Sunteți pe pagina 1din 18

Comparative Analysis of the SPH and ISPH

Methods
K.E. Afanasiev, R.S. Makarchuk, and A.Yu. Popov
Kemerovo State University, Krasnaya st. 6, 650043 Kemerovo, Russia
{afa,mak,a popov}@kemsu.ru
Abstract. Free surface problems in uid mechanics are of a great ap-
plied importance, and thats why they rouse many researchers to ex-
citement. Numerical simulations of such problems, using the so-called
meshless methods, become more and more popular today. One of the
subsets of these methods is the class of meshless particle methods, which
dont use any mesh during the whole numerical simulation process, and
therefore allow solving the problems with large deformations and with
failure of problem domain connectivity. These reasons cause great pop-
ularity of these methods in the sphere of numerical simulation of free
surface problems.
One of the meshless particle methods is the Smoothed Particle Hydro-
dynamics [1]. Numerous computations, mainly of free surface problems,
carried out by dierent scientists, using the mentioned method, proved
its doubtless eciency in receiving the high-quality kinematic images of
ideal as well as viscous uid ows.
However, it has a considerable drawback - it doesnt allow receiving
satisfactory images of pressure distribution.
Recently, with the purpose to deliver such drawbacks, the ISPH (In-
compressible Smoothed Particle Hydrodynamics) method has been de-
veloped [2,3], which is used for simulation the incompressible uid ows,
as its name shows. It uses a split step scheme for integration of basic
equations of uid dynamics.
Comparative analysis of computational results, obtained by the meth-
ods, mentioned above, shows, that the ISPH gives low-grade kinematic
images of uid ows in comparison with the classical SPH, however, it
allows obtaining much better images of pressure distribution, and it will
give an opportunity to compute the hydrodynamics loads in future.
1 Introduction
At present more currency in the eld of numerical simulation of free surface ows
is gained by meshless methods. Among them the subclass of particle methods
is pointed out. These methods dont require using a mesh neither at the stage
of shape functions constructing, nor at the stage of integrating equations of
motion. Their major idea lies in discretization of the computation area by a set
of Lagrangian particles, which are able to move freely within the constraints,
received by means of basic equations of continuum dynamics. Shape functions
E. Krause et al. (Eds.): Comp. Science & High Perf. Computing III, NNFM 101, pp. 206223, 2008.
springerlink.com c Springer-Verlag Berlin Heidelberg 2008
Comparative Analysis of the SPH and ISPH Methods 207
Table 1. Distinctions in the SPH and ISPH algorithms
SPH ISPH
Integration scheme Explicit Implicit
Pressure Equation of state Poisson equation
Courant condition Sound speed Maximum speed of particles
Density Variable Constant
Articial viscosity Applied Not applied
Solid boundary conditions Lennard-Jones potential Morris virtual particles
in this approach are constructed for each time step using a dierent set of nodes
(particles). The meshless nature of these methods and their easy realization and
application have caused their great popularity in the eld of numerical simulation
of free surface ows.
The most widespread for the present moment particle methods are the
smoothed particle hydrodynamics method (SPH) [1,4,5,6,7] and the moving par-
ticle semi-implicit method (MPS) [8,9]. Besides, there are many modications of
the SPH, i.e. the RKPM [10] and MLSPH [11], intended for optimization of its
approximation characteristics.
This work considers the original SPH and one of its modications the In-
compressible Smoothed Particle Hydrodynamics (ISPH) [2,3], which allows, con-
trary to the original SPH method, exactly complying with the incompressibility
condition. The most signicant dierences of the two methods are presented in
Table 1.
2 Governing Equations
The basic equations of uid dynamics, including the Navier-Stokes equations
and the equation of continuity, in case of the Newtonian viscous uid, are of the
following form:
dv
a
dt
= F
a

p
x
a
+

x
b
(T
ab
); (1)
d
dt
=
v
a
x
a
, (2)
where a, b = 1, 2, 3 numerical indices of coordinates, v
a
components of the
velocity vector, F
a
components of the vector of volumetric forces density,
ab

Kronecker symbols, p and pressure and density of the uid, correspondingly,
coecient of dynamic viscosity, while the stress components are calculated
by the formula:
T
ab
=
v
a
x
b
+
v
b
x
a

