You are on page 1of 10

A C O M P U T E R M O D E L FOR T H R E E - D I M E N S I O N A L

FLOW IN F U R N A C E S

S. V. PATANKAR AND D. B. SPALDING

Mechanical Engineering Department, Imperial Collegeof Science and Technology, London, England

The paper describes the development of a computer program for the prediction of the flow,
heat-transfer, and combustion processes in a three-dimensional furnace. The mathematical
model involves the solution of the differential equations for momentum, continuity, chemical-
species concentrations, stagnation enthalpy, radiation fluxes, turbulence quantities, and con-
centration of particles in various size ranges. The mathematical framework is outlined in some
detail, and the results of some recent computations are presented.
The conclusions are: that substantial progress has already been made toward the develop-
ment of a computer model for the transfer processes in a furnace; and that further work is
needed to validate the computer model and to refine the physical inputs.

1. Introduction 2. Solving the H y d r o d y n a m i c P r o b l e m

The authors are developing a computer pro- 2.1. Choice of Method


gram called Mammoth If. [-This stands for
Mathematical model of transfer of Heat I n The velocity field must be computed by solving
furnaces. The program development is being the equations of momentum and continuity.
sponsored by Combustion, Heat and Mass Methods of doing so fall into two classes: direct
Transfer Ltd.] When it is completed, the methods, in which pressure and the three velocity
program will be capable of predicting the flow, components are the dependent variables; and
heat-transfer, and combustion processes inside indirect methods from which pressure is elimi-
industrial furnaces, whose geometry is fully three- nated and attention is concentrated upon the
dimensional. The flow may be steady or unsteady, three components of the fluid vorticity. We have
laminar or turbulent, with or without effects of chosen the direct approach because the equations
buoyancy and particulate suspensions; and the and boundary conditions which it generates are
chemical reaction may be chemically or physically simpler, and because knowledge of the pressure
controlled. Radiative heat transfer is taken is required, during the course of the computations,
account of, as well as the convective and con- whenever the fluid is compressible.
ductive modes. A further sub-division is that between explicit
The first purpose of the present paper is to and implicit formulations of the finite-difference
explain the nature of the decisions which have to equations; the former contain only one unknown
be made during the construction of the model, and for each algebraic equation, while the latter
the arguments which govern them. The second contain several. Despite the necessity for matrix
purpose is to show what progress has been made inversion or iteration entailed by implicit
with the model, by exhibiting some recent methods, we have chosen to use them; the reasons
computations. are that, in unsteady-flow problems, they allow
Before the mathematical model can be accepted the use of larger time-steps and therefore are
as adequate by design engineers, it must be cheaper; and implicit methods allow one to make
validated by careful comparison of its predictions a faster approach to the solution of steady-flow
with experimental data. This phase of the work problems. In this respect our method differs from
will start in the near future. that of Zuber, ~ who has also chosen the pressure-
605
606 FLAMES IN FURNACES AND COMBUSTORS

~"~<'-'~'~ ~Section

Y ~ ~ s e c t i o n
7 ~ A-A
Geometry considered

Section B-B

~c_~_s
P

0 s/c
Section A-A
FIG. 1. Results of computations for a cavity flow in which a strip of solid
material is drawn at uniform velocity W, over the otherwise free surface of the
liquid in the cavity.

