Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
IMECE '02
ASME International Mechanical Engineering Congress & Exposition
November 1722, 2002, New Orleans, Louisiana
Proceedings of IMECE'02
ASME International Mechanical Engineering Congress & Exposition 2002
November 17-22, 2002, New Orleans, Louisiana, November 17-22, 2002
IMECE2002-xxxxx
IMECE 2002
33918
T U R B U L E N C E - R A D I A T I O N INTERACTIONS IN FLAMES: A C H A O T I C - M A P
BASED FORMULATION
Ying Xu 1
J. M. McDonough 2
M. Pinar MengSc: 3
ABSTRACT
Pe
qR
R
Re
~'
Sc
T
u
U
v
In this paper we report initial efforts in developing largeeddy simulation (LES) subgrid-scale (SGS) models capable of
treating turbulence-radiation interactions in sufficient detail to
permit calculation of radiation intensity fluctuations on small
scales. These models are constructed with a fluctuating component consisting of a discrete dynamical system (chaotic map)
and are thus completely deterministic. We present an outline
of the development of this formulation and then employ experimental data to generate large-scale behavior permitting what
might be viewed as part of an a priori test of the SGS model.
We display spatially extensive instantaneous fluctuating temperatures produced by the model as well as time series of fluctuating
intensity calculated from the radiative transfer equation at several heights in a pool fire. We conclude that such results are
physically realistic (and very efficiently computed) and warrant
continued investigations, but we have at this time not yet completely validated the approach due to lack of detailed laboratory
data.
a
fl
#
~,
p
r
NOMENCLATURE
CB
Di
FB
Gr
I~,
k
L
p
specific heat
binary diffusion coefficient
body force
Grashof number
spectral intensity
thermal conductivity
length scale
pressure
P~clet number
thermal radiation
specific gas constant
Reynolds number
radiation propagation direction
chemical source
temperature
u velocity component
referenced velocity magnitude
v velocity component
mass fractions of species i
thermal diffusivity
thermal volumetric expansion coefficient
anisotropy correction
dimensionless temperature
absorptivity
dynamic viscosity
kinematic viscosity
density
time scale
INTRODUCTION
Most industrial scale flames are strongly radiating and
turbulent in nature; they can be viewed as time dependent dynamical systems. Turbulence-radiation interactions
(TRI) need to be accounted for thoroughly in such flames in
order to include the underlying physical mechanisms. Numerical modeling of turbulent diffusion flames requires at
least qualitatively accurate simulation of small-scale fluctuations of velocity, temperature, and concentration fields,
in addition to the corresponding large-scale values. Most
1Ph.D Student
2Professor; jmmcd@uky.edu
3Professor; mengucQengr.uky.edu
1
of the time-dependent information is lost if flow field simulations are not carried out in detail. Even though some
modern techniques such as the direct numerical simulation
(DNS) are mathematically rigorous and do yield a high level
of accuracy, they are not computationally feasible for application to complex practical systems. Instead, it is preferable to "model" the small-scale fluctuations and solve for
the large-scale parameters accurately. This strategy (largeeddy simulation, LES) allows us to save significant amounts
of CPU time, and brings us to the realm of simulating turbulent flames of practical importance.
The first attempt to include effects of radiation on
flow and temperature fields in a turbulent environment was
made by Townsend [1]. Later studies included those by
Cox [2], Tamanini [3], Kabashnikov and Myasnikova [4],
Grosshandler and Joulain [5], Fischer et al. [6], Fischer
and Grosshandler [7], as well as Faeth and his students (see
Gore et al. [8] [9], and Faeth et al. [10] for an overview).
One of the earliest accounts of TRI in the heat transfer
literature is that by Song and Viskanta [11] who investigated a turbulent premixed flame inside a two-dimensional
furnace. More recently, Gore et al. [12] and Hartick et
al. [13] solved the problem for diffusion flames using a k-a
model. They discussed a closure for the governing equations and correlated gas radiative properties using probability density function (PDF) of mixture fraction and total enthalpy. Mazumder and Modest [14] used a velocitycomposition PDF method to investigate TRI; the so-called
P D F / M o n t e Carlo method they developed was accurate,
yet computationally demanding. Li and Modest [15] used
a similar approach; they assumed the species concentrations and temperature/enthalpy are the random variables
and applied this to a methane-air diffusion flame.
