Anda di halaman 1dari 8

Application of lattice Boltzmann method, nite element method,

and cellular automata and their coupling to wave propagation problems


Y.W. Kwon
*
, S. Hosoglu
Department of Mechanical and Astronautical Engineering, Naval Postgraduate School, Monterey, CA 93943, United States
Received 7 January 2007; accepted 20 July 2007
Available online 10 September 2007
Abstract
Three dierent computational techniques were applied to wave propagation problems. Those techniques were the lattice Boltzmann
method, nite element method, and cellular automata. The formulation of each technique was presented, and the coupling procedures of
those techniques were also presented. For example, a part of the problem domain was solved using one analysis technique while the other
part was analyzed by another technique. Such coupled techniques may overcome the diculties that a single technique has, and they may
also provide their own advantages of two dierent methods in a single analysis depending on application problems. For example, one
technique is computationally more ecient while another is useful to model a complex or irregular shape of domain. Combining the two
techniques will be benecial to solve a complex domain shape with computational eciency. The accuracy of the dierent techniques
including the coupled methods was numerically demonstrated by comparing their solutions to other solutions available for wave prop-
agation problems in 1-D and 2-D.
Published by Elsevier Ltd.
Keywords: Finite element method; Cellular automata; Lattice Boltzmann method; Coupled techniques; Wave propagation
1. Introduction
Various numerical solution techniques have been devel-
oped for dierent application problems. The nite element
method (FEM) began more than a half century ago and
now becomes very popular for analyzing linear and nonlin-
ear problems with applications to solid mechanics; uid
mechanics; heat, mass and energy transfer, etc. [1,2]. The
lattice Boltzmann method (LBM) has emerged more
recently with intensive application to thermal-uid prob-
lems [312]. On the other hand, Cellular Automata (CA)
was also developed since the invention by von Neumann
in the late 1940s [13,14]. The CA became one of founda-
tions for developing LBM.
Each of those techniques has own merits depending on
specic applications. The CA is generally simple in terms
of algorithms and computer code development as long as
a simple rule(s) specifying time evolution of a given prob-
lem can be obtained. However, the technique has not been
properly developed yet for solid and energy transfer appli-
cations. The LBM is the next in terms of simplicity of com-
puter coding and computational cost at least for wave
propagation problems studied in this paper. Other advan-
tages of LBM are that multiple species can be modeled rel-
atively with ease [1518], and both continuum as well as
discrete particles can be handled using the technique. As
a result, the technique can be applied to a large span of
length scale. However, LBM has some diculty in apply-
ing boundary conditions and modeling complex shapes of
boundaries even though some techniques for LBM have
been developed to address irregular mesh problems
[1923]. On the other hand, FEM is quite versatile to solve
various complicated problems including highly irregular
shapes of domains with complex boundary conditions.
However, the technique is mostly applicable to continuous
media.
0045-7949/$ - see front matter Published by Elsevier Ltd.
doi:10.1016/j.compstruc.2007.07.013
*
Corresponding author. Tel.: +1 831 656 3468; fax: +1 831 656 2238.
E-mail address: ywkwon@nps.edu (Y.W. Kwon).
www.elsevier.com/locate/compstruc
Available online at www.sciencedirect.com
Computers and Structures 86 (2008) 663670
Combining more than one technique can take advantage
of those that a single technique has depending on case by
case problems. For example, FEM can be used around
the complex boundary while LBM can be selected for the
internal domain. In such a case, a proper coupling tech-
nique must be developed between the two solution meth-
ods. In this paper, coupling techniques were presented
between FEM and LBM, and between FEM and CA,
respectively. The accuracy of the coupled techniques was
numerically demonstrated by comparing the coupled solu-
tions to other solutions obtained using a single technique.
The subsequent sections describe the three dierent tech-
niques in context of wave propagation problems, and their
coupling techniques. Then, some numerical examples are
presented for demonstration of the proposed techniques,
followed by conclusions.
2. Lattice Boltzmann method
The LBM can be developed from either CA or the Boltz-
mann equation. Using the BGK model [24], the lattice
Boltzmann equation is expressed as
f
i
~r ~e
i
Dt; t Dt f
i
~r; t
1
s
f
eq
i
~r; t f
i
~r; t 1
where f
i
~r; t denotes the probability of nding a particle at
lattice site ~r and time t, which moves along the ith lattice
direction with the local discrete particle velocity ~e
i
.
f
eq
i
~r; t is the local equilibrium solution. Furthermore, Dt
is the time increment, and s is the relaxation time.
For the wave equation using the two-dimensional
square lattice, the approach presented in Ref. [25] was
adopted in the present study. As a result, the D2Q5 lattice
(which includes four surrounding nodes at each lattice site)
was used, and its local equilibrium solution is expressed as
[25]
f
eq
i