2
3
divv
ab
(3)
It should be noted, that the original SPH method is applied for simulation of
compressible uid ows, and a certain equation of state is used for closure the
208 K.E. Afanasiev, R.S. Makarchuk, and A.Yu. Popov
equations (1)-(2). For computation of incompressible uid ows, most often the
barotropic equation of state in the Theta form is applied [4]. By selection of the
coecient of volume expansion in the equation of state we can obtain the eect
of incompressible uid. However, certain variations of particles densities occur.
The ISPH method uses the split step scheme for time integration of equations
of motion, what allows calculating the particles pressure values by solving the
Poisson equation without recourse to the equation of state. This approach allows
also to neglect the articial viscosity terms in Navier-Stokes equations, contrary
to the original SPH, because of using the model of incompressible uid.
As a result of the fact, that the ISPH uses the model of incompressible uid,
the equations (2), (3) are represented by the following formulas:
v
a
x
a
= 0; (4)
T
ab
=
v
a
x
b
+
v
b
x
a
. (5)
Using them we can represent the equations of motion (1) in the following form:
dv
a
dt
= F
a

p
x
a
+

x
b
(
v
a
x
b
) (6)
3 Approximation of Functions
The rst stage in the process of constructing approximation formulas for func-
tions, occurring in equations of uid dynamics, is exact presentation of these
functions as the integral:
f(r) =

f(r

)(r r

)dr

, (7)
where the Dirac delta-function.
Then follows substitution of the -function with a certain classic compactly
supported function, what allows receiving the integral formula of the function
approximation on the bounded domain:
f(r) =
_

f(r

)W(r r

, h)dr

, (8)
with the weight function W, which is often called the kernel function. The value
h is a size of support domain of the function W and is called a smoothing length.
At the following stage the problem domain is discretized by a nite number
of Lagrangian particles, and the integral (8) is substituted with the quadrature
[4,5,6,7]:
f
s
(r
i
) =
n

j=1
f(r
j
)
m
j

j
W(r
i
r
j
, h
i
), (9)
Comparative Analysis of the SPH and ISPH Methods 209
Fig. 1. Interaction of particles
where n number of particles, determined as nearest, in the radius h
i
, neighbours
of the i-th particle. Two particles i and j are called neighbouring or interacting
particles, if the distance between their centres does not exceed h
i
+h
j
. r
i
, m
i
,
i
- radius-vector, mass and density of the i-th particle, correspondingly. As the
weight W usually polynomial splines are applied.
On Fig. 1 is shown, that particles 1 and 2 are neighboring to particle i. From
(9) follows, that the gradient of required function is represented by the following
expression:
f
s
(r
i
) =
n

j=1
m
j

j
f
j
W(r
i
r
j
, h) (10)
Using the formula (9), for density approximation we obtain:

i
=
n

j=1
m
j
W(r
i
r
j
, h
i
) (11)
It should be noted, that contrary to the original SPH, where the density value in
particles is most often determined by solving a discrete analogue of the equation
(2), in the ISPH method the density is calculated just by the formula (11).
4 Kernel Function
For numerical simulation, using the considered methods, one can apply various
kernel functions, from the Gaussian function up to splines of dierent orders.
Beside the already known kernel functions the others can be developed, but, at
that, has to follow the minimum requirements specied for them:
W( r, h) = 0, r > h;

W( r, h)dr = 1;
lim
h0
W( r, h) = ( r),
210 K.E. Afanasiev, R.S. Makarchuk, and A.Yu. Popov
Fig. 2. Kernel function (left) and rst derivative of kernel function (right)
where r = ||r

r||. On Fig. 2 shapes of kernel function and its rst derivative are
submitted. Beside the above mentioned requirements, some additional conditions
can be imposed on the kernel function to provide best stability of the method and
the higher degree of approximation of the functions, characterizing ows. Such
additional conditions and ways of constructing the kernel functions, subsequent
upon them, have led, for instance, to development of the RKPM method. For
the problems, considered in this work, the cubic spline from [3,12] is applied:
W( r, h) =
15
7h
2

