North-Holland
1. Introduction
div(u)=0
inD,
-VP/p
u=(0,0)
in D ,
(l.la)
on0D,
(l.lb)
where u = (u, v) is the velocity vector, r = (x, y) is the position vector, t is time, A -V" is the
Laplacian operator, P is the pressure, p is the density and Re is the Reynolds number
associated with the flow.
Equations (1.1) are difficult to solve by the finite difference method, particularly at large
Reynolds numbers. Chorin [1] has developed a grid-free method for modeling turbulent flow
in 1973. In this, the random vortex method, the creation of vorticity along boundaries is
modeled by creation of vortex blobs, discrete quantities of vorticity. The vortex blobs
themselves are not vortices but elements whose unions form vortices. The motion of the fluid
flow is modeled by considering the interaction of these blobs. The vortex sheet method,
presented by Chorin [8] in 1978, is used near the boundary to solve the Prandtl boundary
equations. This method is also a grid-free method and has the advantage that the interaction
between vortex sheet elements is not singular.
Combustion is governed by a complicated interaction of process of advection, diffusion and
reaction. For simplicity, we consider a single step reaction given by A---->B which describes a
1 Current address: Department of Mathematics, Indiana University, South Bend, IN 46634. USA.
0045-7825/91/$03.50 1991 Elsevier Science Publishers B.V. All rights reserved
672
premixed combustible or an oxidation reaction where either fuel or oxidizer is present in small
quantities. When written in dimensionless variables, the equations governing this combustion
process are
O(O,Y A + ( u . V ) Y A ) = L e
A' A Y , t - O K A Y a
e -NA/T ,
p(O,T + ( u - V ) T ) = AT + P Q A K A Y A e -NAIT ,
(1.2)
where T is the temperature, p the density, YA the mass fraction of chemical spices A and u the
velocity vector of fluid flow. Le A is the Lewis number associated with the mixture and Qa is
the heat released by the reaction. The reaction rate is determined by N,a, the activation energy
and K A , the pre-exponential factor. In this form, 0 <~ YA ~< 1 and in absence of external heat
T u < ~ T ~< T ~ + e with T h = I + T , ,
Tu is the ambient temperature and T b the burning
temperature, where e is to accommodate slight overshoot of T possible for a reaction which is
much faster than diffusion.
To solve (1.2) numerically, difficulties arise due to the difference in time scales, especially
in a fast reaction. A very small time step is necessary in order to maintain the stability of the
solution [3]. The random choice method of Glimm [4] has successfully modeled the system of
hyperbolic conservation laws [5, 6] and is suitable for solving differential equations which have
steep gradients in the solution profiles. Sod [7] has shown how the random choice method can
be extended to solve reaction-diffusion equations. The applications of the method can be
found in [8, 9].
2. Vortex methods
Taking the curl (V x) of (1. la), we obtain the scalar vorticity transport equation
0,~ + (u.V)~: = Re -t A~,
div(u) = 0,
(2.1)
where ~ = V x u = O,.v - Oyu is the vorticity. The vorticity field ~ gives rise to a velocity field u
which transports it. A stream function ~b is introduced which satisfies
u -- ~y~0,
v = -O~.
(2.2)
in D .
(2.3)
The vortex method is based upon this stream function-vorticity formulation. To solve (2.1),
we first assume that the flow is inviscid, that is Re -~ = 0, and that there are no boundaries.
Equation (2.1) reduces to the inviscid Euler equations
D~:/Ot = 0,
A~b= - ~ : ,
(2.4)
673
(2.5)
where 6(r) is the Dirac delta function.
The equation A+i = -@(r - ri) for the ith point vertex can be solved to obtain the stream
function
(2.6)
This is the fundamental solution to Laplaces equation in the plane. From this the velocity
field induced by the ith point vortex by (2.2) is
q(r) = -
ti(Y - Yi)
2nllr - ril12
5itx
Jr)=
-4)
27Fllr _ rill?
(2.7)
(2.8)
where u is a cut-off value to be determined later.