c
2
s
2e
2
u
1
2
~e
i

~
f
e
2
if i 6 0 and f
eq
o
1
2c
2
s
e
2
_ _
u
2
where
u

i
f
i
3
and
~
f

i
f
i
~e
i
4
and c
s
is the wave propagation speed.
The local velocity vectors are given as below:
~e
i

0; 0 i 0
cosfi 1p=2g; sinfi 1p=2g i 1; 2; 3; 4
_
5
With the choice of s = 1/2, Eq. (1) is reduced to
f
i
~r Dt~e
i
; t Dt
l
2n
2
u f
i2
~r; t i 6 0
f
o
~r; t Dt 2l
n
2
1
n
2
u f
o
~r; t
6
and recovers the wave equation as given below:
o
2
u
ot
2
c
2
s
o
2
u
ox
2

o
2
u
oy
2
_ _
7
For the above equations, 0 6 l 6 1 is the wave attenuation
factor. l = 0 denotes perfect reection of the wave while
l = 1 indicates perfect transmission. Any in-between value
for l represents that the wave is partially absorbed. Fur-
thermore, n P1 is the refraction index which is dened
as the ratio of the maximum propagation speed of the
model to c
s
[25].
3. Finite element method
The FEM is applied to the wave equation, Eq. (7) using
the weighted residual technique [26] as shown below:
_
X
w
o
2
u
ot
2
c
2
s
o
2
u
ox
2

o
2
u
oy
2
_ _ _ _
dA 0 8
where w is the weighting function, and X is the problem do-
main. As the domain is divided into a nite element mesh, a
set of shape functions are selected for each nite element to
represent the solution within the element. For example, a
four-node quadrilateral element was selected for the pres-
ent study because it closely resembles the grid patterns of
the lattice Boltzmann method and cellular automata. Then,
the solution within a nite element is expressed as
u Hfwg 9
where H h
1
h
2
h
3
h
4
is the row vector consisting
of quadrilateral shape functions [26], and {w} is the nodal
variable vector. For Galerkins method, the weighting
function is chosen such as w = h
i
(i = 1, 2, 3, 4). Eq. (9)
and the weighting function expression are substituted into
Eq. (8) for each nite element. Then, integration of the
resulting expression over each nite element domain gives
the element matrix and vector. The assembly of element
matrices and vectors yields the following matrix equation
Mf

wg Kfwg fF g 10
where superimposed dot denotes temporal derivatives, and
the system matrices [M] and [K] are obtained from sum-
ming up element matrices [k] and [m], respectively, as
shown below:
M

ne
m and K

ne
k 11
Here, the summation is over the number of elements (ne).
The element matrices are computed as
m
_
X
e
H
T
HdX 12
664 Y.W. Kwon, S. Hosoglu / Computers and Structures 86 (2008) 663670
and
k
_
X
e
c
2
s
oH
ox
_ _
T
oH
ox
_ _

oH
oy
_ _
T
oH
oy
_ _
_ _
dX 13
where X
e
is each nite element domain, and
oH
ox
_ _

oh
1
ox
oh
2
ox
oh
3
ox
oh
4
ox
_ _
14
oH
oy
_ _

oh
1
oy
oh
2
oy
oh
3
oy
oh
4
oy
_ _
15
The transient Eq. (10) is solved from initial conditions using
the central dierence technique as summarized below:
f