velocity formulation as the basis of his computer dimensional strip drag, as a function of the ratio
program for furnaces. of strip width to cavity width; and Fig. l ( b )
The details of the present solution procedure illustrates, by velocity vectors, the flow patterns
can be found in two publications.2'3 Briefly, it on two horizontal and vertical planes for a partic-
employs a combination of line-by-line solution, ular geometry and Reynolds Number. The latter
by way of the tri-diagonal-matrix algorithm, with diagram clearly exposes the three-dimensional
block relaxation, which quickly ensures that nature of the flow.
residuals in the difference equations have only Figure 2 illustrates the flow in a packed bed,
small-wavelength variations about a zero mean. into which fluid is blown through an inclined slot
The program contains a number of special tricks in one wall. Figure 2(b) shows velocity vectors
to reduce storage, increase speed, and improve in one vertical central section, and Fig. 2(c) those
the ease and flexibility of boundary-condition in another. The recirculations set up by the inertia
input. of the entering fluid are clearly visible.
In addition to the usual inertia, pressure-
gradient, body-force, and shear-stress terms, dis- 2.3 Discussion
tributed-resistance terms are included in the Physical intuition confirms the plausibility of
momentum equations. This facility makes it the flow patterns displayed in Figs. 1 and 2. Many
possible to solve problems of flow in packed such flow patterns have now been computed, with
beds--for example, blast furnaces and catalyst uniform success. Where exact solutions of the
arrays. equations are available, as for several two-
dimensional laminar flows, good agreement has
2.2 Particular Examples been obtained. Since the computer times involved
To illustrate the flexibility of the method, are quite modest (30-40 seconds on a CDC 6600),
Figs. 1 and 2 are provided. The first represents we can conclude that we now possess a practically
the results of computations of the flow in a usable means of predicting three-dimensional
shallow rectangular cavity, across tile free surface flow patterns.
of which moves a strip of solid having a width The flow in Fig. 1 was supposedly laminar,
which is less than that of the cavity. Such situa- while that of Fig. 2 was dominated by the inertia
tions arise in the "float" process of glass-sheet terms and by the distributed resistance. In
manufacture, in which the strip of glass is drawn general, however, turbulent shear and normal
across the surface of a bath of molten metal, on stresses are of high importance; they must there-
which it floats. fore be included in the mathematical model.
Figure l ( a ) shows the variation of non- Although the techniques for doing so are still
COMPUTER MODEL FOR THREE-DIMENSIONAL FURNACE 607

t t t t t ttttt
O00@O

I
0 @ O0 t t t t ~ t tttt",
Q 0 0 0 0 tttt~
0(300
0 @ 00@ 1 1 ..~i t
0 000 ttz~
@ 0 0@0
0 0 O0
0 0 0 (~0
0 @ O0
\2 z t'.zz'/
FB
(~o~oo I S e c t i o n A-A S e c t i o n B-B

A
oo? ol (a) (b) (c)
0 0 (~BO O(itl
I_ B ~1

FIG. 2. Results of computations for the flow in a packed bed with inflow from
an inclined tuyere; (a) illustrates the geometry of the bed, and (b) and (c) the
velocity vectors on two vertical sections.

under development, the most successful ones including turbulent diffusion, which gives rise to
require the solution of differential equations for second-order differentials; and source and simk
two turbulence quantities--for example, the terms, usually algebraic expressions involving
energy and the length scale. However, these more than one dependent variable each.
differential equations are of the same type as The Mammoth If program solves these equa-
those which govern the velocity components; tions by the same algorithm as is used for the
therefore, they can be solved by substantially velocity distribution. The only points that require
the same procedure. specific mention here are therefore those con-
The subject of turbulence modelling is too large cerning the nature of the source and sink terms.
to review here. An account of the main lines of Because of the complexity of the chemical
current work, with a description of its application kinetics of the combustion of even the simplest of
to flows of many different kinds, has recently been fuels, the chemical-kinetic source terms are
published by Launder and Spalding.4 The numerous and complex. Consequently, the pre-
advanced turbulence models are only now being liminary work on the Mammoth If program has
incorporated into the Mammoth If program, been based on the suppositions: that the fuel gas
having been tested hitherto in two-dimensional and air react chemically in a unique proportion;
contexts; therefore, no results based on them can that the reaction gives rise to a single unique
be supplied with the present paper. product; t h a t the effective diffusivities of all
chemical species are equal; and that the rate of
chemical reaction is so great that it is impossible
3. Flow with Chemical Reaction for both fuel and oxygen to co-exist at any point
in the flame. These presumptions, which are con-
3.1 The Method venient rather than essential, are the common
ones for diffusion-flame theory, and are known to
The concentrations of reactive and unreactive fit industrial-soMe flames quite well; their main
species in a furnace, and the distribution of stag- deficiency is that they neglect the effect of
nation enthalpy, are governed by partial dif- turbulence in delaying chemical reaction. A
ferential equations having the same general form theory for this exists, s however, and its implica-
as those governing the distribution of momentum; tions will be incorporated into Mammoth If at a
the terms of all these equations represent the later date.
effects of: time-dependence and convection, which Because of the assumptions just indicated, there
give rise to first-order differentials; diffusion, is no need to solve the source-containing differ-
608 FLAMES IN FURNACES AND COMBUSTORS

