Professional Documents
Culture Documents
Particle Method
Particle Method
Particle Method
Zhi TAO
Division 402, Department o f Jet Propulsion, Beijing University o f Aeronautics and Astronautics, Beijing
100083, P.R. China
A modified moving particle semi-implicit method (MPS) is presented for incompressible fluids. Modification is
on the removal of imaginary nodes to enforce the boundary conditions. Comparison with the original imaginarynode scheme has proved the validity of the proposed method. Performance of MPS method in general is also
investigated by comparing the results of Lid-driven and natural convection problems with finite volume method
(FVM). It is found that MPS method shows relatively strong numerical diffusion, and for convection problems,
MPS method appears to be less robust than FVM. Though better results could be obtained with different kernel
functions, such improvement is quite limited.
Introduction
Numerical simulation of fluid flow and heat transfer
process still remains a challenging problem, especially
for complex geometrical problems and large deformation
problems. The sticking point is that conventional computational methods require either structured or non-structured
grids for the discretization of governing equations. Need
of grids or meshes causes two major difficulties, i.e. the
generation of grids and the complexity of applying
governing equations upon them.
Recent developments in meshless methods have
gained significant attention in the computation community
for not relying on either structured or non-structured
meshes. Several meshless methods were proposed and
substantially improved, such as Smoothed Particle
Hydrodynamics (SPH), Element-Free Galerkin method
(EFG), Reproducing Kernel Particle method (RKPM),
etc. A comprehensive review of these methods could be
found in Belytschko et al. Ill and Liu 12].
The moving particle semi-implicit (MPS) method
was developed for incompressible fluids I3]. In MPS
method, governing equations are discretized by particle
interaction models and the arbitrary Lagrangian-Eulerian
(ALE) technique is implemented using the fractional step
method for time integration. MPS method has been
Received 2003
Jun GUO: Associate Professor
Nomenclature
thermal diffusivity
gravitational acceleration
characteristic scale
pressure
Pr Prandtl number, Pr = via
Ra Rayleighnumber, R a = g f l H 3 A T /va
Re Reynolds number, Re : uH/v
S
source term
T
temperature
Tc cool wall temperature
Th hot wall temperature
AT
temperature difference, AT = Th - Tc
At time increment
U, V dimensionless velocity component
u,v dimensional velocity component
a
g
H
p
(1)
w(i
'
-~t,~e)
227
(O<r <O.5re)
3.0
(re < r )
(2)
-.
2.5
mW2
"~
.....
....
"%
W3
W5
2.0
-.
1 - 6(r / re ) 2 + 6(r / re ) 3
1.5
10
2(1-r/re) 3
1.0
7re(re / 2 ) 2
0.5
0
(re <- r)
(3)
0.0
0.0
0.2
0.4
0.6
0.8
1.0
478;'r(re / 3) 2
(v~)k" = .k,-S
( 3 - 3r / re) 5
I~j - ~ 12
w(i ~j - ~ I,ro)
0
(Re < r)
(4)
(5)
228
2d
(V2~), =-T-Y'.{(Oj - ~,)w(t fj - 5 [,r~)}
(0)+:..
(7)
"~i j~i
(8)
(9)
j~i
+S
(10)
fj
Other points
(11)
(12)
Ar
(13)
E4w(i
i,r )
Zw(i
l,re)
Jun GUO et al. Modified Moving Particle Semi-Implicit Meshless Method for Incompressible Fluids
where j denote only those inside the computing domain
within the weight radius. The default thermal boundary
condition is adiabatic. Any nodes on surfaces are
effectively adiabatic, since there are no particles beyond,
with which they can exchange energy. While for
Dirichlet boundaries, the prescribed values are simply
imposed to the corresponding nodes, and approximation
calculations for these quantities are skipped.
T2
T1
TO
T-1
T-2
/22
/11
U0
/2 1
/1-2
V2
V1
V0
V I
V 2
~ @
I'~ @
@
@
0
0
0
0
70 =T~,
Sidewall
T=Tw, u =v=0
T -l = 2 T . - T ,
U0 : 0 ,
I1-1 =
V0 = 0 ,
V 1 = _V 1 ,
111 ,
T z =2T.