wtg M
1
fF tg Kfwtg 16
f
_
wt Dt=2g f
_
wt Dt=2g f

wtgDt 17
fwt Dtg fwtg f
_
wt Dt=2gDt 18
4. Cellular automata
The CA method uses the following local rules to repre-
sent the wave equation. Those rules are extension from 1-D
wave equation [27]. For any grid point surrounded by four
neighboring grid points, the evolution in time for u is
expressed as below:
u
C
t Dt u
E
t u
W
t u
N
t u
S
t 2u
C
t=2
19
where subscripts C, E, W, N, and S denote the cen-
ter, east, west, north, and south, respectively. East and
north directions are arbitrarily assumed to be positive.
On the other hand, if the grid point is located at the east
side boundary, the used rule is
u
C
t Dt 2u
W
t a
EW
u
N
t u
S
t 2u
C
t=2
20
For the grid point at the west side boundary
u
C
t Dt 2u
E
t a
EW
u
N
t u
S
t 2u
C
t=2
21
Similarly, for the grid points at the north and south bound-
aries, respectively,
u
C
t Dt 2u
S
t a
NS
u
E
t u
W
t 2u
C
t=2
22
u
C
t Dt 2u
N
t a
NS
u
E
t u
W
t 2u
C
t=2
23
Finally, there is a dierent rule for a corner grid point. For
example, the grid point at the northeast corner of the
boundary, the rule is
u
C
t Dt 2u
S
t a
NS
2u
W
t a
EW
2u
C
t=2
24
Similar rules can be developed for other corners. In above
equations, a
i
(i = EW or NS) is a constant whose value is
usually selected to be the grid spacing. This constant repre-
sents the equilibrium length between grid points. For pre-
venting the grid points from overlapping, this constant
should be at least a half the grid spacing.
The CA model consists of two dierent types of grid
points that will be referred to as black and white particles.
At each time step, only one type of particles are allowed to
move. For example, at t = Dt, Eqs. (19) through (24) are
applied to black particles only and all black particles move
at once. Then, at t = 2Dt, Eqs. (19) through (24) are
applied to white particles only and all white particles move
at once. This process is illustrated in Fig. 1.
In order to apply external forces to the CA model, the
following algorithm was developed. When a force is
applied to a grid point of the CA model, acceleration is
computed as below:
Da
i
t Dt
F
i
t Dt F
i
t
m
i
25
where F
i
is the force applied to the ith grid point and m
i
is
the mass at the grid point. Time integration of the acceler-
ation yields velocity and displacement, respectively, as
below:
v
i
t Dt v
i
t a
i
tDt 26
u
i
t Dt u
i
t v
i
t DtDt 27
Once the displacements at the grid points to which the
forces are applied are computed, the local rules as given
by Eqs. (19) through (24) are applied to all grid points.
Then, the computational cycle repeats the above process.
5. Coupling techniques
Two dierent couplings were studied. One was to couple
LBM to FEM and the other was to couple CA to FEM. In
each case, a part of the problem domain was modeled using
FEM while the other part was modeled using either LBM or
CA. The coupled domain problem was solved in a staggered
manner with an exchange of data at the interface of the two
subdomains which were modeled by two dierent methods,
respectively. First, the coupling between FEM and LBM is
discussed. At the overlapping grid points of the nite ele-
ment mesh and the Boltzmann lattice located at the interface
of the two subdomains, FEM uses the variable u as
unknowns while the LBMuses f
i
as unknowns. When a tran-
sient solution information is transferred from LBM to FEM
at the overlapping grid points, Eq. (3) is utilized. However,
transferring the transient solution from FEM to LBM is not
unique. For this directional coupling, the following tech-
nique is utilized. First of all, f
i
is adjusted as below:
f
i

l
2n
2
u
fem
i 6 0
f
o
2l
n
2
1
n
2
u
fem
28
where subscript FEM denotes the FEM solution at
the interface grid points. Then, for f
i
that is normal to the
Y.W. Kwon, S. Hosoglu / Computers and Structures 86 (2008) 663670 665
interface boundary, a modication is made to incorporate
the ow of f
i
from the neighbor. That is, aDu~e
i
is added
to f
i
, in which Du is the dierence between two neighboring
grids, one in the FEM subdomain and the other in LBM
subdomain, and a constant a is a function of the grid size,
time step, and the relaxation constant. Finally, f
i
is corrected
so that their sum is equal to the FEM solution at the grid.
The coupling between FEM and CA was undertaken as
below. The solution us at the interface and its neighbors
are passed between the two analyses. For the FEM analy-
sis, the CA solution u at the interface grid point is utilized
directly. On the other hand, for CA calculation, the FEM
solution u at the grid point next to the interface is used
to compute the CA solution u at the interface grid point
using the local rule as shown in Eq. (19). For example, if
the grid point at the east lies in the FEM domain, the
FEM solution is used for the grid point. This is illustrated
in Fig. 2.
The procedure for the coupled CA and FEM calculation
is as follows:
1. Divide the physical domain into CA and FEM parts.
Dene the CA and FEM nodes, and make the nodes
at the interface as shared nodes.
2. Apply forces to CA nodes (if any) by using Eqs. (25)
through (27).
3. Apply CA rules to the CA modeled part. For calculating
the next position of the last particle (i.e. shared node
with the FE model), the displacements of two neighbor-
ing particles are necessary. Since this is the last CA node,
there is not any CA particle on the FEM side. As a
result, the value from the second node of FE model is
borrowed.
r