The vorticity field (2.5) is now replaced by
For llrll 2 a, the stream function could be taken from (2.6); for l\rll G 0, we need to solve
- riII + Constant.
(2.10)
674
rill
))
IIr-rill<
,
IIr- r, II >t
(2.11)
u(r) = ~ ~iui(r).
(2.12)
i=l
The vorticity transport equation states that the vorticity is advected by this velocity field, i.e.,
d r J dt = u(ri) = ~ /uj(ri).
(2.13)
j#i
~:jvj(rT),
(2.14)
tl
where r,. = (xi, y~') denotes the position of the ith vortex blob at time n At. Higher order
approximation for (2.13) can be found in [10], in which Sethian has used Heun's method to
obtain second order accuracy in time.
To modify the vorticity field due to the effect of the diffusion, i.e. Re -! # 0, we consider
the diffusion equation
,
(2.15)
A grid-free method for solving this is based on the connection between the diffusion (heat)
equation and the random walk. In one dimension, this reduces to the diffusion equation
= Re-'
(2.16)
(2.17)
This is also the probability density function associated with a Gaussian distribution with mean
0 and variance 2t/Re. The effect of diffusion on the vortex blobs from t = n At to t =
(n + 1) At is modeled by a random walk. Using operator splitting to write (2.15) as (2.16) in
both x and y directions, random variables r/i, ~/2 are picked from a Gaussian distribution of
mean 0 and variance 2 At/Re, and vortex blobs perform a random walk of the form
n + I
Xi
~- Xi -~ 1~il '
n + I _
Yi
- Yi
+ 7qi2
(2.18)
675
xi
y~n +1
(2.19)
where (ui, o~) is from (2.11) and the (Up,, op,) potential flow given in Section 4. (7/l, r/2) is an
independent Gaussian random variable with mean 0 and variance 2 At/Re.
The vortex sheet method, based on the fact that, within the boundary layer, the NavierStokes equation can be approximated by the Prandtl equation [11]
Ozyg,
(2.20a)
div(u) -- O,
(2.20b,c)
0,~ + (u . V ) g = Re -~
= -OyU,
where u = (u, o), u is tangential to 0D and o is normal to #D. The computational element
vortex sheet S~ is parallel to the boundary of length Ah and centered at r,. = (x i, y~). The
intensity sri of a vortex sheet is defined as
~,=-(u(y+)-u(y[)),
the jump of the velocity as y cross the vortex sheet Si. Equation (2.20b) can be integrated in
the form
u(x, y ) = U(x)-
~(x, "r) dr
(2.21a)
(2.21b)
(2.22)
where d r = 1 - [xj - xi[ [Ah and the summation is taken over all j such that y] > YI' and d r > O.
v,=-(l+--l_)/Ah,
(2.23)
676
where
i: =
E.
id-iyj
+* ,
:g
d ~ = l - l x i +- A h - x j l / A h ,
y~ =min(y~, yj).
The sum E+ (respectively, E ) is over Sj such that d~- ~< 1 (respectively, d~" ~< 1).
From (2.22) and (2.23), the position of the vortex sheets are advanced according to
n+l
x~
=x~ + A t u i,
n+l
y~
=y~ + A t e i+,l~,
where % is a Gaussian distribution with mean 0 and variance 2 A t / R e , which only applies to
the y-component since the diffusion in the x-direction is neglected on the Prandtl boundary
layer approximation.
aZ,v + G(v) ,
v(x, O) = f ( x ) ,
(3.1)
(3.2)
A random choice scheme was proposed by Glimm [4] in order to construct solutions to
(3.2) and prove existence under the condition that the total variation of the initial data is
sufficiently small. The method was developed as a computational technique by Chorin [5, 12],
Colella [13] and Sod [6]. For the numerical solution to (3.1), the general flow can be
decomposed into elementary waves (3.2) and steady waves [7], it is in the same spirit as the
method developed by Liu [14, 15] using the boundary value problem to obtain the steady
waves. The steady waves reduce to the system of ordinary differential equations
o.,r(v)
+ G(v).
