Professional Documents
Culture Documents
Engineering
OMAE 2011
June 19-24, 2011, Rotterdam, The Netherlands
OMAE2011-49215
ABSTRACT INTRODUCTION
In modelling incompressible flows using the Smoothed Par- Sloshing is a generic term for the free surface motion of a
ticle Hydrodynamics method (SPH), an equation of state with liquid that occurs in a partially filled container subjected to ex-
a large sound speed is typically used. This weakly compress- ternal excitation. It occurs as the fluid attempts to reach a state of
ible approach (WCSPH), results in a stiff set of equations with equilibrium due to the effective instantaneous acceleration (grav-
a noisy pressure field and stability issues at high Reynolds num- itational, rotational, etc) experienced by the fluid. Understanding
ber. As a remedy, an incompressible SPH technique was intro- sloshing is important in order to avoid structural damage caused
duced [1] (ISPH), which uses a pressure projection technique to by pressure fluctuations near the free surface, acoustic effects due
model incompressibility. In this paper, the incompressible and to the high velocity impact and the impacting loads exerted by the
weakly compressible forms of the SPH method are employed to fluid on the container. Sloshing phenomena occurs in various en-
study sloshing flow. Both methods are compared with experi- gineering applications such as fuel sloshing in launch vehicles,
mental data. The results show the incompressible SPH method Liquified Natural Gas (LNG) sloshing in cargo containers and
provides more accurate pressure fields and free-surface profiles water sloshing in reservoirs [2].
to experiment. In the past five decades, sloshing has been investigated
by many researchers using various techniques. Initial studies
were based on mechanical models of the phenomenon by ad-
justing terms in the harmonic equation of motion [3, 4] when
time–efficient and reasonably accurate results were deemed suf-
∗ Address all correspondence to this author. ficient [5]. Other researchers solved the potential flow problem
1 Copyright
c 2011 by ASME
with a very complicated treatment of free–surface boundary con- tion technique [1]. In the first step, the velocity is integrated
ditions [6]. Although the method is very accurate for specific in time without enforcing incompressibility. Incompressibility is
applications, it cannot handle breaking waves and tanks with baf- then achieved by projecting the intermediate velocity field onto a
fles. Others solved the non–linear shallow water equations [7] by divergent-free space by solving a pressure Poisson equation [1].
means of either Glimm’s method [8] or a gas analogy formula- This approach will be referred to as incompressible SPH (ISPH).
tion [9]. While the WCSPH scheme has been successfully imple-
The most common method of studying sloshing physics is mented at low Reynolds number, the reflection of sound waves
via experimental techniques. Abramson [10] evaluated the re- off boundaries at high Reynolds numbers leads to severe insta-
sults from model tests with tanks of various shapes. The effect bilities in the scheme [30]. In addition the stiff equation of state
of viscosity, tank shape and various damping devices is outlined. can result in large pressure fluctuations which also affects sta-
These studies were mainly related to liquid fuel tanks on space bility. These issues along with the requirement of a very small
vehicles. In the 19700 s and early 19800 s, the sloshing problem time–step in WCSPH due to needing to predict sound waves can
became an important issue in the design of the liquefied natu- be alleviated by using ISPH. However, the solution of the ellip-
ral gas carriers. Several experimental and analytical approaches tic pressure Poisson equation (PPE) increases the computational
were considered during this time. A major set of experiments costs at each time-step. Cummins and Rudman [1] showed the
were performed by Bass et al., [11] during 1980s on the vari- computational costs of solving the PPE can be reduced by use of
ous parameters effecting sloshing. Results indicated that impact a multi-grid technique.
pressure is constant over the range of Reynolds number investi-
In this paper, the WSPH and ISPH techniques are outlined
gated. Thus for large amplitude sloshing, viscous forces are of
including descriptions of free–surface and solid wall bound-
secondary importance. Liquid compressibility was not observed
ary conditions and the SPH–specific multi–grid implementation.
to have a significant effect on the measured impact pressure. Ex-
The accuracy, stability and robustness of WCSPH and ISPH are
periments conducted at reduced ambient pressure inside the tank
studied in sloshing flow simulations. Here the divergence–free
(also known as ullage pressure) showed significant increase in
ISPH is used with corrected density calculation. This has re-
impact pressures. It was concluded that ullage pressure has little
solved the stability issue of divergence–free ISPH [31] while
effect on impact pressures until low pressures near to the vapor
utilising the accuracy of the technique compared to the density–
pressure of the liquid medium were reached.
invariant ISPH approach.The results are compared with exper-
As well as the experimental techniques, numerical simu-
imental data for both global features (free–surface profile) and
lations are also widely adopted for studying highly non–linear
local features (time variation of impact pressure on the bound-
sloshing problems. Frandsen [12] used the Finite Difference
aries). An assessment of their relative merits is also provided.
method for solving the non–linear potential flow in a 2D tank.
