7/30/2019 Lbm Simulation

1/22

International Journal of Numerical Methods for Heat & Fluid FlowEmerald Article: Lattice Boltzmann method simulation gas slip flow in longmicrotubes

Haibo Huang, T.S. Lee, C. Shu

Article information:

To cite this document: Haibo Huang, T.S. Lee, C. Shu, (2007),"Lattice Boltzmann method simulation gas slip flow in long

microtubes", International Journal of Numerical Methods for Heat & Fluid Flow, Vol. 17 Iss: 6 pp. 587 - 607

Permanent link to this document:

http://dx.doi.org/10.1108/09615530710761225

Downloaded on: 15-12-2012

References: This document contains references to 23 other documents

Citations: This document has been cited by 9 other documents

To copy this document: permissions@emeraldinsight.com

This document has been downloaded 759 times since 2007. *

Users who downloaded this Article also downloaded: *

Haibo Huang, T.S. Lee, C. Shu, (2007),"Lattice Boltzmann method simulation gas slip flow in long microtubes", International

ournal of Numerical Methods for Heat & Fluid Flow, Vol. 17 Iss: 6 pp. 587 - 607

http://dx.doi.org/10.1108/09615530710761225

Haibo Huang, T.S. Lee, C. Shu, (2007),"Lattice Boltzmann method simulation gas slip flow in long microtubes", International

ournal of Numerical Methods for Heat & Fluid Flow, Vol. 17 Iss: 6 pp. 587 - 607

http://dx.doi.org/10.1108/09615530710761225

Haibo Huang, T.S. Lee, C. Shu, (2007),"Lattice Boltzmann method simulation gas slip flow in long microtubes", International

ournal of Numerical Methods for Heat & Fluid Flow, Vol. 17 Iss: 6 pp. 587 - 607

http://dx.doi.org/10.1108/09615530710761225

Access to this document was granted through an Emerald subscription provided by INDIAN INSTITUTE OF TECHNOLOGY AT IN

For Authors:

f you would like to write for this, or any other Emerald publication, then please use our Emerald for Authors service.

nformation about how to choose which publication to write for and submission guidelines are available for all. Please visit

www.emeraldinsight.com/authors for more information.

About Emerald www.emeraldinsight.com

With over forty years' experience, Emerald Group Publishing is a leading independent publisher of global research with impact in

7/30/2019 Lbm Simulation

2/22

Lattice Boltzmann methodsimulation gas slip flow

in long microtubesHaibo Huang, T.S. Lee and C. Shu

Fluid Division, Department of Mechanical Engineering,National University of Singapore, Singapore

Abstract

Purpose This paper aims to examine how using lattice Boltzmann method (LBM) aids the study ofthe isothermal-gas flow with slight rarefaction in long microtubes.

Design/methodology/approach A revised axisymmetric lattice Boltzmann model is proposed tosimulate the flow in microtubes. The wall boundary condition combining the bounce-back andspecular-reflection schemes is used to capture the slip velocity on the wall. Appropriate relationbetween the Knudsen number and relax-time constant is defined.

Findings The computed-slip velocity, average velocity and non-linear pressure distribution alongthe microtube are in excellent agreement with analytical solution of the weakly compressibleNavier-Stokes equations. The calculated-friction factors are also consistent with available experimentaldata. For simulations of slip flow in microtube, LBM is more accurate andefficient than DSMC method.

Research limitations/implications The laminar flow in circular microtube is assumed to be

axisymmetric. The present LBM is only applied to the simulation of slip flows (0.01,

Kn0,

0.1) inmicrotube.

Practical implications Lattice-BGK methodis a very usefultool to investigatethe micro slipflows.

Originality/value A revised axisymmetric D2Q9 lattice Boltzmann model is proposed to simulatethe slip flow in axisymmetric microtubes.

Keywords Flow, Numerical analysis, Laminar flow, Fluid mechanics, Approximation theory

Paper type Research paper

1. IntroductionMicro-electro-mechanical-systems (MEMS) devices with dimensions ranging from100 to 1m have found many applications in engineering and scientific researches(Gad-el-Hak, 1999). The fast development of these devices motivated the study of thefluid flow in MEMS (Arkilic et al., 1997). MEMS are often operated in gaseousenvironments where the molecular mean free path of the gas molecules could be the

same order as the typical geometric dimension of the device. Hence, the dynamicsassociated with MEMS can exhibit rarefied phenomena and compressibility effects(Arkilic etal., 1997). Usually, the Knudsen numberKn are used to identify the effects.Knisthe ratio of the mean freepath l to the characteristic lengthL. Generally speaking, thecontinuum assumption for Navier-Stokes (NS) equations may break down ifKn . 0.01.For a flow case0.01 , Kn , 0.1, a slipvelocitywould appear in the wallboundary. Thevalue of 0.1 , Kn , 10 are associated with a transition flow regime. In the slip-flowregime, by introducing a slip velocity at the solid boundary the NS solver can still beused. In the transition regime, theconventional flow solver based on the NS equations isno longer applicable because the rarefaction effect is critical (Lim et al., 2002).

The current issue and full text archive of this journal is available atwww.emeraldinsight.com/0961-5539.htm

Simulation gasslip flow in long

microtubes

587

Received 8 November 2005Revised 20 February 2006

Accepted 8 March 2006

International Journal of NumericalMethods for Heat & Fluid Flow

Vol. 17 No. 6, 2007

pp. 587-607q Emerald Group Publishing Limited

0961-5539

DOI 10.1108/09615530710761225

7/30/2019 Lbm Simulation

3/22

Many analytical studies of rarefied flow in microchannel have been carried out sincethe 1970s. An important analytical and experimental study of gaseous flow intwo-dimensional (2D) microchannels was carried out by Arkilic et al. (1997). Through aformal perturbation expansion of the NS equations under an assumption of 2Disothermal flow, the study demonstrates the relative significance of the contribution ofcompressibility and rarefied effects and good agreements between the analytical andexperimental studies were observed.