(3.3)
Given the approximate solution u'i' to v(ih, nk), where h is the space step size and k is the
time step length, firstly, (3.3) is solved as a two point boundary value problem in each interval
[ih, (i + 1)h] to obtain a steady state solution v ~ with boundary conditions
tl
v~(ih) = ui ,
(3.4)
677
The solution on [ih, (i + 1)h] is sampled at the midpoint to produce a piecewise constant
approximation
n
ih<x<(i+ l)h.
The conservation laws (3.2) are solved using Glimm's method for nk <~t ~ (n + 1)k, using
the piecewise constant states as initial conditions:
v(x, nk)=uT+t/2,
ih<x<(i+ l)h.
Thus a sequence of Riemann problems is produced centered at the grid points x = ih. If the
CFL condition is satisfied, that is, h/k <~1/21Ai(v)l, for all eigenvalues A;, i = 1, 2 . . . . , p of
the Jacobian of F(v), the waves generated by the individual Riemann problem will not
interact. Hence, each is solved separately and the solutions can be combined by superposition
into a single exact solution, denoted by ve(x, t) defined for nk <<-t<~(n + 1)k. Define the
approximate solution for (3.1) at the next time level by
u~'+' = ve((i + ~.)h,(n + 1 ) k ) ,
the condition
k = h2/8u
(3.5)
is necessary to provide the consistent amount of diffusion. For a system of equations, the
extension of (3.12) requires that
k=h2[8Vl=...=h2/8up,
which is impossible to satisfy unless different grid spacing h i are used for each different value
of v~, such that
k=h2/8pi,
i=1,2,...,p.
The two-point boundary value problems can then be solved on the finest grid, with boundary
values interpolated at that grid. The initial values for the Riemann problems are interpolated
from the solution to (3.4) on the finest grid.
The computation required in solving the large number of boundary value problems could be
678
prohibitive, however, Sod [16] proposed the use of a dictionary. Equation (3.3) can be
reduced by omitting the advection term
+ G(v) = 0 ;
this is independent of x and t and the solution is only needed at the midpoint. Therefore it is
necessary to consider different possibilities for the boundary values (3.4) and the solution for
other boundary values can be obtained by using dictionary and linear interpolations.
Let v = (YA, T), then (1.2) can be written as
Otv + (u . V ) v
-~ Av + G ( v ) ,
(3.6)
where -~ = diag(Le~ I , 1), (3.6) is an equation of the form (3.1) except that it is not in
conservation form; however, u is known at each grid point from the solution of the
Navier-Stokes equation, so u can be treated as piecewise constant on intervals centered at the
grid points. The CFL condition becomes
k / h ~ 1/2maxllull.
on O D ,
(3.7)
where a and c are constants. To demonstrate how the boundary conditions are implemented,
without loss of generality, consider the x-sweep with OD = {x I x 0}. Condition (3.7)
becomes
ao(O, t) + c OxV(O, t) ffi b(t)
on OD .
(3.8)
lies outside the physical
(3.9)
In order to find the approximate solution at the next time level at the boundary x = 0, we
need to establish a left and right state for the Riemann problem associated with this point. The
679
......
.........
vo .x
,ooo
vortex blobs
................................ iC)'~O')..................
"
; ~ .........................................................................
Centerline (imaginary)"
~]D2
eymmetrlo to up-half, no computation
appropriate two-point boundary value problem is solved on the interval [ - h , 0] with the left
boundary condition given by (3.9) and the appropriate two-point boundary value problem is
solved on the interval [0, h]. If c = 0, the solution is known at x = 0 and a Riemann problem is
not needed at this point.
The first laboratory premixed-flame burner was invented by Bunsen around 1855. Many
experimental studies of premixed burner have been done since [17]. The difficulty of the
numerical study of the ignition lies in modeling such an ignition laappening in a very short
time.
For simplicity, we consider the cross section of a rectangular burner, so that the fluid
motion is governed by two-dimensional Navier-Stokes equations, and hence can be modeled
by the vortex method. Furthermore, we assume that the flow is symmetric about the
centerline, so that we only need to study the flow motion of the upper half of the cross section
thus saving almost half of the computations (see Fig. 1).
