The tank displacement is given by x = A sin 2 t , T (1)

where A is the displacement amplitude, T the sloshing period and t the elapsed time. In the current case, the tank motion is in the x-direction only, as indicated in Figure 1. The rst part of the investigation is focused on a resonant sloshing ow at 20% lling level, where A = 0.06 m and T = 1.74 sec. Subsequently, a near-resonant sloshing ow with A = 0.015 m and T = 1.404 sec is considered.

A sway-induced resonant sloshing ow in a 1.2 m x 0.6 m rectangular container is investigated using a commer- Figure 1: The sloshing problem used for CFD validation (all dicial Navier-Stokes CFD code. The selected computational mensions in m) model was validated using experimental pressure data from

The uid interaction models for the numerical simulation of sloshing can be implemented using the volume fraction of each uid to determine the uid mixture properties. This is a homogeneous multiphase model. It is analogous to the volume of uid (VOF) method developed by Hirt and Nichols [19], but it includes a simplication as the free surface pressure boundary condition is neglected. A more general but computationally more expensive approach is an inhomogeneous multiphase model, where the solution of separate velocity elds for each uid is matched at the uid interfaces using mass and momentum transfer models [20]. An inhomogeneous viscous compressible multiphase ow with two phases and is governed by the conservation of mass for the compressible phase (r) + (rui ) = m + , t xi (2)

wall function aided convergence when using a viscous ow model [17]. The high resolution scheme for spatial discretization varies between a rst and second order upwind scheme depending on the gradient [21]. The grids used for the various studies are detailed in the sections describing the respective results.
Table 1: Computational models used for sloshing simulations

Parameter Water Air Multiphase model Sloshing motion Turbulence model Spatial discretization Time discretization Timestep control Convergence control

where is mass transfer between the phases and m mass sources, density, r volume fraction and ui velocity of phase . The conservation of momentum for phase is given as p (rui ) + (rui u j ) = r + t x j xi ui u j + r + + M + M + bi , x j x j xi

Setting Incompressible uid Ideal gas inhomogeneous Body force Standard k with scalable wall function High resolution Second order backward Euler Root-mean-square (RMS) Courant number=0.1 RMS residual < 105

The investigation of sloshing in geometrically similar (3) containers required the calculation of an appropriate sloshing excitation. The nature of the excitation, given in Equa where bi are body forces, M forces on the interface caused tion (1), is maintained but the amplitude and frequency are by the presence of phase , the dynamic viscosity and the adjusted. The sloshing period is 95% of the resonant period which depends on the tank size. The resonant freqnecy for term M = ui ui interphase momentum trans- each case is calculated from potential theory as fer caused by mass transfer. If the uid is compressible, g h Equations (2) and (3) are closed using an energy equa2 tanh , (4) n= tion, or an equation of state if the compressible uid can a a be treated as an ideal gas [21]. A discussion of the uid where a is the tank length, g gravity and h the lling level. interface forces is given by Godderidge et al. [22]. The amplitude of the sloshing excitation is adjusted using the sloshing velocity, which may be obtained by differentiAs a full set of conservation equations has to be solved ating Equation (1). Taking the excitation velocity as a charfor each phase, the computational effort required for the acteristic velocity, the following non-dimensional scaling inhomogeneous model has been found to be 2.3 times parameter based on the Froude number [7] can be used greater1 than for the homogeneous model [22]. However, Brennen [23] nds that if two conditions derived from parx l gDl = , (5) ticle size parameter, mass parameter and particle Reynolds xL gDL number are violated, the inhomogeneous multiphase model (Equations 2 and 3) should be used. It is observed that for where Dl and DL are characteristic length scales. the current problem, the particle Reynolds number condition is not satised. This suggests that the use of an inhomogeneous multiphase model is required for the analysis Impact bubble of the current problem. The uid motion caused by sloshing results in static and dynamic pressure loads. The dynamic pressures are usuThe computational models used in the sloshing studies ally conned to small regions, but cause large loads on the are summarised in Table 1. The selection is based on the structure. Sloshing pressure loads can be categorised as sensitivity study by Godderidge et al. [17]. It was found pure uid impact, impact air bubble formation and the imthat the pressure histories of the current uid model com- pact of an air-water mixture formed during a previous uid bination differed by less than 0.1% from the fully com- impact. Pure uid impact has been studied experimentally pressible model but required 20% less computational time. by Peregrine [24] and impact pressures in excess of 100 Kim et al. [12] showed that the sloshing pressure is not times the static uid pressures were observed. The resoinuenced by the inclusion of a turbulence model, but the nant sloshing ow results in a jet impacting the tank wall use of a standard k turbulence model with a scalable and subsequent air bubble entrainment. This tends to result 1 The simulations were run on a 64 bit, 2.2 GHz processor with 2 GB in a longer, oscillating pressure history when compared to of RAM at the University of Southampton Iridis 2 computational facility a pure uid impact.