- I * ~~ " \ _~' \
Air r ~ ~
"\ ~ " \.,,~....
ln

Fuel ~ i , . - T ~ ' ~-' - 2"!"-" T-- "- Products

A B C D

Velocit y vectors in the x-y plane

~ ~ b f b
tl

g
$,

" "ll"
Regions of unburnt fuel
OX OX OX O


fu
O,u @
fu

fu

Temperature T/Tin

-~(/ (r ~
FIO. 3. Predictions for an adiabatic furnace.
COMPUTER MODEL FOR THREE-DIMENSIONAL FURNACE 609

ential equations for the mass fractions of fuel and (b) Because of the symmetrical entry of the
oxygen, m~u and mox. I t suffices instead to solve a fuel and air, and the absence of swirl, there is no
single equation, without sources, for the variable circulation of gas in the sectional plane.
(mf,~- mo,~/i), where i is the mass ratio of (c) The flame, by which here is meant the
oxygen to fuel in a stoiehiometric reaction. envelope within which unburned fueI gas is to be
found, widens at first with distance from the
3.2 Example injection nozzle, and thereafter diminishes to
zero.
Figure 3 shows the results of a computation (d) The maximum of temperature is found at
along these lines, conducted for a flow with dimen- the flame envelope.
sions and conditions which are reminiscent of (e) The nonuniformities of temperature di-
those of the experimental furnace of the In- minish as the furnace exit is approached.
ternational Flame Research Foundation at
Ijmuiden. 6 Methane is supposed to be injected 3.3 Discussion
through a nozzle placed at the center of the left-
end wall of the square-sectioned furnace; the air Since six distinct differential equations have
supply is supposed to be uniform over the been solved simultaneously, and the grid con-
remainder of the end wall. The side walls are tains 1000 points, it might be thought that the
taken to be adiabatic, and of course impervious program is expensive to operate. However, the
to matter; the flow is steady. computer time needed for obtaining a converged
Because no advanced turbulence model was solution was only about 100 seconds on a CDC
incorporated in the program, the effective 6600.
viscosities t~off were computed from the formula: Of course, one eannnot presume that the com-
putations made on a 1000-point grid represent
accurate solutions of the differential equations.
.e~ = Kp~M~2f(x/d, y/d), (1) Accuracy can indeed be established in such
circumstances only by increasing the number of
where p is the local gas density, M the rate of points until the predictions of physically sig-
flow of momentum into the furnace, x, y are, nificant quantities, such as the length of the
respectively, the horizontal and vertical distances flame, are unaffected by further grid refinement.
from the axis of the furnace, d is the width of the This increase in the number of grid points will be
furnace, f(x/d, y/d) is a function which falls from accompanied as a rule by a slightly-more-than-
the value unity near the axis (x = y = 0) to zero proportionate increase in the computer time.
at the walls of the furnace, and simulates the Mathematical models are preferable to phys-
way in which t~eff does in fact fall off in real con- ical ones, or to experiments conducted on full-
fined flows, and K is a constant calculated from scale equipment, only if the information which
the theory of turbulent iets so as to ensure the they provide is both accurate and cheap. Since
correct value of pelf in the jet of the fuel gas. the measure needed to increase the accuracy also
I t is not easy to display the numerous results increases the expense, as we have just seen, the
of a three-dimensional computation on a fiat developer of a model must take great care to
piece of paper. The device adopted in Fig. 3 for ensure its economic competitiveness. Since the
doing this is to provide for a series of sections expense of building and testing a furnace is
across the furnace and for several of the variables, enormous, and since physical models which
contour plots showing the distribution of the simulate real-furnace processes accurately are
variable over the sections. Imaginative inter- almost impossible to construct, computer pro-
polation between these contour plots then grams for furnace design start with a big ad-
permits the building of a complete picture of the vantage; nevertheless, it remains to be established
flow. Vector diagrams, which are also shown in that the expense of prediction can be reduced to a
Fig. 3 for the velocities in the planes of the level which will allow furnace designers to use the
sections, also contribute to this picture. program freely.
Inspection of Fig. 3 reveals the following
features, all of which are physically realistic:
4. Radiative H e a t Transfer
(a) The forward-moving jet, which results
from the fuel injection, soon entrains all the
4.1 The Choice of Method
combustion air which is supplied through the
left-end wall. Consequently, a recireulation of When the mean free path of radiation is very
combustion products is caused, from which the large, and the wall temperature of the furnace is
further entrainment by the jet is supplied. uniform, it is possible to express the influence of
610 FLAMES IN FURNACES AND COMBUSTORS