It is now widely believed that turbulence in a flow field
is not random, but deterministically chaotic in light of the
seminal work of Ruelle and Takens [16]. Moreover, it is
possible to model the turbulent fluctuations via numerical
time series calculated from a linear combination of chaotic
maps, with different parameters and weights (see, e.g., McDonough et al. [17]) with the maps derived from the governing equations (see McDonough and Huang [18], and Yang
et al. [19]). Here, we will present an analysis that investigates TRI using a system of chaotic maps. Fluctuations
of absorption coefficient will be calculated as a function of
temperature modeled in terms of these maps, and the emission from flames will be quantified in terms of parameters
of the chaotic maps.
The concept of chaotic-map based TRI has been discussed before by Mengfi~ and McDonough [20], McDonough
et al. [21], Mengfi~ and McDonough [22], and details of the
chaotic map formulation were reported elsewhere (Mukerji
et al. [231; Hylin and McDonough [24]). Our focus in this
Governing Equations
The equations governing most combustion processes
consist of the Navier-Stokes (N.-S.) equations in a form appropriate for non-constant density (but low-Mach number)
flows, a thermal energy equation and species concentration
equations, along with an equation of state. These can be
expressed as
p, + v . ( p u ) = 0,
(la)
D(pU)
Dt-- - V p + # A U + FB,
(lb)
DT
p C p - ~ - = k A T + V "qR + So,
(lc)
D(pYi)
D--T- - V . Di~7(pYi) + &i,
i = 1,..., Ns.
(ld)
In these equations U K (u, V) T is the two-dimensional velocity vector; p, p, T are, respectively pressure, density
and temperature related through an equation of state, say
p = p R T , with R being the specific gas constant. Tile Yi
are mass fractions of species i, and cbi are corresponding
2
the chemical source term in Eq. (lc), but report that their
approach is not computationally feasible for anything but
the simplest kinetic models. Hence, no details of effects of
species concentration can be deduced, although more global
effects of TRI are produced. It is important to recognize,
however, that there is no true instant-to-instant interaction
between the velocity field and the temperature field, and
thus no detailed turbulence-radiation interaction. Such details cannot be produced when the governing equations are
averaged (RANS) or filtered (usual LES), and then closed
with statistical models representing phenomena on precisely
the scales at which the interactions must occur.
(2)
where ~is the radiation propagation direction; ~x is absorptivity at wavelength I, and Ix is the corresponding spectral
intensity. The formal solution to Eq. (2) is (Viskanta and
Mengfi~ [28]; Modest [29])
~0 8
'
(3)
x E F ~ ~, d = 2 , 3
(4)
with q representing large (computed) scales, and q* denoting the small scales that must be modeled, of any dependent variable vector Q = (Q1,Q2,"" ,QN,) T, where Nv is
the number of variables. In earlier work (e.g., McDonough
and Bywater [30] [31]) components of q* were obtained as
solutions to low-dimensional systems of ordinary differential equations (ODEs) corresponding to high-wavenumber
Fourier modes arising from a Galerkin procedure applied to
the small-scale system of additively split partial differential
equations (PDEs). But beginning with McDonough et al.
ut + uuz + vu u = ~ e A u ,
(6a)
1
Gr
vt + uv~ + vv u = ~eAV - ~e20,
(6b)
(6c)
(5)
was introduced. Here, A is an amplitude factor computed via Kolmogorov scaling (see, e.g., Frisch [33]; ~ is
an anisotropy correction obtained from scale similarity, as
employed in the dynamic SGS models of Germano et al.
[34]; the factor M is analogous to Kolmogorov's stochastic variable, but in the present context is a deterministic
discrete dynamical system.
These ideas are implicitly contained in [32], and much
of the theory required for modeling subgrid-scale physical
variables is presented in [24]. Furthermore, many of these
ideas were applied specifically to the TRI problem in [22].
In these earlier studies, chaotic maps having no particular
connection to the appropriate physics were employed (to
some extent successfully) as subgrid-scale models. More
recently, McDonough and Huang [18] [35] have devised a
method whereby discrete dynamical systems (i.e., chaotic
maps) can be derived directly from the governing PDEs of
essentially any physical system. To date these have been
studied in a comparison of free and forced convection (McDonough and Joyce [36]) and in association with three different H2-O2 and H2-air kinetic mechanisms (McDonough
and Zhang [37] [38]; Zhang et al., [39]). But in all of these
studies emphasis has been placed on the time series of SGS
behavior at a single spatial location; moreover, the bifurcation parameters needed to evaluate the DDSs were chosen
somewhat arbitrarily.
In the present study we employ experimental data from
a spatially extended set of measurements for a pool fire to
automatically set required bifurcation parameters over the
entire physical domain of the experiments. To keep the
problem at a relatively simple level we will not model Eq.
(ld) for the species mass fractions. Hence, the chemical
source term in Eq. (lc) will be ignored (but see the above
cited references for treatments of this). Furthermore, because the pool fire of the Weckman and Strong [25] data is
burning methanol, and soot formation will be at a relatively
low level, we also ignore the radiation source term in Eq.