2(a): Bubble size dependence on grid

2(b): Air bubble formation

Figure 2: Air entrainment bubble formed during sloshing impact

3(a): P2

3(b): P3

Figure 3: Pressure history at P2 (left) and P3 (right)

The sloshing ow is simulated on a hybrid grid with a rened region indicated in Figure 2. Table 2 gives the grid particulars in the rened region. Figure 2(a) shows the grid dependence of the air bubble dimensions and the formation of the air bubble is illustrated in Figure 2(b).
Table 2: Grid renement for sloshing impact

model and initialisation, a signicantly lower pressure than the inhomogeneous model. Figure 4 shows that the direction of the water prior to impact depends on the selected multiphase model. The inhomogeneous ow predicted water velocity is inclined 14.0 from the horizontal, while the homogeneous model estimated the velocity vector inclination at 40.3 .

Grid Grid a Grid b Grid c Grid d

Hex. 408 2552 4602 16284

The grid dependence of the calculated pressures is shown in Figure 5. The characteristic length scales are the length and depth for P1 and P2 are taken where the water surface is above its initial position The length and height of the bubble are the characteristic scales for P3 . Figure 5(a) Figure 3 shows the pressure history during uid impact shows the plot for the grid spacing perpendicular to the botat P2 and P3 for the 20% lling level. In both cases, the ho- tom wall for P1 and P2 and the side wall for P3. The grid mogeneous multiphase model gives, for the identical uid spacing parallel to the wall is shown in Figure 5(b).

horizontal (rst node) (mm) 0.30 0.10 0.05 0.02

vertical (mm) 12 3.5 2.0 0.5

Table 3: Systematic tank size variations

Parameter Size factor Length Height Filling Level A

T10

Case 1 Case 2 Sloshing Tank 1 2 1.2 m 2.4 m 0.6 m 1.2 m 60% 60% Excitation 1 1.961 0.015 m 0.015 m 1.474 sec 2.044

Case 3 4 4.8 m 2.4 m 60% 3.922 0.015 m 2.890

case 1. The computational models used in the simulations are given in Table 1.

5(a): Perpendicular grid spacing

Figure 6: Typical hybrid grid used in CFD investigations. The grid contains 8652 hexahedral and 708 wedge elements

5(b): Longitudinal grid spacing

Figure 5: Pressure dependence on local perpendicular (top) and longitudinal (bottom) ow feature resolution

Figures 7(a) and 7(b) shows the pressure histories at monitor points P4 and P9 respectively for case 3 predicted using grids 1 and 3. At P4, which is dominated by the static pressure component, the pressure histories show reasonable agreement. At P9, the pressure spike captured on grid 3 is not observed using grid 1. Mean uid speed is less susceptible to grid effects. Figures 7(c) and 7(d) show the mean uid velocity, which is computed as Mean uid speed =
mass

mi |vi |
mass

Tank Size Variation

mi

(7)

Figure 7(c) shows acceptable agreement between the mean Equation (1), which describes the sloshing excitation, can uid speed history observed using grids 1 and 3. Finally, be rewritten as mean uid speed appears to be a quantity well suited for scaling with Equation (5) as shown in Figure 7(d). While 2 x = A sin t , (6) the scaled and observed speed histories are out of phase T when using grid 1, the predicted magnitudes show good where is a constant. Using Equation (4) and (5), the agreement with those observed when scaling from grid 2. sloshing excitation can be adjusted for kinematic similitude corresponding to the tank size. The computed values are Concluding Remarks given in Table 3. The grid used for Case 1 which consists of 9360 elements is shown in Figure 6. Grid size and time discretisation parameters were determined from Ref [25]. This grid is then resized using the appropriate size factors for cases 2 and 3. The number of grid cells remains constant but the size of each element increases accordingly. A second set of grids (grid 2 and 3) is constructed for cases 2 and 3 respectively. They contain 38,319 and 153,273 elements respectively and they have the same cell size as the grid used for The faithful discretisation of a sloshing problem in CFD depends on the resolution of local ow features. The most severe pressures were conned to small regions in the problem and occurred as a result of an impact jet and consequent air bubble entrapment. While the air bubble size was estimated accurately using a coarse grid, the grid independence of pressure requires a considerably ner grid. Thus, grid guidelines explicitly specifying grid spacing (e.g. Ref [26]) may not be adequate for sufciently accurate computations. A better approach is to use a coarse