A B C D

u n iform
7.5 + IxlC] 5 7.6 +_ Ixll3 6 = 7.6* Ixl{] 6 : ?.6_+lxlO -6

FIO. 4. Contours of temperature [T/TOnlet)] for the conditions of Fig. 3, except


that the conduction approximation for radiative transfer is used.

radiative heat transfer by way of sinks, propor- cases, the radiation must be calculated by one or
tional to the fourth power of the local temper- the other of two methods which are appreciably
atures, distributed throughout the furnace. The more complex in concept and realization. These
calculation of these sinks rests on the assumption are the "zone method" of Hottel and Sarofim, s
that there is a negligible amount of absorption, and the "six-flux method" which is a generaliza-
within the gas, of radiation emanating from the tion of that employed in astrophysical work by
gas. Their effect on the temperature distribution Schuster, and rendered more readily applicable
through the gas is then deduced by straightfor- to heat transfer by Hamaker. 1
ward solution of the regular differential equations Although the zone method has more advocates
for the stagnation enthalpy. than the six-flux method at the present time, it is
An equally simple means of accounting for the latter that is being incorporated into the
radiation can be employed when the mean free Mammoth If program. The objections to the zone
path of radiation is small compared with the fur- method are that: because each gas wall zone is
nace dimensions. This is the "conduction approxi- connected with each other zone, the finite-
mation", 7 according to which the effect of radia- difference equations for temperature adjustment
tion is the same as that of an additional thermal are extremely lengthy; because the coefficients in
conductivity, proportional to the third power of these equations contain exponentials of integrals
the local absolute temperature. This is easy to involving the variation of absorptivity and
incorporate into the finite-difference equations scattering coefficient along the connecting path,
which govern the enthalpy distribution. these coefficients are time-consuming to compute;
Unfortunately, in most practical furnaces the and because the radiative-exchange process has to
mean free path of radiation has a value which be computed simultaneously with that of
varies with position within the furnace, but which momentum and mass transfer, these coeffi-
is neither large nor small enough for either of the cients must be recalculated at each stage in the
above means of calculation to be valid. In such iterative or time-dependent computation, since

A B C D

FIG. 5. Contours of temperature [T/T(wall)] for the conditions of Fig. 4, except


that the wall temperatures are held at 20C.
COMPUTER MODEL FOR THREE-DIMENSIONAL FURNACE 611

A B C D

lllll
FIG. 6. Contours of temperature [TIT(wall)] for the conditions of Fig. 4, except
that the two side walls are held at 20C.