4.1. Hydrodynamics
We use the hybrid vortex method for hydrodynamics. Within the boundary layer the
computational elements are vortex sheets, outside the boundary layer we use vortex blobs (see
Fig. 1). The thickness of the boundary is assumed to be proportional to 1/V Re. Sheets
moving out the boundary layer become blobs with ~bloh= Ah * ~shcct; however, sheets which
cross the boundary are reflected back to preserve the vorticity.
On the boundary 0D I, the fluid must satisfy the no-slip condition and normal boundary
condition u. n = 0, while on the boundary d D 2 the fluid only needs to satisfy the normal
boundary condition since it is an imaginary boundary. For the normal boundary condition, we
use a single layer potential since the two pieces of boundary are disconnected and ODI is finite
in length while OD E is infinitely long.
The goal is to find the velocity field up = V4~, where
680
(4.1)
where R(q)= ( x - xq)2+ (y _yq)2, q = (X, yq) is a point on the boundary OD, a(q) is a
single-layer source and satisfies the integral equation
a(q) = ~
-2u. n.
(4.2)
We divide the boundary OD into M segments of equal length Ah with centers q~ = (xq,, yq),
i= 1 , . . . , M . The integral equation (4.2) can be approximated by a system of linear
equations. A source of intensity 1 at qi induces at the point qj, j ~ i, a velocity field with
components
1
Xqj -- Xqi
U2(i' j) =
Yq/-- Y~
(4.3)
Aa=b,
where the components of b are the vector values of - u. n evaluated at the points q~, and the
matrix A has the components
{ a,~-~ u ~ ( i / ) n l
al,
+ u2(q)n,
1/2Ah,
i~ / ,
i--1,...,M.
where n = (n 1, n2) is the unit outward normal vector at q~ = (xq,, yq,) on 0D.
One can then derive the discrete form of Up =Vd)using (4.1),
M
up(r)-~ up(i),
where
(4.4)
where r(q~) is the vector joining the point r in D to the point q~ on OD.
At the edge of the nozzle, a shear mixing layer with up stream velocity 0 is formed. Several
studies of a turbulent mixing layer have been made using the random vortex method.
Simulation of a shear layer flow by Ashurst and Meiburg [18] showed good agreement with
experimental results for mixing layer growth and Reynolds stress distribution. In the case of
Bunsen burner, when x, t are large enough, two mixing layers meet on the centerline. We
681
consider that the thickness of the mixing layer is proportional to Vx-/Re and time t, and use
average free stream velocity in the mixing area.
The numerical parameters are chosen to be M = 40, L = 10, Ah = 0.1, U= = 25.0, R = 0.25,
which is the half-width of the nozzle and the cut-off value tr = Ah/~ [19]. The Reynolds
numbers are chosen from 500 to 10000. The time step size of the vortex method is chosen as
At = 0 . 1 .
Figure 2 depicts the streakline plot of blob motions at times t = 4.0, 6.0, 8.0 and 10.0 for
Re = 5000, Figure 3 depicts the velocity profile for corresponding times, which is substracted
by average free stream velocity. From Fig. 2 we can see how the vortices along the centerline
are developed and grow with the time. The blobs just coming out of the boundary layer are
forming a small vortex which grows fairly fast, and later becomes the large leading vortex. As
the leading vortex grows and moves, smaller vortices are formed and moved with it. This
result has good agreement with the experimental observations.
(a)
, ".
.'~,~"
.',.:
(b)
(c)
"
...~
"..
~.'J.~;;~.
"~t.,,
:.'4
--
~.~',
,.:
.
::'~.~.
.~X:'..'.'~l..'~..'.p..:...:.~..''...'..
"..
" ~ .....
"
:.'"
,.. ,. : . " : ;
".
~b~ % ~.
"
".
"~.
~
~ :'.-,:::~,C'.,'."~,'.'.
"
Fig. 2. Streakline plot of vortex biob motion in the Bunsen burner. (a) t = 4.0, (b) t = 6.0, (c) t = 8.0, (d) t = 10.0.
682
::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::
~!!!~ii!