There are also some analytical studies about rarefied flow in circular microtubes.Analytical studies of Prudhomme et al. (1986) and Van den Berg et al. (1993)demonstrated non-constant pressure gradients but their analysis did not incorporaterarefied behavior and the analysis is only one-dimensional (1D) perturbation solutionof the NS equations. Based on assumption of isothermal flow, Weng et al. (1999)obtained the analytical solution for rarefied gas flow in long-circular microtubes. Someexperiments were also carried out to measure the friction constant C f*Re inmicrotubes, which is not equal to 64 as the theoretical prediction for fully developedincompressible flow (Choi et al., 1991; Yu et al., 1995).

In addition to the above analytical and experimental investigations, there are manynumerical studies on rarefied gas behavior in microchannel. Through introducing aslip velocity at the solid boundary, Beskok and Karniadakis (1993) presented numericalsolutions of the NS and energy equations for flows with slight rarefaction. Forsimulations microflow, the direct simulation Monte Carlo method (DSMC) (Bird, 1994)are more popular because the approach is valid for the full range of flow regimes(continuum through free molecular). However, very large-computational effort is

required in the DSMC simulations since the total number of simulated particles isdirectly related to the number of molecules.Besides, numerical solution of NS equation and DSMC, the lattice Boltzmann

method (LBM), which based on meso-scale level and has no continuum assumption,was also applied to simulate the microflows (Lim et al., 2002; Nie et al., 2002).

For LBM simulation micro flow, the boundary condition and correlating relax time twith Kn are important. Nie et al. (2002) simulated a 2D-microchannel flow withbounce-back boundary treatment. However, in the study, a parameter to define Kn wasobtained empirically. Lim et al. (2002) simulated microchannel flow and obtained goodresults with specular and extrapolation boundary treatments. They linked the twiththe molecular free-mean path l by an assumption ofl tdx. Succi (2002) and Tanget al. (2004) showed that a slip velocity on the wall can be captured by using acombination of the bounce-back and specular-reflection conditions. Although the valueof the slip velocity may be highly dependent on the choice of the bounce-back

probability b, the boundary condition is easy to implement. For simplicity, in ourstudy, this boundary condition is applied to capture the correct velocity slip at the wall.

Previous LBM study of microflow is only concentrated in microchannel. Here, wewould like to propose a revised axisymmetric LBM for axisymmetric flows in microtubes.

It is sure that 3D LBM can directly handle the axisymmetric flow problems (Huanget al., 2006). However, for an axisymmetric flow problem, directly 3D simulation is notso efficient. To simulate the problem more efficiently, Halliday et al. (2001) proposed anaxisymmetric D2Q9 model for the axisymmetric flow problems and it seems verysuccessful for simulation steady flow in straight tube. The main idea of the model isinserting several spatial and velocity dependent source terms into the microscopic

HFF17,6

588

7/30/2019 Lbm Simulation

4/22

evaluation equation for the lattice fluids momentum distribution. However, it is foundthat some terms relative to the radial velocity are missing in the axisymmetric D2Q9model of Halliday et al. (2001) and later-developed model (Lee et al., 2005). Although theterms may not affect simulations of the flows in straight circular pipe, they would leadto large error for simulation the constricted or expanded pipe flows.

The main aim of the present paper is to derive a correct D2Q9 axisymmetric modelto numerically investigate the flow in microtubes. We also would like to compare theaccuracy and efficiency between the LBM and the DSMC when simulate the slip flow inmicrotubes.

The structure of this paper is as follows. Firstly, a revised axisymmetric LBM is

proposed and the implementation of the LBM and boundary condition is discussed. Thenthe LBM is applied to simulate the slip flow in microtubes for cases Kn 0.1, 0.05, 0.025with different inlet/outlet pressure ratio. The slip velocity, bulk velocity and pressuredistribution along the tube are compared with analytical solution (Weng et al., 1999) indetail. The friction factors are compared with the available experimental data. Finally, theefficiency and accuracy comparisons between DSMC and LBM are carried out.

2. LBM model and boundary condition2.1 LBM modelIn this part, an axisymmetric D2Q9 model is proposed to simulate the axisymmetricflows in a long microtube. The derivation of our model is illustrated in Appendix 1.

Here, we consider the problems of the laminar internal flow of a weaklycompressible, isothermal flow in circular pipe with an axis inx direction. The geometryis shown in Figure 1. For the axisymmetric flow, the azimuthal velocity u

wand w

coordinate derivatives vanish from the continuity and NS equations. The full 2Dtime-invariant constant viscosity NS equations for a compressible fluid, ignoring bodyforce, are (in the pseudo-Cartesian coordinates (x, r)):

r uu

x v u

r

2 p

x m

2u

x 2 1

r

u

r

2u

r2 1

3

x

u

x vr

vr

!1

r uv

x v v

r

2 p

r m

2v

x 2 1

r

v

r

2v

r22

v

r2 1

3

r

u

x vr

vr

!2

The continuity equation is given by:

ru

x rv

r rv

r 0

3

Figure 1.Geometry of circular tube

and LBM D2Q9 model

x

r

1

2

3

48

56

7

Simulation gasslip flow in long

microtubes

589

7/30/2019 Lbm Simulation

5/22

The equation of state for an ideal gas is given by:

p rRT 4In above equations, u and v are the stream-wise and wall-normal components of velocityu, m is the molecular viscosity,ris the density,p is the pressure andRis the specific gasconstant. In equations (1) and (2), we have assumed a Stokes continuum hypothesis forthe second coefficient of viscosity (Landau and Lifschitz, 1987).