t
Last node of CA
rt Dt
2nd node of FEM
r

t the CA node next to the last CA node


rt D
Last node of CA
r

t r

t rt
Last node of CA
29
Fig. 1. Particle movements in a square membrane clamped at edges, (a) t = 0: initial position, (b) t = Dt: white particles move, (c) t = 2Dt: black particles
move and (d) t = 3Dt: white particles move.
Fig. 2. CA and FEM coupled model of a rectangular membrane.
666 Y.W. Kwon, S. Hosoglu / Computers and Structures 86 (2008) 663670
4. At t = 0 the FEM data at t = Dt is needed. It is
assumed that the data at t = Dt is equal to the data
at t = 0.
5. Pass the displacement data of the last particle of the CA
part to the rst node of the FEM part, and set the accel-
eration of this node to zero so that it can not move dur-
ing the FEM calculations (This works like a boundary
condition).
rt
1st node of FEM
rt
Last node of CA
acct
1st node of FEM
0
30
6. Apply FEM calculations as usual.
7. Go back to Step 2.
During the procedure, the same time step size was used
for both the FEM calculations and the CA calculations so
that FEM and CA calculations can go in synchronized.
The CA model does not have an explicit time step size Dt
other than an integer value. In order to nd a proper Dt
of a CA model that represents the string vibration, for
example, it is used that the travel distance of sound wave
should be the same in both physical model and the CA
model, as expressed below:
c
r
Dt
r
c
CA
Dt
CA
31
where c is the speed of sound, and subscripts r and CA
indicate the physical and CA models, respectively. The
speed of sound in the CA model is
c
CA

L
N 1
32
in which L is the length of the string and N is the number of
nodes in the CA model. Substitution of Eq. (32) into Eq.
(31) yields
Dt
CA

c
r
N 1
L
Dt
r
33
where Dt
r
was used in the FEM model.
6. Numerical examples
Both 1-D and 2-D examples are presented here starting
with 1-D cases. The rst example case was free vibration of
a string. The string was held xed at both ends with a uni-
form tension, and it was plucked at the center initially so
that the initial shape of the string was an isosceles triangle.
Then, the load was removed immediately so that the string
could vibrate up and down. The same problem was solved
using LBM and FEM, respectively, and their solutions
were compared as shown in Fig. 3. For each analysis, 99
grid points were used. The gure shows an excellent agree-
ment between the LBM and FEM solutions. These solu-
tions also agreed with the analytical solution which was
identical to FEM solution in the same gure. As a result,
the analytical solution was not plotted in the gure to
avoid overcrowding. The CA was also applied to the same
example as above. Its solution was the same as LBM so
that it was not plotted in Fig. 3 to avoid any overcrowding.
The next comparison was made for a 2-D wave problem.
Free vibration of a membrane was studied with boundary
edges held xed. A square membrane was initially dis-
placed as shown in Fig. 4 and released. A time history of
the central grid point of the membrane is compared in
Fig. 5 for the LBM, CA and FEM solutions. The three
results agreed very well as seen in the gure. Because the
CA solution is on top of LBM solution, it was omitted in
the plot. The total grid points used for each analysis were
19 19.
Table 1 compares the normalized computational times
between the CA and FEM models for a 2-D wave propaga-
tion problem. It shows that as the grid size increases, the
computational time for FEM increases much faster than
that for CA. As the number of algebraic operations in each
technique is compared for the 2-D wave problem, the
approximate number of operations for addition/subtrac-
tion and multiplication/division are shown in Table 2 for
a given number of degrees of freedom in the problem.
The table shows that the computational time for CA is
0 100 200 300 400 500 600 700 800
0.1
0.05
0
0.05
0.1
0.15
Time (no unit)
D
i
s
p
l
a
c
e
m
e
n
t