Celebi and Akyildiz [13] used the finite difference method along
with Volume of Fluid technique (VOF) for tracking the free–
surface. Sames et al. [14], applied a commercial VOF technique
to rectangular and cylindrical tanks. THE SPH METHOD
In contrast to mesh–based methods, the Smoothed Parti- The SPH method uses smoothing kernels to express a
cle Hydrodynamics (SPH) method is a meshless, purely La- smooth function in terms of its values at a set of disordered
grangian technique which was originally developed in 1977 by points. The kernel function or weighting function, specifies the
Lucy [15] and Monaghan and Gingold [16, 17]. It has subse- contribution of a typical field variable, A(r), at position, r, in
quently been successfully employed in a wide range of problems, space. The kernel estimate of A(r) is defined as [32, 33]
e.g. astrophysics [18–20], fluid mechanics [21, 22], solid me-
chanics [23–25], fluid–structure interaction [26–28] and many Z
more (see [29] for a recent review). A(r) = A(r0 )W (r − r0 , h)dr0 (1)
In SPH, the ”particles” are moving nodes that are advected V
with the local velocity and carry field variables such as pressure
and density. As the fields are only defined at the set of dis-
crete points, smoothing (interpolation) kernels are used to define where V represents the solution space and the smoothing length
a continuous field and to ensure differentiability. Incompress- h represents the effective width of the kernel W . The kernel has
ibility is typically satisfied in two ways. In the most common the following properties
approach, termed weakly compressible SPH (WCSPH), the fluid
is assumed compressible with a large sound speed (such that the
Mach number M ≈ 0.1 and the density of the fluid varies less Z
than 1%). An alternative approach uses a fractional-step projec- W (r − r0 , h)dr0 = 1, lim W (r − r0 , h) = δ (r − r0 ). (2)
V h→0
2 Copyright
c 2011 by ASME
If A(r0 ) is known only at a discrete set of N points r1 , r2 , ..., rN , where s = |r|/h and in two-dimensional problems the normaliza-
then we approximate A(r0 ) as, 7
tion factor takes the value W0 = 64πh 2.
N
A(r0 ) = ∑ δ (r0 − r j )A(r j )(dV ) j (3) Weakly Compressible SPH (WCSPH)
j=1 In the WCSPH the pressure is calculated through an equa-
tion of state. In this work, the most common form of equation of
where the index j denotes the particle label and particle j has a state for water is used [33, 37]
mass m j and density ρ j at position r j . The differential volume el-
ement around the point r j is called (dV ) j . By inserting equation ρ0 ci 2
ρi
(3) into equation (1), we obtain, Pi = ( )7 − 1 . (10)
7 ρ0
N Z
A(r) = ∑ δ (r − r j )A(r j )(dV ) jW (r − r j , h) dr j (4) where ρ0 is the reference density. In order for density fluctua-
j=1 tions to be around 1%, the Mach number should be M ≈ 0.1. To
ensure this condition, the speed of sound is set to c = 10V , where
and upon integrating, we have: V is the maximum expected velocity inside the flow. The density
ρi is evolved at each time–step using the continuity equation
N
A(r) = ∑ A(r j )W (r − r j , h)(dV ) j (5) dρi mj
= ρi ∑ (ui − u j ) · ∇Wi j . (11)
j=1
dt ρj
3 Copyright
c 2011 by ASME
time–step using
1
ui n+1 = ui n + 4tFi n+ 2
1
xi n+1 = xi n+ 2 + 4t
2 ui
n+1 (14)
1
= ρi n+ 2 + 4t
n+1 n+1 .