T2
ll 2 : _112
v-------L"- (v v~+s~.
15"At
V2p.+~=v./5"
At
~L _~.
At
Supportof boundary
nodeusingimaginarynodes
__@
'x:;I o
0 "0
0
1 Vp n+l
(16)
Boundary
Imaginary
O Fluid
(15)
(14)
V-2=_V 2
229
At
- aVeT"
(17)
"o o~
vn+l
"~'U0
_ ~'L
At
*2
~(/?_/?c).v/SL=0
(18)
~(V-/?c).VTL =0
(19)
o
e~,-~
T,,+I _ T L
At
Supportof nodeP
Boundary
230
1.0 Re=1.0E2
w5,x=0.5
0.8 [] Imaginray
0.6
0.12
0.2
.
~
Vj~
J
~
0.10
Re=I.0E2'
'
W5, Y=0.5
[] Imaginary
Non-Imaginary
0.08
0.0
0.06
.~
0.2
0.04
"~
> -0.1
0.0
0.02
-0.2
0.00
014
-0.2
-0.3
1'.0
0'.0
Y
(a) W5 results at section X = 0.5
i
0.1
0.4
1.0 Re=1.0E2
W2oX=0.5
0.8
[] Imaginary.
0.2
0.10
V
~t~
0.6
0.08
0.1
0.06
0.0
0.04
> -0.1
0.2
016
018
110
X
(b) W5 results at section Y= 0.5
,
0.4
Re=I.0E2'
W2, Y=0.5
[] Imaginary
Non-Imaginar.
0.4
0.2
0.0
0.02
-0.2
0.00
010'0'.2'014
0'.6'0'.8'1.0
Y
(c) W2 results at section X = 0.5
-0.2
-0.3
018
X
(d) W2 results at section Y= 0.5
Fig.5 Velocity profiles for Lid-driven problem using imaginary nodes or not using
(Re = 1.0E2, At = 1.0E-3)
1.0
Jun GUO et al. Modified Moving Particle Semi-Implicit Meshless Method for Incompressible Fluids
0.4
0.4
0.2
0.2
~ 0.0
_o
>
-0.2
0.0
>
231
018
1.0
-0.2
-0.4
-0.4
I
0.0
012
0.4
016
018
ll.O
0.4
016
Xor Y
Xor Y
Fig.6 Comparison of results for natural convection in a square cavity using imaginary nodes or not using
(Ra = 1.0E2, At = 1.0E-4)
i
1.0
0.10
....... W2
\ /
0.2
0.08
0.1
0.6
0.06
0.0
0.4
0.04
"~ o0.1
>
0.8
0.2
....... W2
0.02
-0.2
0.0
W5
0.00
-0.2
0'.0 ' 0:2 ' 0'.4 ' 0:6 ' 0'.8 ' l'.O
Y
(a) at section X = 0.5
-0.3
014
ll.0
X
(b) at section Y= 0.5
232
0.4
0.5
0.3.
0.2-
0.8
1.0
o.o
f"~--
-0.2
,,~
k
-0.3
V(X-0.6)
-0.2!
+ u(x-o.4)
.~
V(X=0.4)
-0.3-"
--MRS
-0.4-
0:0
0:2
0'4
0:6
0:8
l:o
0.0
0.2
OI
0.2
0.4
0.6
0.5
U(Y=0.4)
o V(Y=O.4)
V U(Y=0.6)
+
v(r=0.6)
0.3
[]
0.3
~)
v(r=0.2)
, uo,=o
0.4~ o
a
u(r=o.8) ~
~,.
+__ v(r 8) y
-,,.
0.2
0.1
~
"-6
> -0.1.
~ -0.1
0.4
.~ 0.1 i
0.0
-0.4
v(x=o8)/
0.2-"
0.1
~
D u(x=o.2)
0.4- o V(X=0.2)
0.3: + u(x-o.8) .o~
0.1
0.0
0.0
"~
-o.1 -.
>
-0.1i
-0.2-
-0.2-
-0.3 -~
-0.3-'
-o.41
i,
00
0.2
0.4
0.6
0.8
-0.4-
1.0
'
0.0
.
0.4
0'.2
. 0.6.