Our present axisymmetric D2Q9 model is proposed to simulate the microtube flowsdescribed by above equations. Among different lattice Boltzmann equation (LBE)models in application, the lattice Bhatnagar-Gross-Krook (LBGK) model is the simplestone because it only has one scalar relaxation parameter and a simple equilibriummomentum distribution function. Here, our axisymmetric LBM is derived from LBGKD2Q9 model. In our axisymmetric D2Q9 model, the nine discrete velocities of our modelare defined as following:

ei 0; 0 i 0cosi2 1p=2; sini2 1p=2c i 1; 2; 3; 4ffiffiffi

2p cosi2 5p=2 p=4; sini2 5p=2 p=4c i 5; 6; 7; 8

8>>>: 5

where c dx/dt, and in our studies c 1. dx and dtare the lattice spacing and time stepsize, respectively.

In our model, fi(x, r, t) is the distribution function for particles with velocity ei atposition (x, r) and time t. The macroscopic density rand momentum ru are defined as:

X8i0

fi r;X8i0

fieia rua 6

The equilibrium distributionfieq of D2Q9 model (Qianetal., 1992) is defined by equation(7):

feqi x; r; t vir 1 ei u

c2s ei u

2

2c4s2

u2

2c2s

!i 0; 1; 2; . . . ; 8 7

where, cs c=ffiffiffi

3p

;v0 4=9; vi 1=9; i 1; 2; 3; 4;vi 1=36; i 5; 6; 7; 8.The two main steps of lattice BGK model are collision and streaming. In the

collision step, a group of calculations (8) and (9) are implemented:

fnei

fi

x; r; t

2f

eqi

x; r; t

8

fi x; r; t feqi x; r; t 12

1

t

fnei dth1i d2th2i 9

In above equations, fnei is the non-equilibrium part of distribution function. fi is the

post-collision distribution function. h1i and h2i are the source terms added into

the collision step, which can be calculated through below equations (10) and (11),respectively. The brief derivation of the equations is illustrated in Appendix 1:

h1i 2virur

r10

HFF17,6

590

7/30/2019 Lbm Simulation

6/22

h2i

vi

2rrr

3 xruxur rrurur

3virn

reixrux eirrur2 3virn

r2ureir

2 3viruxur

reix

urur

reir

2 12 tvi 1

rxrureix 2

rur

r2eir rr

ur

reir

11

In above formulas, the relax time constant tand the fluid kinetic viscosity nsatisfiesthe below equation:

n c2sdtt2 0:5 12For the microflow simulation, the tshould be related to the Knudsen number.

In the streaming step, the new distribution function value obtained from equation (9)would propagate to neighbour eight lattices. That procedure can be represented by thefollowing equation (13):

fix eix; dt; r eir; dt; t dt fi x; r; t 13For the velocity derivations in equation (11), the terms rux xur, xux and rur canall be obtained through equation (14) with a x,b r; a b x; a b r,respectively:

rnb

ua

a

ub

2 121

2t X8

i0f1

ie

iae

ib

2 12 12t

X8i0

fnei eiaeib o1214

For the term rux in equation (11), it is equal to (rux xur) 2 xur. Since,(rux xur) can be easily obtained by equation (14), only value ofxur is left unknownto determine rux. Here, we recourse to finite difference method to obtain xurat latticenode (i, j), which can be calculated by equation (15):

xuri;j uri1;j 2 uri21;j

2dx15

The values of rux

xur,xux, rur, rux and xur for the lattice nodes which just

on the wall boundary can also be calculated from equations (14) and (15). Obtainingthese values for lattice nodes on the periodic boundary is also easy. However, toobtain these values for the nodes on the inlet/outlet pressure-specified boundary, thesevalues are extrapolated from those of the inner nodes.

2.2 Knudsen number and boundary conditionCorrelating the parameter twith Kn is important for LBM application in simulationmicro-flows. (Nie et al., 2002; Tang et al., 2004) Here, an expression (Tang et al., 2004)between Kn and twhich based on the gas kinematics is used in our simulation but wederived it in a simpler way in the following.

Simulation gasslip flow in long

microtubes

591

7/30/2019 Lbm Simulation

7/22

Form the kinetic theory of gases, the density can be determined by:

r mpkBT

16

where m represents the molecular mass and kB is the Boltzmann constant. On the otherhand, in LBM, the density and pressure have the relationship (Qian et al., 1992):

r pc2s

17Hence, in LBM, we have:

kBT

m c2s 18

For an ideal gas modeled as rigid spheres, the mean free path l is related to theviscosity nas:

n 0:5nml 19where the mean velocity of the molecular vm

ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ffiffiffiffiffi8KBT=pm

p. Hence, form

equations (12) and (19), we get:

Kn lD

2nvmD

ffiffiffiffip

6

r t2 0:5ND

or t KnNDffiffiffiffiffiffiffiffip=6

p 0:5 20where ND is the lattice number in the tube diameter, Kn is local Knudsen number.

Since, the mean free path is inversely proportional to the pressure, the local Kn can becalculated by:

Kn Knopopx; r 21

where Kno and po are the Kn and the pressure at the outlet. So, in equation (20),t is variable along the microtube and the corresponding n can be obtained fromequation (12).

Another important issue about using LBM to simulate the micro flows is wallboundary condition. For wall boundary condition, bounce-back scheme is usually usedto realize non-slip boundary condition when simulate continuum flow. On the otherhand, specular reflection scheme (Lim et al., 2002) can be applied to free-slip boundarycondition where no momentum is to be exchanged with the wall along the tangential

component. For real gas flow in microtubes, a combination of the two schemes isconsidered here. To describe boundary condition treatment, a wall V is completelyspecified. For a point xx [ V, n is the inward unit normal vector of the wall. Afterstreaming step implemented, the unknown distribution functions of fix; t; ei n . 0,can be evaluated by Succi (2002) and Tang et al. (2004):

fix; t bfjx; t 12 bfkx; t 22wherefj(x,t) is the distribution function in ej direction, where ei2 ej 2ei, andfkx; tis the distribution function in ek direction, where ei2 ek 2n. b is the bounce-backprobability chosen as 0.7.