(
n
o

u
n
i
t
)
LBM
FEM
LBMFEM Coupled
Fig. 3. Plot of displacement at the center of a free vibrating string held at
both ends. The LBM, FEM and LBMFEM coupled solutions are
compared. The analytical solution overlaps the FEM solution so that it
was not plotted in the gure.
0
5
10
15
20
0
10
20
0
0.002
0.004
0.006
0.008
0.01
0.012
X axis (no unit)
Y axis (no unit)
D
i
s
p
l
a
c
e
m
e
n
t

(
n
o

u
n
i
t
)
Fig. 4. Initial displacement of a square membrane.
Y.W. Kwon, S. Hosoglu / Computers and Structures 86 (2008) 663670 667
linearly proportional to the number of degrees of freedom
while that for FEM is a quadratic function of the number
of degrees of freedom, as also demonstrated in Fig. 6.
Considering the computational time of the wave propaga-
tion problem, coupling of CA and FEM reduces the
computational time compared to the FEM only model.
Furthermore, a coupled model can represent a complex
boundary better than the CA only model.
The next set of examples are those of forced vibration.
The rst example compared the CA solution to the FEA
solution for a forced string vibration. A periodic step force
function, as sketched in Fig. 7, was applied to the center of
the string. The problem was solved using CA and FEM,
respectively. Their results compare excellently as seen in
Fig. 8. Fig. 9 compares the CA solution to the FEM solu-
tion for a 2-D square membrane vibration with a sinusoidal
force applied at the center of the membrane while all edges
were clamped. The two solutions agree very well, too.
In the subsequent numerical examples, coupled tech-
niques were applied to the same problems presented above.
By doing so, the solutions obtained by coupled techniques
could be compared to the results obtained using a single
technique. The rst coupled example was a free string
vibration plucked at the center as before. A half of the
domain was modeled using LBM while the other half was
modeled using FEM with the total grid points equal to
99. The result is compared to the LBM solution in Fig. 3.
The solution from the coupled technique agrees well, in
general, with the solution from a single technique. How-
ever, the coupled technique resulted in a slightly dierent
solution compared to those obtained by a single analysis
technique. This might be explained as below even if more
investigation is needed in the future. The coupled analysis
technique solved the problem in a staggered manner. In
other words, one technique was applied to one part of
0 100 200 300 400 500
0.01
0.008
0.006
0.004
0.002
0
0.002
0.004
0.006
0.008
0.01
Time (no unit)
D
i
s
p
l
a
c
e
m
e
n
t

(
n
o

u
n
i
t
)
LBM
FEM
Fig. 5. Plot of displacement at the center of a free vibrating membrane
held along the boundary. The LBM and FEM solutions are compared.
Table 1
Computational times for CA and FEM for 2-D wave problems
a
Number of nodes CA model FEM model
11 11 = 121 1.00 1.58
21 21 = 441 7.08 20.4
31 31 = 961 24.4 236.8
41 41 = 1681 59.6 1042.1
51 51 = 2601 111.3 3231.3
a
Normalized based on the case of the CA model with 121 nodes.
Table 2
Approximate number of required algebraic operations
CA FEM
Addition and subtraction 4n 2n
2
+ 6n
Multiplication and division 2n 2n
2
+ 8n
n = number of system degrees of freedom.
0 500 1000 1500 2000 2500 3000
0
100
200
300
400
500
600
700
800
No. of Degrees of Freedom
C
P
U

T
i
m
e

(
s
e
c
.
)
CA
FEM
Fig. 6. Plot of computational time vs. the number of degrees of freedom
for CA and FEM models of a 2-D wave propagation problem.
0 1 2 3 4 5
2
0
2
4
6
8
10
12
Time (no unit)
F
o
r
c
e

