Turbulent Flows Invol Ving Chemical Reactions: Paul A. Libby and F. A. Williams

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 26

ANNUAL

REVIEWS Further
Copyright 1976. All rights reservea Quick links to online content

TURBULENT FLOWS x8092


INVOLVING CHEMICAL
REACTIONS
Annu. Rev. Fluid Mech. 1976.8:351-376. Downloaded from www.annualreviews.org
Access provided by University of York on 01/27/15. For personal use only.

Paul A. Libby and F. A. Williams


Department of Applied Mechanics and Engineering Sciences, University of California,
San Diego, La Jolla, California 92093

1 INTRODUCTION

An important problem from both the theoretical and applied points of view arises
when turbulent flows involve chemical reactions. This occurs in many situations
ranging from stirred reactors in industrial chemical processes, to burners in power
plants, to jet-engine combustors, to chemical lasers, and to the wakes of bodies
entering the atmosphere at hypersonic speeds. In these flows the questions of
interest usually center either directly or indirectly on the ways in which
heterogeneities in composition and fluid properties associated with turbulence
interact with the chemical behavior of reactants and products. The fluid-mechanical
aspects, while far from being fully understood, have received greater clarification
than the chemical aspects.
In recent years, for a variety of reasons there has been a quickening of attention
to the phenomenology of the subject. The improved predictive methods associated
with so-called second-order closure lead to explicit consideration of various second­
order correlations that also arise in the description of reacting flows; thus their
treatment as an extension from purely fluid-mechanical considerations appears
natural. Improved experimental techniques of flow diagnostics presently' or will
shortly make possible more detailed and higher quality measurements of gas
composition and state in turbulent flows, thereby providing guidance for more
sophisticated theoretical developments. Finally, there have been technological
developments, such as the chemical laser and the requirement that lower levels of
contamination be attained in the exhaust gases of combustors, focusing attention
on the interaction between turbulence and the chemical processes involved.
Given this renewed interest, it now seems appropriate to review the present status
of our understanding of chemically reacting turbulent flows. The contributions on
this subject are scattered among the aerospace, chemical, .combustion, and fluid­
mechanical literature. As might be expected there is little transfer among the various
elements comprising the total literature, due in part to widely differing points of

351
352 LIBBY & WILLIAMS

view and differences in terminology. Here we discuss what we consider to be some


of the significant contributions of lasting value, attempting to place them in a
common framework.

2 THE UNDERLYING CHEMICAL KINETICS

In starting this review it seems worthwhile to discuss first the role of chemical
kinetics at the fundamental level in turbulent reacting flows. Consider a typical
elementary step for a chemical reaction,

(1)
Annu. Rev. Fluid Mech. 1976.8:351-376. Downloaded from www.annualreviews.org
Access provided by University of York on 01/27/15. For personal use only.

where Mi denotes a molecule or atom of species i. Vi is the molecularity of this


reaction with respect to species i. and kf and kb are the forward and backward
rate constants. When this reaction occurs in a flowing system, the instantaneous
volumetric rate of production or destruction of Mj• e.g. in g cm-3 sec- 1, appears in
the conservation equation for species i and is given by
(2)
where p is the mass density of the mixture and Yj is the mass fraction of species
i. Similar equations prevail for the other species and for other reaction steps
contributing to the production of species i. The rate constants kJ and kb in general
are functions of the temperature.
In complex chemical systems many individual reactions of the type given by
equation (1) may be involved so that WI contains a large number of pairs of terms
like those given in equation (2). Because it is frequently not possible to take into
account all of these individual reactions and their related rate constants, it is
commonplace in the analysis of complex chemical systems to consider only a few
reaction steps, which in general do not correspond to the proper mechanism of
the reaction on a molecular scale but which in a gross sense do describe the
overall chemical processes. When this is done, nQnintegral molecularities and
pressure exponents differing from the sum of molecularities often arise.
Although the fact that gases are not continua but instead are composed of
moving molecules could influence certain aspects of turbulent dynamics (Tsuge 1974),
there is little doubt that the complete description of the rates of elementary
chemical steps that prevail at the molecular level applies at each point in space and
time in a turbulent flow. The smallest length scales of interest in such flows, i.e.
lengths dominated by molecular effects, are always large compared with molecular
path lengths. However, of major concern is the applicability to turbulent flows of
overall reaction mechanisms and rates; since they do not correspond to chemical
behavior at the molecular level, their utilization in turbulent flows with hetero­
geneities in composition is suspect, especially if some of the characteristic chemical
times of elementary steps are long compared with a representative time scale of the
turbulence. In principle, turbulence may modify reaction mechanisms. In particular,
overall reaction models derived from experiments with homogeneous systems may
TURBULENT FLOWS INVOLVING CHEMICAL REACTIONS 353

be totally inapplicable to turbulent flows under certain conditions. To ascertain


those conditions should be one important objective of research on turbulent reacting
flows.