2/3 q
2
+ q
3
/2, 0 q 1;
(2 q)
3
/6, 1 < q 2;
0, q > 2,
(12)
where q =
r
h
.
5 Articial Viscosity
For additional stability of the original SPH the additional term, called articial
viscosity, is added to the right part of the Navier-Stokes equations of motion
[4,7,13]:

ij
=

ij
(
ij
c)
1
/
2
(
i
+
j
)
,

d=1
_
v
d
i
v
d
j
_ _
x
d
i
x
d
j
_
< 0;
0, otherwise,
(13)
where dimension of the problem domain, and
ij
are computed by the
formula:

ij
=

d=1
h(v
d
i
v
d
j
)(x
d
i
x
d
j
)
(x
d
i
x
d
j
) +
2
h
2
; (14)
c =
1
2
(c
i
+ c
j
), h =
1
2
(h
i
+ h
j
), = = 1, = 0.1 (15)
Since the ISPH method was developed for incompressible uids it doesnt
need any articial viscosity term to be added to the approximated equations of
motion.
Comparative Analysis of the SPH and ISPH Methods 211
6 Pressure and Viscosity
For approximation of the Poisson equation one can apply SPH formulas for the
gradient of function and the vector eld divergence sequentially. However, on
account of the fact, that approximation of the second-order derivative, obtained
in this case, is too sensitive to the particles disorder, approximation of the rst
derivative in terms of the SPH and its nite dierence analogue are usually
applied together [3]:

_
1

p
_
i
=
n

j=1
m
j
8
(
i
+
j
)
2
(p
i
p
j
) (r
i
r
j
)
i
W(r
i
r
j
, h
i
)
r
i
r
j

2
(16)
We can also obtain other formulas for approximation of the Poisson equation,
particularly, the following springs out of the work [14]:

_
1

p
_
i
=
n

j=1
m
j
2

j
(p
i
p
j
) (r
i
r
j
)
i
W(r
i
r
j
, h
i
)
r
i
r
j

2
(17)
The calculations of the model problems have been executed using as the formula
(16), so (17), however any essential dierences in the obtained results have not
been discovered.
The formula for approximation of viscous forces for the ISPH is obtained in
a similar way and takes on the form:
_

2
v
_
i
=
n

j=1
m
j
4
(
i
+
j
)
2

(
i
+
j
) (r
i
r
j
)
i
W(r
i
r
j
, h
i
)
r
i
r
j

2
( v
i
v
j
)
(18)
The pressure gradient term in the equation (1) can also be approximated in
dierent ways. One can directly apply the SPH approximation formulas to this
term, but it was found by researchers, that the most stable approximation can
be obtained using the formula:
p

=
_
p

_
+
p

2
(19)
Now one can apply the SPH approximation formulas to the right-hand side of
(19) and obtain the following approximation of pressure gradient:
_
p

_
i
=
n

j=1
m
j
_
p
i

2
i
+
p
j

2
j
_

i
W(r
i
r
j
, h
i
) (20)
This approximation of pressure gradient, because of its symmetrical form, pro-
vides the total momentum conservation law fullment.
212 K.E. Afanasiev, R.S. Makarchuk, and A.Yu. Popov
Based on considered approximations it is possible to rewrite equations (1)-(2)
in the following form for the original SPH:
dv
a
i
dt
= F
i

n

j=1
m
j
_
p
i

2
i
+
p
j

2
j
+
ij
_
W
a
(r
i
r
j
, h
j
)+
+
n

j=1
m
j

d=1
_

i
T
i
da

2
i
+

j
T
j
da

2
j
_
W
d
(r
i
r
j
, h
j
);
(21)
d
i
dt
=

d=1
n

j=1
m
j
(v
d
i
v
d
j
)W
d
(r
i
r
j
, h
j
), (22)
where p
i
,
i
,
i
pressure, density, and coecient of dynamic viscosity accord-
ingly for particle i. Normal and tangent components of viscous stress tensor T
i
are dened by following expressions:
T
i
aa
=
n

j=1
m
j

j
[2 (v
a
j
v
a
i
)W
a
(r
i
r
j
, h
j
)
(v
b
j
v
b
i
)W
b
(r
i
r
j
, h
j
)];
(23)
T
i
ab
=
n

j=1
m
j

j
[(v
a
j
v
a
i
)W
b
(r
i
r
j
, h
j
)
(v
b
j
v
b
i
)W
a
(r
i
r
j
, h
j
)]
(24)
7 Model of Incompressibility in the SPH
Initially method SPH was applied only to modeling strongly compressed envi-
ronments. Incompressibility or weak compressibility is reached by a choice of the
suitable equation of a state to solve the pressure. The following equation, as a
rule, is applied to simulation of liquid dynamics [4]:
p = B
__