absorptive and scattering properties of the for which the Mammoth If algorithm has been
medium suffer change at each stage. constructed. I t is therefore a simple matter to
In the six-flux method, attention is focussed on incorporate the equations for ( I + J ) , (K ~- L),
the quantities I -~- J, K -~- L, M + N, where I ( M ~ - N) into the program, and to solve them
and J are, respectively, the forward and back- simultaneously with the other equations. Of
ward radiation fluxes in the x-direction, and K course, the absorption and emission of radiation
and L, and M and N, are the corresponding fluxes appropriate to these flux-sum distributions appear
in the y and z directions, respectively. The in the equation for stagnation enthalpy as source
second-order differential equations governing and sink terms. I t is easily possible to give
these quantities are derived from the addition separate treatment to the fluxes in various wave-
and subtraction of first-order terms describing length intervals.
the variation with distance of the individual
fluxes. Thus
4.2 Example

dI/dx = a E - - (a + s ) I + ( # 6 ) The incorporation of the flux model into the


Mammoth If program is still in progress at the
X (I--F-J-~-K-F-L-~-M+N), (2) time of finalization of the text of this paper. Its
results will be presented orally at the Symposium.
where E stands for the blackbody emissive power, As a stop-gap, there will now be presented some
a for absorptivity per unit length, and s for computations for the same flow as was represented
scattering per unit length, in Fig. 3, but based on the conductive approxima-
tion with an absorption coefficient of 1 m -1.
dJ/dx = -- aE + (a + s ) J -- @/6) Figure 4 shows the temperature distributions at
the various sections when the additional con-
X (I+J+K+L+M+N), (3) ductivity is incorporated, but the wall is still
adiabatic. Figure 5 shows further temperature
whence distributions, with the same input values of fuel
and air properties when the wall temperature is
d ( I ~ - J ) / d x = -- (a~- s) ( I - - J ) , (4) held to 20C; and Fig. 6 shows the temperature
distributions which result when two of the walls
( d/dx) { ( a -~- s)-~[d( I Jr- J) / dx-]} are adiabatic and the other two are at 20C.
= -- ( d / d x ) ( I -- J) Figure 7 shows distributions along the duet of
maximum velocity and maximum temperature
= -- 2aE Jr- (a + s) (I -~- J) -- (s/3) for the four different cases.
X {(I+g)+ (K+L) + (M+N)}.
4.3 Discussion
(5)
Comparison of Figs. 4 and 3 shows that the
The finite-difference equivalents of these equa- action of radiation is to make the temperatures
tions are easy to derive; and, moreover, they within the furnace and at its walls significantly
possess forms which are simple versions of those more uniform. Figures 5 and 6 show the effects of
612 FLAMES IN FURNACES AND COMBUSTORS

~2 1
2800 i
70 /
/ 4/
8 2000 fig.3
Wmax6 Tmax
(m/s) (K)
1200

0 I I I I I 4OO I I I I I

0 2 4 6 0 2 4
Z (m) z (m)
FIG. 7. Distributions along the furnace of the maximum z-direction velocity and
the maximum gas temperature for the computations of Figs. 3, 4, 5, and 6.