3 THE PROBLEM OF THE CREAnON TERM

If the description of a turbulent flow involving chemical reaction is formulated in


terms of the usual time averaging, i.e. in terms of mean and fluctuating components,
the conservation equation for the average mass fraction of species i involves the
usual convection terms, molecular and turbulent-diffusion terms, and the mean value
Annu. Rev. Fluid Mech. 1976.8:351-376. Downloaded from www.annualreviews.org

of the creation term, Wi. If this term could be simply and convincingly expressed
in terms of quantities whose behaviors are understood and predicted by the usual
Access provided by University of York on 01/27/15. For personal use only.

methods of analysis, an essential difficulty would be overcome. Unfortunately, such


expressions do not exist except for rather uninteresting chemical systems.
To see some of the difficulties arising from the creation term, consider a one-step,
unidirectional reaction

M1+M2"�M�. (3)

In this case

WI = -p2kY1 Yz· (4)

The influence of the right side of equation (4) on the mean production rate can be
expressed in a variety of forms depending on the averaging procedure adopted,
specifically on how the density is treated. With Favre averaging (Favre 1969) most
dynamical variables are mass averaged, e.g. Y1 = Y1 + Y{' = (pY/p) + Y{" and ex­
plicit density fluctuations are absent. In this treatment equation (4) leads to the
following time-averaged form :

W1 =
2 _

-P /ZY1 Yz 1+
(p
Zy!.' pZy.;: pZy!.'y.;.'
� +- + =--
PZY1 pZyz PZY1 Yz

p2k" p2k"Y" pZk"Y"z p2kIlY1"YZII


+=-+�+_ +
) •
(5)
p2/Z p2/ZY1 p2/ZY2 p2/ZY1 Yz
If the density is treated in more conventional fashion, i.e. so that p = p + p', then
even more terms are involved. Note that if k(T) is of the Arrhenius form, the k"
terms cannot be simply related to temperature fluctuations unless those fluctuations
"
obey the often stringent requirement (1;, T'ITZ) � 1 where T. is the activation
temperature of the reaction.
Inspection of equation (5) reveals that complications arise in averaging the
creation term even for the relatively simple chemical system of equation (3). We
see that a variety of terms describing the heterogeneity of the composition and
fluid properties contribute to the effective chemical behavior in turbulent flows. To
focus on the physical significance of one of these terms, consider a further simplifi­
cation that occurs if the reactants are highly diluted and the system is essentially
354 LIBBY & WILLIAMS

isothermal. Then the density and the reaction rate may be considered constant,
and we obtain
- 2 - -(
YiY2
wl=-p kY1Y 2 1+-=-=.
) (6)
Y1 Y2
The effect of turbulence on the mean rate of destruction of species 1 is now seen
to be contained i� the term (Y; y1/Yl172), which is related to the correlation
2
coefficient (Y{ Y2/(Y? Yi2)1/ ), a measure of the heterogeneities in the concentrations
of the two reactants. If there are no inhomogeneities in composition, as might be
the case prior to any significant chemical reaction in some flow situations, then this
Annu. Rev. Fluid Mech. 1976.8:351-376. Downloaded from www.annualreviews.org

coefficient is negligible initially. However, production of product as time proceeds


will lead to turbulent interference in the chemical behavior associated with non­
Access provided by University of York on 01/27/15. For personal use only.

negligible values of this coefficient. Limiting cases of chemical behavior are those
in which this coefficient has the value plus one (perfect correlation) and those in
which it has the value minus one (the fluctuations of the two reactants are
perfectly anti-correlated); the intermediate case in which the coefficient is zero
(uncorrelated fluctuations) serves as a limiting case that defines maximum mixing
in certain initiaJly unmixed flows.
From these simple examples it is seen that the crux of the problem of under­
standing and predicting chemical reactions in turbulent flows lies in the multiplicity
of ways in which the heterogeneities in composition and state within the turbulence
influence the effective chemical-reaction rate. Also worthy of consideration is the
counterinfluence of chemical reactions on the turbulence, discussed in Section 6.