A
m
p
l
i
t
u
d
e

(
n
o

u
n
i
t
)
Fig. 7. Step function force applied to the center of a string held at both
ends.
668 Y.W. Kwon, S. Hosoglu / Computers and Structures 86 (2008) 663670
the domain. Then, subsequently another technique was
applied to the other part of the domain. On the other hand,
when a single solution technique was used, the whole
domain was solved at the same time. This might cause
some dierence between the coupled technique solution
and a single technique solution.
Another study was to investigate the eect of overlap-
ping the CA and FEM portions of the problem at the inter-
face so that more than one grid points could be shared
between the two models across the interface. The result
showed that overlapping of a couple of nodes across the
CA/FEM interface has a negligible eect on the solution
accuracy.
A mesh renement study was also conducted for the
same problem presented above using the CA and FEM
coupled technique. The result showed that the reduction
of error was linearly proportional to the reduction of the
mesh size. Therefore, the error in the coupled technique fol-
lowed the same trend as that in a single analysis technique.
This means the coupling did not aect the convergence rate
of the solution.
The second example of the coupled technique was a 2-D
wave problem as previously. The coupling of LBM and
FEM was again tested. A bottom half was modeled using
LBM and the top half was modeled using FEM with the
same total number of grid points as before. The result is
shown in Fig. 10. The solution from the LBMFEM cou-
pling agrees well with the LBM solution except for some
small discrepancies at the peaks of the curves. The LBM
domain was also modeled by CA so that the coupled CA
and FEM technique could be applied. The result was very
comparable to the solution obtained by the coupled LBM
and FEM technique. As a result, the CA/FEM solution
was not plotted in Fig. 10 in order not to make it too
crowded.
0 20 40 60 80 100
6
4
2
0
2
4
6
8
10
x 10
3
Time (no unit)
D
i
s
p
l
a
c
e
m
e
n
t

(
n
o

u
n
i
t
)
LBM
Coupled LBM/FEM
Fig. 10. Displacement at the center of a free vibrating square membrane
with all edges xed. LBM and coupled LBMFEM solutions are
compared.
0 1 2 3 4 5
20
15
10
5
0
5
10
15
20
25
30
35
Time (no unit)
D
i
s
p
l
a
c
e
m
e
n
t

(
n
o

u
n
i
t
)
CA Solution
FEM Solution
Fig. 8. Displacement at the center of a string under forced vibration with
a period step function. CA and FEM solutions are compared.
0 2 4 6 8 10 12 14 16 18 20
0.04
0.03
0.02
0.01
0
0.01
0.02
0.03
0.04
0.05
0.06
Time (no unit)
D
i
s
p
l
a
c
e
m
e
n
t

(
n
o

u
n
i
t
)
CA Solution
FEM Solution
Fig. 9. Plot of displacement at the center of a vibrating membrane held
along the boundary and under sinusoidal force at the center. The CA and
FEM solutions are compared.
0 5 10 15 20
4
3
2
1
0
1
2
3
4
5
6
Time (no unit)
D
i
s
p
l
a
c
e
m
e
n
t