the various thermal boundary conditions; and to suffice; some will be illustrated by slides during
Fig. 7 serves as a reminder that the calculation the oral presentation.
of radiative heat transfer cannot be separated
from the calculation of the flow field, because the Buoyancy
distribution of sources and sinks of heat has a
significant effect upon the distributions of Many industrial furnaces are large enough for
velocity. the interactions of the gravitational force field
Of course, an absorption coefficient of 1 m - t is with the variations of gas density to have some
not sufficiently large to justify the use of the con- influence on the distributions of velocity. Thus,
duction approximation for a furnace of 2-m in a furnace as large as that of Fig. 3, the flame
width; so the results will not be quantitatively would certainly rise from its position along the
correct. However, their qualitative trends are horizontal axis and cause the heat flux to the top
correct; and the computations which have been wall to differ from that to the bottom.
reported will serve as useful points for comparison Mammoth If can take account of this process
with the results of the six-flux model. by the simple inclusion of the additional body-
Further, it should not be lost sight of that the force terms in the momentum equation for the
six-flux model is itself an approximation. In vertical direction. These terms bring an additional
reality, radiation is distributed through all 47r linkage between the various differential equa-
radians of solid angle, and is not neatly tied in the tions; and this linkage can, on occasion, induce
six kinds of bundles which our model takes into numerical instability. A special form of under-
account. To establish the accuracy which the six- relaxation has been devised to combat this.
flux model can provide, it will be necessary, at
least in a few simple situations, to make compari- Swirl
sons with exact solutions of the integro-differ-
ential equations. If the accuracy proves not to be In many furnaces, at least a part of the com-
adequate, possibly some less drastic discretization bustion air is injected in such a way as to impart
of the "angle space" can be employed. to the gases a swirling motion. Once again, it is
easy to introduce such effects into the Mammoth
If program; for they are expressible simply by way
5. Other Effects of the boundary (i.e., entry) conditions of the
velocity equations.
In a paper short enough for inclusion in the There are some sources of doubt, however.
proceedings of this Symposium, it is not possible There is evidence 1I which suggests that existing
to describe in detail the further physical processes turbulence models are not yet able to predict
which are being incorporated into the Mammoth correctly the processes of nonisotropic momentum
If program. The following few remarks will have transport in turbulent swirling flows. This is an
COMPUTER MODEL FOR THREE-DIMENSIONAL FURNACE 613

important research area for fluid dynamicists and of break-up and coalescence of liquid drop-
interested in combustion. lets? The answers to these questions must be
provided by detailed research. Computer pro-
The Presence of Particles grams of llae Mammoth of type merely use the
answer; they cannot provide them.
Most industrial furnaces embody two-phase
flow effects in some degree. Even if the fuel is
gaseous, particles of soot are often formed; and 6. C o n c l u s i o n s
knowledge of their size distribution is needed both
to assist in the computation of radiative heat The foregoing discussion shows that sub-
transfer and to allow prediction of what propor- stantial progress has already been made towards
tion of the incoming carbon will pass unburned the development of a mathematical model for the
into the flue gases. transfer of heat in furnaces. It is now just a matter
In some furnaces, the rate of vaporization of of programming craftsmanship to put together
the injected liquid-fuel droplets is slow enough to the various program elements (e.g., turbulence
exert some influence on the length of the flame; models, particle-size equations, the six-flux radia-
and, of course, in pulverized-coal-burning fur- tion model) and so to complete the required
naces, the rate of diminution of particle size by computer program.
oxidation of the surface layers of the particles However, there will still be a need for much
dominates the processes within the combustion further research, of two main kinds. The first
space. concerns the validation of the mathematical
The distribution of particle size will be cal- model, by comparison of predictions with experi-
culated in Mammoth If by way of equations mental data for furnaces. The second concerns
having as dependent variables the mass of the elementary processes of radiation absorption,
particulate matter in fixed particle-size intervals, particle-size growth, turbulence-energy dissipa-
per unit mass of local mixture. Thus, if fl is the tion, etc., which the more complex of the furnace
mass fraction for the ith size interval, the calculations will require as an input.
differential equation for fi wilt be of the form Applied science rarely proceeds in an orderly
way. Sometimes engineering practice has per-
[O(pf)/Ot~-[- div ( f i g - F/, gradf~) = S),, (6) force to venture where science lacks confidence
to tread; and sometimes a surge of scientific
wherein the first term expresses the time de- knowledge precipitates advances for which in-
pendence, the div term expresses the influences of dustry is not ready. The application of mathe-
convection (G ~- mass-velocity vector) and dif- matical models to furnace design exhibits this
fusion (Fi-~ relevant exchange coefficient), and typical disorderliness, which it is useless to
S/~ stands for the source. deplore. The task is to develop and refine, to
All the knowledge of the physics of particle- apply and research, all at the same time, and the
size change is concentrated into Ss,, which may motive for performing this task can be equally
be replaced by provided by its economic significance and its
intellectual challenge.
s~, = - p v i ( o f i / O s ) , (7)
Acknowledgment
where V~ stands for the rate of increase of particle
size s with time; i.e., it is a velocity in particle- The authors are grateful to Mr. D. Sharma for
size space. Since it is easy to derive finite- his assistance in the preparation of the figures.
difference analogs of (7), and since the terms on
the left of (6) are of the kind possessed by all the REFERENCES
other equations which Mammoth If solves, there
is no difficulty about handling particle-size effects 1. ZUBER, I.: Ein mathematisches Modell des
within the general framework of the program. Brennraums, Monographs and Memoranda No.
It should be mentioned that there is still need 12, Staatliche Forsehungs Institut ftir Maschin-
for answers to questions such as: How is rate of enbau, Bechovice, Czechoslovakia, 1972.
dissipation of turbulence energy influenced by the 2. PATANKAR, S. V. AND SPALDING, D. B.: Intern.
presence of particles? Under what conditions J. Heat Mass Transfer 15, 1787 1972.
must allowance be made for the "slip" between 3. PATANKAR, S. V. AND SPALDING, D. B. : Numeri-
the particles and the gas? What are the laws of cal prediction of some three-dimensional fluid
nucleation and flocculation of carbon particles, flows. To appear.
614 FLAMES IN FURNACES AND COMBUSTORS