0
_

1
_
, (25)
where
0
- initial density of a uid, = 7 - adiabatic index. For problems with
failure of a weighty uid constant B gets out under the following expression:
B =
200
0
g(H y)

, (26)
where g gravity acceleration, H initial height of a uid column, y vertical
coordinate of the particles. In case of a choice of constant B under the formula
(26) density of particles during calculation dier from initial density
0
no more
than on 2-3 %, that allows to take a uid weakly compressed.
Comparative Analysis of the SPH and ISPH Methods 213
8 Time Integration
In the SPH method for time integration the predictor-corrector scheme is used.
Predictor:

n
i
=
n1/2
i
+ (t/2)(d
n1
i
/dt);
v
n
i
= v
n1/2
i
+ (t/2)(dv
n1
i
/dt).
(27)
Corrector:

n+1/2
i
=
n1/2
i
+ t(d
n
i
/dt);
v
n+1/2
i
= v
n1/2
i
+ t(dv
n
i
/dt);
x
n+1
i
= x
n
i
+ t(v
n+1/2
i
/dt).
(28)
For time integration of motion equations in the ISPH the split step scheme is
applied. Initially under consideration is only the eect of mass forces and viscous
forces upon the particles, what allows receiving preliminary values of densities,
velocities and coordinates of the particles. At this stage the pressure is not taken
into account, and in the formula (3), due to incompressibility of the uid, the
term with the velocity divergence disappears. The mentioned restrictions lead
to the equations:
v

=
_
g +

2
v
_
t;
v

= v
n
+ v

;
r

= r
n
+ v

t.
(29)
Here v

velocity change caused by mass forces and viscous forces, v

and r

preliminary values of the velocity vector and the radius-vector of the particles
centers, correspondingly, v
n
and r
n
values of the velocity vector and the radius-
vector of the particles centers on the n-th time step.
Using the obtained preliminary values of the physical characteristics, we de-
termine intermediate values of the particles densities

i
by the formula (11).
Variations of the obtained particles densities from their initial values are used
later to comply with the incompressibility condition, and the nal values of
the characteristics on the (n + 1)-th time step are determined by the following
formula:
v

=
1

p
n+1
t;
v
n+1
= v
n
+ v

;
r
n+1
= r
n
+
_
v
n
+ v
n+1
_
2
t.
(30)
214 K.E. Afanasiev, R.S. Makarchuk, and A.Yu. Popov
At this stage we have to solve the pressure Poisson equation, which is of the
following form [3]:

_
1

i
p
_
i
=

0

0
t
2
(31)
This equation can be nally reduced to the system of linear algebraic equations,
and the conditions on the free surface, represented here by the Dirichlet condi-
tion, are introduced to the matrix of coecients of the resulted system on the
same principle, as in the nite element method [15].
The time step gets out proceeding from Courant condition [5,3]:
t C
min
i
(h
i
)
max
i
(c
i
+v
i
)
(SPH), t C
h
max
i
(v
i
)
(ISPH), (32)
where c
i
- speed of sound for particle i, C (0; 1) - Courant constant. In calcu-
lations it was used C [0.25 ; 0.5].
9 Solid Boundary Conditions
In the original SPH method the most frequently used way of imposing conditions
at the solid boundaries is application of virtual particles, divided into two
types.
The rst type Monaghan virtual particles [4]. These particles are located
along the solid boundary in a single line, dont change their characteristics in
time, and eect on the uid particles by means of a certain interaction potential.
The most popular among researchers is the Lennard-Jones potential, though
selection of this is not imperative. In his work [6] Monaghan proposes a new
interaction potential, taking into account peculiarities of the SPH method.
The most frequently used potential in a method of the smoothed particles
Lennard-Jones potential:
U(r) =
D
r
_
_
r
0
r
_
12