HFF17,6

592

7/30/2019 Lbm Simulation

8/22

For the inlet/outlet boundary conditions, the pressure is specified and thecorresponding velocity value in these boundaries is extrapolated from the next innernodes (Fang et al., 2002). Hence, the equilibrium part of distribution function can bedetermined and the non-equilibrium part of distribution function can be obtainedthrough extrapolation (Fang et al., 2002). So, the collision step for boundary nodes canbe implemented normally as inner nodes.

The axisymmetric boundary condition is also applied in most simulations here. Inmost simulations, the computational domain is an axisymmetric plane above the axis.To implement this boundary condition, an extra row of grids below the axis is added.The variables in the grid of this row can be evaluated from that of its symmetric node.

For example, ifj 2 and j 1 represents the row index of the axis and the extra row,velocities in the extra grids can be obtained by uxi;1 uxi;3; uri;1 2uri;3, andthe corresponding source terms in equations (10) and (11) can be evaluated ashi;1 hi;3. In this way, the collision and streaming steps for lattices in the extra rowcan be implemented as that of inner lattices.

3. Results and discussion3.1 Distributions of pressure and velocityIn our simulation, the radius is represented by 11 lattice nodes (ten lattice space) andthe length of the tube is 20 times of the diameter except for specially noted cases. In allof cases, the Mach number in tube is very low. Even for case of Pr 3.0, maximumMach number in tube is M 0.15/cs ! 1, which satisfy the requirement of ouraxisymmetric D2Q9 model. The stream-wise momentum accommodation coefficients

1 has been used for almost all engineering calculations (Weng et al., 1999).

Therefore, we take s 1 throughout the paper.Figures 2 and 3 show the axial and radial-velocity distribution along the tube

(Pr 2, Kno 0.1), respectively. The u, v velocity contour are also shown in Figures 2and 3, respectively. From the Figure 2, we can see that the axial-velocity profile isparabolic type and the slip velocity at the wall and the central velocity increase towardthe exit. Owing to the pressure decreasing, the density of gas also decreases along thetube. To satisfy mass conservation, the average velocity must increase toward the exit.In Figure 3, the magnitude of the radial velocity is much smaller than that of axialvelocity. These results are consistent with previous studies on microchannel (Arkilicet al., 1997; Lim et al., 2002).

The pressure distribution along the tube predicted from the first slip boundarycondition is illustrated in equation (23), which is originally given by Weng et al. (1999).The derivation of present expression is illustrated in Appendix 2:

~p~x 28Kno ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ffiffiffiffiffiffiffi8Kno2 1 16Kno~x Pr2 16KnoPr12 ~x

q23

In equation (23), ~p is the pressure normalized by outlet pressure, ~x x=L, L is the tubelength. Pr is the ratio of the inlet and outlet pressure.

The pressure drop along the tube which deviate from linear pressure drop fordifferent Pr with the same outlet Knudsen number Kno 0.1 are shown in Figure 4.When Princrease, the compressibility effect within the tube is also increase, results ina larger deviation from the linear pressure distribution. In Figure 4, our results agreewell with equation (23).

Simulation gasslip flow in long

microtubes

593

7/30/2019 Lbm Simulation

9/22

The pressure drops along the tube for different outlet Kno are shown in Figure 5.Compared with the analytical solution equation (23), the results of LBM is quite good.Figure 5 shows that the larger Kno, the smaller the deviation from the linear-pressuredistribution. It seems that the rarefaction effect (indicate by Kno) can decrease thecurvature in the pressure distribution which caused by the compressibility effect.Maybe that means the compressibility effect and the rarefaction effect on the pressuredistribution are contradictory.

The Knudsen numbers along the stream-wise direction are shown in Figure 6. Kn isa function of the local pressure. With the decreasing pressure along the tube, theKnudsen number increases and reaches its maximum value at the outlet. For differentoutlet Kno, the slope ofKn curve along the tube is different. For smaller Kno, the slopeof Kn curve is smaller although Pr is same.

In Figure 7, the variation of slip velocity along the microtube wall is shown. Firstly,we obtained the analytical solution of slip velocity from results of Weng et al. (1999)(Appendix 2). Equation (B2) is can be normalized by the central velocity at outlet Uoc:

Ux; rUoc

d~p=d~x

d~p=d~xo0:25 Kn2 r2=D2

0:25 Kno 24

where d~p=d~x is the non-dimensional pressure gradient and the d~p=d~xo means thepressure gradient at exit, which can be referred to equation (B7) in Appendix 2.

Figure 2.Axial-velocitydistributions in the tube

0.005

0.01

0.015

0.02

0.025

0.03

0.035

0.04

0

0.10.2

0.30.4

0.50.6

0.70.8

0.91

x/L 00.1

0.20.3

0.40.5 r/D

u

HFF17,6

594

7/30/2019 Lbm Simulation

10/22

Figure 3.Radial-velocity

distributions along thetube

1E-05

0

1E-05

2E-05

3E-05

4E-05

00.1

0.20.3

0.40.5

0.60.7

0.80.9

1

x/L 00.1

0.20.3

0.40.5

r/D

v

Figure 4.Pressure distribution

along the tube for differentPr (Kno 0.1)

analytical solution

Pr=3

Pr=2.5

Pr=2

0.20

0.18

0.16

0.14

0.12

0.10

0.08

0.06

0.04

0.02

0.00

0.0 0.1 0.2 0.3 0.4 0.5

x/L

0.6 0.7 0.8 0.9 1.0

(p-pincomp

)/po

Simulation gasslip flow in long

microtubes

595

7/30/2019 Lbm Simulation

11/22

Hence, the analytical solution for slip velocity on the wall and average velocity inmicrotube are equations (25) and (26), respectively:

UslipxUoc

d~p=d~x

d~p=d~xoKn

0:25 Kno 25

Figure 5.Pressure distributionalong the tube for differentKnudsen number (Pr 2)

analytical solution

Kno=0.025

Kno=0.025

Kno=0.05

Kno=0.05

Kno=0.1

Kno=0.1

0.08

0.07

0.06

0.05

0.04

0.03

0.02

0.01

0.00

0.0 0.1 0.2 0.3 0.4 0.5

x/L

0.6 0.7 0.8 0.9 1.0

(p-pinco

mp

)/po

Figure 6.Local Kn distributionalong the tube for differentKno (Pr 2)

analytical Kn=Knopo/p

Kno=0.025

Kno=0.05

Kno=0.1

numerical0.08

0.09

0.10

0.07

0.06

0.05

0.04

0.03

0.02

0.01

0.0 0.1 0.2 0.3 0.4 0.5

x/L

Kn

0.6 0.7 0.8 0.9 1.0

HFF17,6

596

7/30/2019 Lbm Simulation

12/22

UavxUoc

d~p=d~x

d~p=d~xo1=8 Kn1=4 Kno 26

From equation (25), we can see that since the local Knudsen number increases and theslope of pressure drop also increases along the tube, the slip velocity on the wall wouldincrease along the microtube. Figure 8 shows the average velocity variations along thestream-wise direction. The average velocity increases as the flow proceeds down thetube since density decrease along the microtube. In Figures 7 and 8, both the slip

Figure 8.Average axial velocity Uavalong the tube for different

Kno (Pr 2)

Kn=0.025

analytical solution

Kn=0.05

Kn=0.1

0.25

0.30

0.35

0.40

0.45

0.50

0.55

0.60

0.65

x/L

0.0 0.1 0.2

Uav

/Uoc

0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0

Figure 7.Slip velocity in wall alongthe tube for different Kno

(Pr 2)

Kno=0.025

Kno=0.025

analytical solution

Kno=0.05

Kno=0.1

0.00

0.05

0.10

0.15

0.20

0.25

0.30

Us

/Uoc

0.0 0.1 0.2 0.3 0.4 0.5

x/L

0.6 0.7 0.8 0.9 1.0

Simulation gasslip flow in long

microtubes

597

7/30/2019 Lbm Simulation

13/22

velocity on wall and local bulk velocity along the microtube agree well with theanalytical solution equations.

3.2 Mass flow rate and normalized friction constantThe effect of rarefaction on mass flow rate is investigated by comparison of the LBEresult with analytical predictions. The non-dimensional mass flow rate ~Q can beexpressed as a function of pressure ratio (Appendix 2):

~Q _q

_q

continuum

1 16KnoPr

1

27

In Figure 9, the non-dimensional mass flow rate computed by the LBE method forKno 0.1 is compared with the first order analytical prediction equation (27). For allcases, slip effects become less pronounced with increasing pressure ratio. The LBEresults agree well with analytical results and the deviation is less than 4 percent.

Then in Figure 10, the friction factors predicted by present LBM simulations arecompared with experimental results of Kim et al. (2000). The theoretical friction constant(C0 f*Re 64) for fully developed incompressible flow is used to normalize frictionconstant C f*Re. The microtubes used in the experiment are also shown in Figure 10.Here, our numerical data were taken from results of cases Kno 0.013 with differentinlet/outlet pressure ration. In these cases, for Kn l/D 0.013, the correspondingsimulated diameters D of microtubes for Nitrogen, Argon and Helium are listed inTable I. The diameters of our simulation are all close to that of correspondingexperimental facility. Hence, our numerical results are valid to compare with the

experimental data. In Figure 10, the normalized friction constant C* obtained by LBMranges from 0.80 to 0.86, which agree well with the experiment data.

Besides, experiment of Kim et al. (2000) and Choi et al. (1991) also found that fornitrogen flow in microtube with diameters smaller than 10mm, C f*Re 53.Another experiment conducted by Yu et al. (1995) concluded that C f*Re 50.13

Figure 9.Mass flow rate normalizedo non-slip mass flow rate

as a function of Pr atKno 0.1

analytical solution

LBM

1.30

1.35

1.40

1.45

1.50

1.55

1.60

1.6 1.8 2.0 2.2 2.4 2.6 2.8 3.0

Q~

Pr

HFF17,6

598

7/30/2019 Lbm Simulation

14/22

for laminar nitrogen flow in microtubes with diameter 19 mm. In Figure 10, it wasobserved that our numerical data are also in consistent with their experimental results(Choi et al., 1991; Yu et al., 1995).

3.3 Comparison with DSMCTo demonstrate the efficiency of the LBM, we compared the accuracy and efficiency ofthe LBM and DSMC. It is well known that DSMC is the most popular model forsimulation of micro flows. DSMC is a particle-based method proposed by Bird (1994).Unlike the molecular dynamics (MD) method which takes each individual molecule intoconsideration, DSMC method assumes that a group of molecules have the same

properties such as velocity and temperature which can be obtained by statisticalanalysis. In this way, the computational effort can be greatly reduced compared withthe MD method (Bird, 1976; Bird, 1994). Here, the developed DSMC code (Mao et al.,2003) was used to simulate the slip flow in microtubes.

In the DSMC simulation, the working gas is nitrogen. The physical geometry is200mm long and radius of the tube is 2.5mm. The computational region is anaxisymmetric plane divided into 400 30 sampling cells and each cell contains foursubcells. The total number of simulated particles is about 4.8 105. That meansnearly 40 particles in a sampling cell (Mao et al., 2003). In this part, the case of

Kno 0.0134 and Pr 2.5 was simulated.