!~ El ~
ii!iiiiiiiiiiii!!iii!!iiii!!iii!!iii!!!!!!ii!!!
==========================================================================
~]~--~.~- : - : " l - i - ~ f - - _ :
========================================================
,.,'~:$=~,
(a)
:::::::::::::::::::::::::::::::::::::
=-":=~'.. : ::::::::::::::::::::::::::::::::::::::
~=='=-~==~==.=:.-[==:~i!i:i
ii!iii!iiiF"-:iii!~iiiii~!!i~!!!~!!!!!!'!i!~!i!iii!iiii
....,s,. o ~ . ~ . . ~ . ,
o ....
o.ooo.,.oooo,,
.....
o ........o
~a
~,' p ~ , ,
...............................
,==J'~,=,
(b)
~ ' : : : :"= : : : : : : : : : : : : - - = - " = - = = - ~ - : : : : : : : ~ : - - ~ = L ~ . : : = : : : : = : : : : : : : : :
. . . . . . . .
. _ _
7,._.-..,.,~..~ , , . , - ~ - -
., ., ., ., . ., .
.,,,,,,,~,~.., . . . . . . . . . . . . . .
(c)
,,,,_.,.,~,
H#UKINL~NYKCTON
.... = = - = = = = = ~
= --====.
:- = = = o . = = ,
:::::
H =HHHHF..H"HHE='HHHHHHHH"
":-=HqH~:H~H~'HHHH~HH~ HHHHHHHH~HH~H"'Hii:
Fig. 3. Velocity profile in the Bunsen burner. (a) t = 4.0, (b) t = 6.0, (c) t = 8.0, (d) t = 10.0.
4.2. Combustion
ODo={x, ylx=-l,O~y<-R},
683
aD~={x,yl-l<-x<-O,y=R},
JiiiiiiiiiiiiiiIIl|llllllllllllllllllllllllll|llllllllllllllllllllllllllllllk
(b)
(a)
i
i
l
G011ll~lRrlll~ I . l l I
~ I,Illl
11111
DIUINTI~AL~ I.lUll
I
PT$.:I*
t.17137
C~TO~ I.MliI
q 1.3B
JTIII)IIIIItlIItI~IIIT~IIIIII~"~'I'I'TT'['I'~TI]TII
IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIlllIIIIIIIIIIIIII'IIIItlIIIIIL
I
PTIl.li,
1.434o4~
C~I~LIRI~UYAG
~ IllNfm
~ F T T ' r T T ~ ~
(d)
(el
L
.200
A
; j
IIIIllIil~~illlllllllll
(~(~4|O~Li~C~|.llilaaI
~ |,|~
Illll
L
,269
...............
COIfOUrl|fllr.iIiK~ |.l~li
....
i
PYiI,|I,
1.11~1
i[itll
~TO~IR~
l.i I
ii l i l l l l l l lllltl11111111F
I[0 1.31~ ~IaTOLIR
IHTI~Yil.Of l . i i l
1
PT(i,$1, I.&3@47
Fig. 4. Flame propagation temperature contour in the Bunsen burner. (a) t=0.01250, (b) t =0.0250,
t = 0.03125, (d) t = 0.0500.
(c)
684
40
'
35
36
25
15
0
.J
== 5
gD
----~--,,~_
~.
o I
.-
(a)
*
-5
0
.1
,2
.3
.4
.5
,5
.7
.8
.9
,8
.9
.I
X
86
7~
6;D
BO
4iD
k-
36
d
=D
20
Z
3z
[l:
m
16
m(b)
-llJ
e
,1
,2
,3
,4
.5
.0
.7
1 .g
Fig. 5. F l a m e burning velocity in the Bunsen burner. (a) U~ = 25, (b) U~ = 50; with A: at t --- 0.012, B: at t -- 0.024,
C: at t = 0.036, D: at t = 0.048.
T=T u=0.2,
685
KA
= 2.5 103, QA = 1,
every where.
The premixed gas is ignited at the nozzle edge {x =0,0 ~<y ~< R} by raising the temperature to T = T b = 1.2 and holding at this temperature untilt = 0.003125.