ρi 2 Di
4 Copyright
c 2011 by ASME
In contrast to WCSPH, where the particle density is evaluated
using Eq.11,in ISPH the density of each particle is calculated
from the SPH summation of density
ρi = ∑ m jWi j . (21)
j
Wi j
ρi = ∑ m jW̃i j , where W̃i j = m . (22)
j ∑k ρkk Wik (a)
ui n + ui n+1
ri n+1 = ri n + ∆t . (24)
2 (b)
Following [41], the Laplacian pressure term in the PPE (Eq. 18) FIGURE 2. The ISPH results for sloshing flow, without kernel correc-
is approximated as tion (a) and with kernel correction (b). Particle spacing is l0 = 10−2 m.
1 8 Pi j ri j · ∇Wi j
∇· ∇P = ∑ m j . (25)
ρ (ρi + ρ j )2 |ri j |2
Boundary Conditions in ISPH
Discretisation of Eq. 25 leads to a linear system of equations for
pressure with a symmetric and positive definite matrix of coef- In ISPH, one layer of fixed particles are placed on the wall.
ficients. In this work, a multi–grid technique with Gauss–Seidel These particles contribute to the construction of the PPE matrix.
method is employed to solve the corresponding Poisson equation. The pressure from them allows the Neumann boundary condition
In ISPH the time–step is governed by the following condi- to be satisfied at the wall. In order to avoid accuracy degrada-
tion tion due to truncation of the kernel near the wall, several layers
of boundary particles are placed outside the wall. These parti-
s ! cles do not contribute to the solution of the PPE. Instead their
h pressure is calculated from the wall particles by assuming a zero
4t ≤ 0.25min (26)
|fi | pressure gradient in the normal direction to the wall. Figure 3
shows a schematic of the particle positions. The fluid particles,
where fi is force per unit mass and equal to magnitude of particle wall particles and external boundary particles are illustrated by
acceleration [30]. circles, stars and rectangles, respectively.
5 Copyright
c 2011 by ASME
used. Finally one last Gauss–Seidel iteration is performed at the
particles to calculate the pressure and the residual. The proce-
dure represents a one V–cycle.
SLOSHING FLOW
To compare the capability of WCSPH and ISPH methods
and study the incompressibility effects in sloshing flow, an exper-
iment was carried out to measure the impact pressure and capture
the flow pattern. Here, the experimental setup is introduced and
then the results are compared with numerical results from ISPH
FIGURE 3. Schematic of the boundary particles in the ISPH simu- and WCSPH.
lations. The wall particles (stars) contribute to the PPE but the outer
boundary particles (rectangles) do not.
Experimental Setup
The sloshing experiments are carried out in a rectangular
Free–surface tank partially filled with water at atmospheric pressure and room
Since the number of neighboring particles are smaller for temperature. The tank dimensions are 130 cm × 90 cm × 10 cm,
particles on the free surface the value of ∇ · r decreases for the corresponding to the length, height and width of the tank. Three
free–surface particles [30]. In 2D simulations the value of ∇ · r of the walls are made of transparent plexiglass for optical ac-
is equal to 2 inside the fluid domain and is significantly below 2 cess, while the top and one of the side walls are made of alu-
for the surface particles. Therefore a criterion of minum and host pressure sensors. On the side wall the sensors
are spaced 15 mm apart in the vertical direction and 25 mm apart
∇ · r = 1.5 (27) in the horizontal direction. There are 59 vertical positions avail-
able in the central column and 31 on each of the side columns.
On the top of the tank there are 3 lines of 22 holes each spaced
is used to recognize the particles on the free–surface where the
15 mm apart and the lines are 25 mm apart. The time variation
divergence of the particle position is calculated using
of the impact pressure at the position of initial water level on the
left wall is recorded at the frequency of 40 kHz during the exper-
mj
∇·r = ∑ ri j · ∇Wi j . (28) iment. The tank is mounted on top of a motion system platform.
ρj This device is capable of performing motions with six degrees
of freedom [40]. Figure 4 shows the experimental setup. For
A Dirichlet boundary condition of p = 0 is assigned to these par- the experiment considered here, the tank was filled 20% with
ticles. water, corresponding to H = 0.18 m. The motion is sinusoidal
with the amplitude of A = 0.1 m at the resonant frequency of
f = 0.496 Hz, which is the frequency of a shallow water wave
Multi–grid solver
with wavelength of λ = 1.3 m and depth of H = 0.18 m. Figure 5
In this work, a multi–grid scheme [45] is used for efficient
shows different snap shots of impact at three different times.