(lc) - - despite its general importance. This permits us to
focus on the effects of fluctuating temperature in producing
radiation intensity fluctuations without needing to separate
the feedback from thermal radiation via the radiation source
term in Eq. (lc).
Thus, as in [36] we employ the following system of 2-D
equations, utilizing the Boussinesq approximation to obtain the form of the body force term in Eq. (lb) and Leray
projection [40] to remove the pressure gradient from this
equation. This produces the following system written in
dimensionless form:
I~g6TL 3
u~~
,
R e =-
UL
Pe ~
UL
(7)
C~
where a is thermal diffusivity, /3 is thermal volumetric expansion coefficient, and u is kinematic viscosity.
The velocity components (u, v) have been scaled with
a reference velocity magnitude U, and distances are scaled
by a fixed length scale L. The dimensionless temperature
is defined as
T-To
0 =- Ti- ~ 0 '
(8)
where To and T1 are reference temperatures, and ~T = T1 To. The subscripts x, y, t denote partial differentiation
with respect to the spatial coordinates (x,y) and time t
respectively, and A is the Laplacian.
We apply the Galerkin procedure to this system and
represent all dependent variables by Fourier series as
(9a)
(9b)
O(x, y, t) = E
(9C)
ck(t)cpk(X, y).
Application of the Galerkin procedure consists of substituting Eqs. (9) into Eqs. (6), and then computing inner products with each of the countably infinite (pks. The result of
this (see [36]) is reduced to
/~+ A(1)a 2 + A ( 2 ) a b -
+ B(1)b2 + B(2)ab _
C[k[2 b _
d 4- C(1)ac 4- C(2)bc 4
Clkl 2
R e a,
Re
Gr
~ 7c'
Clkl~
~e c,
(10a)
(10b)
(10c)
(lla)
(llb)
It should be noted that Eq. (11c) is linear in c, implying that unless a reference level is prescribed, the DDS
will either approach zero or infinity after sufficiently long
time. This is remedied with the term Co, which is set by
the high-pass filtered resolved-scale solution in the context
of a complete LES implementation. In the present study, it
is assigned a fixed value.
Choosing Parameter values
Re
]'
i=1,2,
and since we have based our derivations on a single wavevector k it is natural to set j31 = j32. By comparing Eqs. (10)
and (11) one sees that
8T =
TCIkl 2
Pe '
also, aT
vGr
Re 2
~o
~8
~6
E
N
~o
8
e
4
2
-le
-12
Radius, r
12
12
[cml
18
16
E
12
~1o
e
6
4
2
-16
-~2
-a
-4
Radius, r
le
[cm]
Figure 1. RECONSTRUCTED AXIAL AND RADIAL MEAN VELOCITY CONTOURS FROM EXPERIMENTAL DATA OF WECKMAN AND
STRONG [25]: (a) AXIALVELOCITYCOMPONENT;(b) RADIALVELOCITY COMPONENT
time series for intensity display a periodic behavior throughout the flame. If the range of/3 is increased further from
3.4 (base) to 3.6 (tip), quasiperiodic intensity fluctuations
are observed, as seen in Fig. 3c. Finally, if the/3 range is
from 3.6 (base) to 3.8 (tip), the behavior is chaotic (see Fig.
3d). This change in qualitative behavior corresponds to the
well-known Ruelle and Takens [16] bifurcation sequence.
In Figs. 3, the level of fluctuations in radiation intensity is higher at the base than at the tip of the flame. The
reason is that the width across which intensity is measured,
is greater at the base. In these figures, we also show the
mean radiation intensity (in red) based on the mean temperature and absorption coefficient values. Note that, if
turbulence-radiation interactions are accounted for, the resulting intensities are larger. This result is expected and
consistent with those published earlier (see e.g., [10], [11],
[15]). It should be understood, however, that the exact
effect of TRI on mean radiation intensity is a function of
flame conditions considered, and may vary from one flame
to another.
Another parameter that affects the extent of TRI is the
magnitude of small-scale fluctuations in the turbulent flow
field, which is a function of the actual flow conditions. Here,
however, we investigate the affect of fluctuations by simply
6
80QO
~o,
h= 3cnn
m'
70QO
m~
__
h=11crn
600O
la'
h=14cm
5000
h=19cxn
4~
(a)
400O
-~
-8
12
16
80OO
~K
h=3cm
Radius, r [can]
2oJ
7000
h=11orn
`zA~AAAaA&AA~AAAA|~LkAAAAAAkAAaAAAAkAAAAA~ahaA~AkAAA~AAa"AaAAAAA~AAa~AA~AAAuAAka~AAAAAAAk"
6O0O
h=14cm
||||AA|AAAAAAAAAAAAAAAAA/A|AAu|AAAAAi
rf~T~y~|~;~TT~T~|~v~T~T~T~T~T~|~Tf~T~v~T~v~|~
500O
_ _ . - - _ . - : -.- =- - - - : . . . . . . . . .