Gas (105 Pa) Nitrogen Argon Helium

Mean free path (nm) 67 72 196Diameter of tube (mm) 5.2 5.5 15.0

Table I.Simulated diameter of

microtubes for differentgas flow (Kno 0.013)

Figure 10.Normalized friction

constant C* of gas flow inmicrotube as a function of

Re (Kno 0.013)0.0

0.2

0.4

0.6

0.8

1.0

1.2

0.1 1

experiment N2 5mm (Kim et al. 2000)

experiment Ar 5mm (Kim et al. 2000)

experiment He 15mm (Kim et al. 2000)

numerical LBM

10 100

Normalized

frictionconst.

C*

Re

Simulation gasslip flow in long

microtubes

599

7/30/2019 Lbm Simulation

15/22

In the LBM simulation, the uniform square lattices 801 21 is used to simulate thesame microtube flow. For this case, if the computational domain is an axisymmetricplane and the axisymmetric boundary condition is applied, the calculation is unstable.However, when the computational domain is a whole plane passing through the axis,the calculation is stable with the slip wall boundary condition. Hence, here thecomputational domain is a whole plane passing through the axis and the diameter isrepresented by 21 lattice nodes.

The present DSMC and LBM calculations were performed on a single-CPU of thecomputer Compaq ES40 supercomputer. To make the efficiency comparison, the sameconvergence criterion was set as:

i

X kuxi; t2 uxi; t2 1kkuxi; tk

, 1026 28

The velocity field error is measured by uwhich is defined as:

u iPuri2 uari2

i

Pu2ari

29

where ua(ri) is the analytical solution obtained by Weng et al. (1999) and ri is the meshpoint at intersection x/L 0.375 where the microflow is supposed to be in fullydeveloped region.

The efficiency and accuracy comparison is listed in Table II. The mesh or cell number

is comparable for LBM and DSMC simulations. However, since DSMC still has tosimulate 4.8 105 particles, it used much larger memory than LBM in the simulation.To obtain the well-converged results, DSMC takes more CPU time than LBM.

The velocity profiles at intersection x/L 0.375 obtained by analytical solution(Weng et al., 1999), LBM and DSMC are shown in Figure 11. The velocity U isnormalized by outlet Uoc ris normalized by the diameter. Compared with the analyticalsolution, the result of LBM seems more accurate than that of DSMC.

4. ConclusionIn this paper, a revised axisymmetric D2Q9 model was applied to investigate gaseousslip flow with slight rarefaction through long microtubes. With limit of small Machnumber, this axisymmetric LBGK model successfully recovered the weaklycompressible NS equation in the cylindrical coordinates through Chapman Enskogexpansion (refer to the Appendix 1). For the additional source term in our model, most

velocity gradient terms can be obtained from high order momentum of distributionfunction, which is consistent with the philosophy of the LBM. For the slip wall boundarycondition, the wall boundary condition combined the bounce-back andspecular-reflection scheme was applied for microtube flows withKno in range(0.01, 0.1).

Method CPU time (s) Mesh or cells Memory (M) u

LBM 4.52 102 1.6 104 9.2 4.31 1024

DSMC 3.22 104 1.2 104 31.4 3.3 1023

Table II.Efficiency and accuracycomparison (LBM andDSMC) (Kno 0.0134,Pr 2.5)

HFF17,6

600

7/30/2019 Lbm Simulation

16/22

The distributions of pressure, the slip velocity and the average velocity along themicrotube all agree well with the analytical results. The friction factors are comparedwith experimental results and good agreements are also observed. The axisymmetricLBGK model was successfully applied to simulate the laminar flow in microtubes.

Through comparison, it was found that our LBM is more accurate and efficient thanDSMC when simulate the slip flow in microtube. Although present LBM is only appliedto the slip flow simulation (0.01 , Kn0, 0.1) in microtube, the LBM may be extendedto study the transition flow or higher Knudsen number cases in the future.

References

Arkilic, E.B., Schmidt, M.A. and Breuer, K.S. (1997), Gaseous slip flow in long micro-channels,J. MEMS, Vol. 6 No. 2, pp. 167-78.

Beskok, A. and Karniadakis, G. (1993), Simulation of heat and momentum transfer inmicro-geometries, AIAA Paper 93-3269.

Bird, G.A. (1976), Molecular Gas Dynamics, Clarendon, Oxford.

Bird, G.A. (1994), Molecular Gas Dynamics and the Direct Simulation of Gas Dynamics,Clarendon, Oxford.

Choi, S.B., Barron, R. and Warrington, R. (1991), Fluid Flow and heat transfer in microtubes,Micromechanical Sensors, Actuator and Systems, ASME DSC, Vol. 32, pp. 123-34.

Fang, H.P., Wang, Z.W., Lin, Z.F. and Liu, M.R. (2002), Lattice Boltzmann method for simulatingthe viscous flow in large distensible blood vessels, Physical Review E, Vol. 65, 051925.

Gad-el-Hak, M. (1999), The fluid mechanics of microdevices-the Freeman scholar lecture,J. Fluids Eng., Vol. 121 No. 1, pp. 5-33.

Halliday, I., Hammond, L.A., Care, C.M., Good, K. and Stevens, A. (2001), Lattice Boltzmannequation hydrodynamics, Phys. Rev. E, Vol. 64, 011208.

He, X. and Luo, L. (1997), Lattice Boltzmann model for the incompressible Navier-Stokesequation, J. Stat. Phys., Vol. 88 Nos 3/4, pp. 927-43.

Figure 11.Velocity profiles at

x/L 0.375 obtained byanalytical solution, LBM

and DSMC

0.6Weng et al.

LBM

DSMC0.5

0.4

0.3U

0.2

0.5 0.4 0.3 0.2 0.1 0.0 0.1 0.2

r/D

0.3 0.4 0.5

0.1

0.0

Simulation gasslip flow in long

microtubes

601

7/30/2019 Lbm Simulation

17/22

Huang, H., Lee, T.S. and Shu, C. (2006), Lattice-BGK simulation of steady flow through vasculartubes with double constrictions, Int. J. Num. Methods for Heat & Fluid Flow , Vol. 16 No. 2,pp. 185-203.