solution of the pressure Poisson equation. The multi–grid is im-
plemented in a similar way to [1]. First an equivalent grid resolu-
tion is defined at the particle level. This is achieved by extending SPH Comparisons With Experiment
the domain size to ensure 2n grid cells fits in the domain with a WCSPH and ISPH simulations were run to compare both
grid cell size of 2h × 2h. Initially several Gauss–Seidel iterations global and local features of the sloshing phenomenon with exper-
(typically ∼ 3 − 5) are performed at the particles. Then the error imental data. For the SPH calculations, the particles were placed
is mapped to the finest grid level which has a cell size of 2h × 2h. on a square grid with initial spacing of l0 = 5 × 10−3 m and the
This corresponds to the size of the search grid used in assem- smoothing length was set to h = 1.5l0 . The Reynolds number
bling the list of nearest neighbour particles. The mapping from is calculated using the depth of water as the typical length and
the particles to the grid is done by using the kernel employed in is Re = (2g)1/2 H 3/2 /ν, where ν = 10−6 m2 /s is the kinematic
the simulations (Wendland kernel, see Eq. 9). Similar to [1], a viscosity of water. Using these dimensions the Reynolds number
standard defect correction multi–grid technique is used at coarse is Re ≈ 338000. Figure 6 shows the shape of free–surface pro-
levels and bilinear interpolation is used to prolong the correction files for ISPH and WCSPH. The free–surface profile has been
to the finer level. To project back the correction from the finest digitized from the experimental data and superimposed on to the
grid level to particles level, the bilinear interpolation scheme is numerical results for better comparison. The particle positions
6 Copyright
c 2011 by ASME
FIGURE 5. The snapshots of the sloshing at a random impact.
WCSPH, where the decay time is in fact hidden due to the highly
noisy pressure field.
Efficiency comparisons
Figure 8 compares the CPU time for 2 s of sloshing simu-
lation for ISPH and WCSPH at three different resolutions, cor-
responding to initial particle spacing of 10−2 m, 5 × 10−3 m and
3 × 10−3 m. The CPU wall-clock time was measured on an In-
tel Core2 Duo PC with 4 GiB memory. Despite an increase in
the time–step ISPH is ≈ 3 − 4 times slower than WCSPH at
the finest resolution. Further work is needed to accelerate the
PPE solver such as investigating pre–conditioned linear solvers
to make ISPH method more efficient.
CONCLUSION
FIGURE 4. The tank installed on top of the motion platform.
In this paper, a two-dimensional sloshing flow has been sim-
ulated with WCSPH and ISPH and the results are compared
against experimental data. It has been observed that ISPH gives
more accurate, stable and reliable results in comparison to WC-
are shown coloured by pressure, to compare the pressure field SPH at high Re flows. This is credited to a more accurate pres-
obtained from both techniques. It can be seen in the Fig. 6 that sure calculation in ISPH which is derived from the an incom-
the pressure field from WCSPH is highly noisy compared to the pressible flow constraint rather than an equation of state. Despite
very smooth pressure gradient in the ISPH. Using stiff equation the larger time–steps in ISPH compared to WCSPH, the compu-
of state in the WCSPH leads to development of negative pressure. tational cost is still greater in ISPH. This cost has been reduced
This high negative pressure in WCSPH causes an unrealistic ten- by using a multi–grid scheme to solve the pressure Poisson equa-
sile force between particles which can lead to creation of clusters tion but further work needs to be performed in this area.
of particles in some regions of the field. This effect is not seen
in the ISPH. On the other hand the particles in ISPH results are
more scattered. REFERENCES
The time variation of the impact pressure at the initial water [1] Cummins, S. J., and Rudman, M., 1999. “An SPH projec-
level on the left wall is compared against experimental data. The tion method”. J. Comput. Phys., 152, pp. 584–607.
results are shown in Fig. 7. The ISPH results are in good agree- [2] Wu, C. H., and Chen, B. F., 2009. “Sloshing waves and
ment with the experimental data during the decay period while resonance modes of fluid in a 3d tank by a time-independent
the WCSPH results are highly oscillatory during the decay. The finite difference method”. Ocean Eng., 36, pp. 500–510.
decay time is predicted more accurately in ISPH compared to [3] Graham, E. W., and Rodriguez, A. M., 1952. “The charac-
7 Copyright
c 2011 by ASME
FIGURE 6. The snapshots of the SPH results for sloshing flow for ISPH (left column) and WCSPH (right column). The black dots represent the
experimental free–surface.
teristics of fuel motion which affects airplane dynamics”. stabiliser tanks”. RINA Transactions and Annual Report,
Journal of Applied Mechanics, 19, pp. 381–388. 19, pp. 31–45.