_
r
0
r
_
6
_
, (33)
where D - depth of a potential hole, r
0
- distance on which the potential of
interaction addresses in a zero. If r < r
0
there are repulsive force, otherwise -
attractive force. The found potential is included in the momentum equation as
additional force which turns out as a result of application of SPH approximations:
U
a
i
=
n

j=1
D(x
a
i
x
a
j
)
_
r
0
r
_
12

_
r
0
r
_
6
r
2
(34)
For simplication the potential (34) is calculated only for repulsive forces, and
forces of an attraction at r > r
0
can be neglected.
The second type Morris virtual particles [5]. These particles are located
along the solid boundary in several lines. The number of the lines depends on
Comparative Analysis of the SPH and ISPH Methods 215
the smoothing length of particles of the uid. This allows solving one of the
main problems of the SPH method asymmetry of the kernel function near the
boundaries. The eect of the Morris particles on the uid particles diers from
the eect of Monaghan particles by the fact, that there is no need in using any
interaction potential. Instead of this, values of the characteristics in the Morris
particles are calculated on the basis of their values in particles of the uid.
In the ISPH method, for imposing solid boundary conditions the Morris vir-
tual particles are used. In the work [3] the approach is presented, which is applied
in solving problems by the MPS method [8,9]. For solving the pressure Poisson
equation, on the solid boundaries the Neumann condition is imposed, specify-
ing equality of pressures in the particles, located along the normal to the solid
boundary. In this work 3 lines of virtual particles are applied for numerical
simulation. The pressure Poisson equation is solved using only one line of vir-
tual particles and also only one line is used in the gradient formula (20). The
other two lines are necessary to keep particles densities near the solid boundary
the same as for inner ones.
10 Free Surface Conditions
For identication of particles, belonging to the free surface, one can apply dier-
ent ways. One of such ways is using the Dilts algorithm [11], based on the fact,
that each particle has its size, which is determined by the smoothing length in
the ISPH method. However most often, as in the MPS method, so in the ISPH,
the particles, belonging to the free surfaces, are identied by their densities. This
approach is based on the fact, that the density of particles, belonging to the free
surface, is less than the density of inner uid particles, because of the lack of
neighbours from one side of the boundary.
For solving the pressure Poisson equation, in the particles, belonging to the
free boundary, the Dirichlet condition is imposed: p = 0. Besides, this condition
is also applied in calculation of the pressure gradient, used in the formula (20). It
is done as follows. For each particle, belonging to the free surface, all its nearest
neighbours should be determined. It is evident, that the neighbours are only
inner particles of the uid, since there are no particles above the free surface.
To keep the symmetry of support domain of kernel function for the particles,
belonging to free surface, above the free surface we introduce additional ghost
particles. They are located symmetrically about the free surface to the inner
particles of the uid, and the pressure in them is converse. In this work the
approach, used in [3], is applied.
11 Nearest Neighbours Search
Since nearest neighbours search algorithm is carried out on each time step and
requires high computing expenses it is necessary to use eective algorithms of
search of neighbours. In our calculations we used grid algorithm.
216 K.E. Afanasiev, R.S. Makarchuk, and A.Yu. Popov
Fig. 3. Direct search
Fig. 4. Grid algorithm
Fig. 5. Acceleration of computations
Comparative Analysis of the SPH and ISPH Methods 217
To dene eciency of neighbouring particles search algorithm we compared
direct search and grid algorithm. Test calculations were carried out on uniproces-