. 0.8
1.0
0.4
0.012 I
0.010 I
0.2
0.008 I
0.006 I
"~ 0.0
>O
>
-0.2
,~
I~VM '
-'
....... W5 41"41~
.---. o W5 81"81"
-'"/
iit
%%% Ij
"
~1
0.004
0.002
0.000
-0.002
-0.4
-0.004
0'.0
012
1'.0
0.0
0.2
'
'o'. 4
0.6
'
0 '.8
Xor f
(b) secondary velocity component
Fig.9 Velocity profiles for natural convection in a square cavity at sections X = 0.5 & Y= 0.5
(Ra = i.0E2, At = 1.0E 4)
1.0
'
Jun GUO et al. Modified Moving Particle Semi-Implicit Meshless Method for Incompressible Fluids
~"
60-
**
V( Y-O, 5 )
./
401
"
I in
'----:- w'2
l II
i I
n
U(Y=0.5)
. ~
Ill
V(X=0.5)
i***.
u(x=o.5)
illll/
201
233
0.
mm
"
W3
W5
W5 8t'81
--
0
>O
-20
mm
-4
-40-
[] W3
- W5
-600
* L.W*
~,
0.2
0.4
0.6
0.8
Xor Y
(a) major velocity component
-8
0.0
1.0
FVM_V
**"* -
FVMU
I
0:2
014
0:6
018
1.0
1.2
X
(b) secondary velocity component
Fig.10 Velocity profiles for natural convection in a square cavity at center sections X = 0.5 & Y= 0.5
(Ra = 1.0E5, At = 1.0E-5 for 4l )<41 points, and At = 2.0e-6 for 81 81 points)
i
1.0
1.0.
0.53
~"
052~/
0.8
0.6
0.54
rw
. . . . . . W2
-~
l~a
" ~
~-
0.8.
x=o5
a-
0.4
0.50 ~
0.4.
0.49~
....
~" 0.2
0.0
o'.oo'.2o'.4o'.6o'.811o
0.48 ~
0.2.
0.47
0.0.
0.46
0'.0
Xor Y
0.2
'
014 . 0.6
. . . 0.8
. .
1.0
Xor Y
Fig.ll Temperature profiles for natural convection in a square cavity at central sections
Conclusion
This paper extensively discussed various aspects of
applying MPS method to incompressible fluid dynamics
problems. The performance of the present method is
studied through the analysis of lid-driven and natural
convection problems in square cavities. Several conclusions
can be drawn as follows:
(1) Imaginary nodes are not necessary for boundary
condition disposal. Numerical analysis shows using or
not using images always reach nearly the same results.
(2) MPS method has much room to improve its
accuracy, numerical tests show its noticeable discrepancy
with the conventional FVM prediction. For natural
convection problem, MPS results of coarse point
configuration even appear to have oscillatory behavior
near boundaries, finer node distribution could reduce the
oscillation greatly, indicating that MPS is less robust than
FVM.
(3) Analysis of the effects of different kernels shows
References
[1] Belytschko, T, Krongauz, Y, Organ, D, et al. Meshless
Methods: An Overview and Recent Developments.
Comput. Methods Appl. Mech. Engrg., 1996, 139:3--47
234
[2] Liu, W K, Jun, S, Sihling, D T, et al. Multiresolution
Reproducing Kernel Particle Method for Computational
Fluid Dynamics. Int. J. Numer. Methods fluids, 1997, 24:
1391--1415
[3] Yoon, H Y, Koshizuka, S, Oka, Y. A Particle-gridless
Hybrid Method for Incompressible Flows. Internat. J.
Numer. Meth. Fluids, 1999, 30:407--424
[4] Koshizuka, S, Yoon, H Y, Yamashita, D, et at. Numerical
Analysis of Natural Convection in a Square Cavity Using
MPS-MAFL. Comp. Fluid. Dynamic J., 2000, 8: 485-494
[5] Chikazawa, Y, Koshizuka, S, Oka, Y. A Particle Method
for Elastic and Visco-plastic Structures and Fluid-structure
Interactions. Comp. Mech., 2001, 27: 97--106
[6] Yoon, H Y, Koshizuka, S, Oka, Y. Direct Calculation of