4. LAUNDER B. E. AND SPALDING D. B. : Turbulence 7. ROSSELAND,S.: Astrophysik und Atom-Theo-


models and their application to the prediction retisches Grundlage, p. 41, Springer-Verlag,
of internal flows, University of Salford Sym- Berlin, 1931.
posium on Internal Flows, Paper 1, 1971. 8. HOTTEL, I~. C. AND SAROFIM, A. F.: Radiative
5. SPALDING,D. B.: Thirteenth Symposium (Inter- Transfer, McGraw-Hill, New York, 1967.
national) on Combustion, p. 649, The Combus- 9. SCHUSTER,A.: Astrophys. J. 21, 1 (1905).
tion Institute, 1971. 10. HAMAKER, H. C.: Radiation and heat conduc-
6. FRICKER, N., VAN I~EYDEN, L., AND MICHEL- tion in light-scattering material, Philips Res.
FELDER, S. : Investigations into the combustion Rept. 2, 55 (1947).
of natural gas in multiple burner systems, 11. ROBERTS, L. W.: Turbulent Swirling Flows
International Flame Research Foundation, Doc. with Recirculation, Ph.D. thesis, University of
No. F 35/9/5, 1971. London, 1972.

COMMENTS

David T. Pratt, Washington State University, constructed so as to fit the experimental facts of
Pullman, Wash. I understand the origins of the confined-jet spread in their main features.
differential equation for the turbulent kinetic More complete accounts of turbulence models
energy, or for the turbulent vorticity, b y taking used by the hnperial College group can be found
higher moments of the Reynolds equations; but in a book by B. E. Launder and D. B. Spalding,
I do not understand the physical meaning or "Mathematical Models of Turbulence," Aca-
origin of the differential equation for the turbu- demic Press, London, 1972, and in the contribu-
lence length scale. tion of Launder, Morse, Rodi, and Spalding to
the N A S A - - L a n g l e y Conference on Free Turbu-
Authors' Reply. Strictly speaking, our second lent Flows, July, 1972.
differential equation is for the vorticity fluctua-
tion squared, for the energy-dissipation rate, or
for some other correlation. The length scale is
then deduced from the values of the energy and
this quantity. R. S. Levine, N A S A--Langley Research
Center, Hampton, Va. When you mentioned
including "unmixedness," did you mean you
could consider, for instance, first a fuel-rich
Charles J. Schorr, Bell Aerospace, Buffalo, eddy, then an oxidizer-rich eddy in a zone
N.Y. What is the source of the empirical co- whose average condition is fuel-rich?
efficients and exponents used in the eddy- Authors' Reply. Yes, we solve a special dif-
viscosity mixing models? ferential equation for the root-mean-square
concentration fluctuation. The unmixedness is
Authors' Reply. The eddy-viscosity model for deduced from the local value of this quantity,
the calculations presented in the paper was which is computed for every point in the field.

You might also like