sor system: AMD Athlon 2000+ / 512 Mb RAM / Windows XP SP2 in Fortran
PowerStation 4.0. Time of search neighbouring particles depending on number
of particles for 1000 time steps was measured.
Apparently from Fig. 3 and 4, dependence for direct search is square-law and
for grid algorithm is linear that corresponds to theoretical assumptions.
Dependence of acceleration of neighbouring particles search procedure on
number of particles is resulted in the gure 5.
12 Parallelization
A great amount of problems, simulating with the ISPH, is restricted by the capa-
bilities of the computers. However, thanks to the appearance of new more power-
ful computers, clusters of computers and to the organization of high-performance
computations using them, the opportunity to expand of class of problems, which
can be simulated using the ISPH appeared.
While simulating the problems on clusters the algorithm runs as an MPI-
application. MPI is a package of subprograms, giving the opportunities of com-
munications between the processes, running on dierent CPUs.
One of the most weak places in the ISPH algorithms is a necessity to solve
high-dimensional systems of linear algebraic equations, which appear when ap-
proximating the pressure Poisson equation.
At rst the algebraic system was solved using the standard Gauss procedure,
then the serial algorithm was parallelized.
For demonstration of eciency of obtained parallel algorithm some test com-
putations were performed. The dimensions of matrices in computations varied
from 200 to 4600.
For comparison of computation time of serial and parallel algorithms and for
demonstration of parallel algorithm advantages the coecients of acceleration
and eciency are used:
S
m
=
T
1
T
m
, E
m
=
S
m
m
,
where T
m
is the computation time using the parallel algorithm on the cluster
with m (m>1) processors, T
1
is the computation time using the serial algorithm.
Parallel computations were carried out on the cluster, consisted of units with
the characteristics: Intel Pentium 4 2.80 GHz / 1 Gb RAM / Red Hat Linux
with Intel Fortran compiler and MPI package.
On Fig. 6 the dependencies between the dimensions of matrices and compu-
tation time for dierent number of processors are presented. Fig. 7 presents the
dependencies between the dimensions of matrices and acceleration of computa-
tions. The dependencies between the dimensions of matrices and eciency of
parallel algorithm are presented on Fig. 8.
The developed parallel realization of the SLAE Gauss solver allows receiving
considerable accelerationof the computations as can be seen fromthe Table 2. Also
218 K.E. Afanasiev, R.S. Makarchuk, and A.Yu. Popov
Fig. 6. Dependencies between the dimensions of matrices and computation time for
dierent number of processors
Fig. 7. Dependencies between the dimensions of matrices and acceleration of compu-
tations for dierent number of processors
it gives an opportunity to use much more particles in computations for obtaining
more accurate results. Nevertheless, the uniprocessor PGMRES solver shows more
signicant results in computation time than parallel Gauss solver. However the full
computation time (Table 3) of the ISPH method remain rather great. Therefore
Comparative Analysis of the SPH and ISPH Methods 219
Fig. 8. Dependencies between the dimensions of matrices and eciency of parallel
algorithm for dierent number of processors
Table 2. Computation time of dierent SLAE solvers (in sec.) for one time step
Particles number PGMRES Gauss procedure Parallel Gauss procedure (32 CPU)
1409 0.032 0.984 0.031
2009 0.047 3.421 0.108
2709 0.094 8.625 0.272
3509 0.203 20.203 0.638
4409 0.719 43.125 1.361
X
Y
0 0.5 1 1.5 2 2.5 3
0
0.5
1
1.5
2
2.5
3
X
Y
0 0.5 1 1.5 2 2.5 3
0
0.5
1
1.5
2
2.5
3
Fig. 9. Dam breaking problem at t = 0.0s (left - SPH, right - ISPH)
220 K.E. Afanasiev, R.S. Makarchuk, and A.Yu. Popov
parallelization of PGMRES algorithm is planed for future work. Possibly more ef-
cient will prove to be the defect correction multi-grid method [2].