[4] Lewison, G. R. G., 1976. “Optimum design of passive roll [5] Aliabadi, S., Johnson, A., and Abedi, J., 2003. “Compari-
8 Copyright
c 2011 by ASME
WCSPH .vs. Experiment
[7] Lee, T., Zhou, Z., and Cao, Y., 2002. “Numerical simu-
lations of hydraulic jumps in water sloshing and water im-
pacting”. ASME. J. Fluids Eng., 124, pp. 215–226.
[8] Stoker, J. J., 1957. Water waves, Pure and Applied Mathe-
matics, Vol. IV. Interscience, New York.
[9] Verhagen, J. H. G., and Wijngaarden, L., 1965. “Non–
linear oscillations of fluid in a container”. Journal on Fluid
Mechanics, 22, pp. 737–751.
[10] Abramson, H. N., 1966. The dynamic behavior of liquids
in moving containers, with applications to space vehicle
technology. Technical Report SP-106, National Aeronau-
tics and Space Administration.
[11] Bass, R. L., Bowles, E. B., Trundell, R. W., Navickas, J.,
Peck, J. C., Yoshimura, N., Endo, S., and Pots, B. F. M.,
ISPH .vs. Experiment 1985. “Modeling criteria for scaled lng sloshing experi-
ments”. Transactions of the American Society of Mechani-
FIGURE 7. Comparison of time history of impact pressure of WC- cal Engineers, 107, pp. 272–280.
SPH and ISPH with experimental data. [12] Frandsen, J. B., 2004. “Sloshing motions in excited tanks”.
J. Comput. Phys., 196, pp. 53–87.
[13] Celebi, M. S., and Akyildiz, H., 2002. “Nonlinear mod-
elling of liquid sloshing in a moving rectangular tanks”.
son of finite element and pendulum models for simulation Ocean Eng., 29, pp. 1527–1553.
of sloshing”. Computational Fluids, 32, pp. 535–545. [14] Sames, P. C., Marcouly, D., and Schellin, T. E., 2002.
[6] Faltinsen, O. M., Rognebakke, O. F., and Timokha, A. N., “Sloshing in rectangular and cylindrical tanks”. Journal
2005. “Resonant three-dimensional nonlinear sloshing in a on Ship Research, 46, pp. 186–200.
square-base basin. part 2: Effect of higher modes”. Journal [15] Lucy, L. B., 1977. “A numerical approach to the testing of
on Fluid Mechanics, 523, pp. 199–218. the fission hypothesis”. Astron. J., 82, pp. 1013–1020.
9 Copyright
c 2011 by ASME
[16] Gingold, R. A., and Monaghan, J. J., 1977. “Smoothed par- projection method and a new approach”. Journal of Com-
ticle hydrodnamics: Theory and application to nonspherical putational Physics, 228(18), pp. 6703–6725.
stars”. Mon. Not. Roy. Astr. Soc., 181, pp. 375–389. [32] Monaghan, J. J., 1992. “Smoothed particle hydrodynam-
[17] Monaghan, J. J., and Gingold, R. A., 1983. “Shock simu- ics”. Annul. Rev. Astron. Astrophys., 30, pp. 543–574.
lation by the particle method SPH”. J. Comput. Phys., 52, [33] Monaghan, J. J., 1994. “Simulating free surface flows with
pp. 374–389. SPH”. J. Comput. Phys., 110, pp. 399–406.
[18] Benz, W., 1988. “Applications of smoothed particle hy- [34] Cleary, P. W., 1998. “Modelling confined multi-material
drodynamics (SPH) to astrophysical problems”. Computer heat and mass flows using sph”. Appl. Math. Modelling,
Physics Communications, 48, pp. 130–139. 22, pp. 981–993.
[19] Monaghan, J. J., 1990. “Modeling the universe”. In Pro- [35] Wendland, H., 1995. “Piecewise polynomial, positive def-
ceedings of the Astronomical Society of Australia, 18, inite and compactly supported radial functionsof minimal
pp. 233–237. degree”. Adv. Comput. Math., 4, pp. 389–396.