--:-:--.-
=lgcm
- - - - : - - : : - : - :h
-: - -:.--: -. - .:--::-
-: - -- -...::.:.:....
400O
|
8O0O
4:
..........
_ .-::
(b)
e,
!
h=
=*
3cm
o)
-~2
-e
-4
12
1~
7000
h=11cm
I~zdlus, [cxnt
~
6O00
h=14crn
500O
,
r~
h = l 9c,m
4000
(c)
i
80OO
-" " 'lq"W "'v V " "y v v v ~ ' ' w l T ' T '
"'~
,iv vvy,'v ,
7O00
-ul,
-12
-,4
12
aa(l~*, r [cm]
~v~,~ . . . . .
600O
p|-
"~ vv r , , , - r ~/ q , , w
h=14cm
5OOO
r-,'
400O
(o)
m
0.380
0,385
0.395
0.390
,Scaled Time
(Arbitrary
0.400
Units)
TIP); (b) ~ RANGE 3.0--3.2; (c) ,/3 RANGE 3.4--3.6; (d) /~ RANGE 3.6-3.8.
for part (a) except the amplitude is set at 75%. These time
series are taken from detailed results as shown in Figs. 2(b)
and (c) for a single instant of time.
Also shown in Figs. 4 are the mean intensities (shown
in red) calculated from data corresponding to Fig. 2(a),
i.e., with no turbulent fluctuations. It is clear from the
figures that although the temperature fluctuates about its
mean (not shown), the intensities fluctuate about a different
(higher) mean value. This is a consequence of the nonlinear
7
,e
8DQO~
h=14cm
gOOD
8000!
6OO@
h=lgcm
4OQO
REFERENCES
0.3~
0.38,5
0.390
0.395
0.400
[2] Cox, G., 1977, "On radiant heat transfer from turbulent flames," Combust. Sci. and Tech., 17, pp. 75-78.
CONCLUSIONS
In this paper we have presented a new approach to SGS
modeling of turbulence-radiation interactions in the context of large-eddy simulation. We outlined the derivation
of model equations and then conducted what can be viewed
as (a portion of) an a priori test of this new discrete dynamical system technique. Computed temperature and radiation intensity fluctuations were provided to characterize
behavior of the model as bifurcation parameters of the DDS
were varied. These simulations represent the first application of this approach over a spatially-extended region, and
furthermore the first direct use of large-scale results (experimental in the present case) to automatically set bifurcation
parameters of the DDS.
The results produced by the model are physically reasonable and self consistent; however, the nature and quantity of extant experimental data, and absence of detailed
DNS data for turbulence-radiation interactions, make it dif-
[9] Gore, J. P. and Faeth G. M., 1988, "Structure and radiation properties of luminous turbulent acetylene/air diffusion
flames," ASME J. Heat Transfer., 110, No. 1, pp. 173-181.
[14] Mazumder, S. and Modest, M. F., 1999, "A probability density function approach to modeling turbulenceradiation interactions in nonluminous flames," Int. J. Heat
Mass Transfer, 40, 1999, pp. 971-991.
[25] Weckman, E. J. and Strong, A. B., 1996, "Experimental investigation of the turbulence structure of medium-scale
methanol pool fires," Combust. Flame, 105, pp. 245-266.
[15] Li, G. and Modest, M. F., 2002, "Application of Composition PDF methods in the investigation of turbulenceradiation interactions," J. of Quantitative Spectroscopy and
Radiative Transfer, pp. 461-472.
[26] Libby, P. A. and Williams, F. A., 1994, Turbulent Reacting Flows, Academic Press, San Diego, CA.
[27]Peters, N., 2000, Turbulent Combustion, Cambridge
University Press, Cambridge.
[30] McDonough, J. M. and Bywater, R. J., 1986, "Largescale effects on local small-scale chaotic solutions to Burgers' equations," AIAA J., 24, pp. 1924-1930.
[31] McDonough, J. M. and Bywater, R. J., 1989, "Turbu9
[43] McDonough, J. M., 2002, "A 'synthetic scalar' subgridscale model for large-eddy simulation of turbulent combustion," presented at Spring Tech. Mtg. Central States Sea
Combust. Inst., Knoxville, TN, April 7-9, 2002.