Further optimization of algorithms of the both presented methods lies in par-
allelization of the grid algorithm for nearest neighbours search.
13 Model Problem. Dam Breaking
The equations (1)-(2) are solved. In the initial moment t = 0 the viscous uid
column gets broken under the inuence of gravity. The following values of phys-
ical characteristics are used for this problem: = 1000 kg/m
3
uid den-
sity, = 1 kg/m s
2
dynamical viscosity. Fig. 9-14 presents ow pictures in
X
Y
0 0.5 1 1.5 2 2.5 3
0
0.5
1
1.5
2
2.5
3
X
Y
0 0.5 1 1.5 2 2.5 3
0
0.5
1
1.5
2
2.5
3
Fig. 10. Dam breaking problem at t = 0.21s (left - SPH, right - ISPH)
X
Y
0 0.5 1 1.5 2 2.5 3
0
0.5
1
1.5
2
2.5
3
X
Y
0 0.5 1 1.5 2 2.5 3
0
0.5
1
1.5
2
2.5
3
Fig. 11. Dam breaking problem at t = 0.65s (left - SPH, right - ISPH)
Comparative Analysis of the SPH and ISPH Methods 221
Table 3. Full computation time (in sec.) of the SPH and ISPH methods with dierent
number of particles (2000 time steps)
Particles number SPH ISPH
1409 12.2 52.2
2009 17.4 111.4
2709 23.4 211.4
3509 30.2 436.2
4409 38.4 1476,4
X
Y
0 0.5 1 1.5 2 2.5 3
0
0.5
1
1.5
2
2.5
3
X
Y
0 0.5 1 1.5 2 2.5 3
0
0.5
1
1.5
2
2.5
3
Fig. 12. Dam breaking problem at t = 1.31s (left - SPH, right - ISPH)
X
Y
0 0.5 1 1.5 2 2.5 3
0
0.5
1
1.5
2
2.5
3
X
Y
0 0.5 1 1.5 2 2.5 3
0
0.5
1
1.5
2
2.5
3
Fig. 13. Dam breaking problem at t = 1.5s (left - SPH, right - ISPH)
222 K.E. Afanasiev, R.S. Makarchuk, and A.Yu. Popov
X
Y
0 0.5 1 1.5 2 2.5 3
0
0.5
1
1.5
2
2.5
3
X
Y
0 0.5 1 1.5 2 2.5 3
0
0.5
1
1.5
2
2.5
3
Fig. 14. Dam breaking problem at t = 1.64s (left - SPH, right - ISPH)
dierent moments of time with the calculation number of uid particles 1600.
The numerical results, obtained solving this problem by the SPH and MPS meth-
ods, are provided in the work [16]. Computation time for the SPH and ISPH
methods with dierent number of particles is presented in the Table 3.
14 Conclusion
On the problems submitted in the given work in 2D-dimensional statement the
basic advantages and opportunities of the SPH and ISPH methods can be re-
vealed. The shown results demonstrate wide applicability of the methods in
simulation of problems with large deformations of problem domains. However,
the presented methods have the following disadvantages: applying the equation
of state for pressure eld calculation in the original SPH doesnt allow comput-
ing dynamic loads; in the obtained with the ISPH results a particle recession
can be observed after uid impact on solid wall. Nevertheless, the SPH shows
pronounced kinematics of ows. The ISPH due to using the pressure Poisson
equation can be applied to calculation of dynamic loads.
References
1. Gingold, R.A., Monaghan, J.J.: Mon. Not. Royal Astron. Soc. 181, 375389 (1977)
2. Cummins, S.J., Rudman, M.: J. Comp. Phys. 152, 584607 (1999)
3. Shao, S., Lo, E.Y.M.: Adv. Water Res. 26, 787800 (2003)
4. Monaghan, J.J., Thompson, M.C., Hourigan, K.: J. Comput. Phys. 110, 399406
(1994)
5. Morris, J.P., Fox, P.J., Zhu, Y.: J. Comput. Phys. 136, 214226 (1997)
6. Monaghan, J.J.: Rep. Prog. Phys. 68, 17031759 (2005)
Comparative Analysis of the SPH and ISPH Methods 223
7. Liu, G.R., Liu, M.B.: Smoothed particle hydrodynamics: a meshfree particle
method. World Scientic, Singapore (2003)
8. Koshizuka, S., Tamako, H., Oka, Y.: Comp. Fluid Dyn. 4(1), 2946 (1995)
9. Koshizuka, S., Nobe, A., Oka, Y.: Int. J. Numer Methods Fluids 26, 751769 (1998)
10. Liu, W.K., Jun, S., Sihling, D.T., Chen, Y., Hao, W.: Int. J. Numer Methods Fluids
24, 1 25 (1997)
11. Dilts, G.A.: Int. J. Numer Methods Eng. 48, 15031524 (2000)
12. Monaghan, J.J., Lattanzio, J.C.: Astron Astrophys. 149, 135143 (1985)
13. Monaghan, J.J., Poinracic, J.: Appl. Numer Math. 1, 187194 (1985)
14. Brookshaw, L.A.: Proc. ASA 6, 207210 (1985)
15. Connor, J.J., Brebbia, C.A.: Finite element techniques for uid ow. Sudostroenie,
Leningrad (in Russian) (1979)
16. Afanasiev, K.E., Iliasov, A.E., Makarchuk, R.S., Popov, A.Yu.: Comput. Technolo-
gies 11, 2644 (in Russian) (2006)

S-ar putea să vă placă și