The boundary conditions are
YA= I ,
T = I",, onOD o,
OT/On = 0
Figure 4 depicts the flame propagation contours at timest = 0.0125, 0.025, 0.0375 and 0.05.
After time t = 0.05, the flame is stabilized. The height 0fthe cone depends on the free stream
velocity. The flame burning velocity is calculated by measuring the distance between the
temperature profiles at same time interval. Figure 5 depictsthe flame velocity for two different
free stream velocities U~ = 25 and U~ = 50 at several different time level. The numerical
results have good agreement with experimental results [17,20]. The maximum velocity occurs
at the center of the burner and at time t = 0.024. The velocitydecreases to zero at the edge of
the nozzle. As time reaches t = 0.48 the burning velocitybecomes close to zero and the flame
is stabilized.
Acknowledgment
The author wishes to thank Professor Gary A. Sod for his many helpful discussions and
comments.
References
[1] A.J. Chorin, Numerical study of slightly viscous flow, J. FluidMech. 57 (1973) 785-796.
[21 A.J. Chorin, Vortex sheet approximation of boundary layers, J.Comput. Phys. 27 (1978) 428-442.
[3] R.A. Dwyer and G.R. Otey, A numerical study of the interacti0aof fast chemistry and diffusion, AIAA J. 7
(1979) 606-617.
[4] J. Glimm, Solution in large for nonlinear hyperbolic systems 0fequations, Commun. Pure Appl. Math. 18
(1965) 697-715.
[5] A.J. Chorin, Random choice solution of hyperbolic system, J. C0mput. Phys. 22 (1976)517-533.
[6] G.A. Sod, Numerical study of converging cylindrical shock, J. Fluid Mech. 83 (1977) 785-794.
[7] G.A. Sod, A random choice method with application to reacti0~.diffusionsystem in combustion, Internat. J.
Comput. Math. Appl. 11 (1985) 129-144.
[8] G.A. Sod, A dictionary approach to reaction-diffusion systemswithnonlinear diffusion coefficients, Internat.
J. Comput. Math. Appl. 13 (1987) 771-783.
686
[9] Y. Song, Numerical simulation of flame propagations in circular cylinders, Appl. Numer. Math. (1991) to
appear.
[10] J.A. Sethian, Turbulent combustion in open and closed vessels, J. Comput. Phys. 54 (1984) 425-456.
[11] H. Schlichting, Boundary Layer Theory, 7th Edition (McGraw-Hill, New York, 1979).
[12] A.J. Chorin, Random choice methods with application to reacting gas flow, J. Comput. Phys. 25 (1977)
253-272.
[13] P. Colella, Glimm's method for gas dynamics, Siam J. Sci. Statist. Comput. 3 (1982) 76-110.
[14] T.P. Liu, Quasilinear hyperbolic systems, Commum. Math. Phys. 68 (141) (1979).
[15] T.P. Liu, Nonlinear stability and instability of transonic gas flow through a nozzle, Commun. Math. Phys. 83
(243) (1982).
[16] G.A. Sod, A flame dictior,',ry approach to modeling unsteady combustion phenomena, Math. Appl, Comput.
3 (1984) 157-182,
[17] B. Lewis and G. von Elbe, Combustion, Flames and Explosions of Gases, Third Edition (Academic Press,
New York, 1987).
[18] W.T. Ashurst and E. Meiburg, Three-dimensional shear layers via vortex dynamics, J. Fluid Mech. 189 (1988)
87-116.
]191 A.Y. Cheer, Numerical study of incompressible slightly viscous flow past blunt bodies and airfoils, Si~amJ. Sci.
Statist. Comput. 4 (1983) 81-92.
[20] K.K. Kuo, Principles of Combustion (Wiley, New York, 1986).
[21] C. Anderson, A method for local corrections for computing the velocity field held due to a distribution of
vortex blobs, J. Comput. Phys. 62 (1986) 111-123.
[22] G.K. Batchelor, An Introduction to Fluid Dynamics (Cambridge Univ. Press, Cambridge, 1967).