(
n
o

u
n
i
t
)
CAFEM Coupled Solution
FEM Solution
Fig. 11. Displacement at the center of a vibrating string under a sinusoidal
force. FEM and coupled CAFEM solutions are compared.
Y.W. Kwon, S. Hosoglu / Computers and Structures 86 (2008) 663670 669
The nal example cases were the coupled CA and FEM
technique that was applied to string and membrane vibra-
tion problems. Both cases were forced vibrations with a
sinusoidal force at the center. One third of the domain
was modeled using FEM and the other two thirds of the
domain was modeled using CA for the string and mem-
brane, respectively. The center displacements of the string
and the membrane are compared between the FEM solu-
tions and the coupled CA and FEM solutions in Figs. 11
and 12, respectively. Both solutions agree very well for
both cases.
7. Conclusions
In this paper, wave propagation problems were analyzed
using three dierent numerical modeling techniques like
FEM, CA, and LBM and their solutions were compared.
Furthermore, coupling techniques were presented between
FEM and LBM, and FEM and CA. Depending on appli-
cation problems, coupling dierent methods for a single
analysis may be benecial to overcome limitations of a sin-
gle method. In other words, the coupled methods may be
applicable to more complex and diverse problems for
which a single analysis method has diculty, with less com-
putational costs. The accuracy of the coupled techniques
was demonstrated using dierent wave problems in 1-D
and 2-D. The presented work will be further developed
and investigated for more complex problems such as cou-
pled multi-dimensional problems.
References
[1] Zienkiewicz OC, Taylor RL. The nite element method. 4th
ed. London: McGraw-Hill; 1988.
[2] Bathe K-J. Finite element procedures. Englewood, NJ: Prentice
Hall; 1996.
[3] Qian YH. Simulating thermodynamics with lattice BGK models. J
Sci Comp 1993;8:23141.
[4] Chen H. Discrete Boltzmann systems and uid ows. Comp Phys
1993;7:6327.
[5] Cali A, Succi S, Cancelliere A, Benzi R, Gramingnani M. Diusion
and hydrodynamic dispersion with the lattice Boltzmann method.
Phys Rev A 1992;45:57714.
[6] Chen S, Doolen GD. Lattice Boltzmann method for uid ow. Annu
Rev Fluid Mech 1998;30:32964.
[7] Flekkoy EG. Lattice Bhatnagar-Gross-krook models for miscible
uids. Phys Rev E 1993;47(June):424757.
[8] Swift MR, Orlandini E, Osborn WR, Yeomans JM. Lattice
Boltzmann simulations of liquid-gas and the binary uid systems.
Phys Rev E 1996;54(November):504152.
[9] Soe M, Vahala G, Pavlo P, Vahala L, Chen H. Thermal lattice
Boltzmann simulations of variable Prandtl number turbulent ows.
Phys Rev E 1998;57(April):422737.
[10] Peng Y, Shu C, Chew YT. Simplied thermal lattice Boltzmann
model for incompressible thermal ows. Phys Rev E 2003;68:026701-
11-8.
[11] Krafczyk M, Tolke J, Rank E, Schulz M. Two-dimensional simula-
tion of uidstructure interaction using lattice-Boltzmann methods.
Comput Struct 2001;79:20317.
[12] Frisch U, Hasslacher B, Pomeau Y. Lattice-gas automata for the
NavierStokes equations. Phys Rev Lett 1986;56:15058.
[13] Doolen G, editor. Lattice gas method for partial dierential
equations. Addison-Wesley; 1990.
[14] Perdang J, Lejeune A, editors. Cellular automata: prospect in
astrophysical applications. World Scientic; 1993.
[15] Gunstensen AK, Rothman DH, Zaleski S, Zanetti G. Lattice
Boltzmann model of immiscible uids. Phys Rev A 1991;43:43207.
[16] Shan X, Doolen G. Multicomponent lattice-Boltzmann model with
interparticle interaction. J Stat Phys 1995;81:37993.
[17] Guo Z, Zhao TS. Discrete velocity and lattice Boltzmann models for
binary mixtures of nonideal uids. Phys Rev E 2003;68:035302-12-4.
[18] Dupin MM, Halliday H, Care CM. Multi-component lattice
Boltzmann equation for mesoscale blood ow. J Phys A Math Gen
2003;36:851734.
[19] Peng G, Xi H, Duncan C, Chou S-H. Lattice Boltzmann method on
irregular meshes. Phys Rev E 1998;58:R41247.
[20] Xi H, Peng G, Chou S-H. Finite Volume scheme for the lattice
Boltzmann method on unstructured meshes. Phys Rev E
1999;59:62025.
[21] Mei R, Luo L-S, Shyy W. An accurate curved boundary treatment in
lattice Boltzmann method. NASA/CR-2000-209854. ICASE report
No. 2000-6, February 2000.
[22] Lee T, Lin C-L. A characteristic Galerkin method for discrete
Boltzmann equation. J Comput Phys 2001;171:33656.
[23] Li Y, LeBoeuf EJ, Basu PK. Least-squares nite-element lattice
Boltzmann method. Phys Rev E 2004;69:065701-11-4.
[24] Bhatnagar P, Gross E, Krook M. A model for collision process in
gases. I: small amplitude processes in charged and neutral one-
component system. Phys Rev 1954;94:51125.
[25] B. Chopard, P. Luthi, S. Marconi, A lattice Boltzmann model for
wave and fracture phenomena, 1998. Available from: cond-mat/
98122201.
[26] Kwon YW, Bang H. The nite element method using Matlab. 2nd
ed. Boca Raton, FL: CRC Press; 2000.
[27] Chopard B. A cellular automata model of large-scale moving objects.
J Phys A Math Gen 1990;23:167187.
0 0.2 0.4 0.6 0.8 1
0.025
0.02
0.015
0.01
0.005
0
0.005
0.01
0.015
0.02
0.025
Time (no unit)
D
i
s
p
l
a
c
e
m
e
n
t

(
n
o

u
n
i
t
)
CAFEM Coupled
FEM
Fig. 12. Displacement at the center of a vibrating membrane under a
sinusoidal force. FEM and coupled CAFEM solutions are compared.
670 Y.W. Kwon, S. Hosoglu / Computers and Structures 86 (2008) 663670

Anda mungkin juga menyukai