[20] Monaghan, J. J., and Lattanzio, J. C., 1991. “A simula- [36] Crespo, A. J. C., 2008. “Application of the smoothed par-
tion of the collapse and fragmentation of cooling molecular ticle hydrodynamics model sphysics to free–surface hydro-
clouds”. Astrophysical Journal, 375, pp. 177–189. dynamics”. Ph.d thesis, Department of Applied Physics,
[21] Monaghan, J. J., Cas, R. F., Kos, A., and Hallworth, M., University of Vigo.
1999. “Gravity currents descending a ramp in a stratified [37] Batchelor, G. K., 1974. An Introduction to Fluid dynamics.
tank”. Journal Fluid Mechanics, 379, pp. 39–70. Cambridge University Press.
[22] Swegle, J. W., Hicks, S. W., and Attaway, S., 1995. [38] Molteni, D., and Colagrossi, A., 2009. “A simple procedure
“moothed particle hydrodynamics stability analysis”. J. to improve the pressure evaluation in hydrodynamic context
Comp. Phys., 116, pp. 123–134. using the SPH”. Comput. Phys. Comm., 180, pp. 861–872.
[23] Libersky, L. D., and Petscheck, A. G., 1993. “High strain [39] Morris, J. P., Fox, P. J., and Zhu, Y., 1997. “Modeling
lagrangian hydrodynamics- a three–dimensional SPH code low reynolds number incompressible flows using SPH”. J.
for dynamic material response”. J. Comput. Phys., 109, Comput. Phys., 136, pp. 214–226.
pp. 67–75. [40] Rafiee, A., Pistani, F., and Thiagarajan, K. P., 2010. “Study
[24] Benz, W., and Asphaug, E., 1995. “Simulations of brittle of liquid sloshing: Numerical and experimental approach”.
solids using smoothed particle hydrodynamics”. Computa- Computational Mechanics, In-press.
tional Physics Communications, 87, pp. 253–265. [41] Shao, S., and Lo, E. Y. M., 2003. “Incompressible SPH
[25] Bonet, J., and Kulasegaram, S., 2000. “Corrections and method for simulating newtonian and non–newtonian flows
stabilization of smooth particle hydrodynamics methods with a free surface”. Advances in Water Resources, 26,
with applications in metal forming simulations”. Interna- pp. 787–800.
tional Journal for Numerical Methods in Engineering, 47, [42] Koshizuka, A., Nobe, A., and Oka, Y., 1998. “Numeri-
pp. 1189–1214. cal analysis of breaking waves using the moving particle
[26] Rabczuk, T., Garcie, R., Song, J. H., and Belytschko, T., semi–implicit method”. Int. J. Numer. Methods Fluids, 26,
2009. “Immersed particle method for fluid.structure inter- pp. 751–769.
action”. Int. J. Numer. Meth. Engng, 81, pp. 48–71. [43] Chorin, A. J., 1968. “Numerical solution of the navier–
[27] Maurel, B., Potapov, S., Fabis, J., and Combescure, A., stokes equations”. J. Math. Comp., 22, pp. 745–762.
2009. “Full SPH fluid–shell interaction for leakage sim- [44] Bonet, J., and Lok, T. S. L., 1999. “Variational and momen-
ulation”. Int. J. Numer. Meth. Engng, 80, pp. 210–234. tum preservation aspects of smooth particle hydrodynamic
[28] Rafiee, A., and Thiagarajan, K. P., 2009. “An SPH projec- formulations”. Comput. Methods Appl. Mech. Engrg., 180,
tion method for simulating fluid–hypoelastic structure in- pp. 97–115.
teraction”. Computer Methods in Applied Mechanics and [45] Hackbusch, W., 1985. Multi–Grid Methods and Applica-
Engineering, 198, pp. 2785–2795. tions. Springer–Verlag Berlin.
[29] Monaghan, J. J., 2005. “Smoothed particle hydrodynam-
ics”. Rep. Progr. Phys., 68, pp. 1703–1759.
[30] Lee, E. S., Moulinec, C., Xu, R., Violeau, D., Laurence, D.,
and Stansby, P., 2008. “Comparisons of weakly compress-
ible and truly incompressible algorithms for the SPH mesh
free particle method”. J. Comput. Phys., 227, pp. 8417–
8436.
[31] Xu, R., Stansby, P., and Laurence, D., 2009. “Accuracy
and stability in incompressible SPH (ISPH) based on the
10 Copyright
c 2011 by ASME