7(a): Pressure at P4

7(b): Pressure at P9

7(c): Grid dependence of uid momentum

7(d): Predicted and observed uid momentum

Figure 7: Grid inuence on and scaling of uid momentum

grid to observe critical ow features and repeat the simula- [12] Yonghwan Kim, Jungmoo Lee, Young-Bum Lee, and YongSoo Kim. Sensitivity study on computational parameters for tion on a grid that adequately resolves local ow features the prediction of slosh-induced impact pressures. In Proby including information from e.g. Ref [27]. When increasing the tank size, local impact pressures are not captured unless the grid is rened according to the ow [13] Shin Hyung Rhee. Unstructured grid based ReynoldsAveraged Navier-Stokes method for liquid tank sloshing. eld. Moreover, the scaling of sloshing pressures remains Transactions of the American Society of Mechanical Engia task of some difculty. The mean uid velocity dened neers, 127:572582, 2005. in Equation (7) appears to be a quantity better suited to scaling. The magnitude of the mean uid velocity of case [14] Ould El Moctar. Assessment for tankers. Shipping World and Shipbuilder, 204:2831, 2006. 3 is estimated with good accuracy based on grids 1 or 2. However, a lag develops between the solution estimated [15] Mateusz Graczyk, Torgeir Moan, and Olav Rognebakke. Probabilistic analysis of characteristic pressure for LNG from grid 1 and the mean uid momentum obtained from tanks. Journal of Offshore Mechanics and Arctic Engineergrid 3.
ing, 128:133144, 2006. ceedings of the 15th International Offshore and Polar Engineering Conference, 2005.

The scaling of mean uid velocity requires further study [16] Munehiko Hinatsu. Experiments of two-phase ows for the joint research. In Proceedings of SRI-TUHH minifor additional validation. The simulations for the systemWorkshop on Numerical Simulation of Two-Phase Flows. atic variations of tank size should be extended to at least 10 National Maritime Research Institute & Technische Univeroscillations. A further tank size of 9.6 m by 4.8 m should sit at Hamburg-Harburg, NMRI, 2001. be included to conrm the scaling properties of mean uid momentum. Ultimately, the mean uid velocity may pro- [17] Bernhard Godderidge, Mingyi Tan, and Stephen Turnock. A verication and validation study of the application of comvide an alternative design criterion more suitable for scalputational uid dynamics to the modelling of lateral sloshing when assessing the safety of LNG tanks. ing. Ship Science Report 140, University of Southampton,
2006.

References
[1] Harald Olsen. What is sloshing? In Seminar on Liquid Sloshing. Det Norske Veritas, 1976. [2] E Rizzuto and R Tedeschi. Surveys of actual sloshing loads onboard of ships at sea. In NAV 97: International Conference on Ship and Marine Research, 1997. [3] Robert L Bass, E B Bowles, and P A Cox. Liquid dynamic loads in LNG cargo tanks. SNAME Transactions, 88:103 126, 1980. [4] T Knaggs. New strides in ship size and technology. Gas Ships: Trends and Technology., 2:14, 2006. [5] Sungkon Han, Joo-Ho Heo, and Sung-Geun Lee. Critical design issues of new type and large LNG carriers. In Proceedings of the 15th International Offshore and Polar Engineering Conference, 2005. [6] James Card and Hoseong Lee. Leading technology for next generation of LNG carriers. In Proceedings of the 15th International Offshore and Polar Engineering Conference, 2005. [7] H Norman Abramson. The dynamic behavior of liquids in moving containers, with applications to space vehicle technology. Technical Report SP-106, National Aeronautics and Space Administration, 1966. [8] Raouf A Ibrahim. Liquid Sloshing Dynamics. Cambridge University Press, 2005. [9] I Hadzic, Frank Mallon, and M Peric. Numerical simulation of sloshing. Technical report, Technische Universit at Hamburg-Harburg, 2002. [10] Shahrouz Aliabadi, Andrew Johnson, and Jalal Abedi. Comparison of nite element and pendulum models for simulation of sloshing. Computers and Fluids, 32:535545, 2003. [11] R G Standing, S Amaratunga, F Lopez-Calleja, S Orme, and R Eichaker. Marine hydrodynamics modelling using CFD. In CFD 2003: Computational Fluid Dynamics Technology in Ship Hydrodynamics, 2003.

[18] Bernhard Godderidge, Mingyi Tan, Stephen Turnock, and Chris Earl. Multiphase CFD modelling of a lateral sloshing tank. In Numerical Towing Tank Symposium, 2006. [19] C W Hirt and B D Nichols. Volume of uid (VOF) method for the dynamics of free boundaries. Journal of Computational Physics, 39:201225, 1981. [20] M Ishii and T Hibiki. Thermo-Fluid Dynamics of TwoPhase Flow. Springer Verlag, 2006. [21] Ansys Inc. CFX-10 Users Guide, 2005. [22] B Godderidge, S Turnock, M Tan, and C Earl. An investigation of multiphase CFD modelling of a lateral sloshing tank. Computers and Fluids (submitted). [23] C E Brennen. Fundamentals of Multiphase Flow. Cambridge University Press, New York, 2005. [24] Wu-Ting Tsai and Dick K. P Yue. Computation of nonlinear free surface ows. Annual Review of Fluid Mechanics, 28:249278, 1996. [25] B Godderidge, M Tan, C Earl, and S Turnock. Boundary layer resolution for modeling of a sloshing liquid. In International Society of Offshore and Polar Engineers, 2007. [26] American Bureau of Shipping. Guidance notes on strength assessment of membrane-type LNG containment systems under sloshing loads. Technical report, American Bureau of Shipping, 2006. [27] Joe F Thompson, Bharat K Soni, and Nigel P Weatherill, editors. Handbook of Grid Generation. CRC Press, 1999.