Kim, M.S., Araki, T., Inaoka, K. and Suzuki, K. (2000), Gas flow characteristic in microtubes,JSME International Journal Series B, Fluid and Thermal Engineering, Vol. 43 No. 4,pp. 634-9.

Landau, L.D. and Lifschitz, E.M. (1987), Fluid Mechanics, 2nd ed., Pergamon, Oxford.

Lee, T.S., Huang, H. and Shu, C. (2005), An axisymmetric incompressible Lattice-BGK model forsimulation of the pulsatile flow in a circular pipe, Int. J. Numer. Meth. Fluids, Vol. 49 No. 1,pp. 99-116.

Lim, C.Y., Shu, C., Niu, X.D. and Chew, Y.T. (2002), Application of lattice Boltzmann method tosimulate microchannel flows, Phys. Fluids, Vol. 14 No. 7, pp. 2299-308.

Mao, X.H., Shu, C. and Chew, Y.T. (2003), Numerical and theoretical study of a micro tube flow,Int. J. Nonlinear Sciences and Numer. Simulation, Vol. 4 No. 2, pp. 187-200.

Nie, X., Doolen, G.D. and Chen, S.Y. (2002), Lattice Boltzmann simulation of fluid flows inMEMS, J. Stat. Phys., Vol. 107, pp. 279-89.

Prudhomme, R., Chapman, T. and Bowen, J. (1986), Laminar compressible flow in a tube,Appl. Sci. Res., Vol. 43, pp. 67-74.

Qian, Y.H., dHumieres, D. and Lallemand, P. (1992), Lattice BGK models for Navier-Stokesequation, Europhys. Lett., Vol. 17, pp. 479-84.

Succi, S. (2002), Mesoscopic modeling of slip motion at fluid-solid interfaces with heterogeneouscatalysis, Phys. Rev. Lett., Vol. 89 No. 6, 064502.

Tang, G.H., Tao, W.Q. and He, Y.L. (2004), Lattice Boltzmann method for simulating gas flow in

microchannels, Int. J. Mod. Phys. C, Vol. 15 No. 2, pp. 335-47.Van den Berg, H., Tenseldam, C. and VanderGulik, P. (1993), Compressible laminar flow in a

capillary, J. Fluid Mech., Vol. 246, pp. 1-20.

Weng, C.I., Li, W.L. and Hwang, C.C. (1999), Gaseous flow in microtubes at arbitrary Knudsennumbers, Nanotechnology, Vol. 10, pp. 373-9.

Yu, D., Warrington, R., Barron, R. and Ameel, T. (1995), An experimental and theoreticalinvestigation of fluid flow and heat transfer in Microtubes, ASME/JSME ThermalEngineering Conference, Vol. 1, pp. 523-30.

Appendix 1. Brief derivation of the axisymmetric modelHere, we would show how continuity equation (A1) and momentum equation (A2) in thepseudo-Cartesian coordinates (x, r) can be recovered from our axisymmetric D2Q9 model:

r

t rub

xb 2rur

r A1

rua

t rubua

xb ruaur

r pxa2 m

2ua

x2b2 ma7 u m

r

ua

r2

ur

rdar

A2

where 7 u bub ur=r. Here, we adopt the Einstein convention that the same index appearstwice in any term, summation over the range of that index is implied. ua, ub is the velocity uxorur. a, b is x or r, xa, xb means x or r.

It is noticed that the small discrepancy between equations (1), (2) and equation (A2) isthe coefficient before term 7 u, in equation (1) that is 1/3, here in equation (A2) is unit.

HFF17,6

602

7/30/2019 Lbm Simulation

18/22

Since, we consider a slight rarefaction in long microtubes, which means the weakly compressibleflow, this term 7 u should be very small. The coefficient difference can be neglected. Ournumerical results also verified this opinion.

To recover above equations, the Chapman-Enskog expansion is applied. The evaluationequation to describe 2D flow in (x, r) pseudo-Cartesian coordinates is illustrated as equation (A3):

fix eixdt; r eirdt; t dt2fix; r; t 1t

feqi

x; r; t2fix; r; t hix; r; t A3

equation (A3) is similar to the evaluation equation in 2D (x, y) Cartesian coordinates except that asource term hi(x, r, t) was incorporated into the microscopic evaluation equation (Halliday et al.,

2001).Here, we introduce the following expansions (He and Luo, 1997):

fix eixdt; r eirdt; t dt X1n0

1 n

n!Dnfix; r; t A4

fi f0i 1f1i 1 2f2i t 11t 1 22t

b 11bhi 1h1i 1 2h2i

8>>>>>>>>>>>:

A5

where, 1 dt and D; t eibb. In equation (A5), there is no equilibrium hi term.Retaining terms up to O(1 2) in equations (A4) and (A5) and substituting into equation (A3)

results in equations in the consecutive order of the parameter 1:

O1 0 :f

0i 2f

eqi

t

0 A6

O1 1 : 1t eib1bf0i f1i

t2 h1i

0 A7

O1 2 : 2tf0i 121

2t

1t eib1bf1i

1

21t eib1bh1i

1

tf2

i2 h2

i 0 A8

The distribution function fi is constrained by equation (6) and the following equation (A9):

X8

i0f

mi

0; X

8

i0e i f

mi

0 for m . 0

A9

Note thatE2n1 0 for n 0,1,. . . where E(n) are the tensors defined asEn Paea1ea2. . .eanand:

X4i1

eiaeib 2dab A10

X8i5

eiaeib 4dab A11

Simulation gasslip flow in long

microtubes

603

7/30/2019 Lbm Simulation

19/22

X4i1

eiaeibeigeiz 2dabgz A12

X8i5

eiaeibeigeiz 4Dabgz2 8dabgz A13

where dab and dabgz are the Kronecker tensors, and:

Dabgz dabdgz dagdbz dazdbg A14With above properties of the tensor E(n), we have:

X8i0

eiaeibf0i r0uaub pdab A15

X8i0

eiaeibeik f0i

r0c2s djkdba djadbk djbdakuj A16

Mass conservation and h1i

Summing on i in equation (A7), we obtain at O(1):

1tr

brub

iXh

1i

A17

which motivates the following selection of h

1i when comparing with the target dynamics

(of equations (A1) and (A2)). To recover the continuity equation (A1), becauseP

ivi 1, thefollowing selection of h

1i is reasonable (Halliday et al., 2001):

h1i

2virurr

A18

Then, we proceed to O(1 2) now. Summing on i in equation (A8) gives:

2tri

X 12

1t eib1b

h1i2

i

Xh2

i 0 A19

With our target dynamic in view, we obtain the equation (A20):

i

X 12

1t eib1b

h1i 2i

Xh2i 0 A20

equation (A20) can also be rewritten as equation (A21):

i

Xh2

i 1

2i

X1t eib1bh1i

1

21t

i

X2virurr

24

35 2 1

2

1trur

r

!A21

HFF17,6

604

7/30/2019 Lbm Simulation

20/22

Momentum conservation and h2i

Multiplying equation (A7) with eia and summing over i, gives:

1trua 1bP0ab i

Xh1i eia 0 A22

where, P0ab X8i0

eiaeibf0i is the zeroth-order of momentum flux tensor. With P

0ab given by

equation (A15), using equation (A22) with a r, and substituting into (A21), we have acondition on h

2i :

Xh

2

i 1

2rb c

2

srdrbrubur A23

The error in previous model of Halliday et al. (2001) partly lies in their opinion about equation(A22). It seems that Halliday et al. simply regarded terms 1trua 1bruaub as the terms

Dtrua trua brubua ruaur=r. Unfortunately, that is not true.Multiplying equation (A8) with eia and summing over i gives:

2trua 12 12t

1bP

1ab 2

1

21t

i

Xeiah

1i

1bi

Xeiaeibh

1i

0@

1A

i

Xh

2i

eia A24

where,P1ab

i

Peiaeibf

1i is the first-order momentum flux tensor. With the aid of equations (A7)

and (A16), we have:

P1ab

iXeiaeibf

1i 2t

iXeiaeibD1tf

0i t

iXeiaeibh

1i

2ti

X1tP

0ab k

i

Xeiaeibeik f

0i

0@

1A

24

35 t

i

Xeiaeibh

1i

2ti

X1tP

0ab c2s dabjruj brua arub

24

35 t

i

Xeiaeibh

1i

7/30/2019 Lbm Simulation

21/22

Solving equation system (A23) and (A27), we can obtain the expression of h2i :

h2i vi

2rb c

2srdrb rubur

3vi rnr

rub21

rurdrb

eib2

rubur

reib

!

2 12 tvib rurr

eib

A28

The expression of h1i (equation (A18)), h

2i (equation (A28)) are successfully derived and the

continuity equation (A1) and NS equation (A2) can be fully recovered. In the model of Hallidayet al. (2001), the mainly missing terms are relative to ur. Although these terms may only slightlyaffect results of straight pipe flow, without these terms, the flows in constricted pipes cannot besimulated correctly.

Appendix 2The stream-wise velocity profile (first-order slip-flow model) in a long microtube with rarefactioneffect is given by Weng et al.(1999):

U x; r 2 r20

4m

p

x12

r

r0

2 2l

r0

" #A29

where l is the molecular mean free path, r0 is the radius of the microtube. Since, r0 D=2 andlocal Kn l=D, using equations (21) and (A29), we have:

Ux; r 2 D2

16m

dp

dx12 4

r

D

2 4Kno

~p

A30

where ~p p x; r =po .The pressure distribution in a long microtube is given by Weng et al. (1999) as:

S 2 8ffiffiffiffip

p 64p

S2in 16ffiffiffiffip

p Sin S2out 2 S2in

16ffiffiffiffip

p Sout 2 Sin !

~x

& '12

A31

where:

S Knoffiffiffiffip

p 21~p; Sin Kno

ffiffiffiffip

p 21Pr; Sout Kno

ffiffiffiffip

p 21;Pr pin=po;

and ~x x=L. L is the tube length. Hence, equation (A31) can also be rewritten as:

~p~x 28Kno ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi8Kno2 1 16Kno~x Pr2 16KnoPr

12 ~xq A32From equation (A32) we can see that gas flowing in a long microtube with a significant pressure

drop will also exhibit compressibility effects.The mass-flow rate is computed by multiplying equation (A29) by the density and integrating

across the tube. The dimensional mass-flow rate is given by Weng et al. (1999):

_q 2prr40

2m

p

x

1

4 2Kn

!A33

Hence, the dimensional mass flow rate at outlet of microtube is:

_q 2proD4po

16mL

~p

~x

o

1

8 Kno

!A34

HFF17,6

606

7/30/2019 Lbm Simulation

22/22

The non-dimensional pressure gradient along the tube can be calculated from equation (A32) as:

d~p

d~x 12Pr

2 16Kno12Pr2

ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi8Kno2 1 16Kno~x Pr2 16KnoPr

12 ~xq A35Hence, equation (A34) can also be written as:

_q pD4p2o

256mLRTPr2 2 1 16KnoPr2 1 A36

In addition, the mass-flow rate for the continuum gas (without the rarefaction effect) is:

_qcontinuum 2pD4

256mRT

p 2x

pD4p2o

256mRT

Pr2 2 1L

A37

Corresponding authorHaibo Huang can be contacted at: g0301108@nus.edu.sg

Simulation gasslip flow in long

microtubes

607

To purchase reprints of this article please e-mail: reprints@emeraldinsight.comOr visit our web site for further details: www.emeraldinsight.com/reprints