4 LIMITING CASES AND THEIR REALIZATION

In subjects with complexities as great as those indicated above it is always helpful


to attempt to identify limiting cases in which some degree of simplification is
achievable. Certain limits are defined here, while others are identified later. In
cO'llbustion, two major limiting cases are those of premixed and non-premixed
systems. In the non-premixed case, the reactants initially are in separate flows and
in some fashion subsequently mix and react. Under some circumstances this
situation leads to the case variously termed "the diffusion limit," "rapid chemistry,"
or "fast chemistry." As an example of a boundary-layer type of flow, the two­
dimensional mixing layer provides an ideal situation to consider the non-premixed
limit; each flow contains one reactant that meets the other at the end of a splitter
plate, downstream from which reaction occurs in a mixing layer.
The second limiting case involves premixed reactants at sufficiently low tempera­
ture so that their reaction rates are negligible. These reacta..'lts
of elevated but nonuniform temperature. The temperature inhomogeneities lead to
heterogeneities in composition because chemical reactions occur more rapidly in
hotter regions. If the reactions are exothermic, then heat release further elevates
temperatures. Cooler regions of the flow must be heated by diffusion until their
temperature is raised sufficiently for heat release to occur within them. An example
of such a flow would arise if a heated plate were inserted in a flow consisting of
TURBULENT FLOWS INVOLVING CHEMICAL REACTIONS 355

reactants that at the low temperature of the external stream are inert but that at a
temperature below that of the plate will begin to react significantly.
We have described shear-flow situations to illustrate both limiting cases. How­
ever, these cases can also be understood in terms of grid flows, which play such an
important role in developing understanding of turbulence. Consider a simple reaction
occurring in an isothermal system ; if the reaction involves significant heat release,
we imagine the reactant concentrations to be small. In the non-premixed case,
upstream from the grid only one reactant is flowing but at the grid the second
reactant is injected. Downstream of the grid, the inhomogeneities in concentration
of the two reactants, measured by Yi'2, decay due to both turbulent mixing and
chemical reaction, while inhomogeneities in product arise initially from inhomo­
Annu. Rev. Fluid Mech. 1976.8:351-376. Downloaded from www.annualreviews.org

geneities in chemical production but eventually decay due to mixing.


Access provided by University of York on 01/27/15. For personal use only.

Sub-limiting cases of this non-premixed limit relate to the chemical behavior in


such a flow. If the temperature and pressure levels are such that the chemical
reactions are infinitely fast, the case of fast chemistry prevails. In this limit, which
we discuss in some detail below because it corresponds to an especially simple but
informative case, the two reactants cannot coexist, i.e. Y1 Y2 0, and the hetero­
=

geneities in reactants are separated by reacting surfaces at which reactants are


destroyed and product is formed. If the temperature and pressure are so low that
no reaction takes place, the situation corresponds to pure mixing of scalars down­
stream from a grid. Clearly, temperatures and pressures intermediate between those
corresponding to these sub-limiting cases lead to a range of interaction of mixing
and chemical reaction. When the reaction rates are relatively slow, the situation
approaches that of the premixed limit, but in general it displays features of both
limiting cases.
In the other limiting case of premixed reactants, the flow upstream from the grid
may be thought to consist of a homogeneous mixture of reactants at a low
temperature, traveling at a velocity exceeding their flame speed. The grid is heated
to a temperature sufficiently high so that reactants at that temperature will combine
chemically and produce in the downstream flow further temperature heterogeneities,
which may grow or decrease depending on the extent of chemical reaction and on
the exothermicity of the reaction. Far downstream, after all of either reactant has
been consumed, the temperature fluctuations will decay due to mixing. Associated
with chemical reaction in this case will be heterogeneities in composition of
reactants and of product. Again in this limiting case the pressure level of the flow
and the temperature of the grid determine a range of chemical behavior ; if the
temperature and the pressure are low, no chemical reaction occurs and we deal with
a conventional grid flow, but if the temperature and pressure are high, the entire
chemical reaction will take place close to the grid.
These various grid flows provide situations sufficiently simple for quantitative
descriptions to be developed in certain of their limiting cases, if the chemistry is
simple enough. Work of Corrsin (1958) and O'Brien (1966, 1971) may be consulted
in this regard. One interesting result from these studies relates to the effect of
chemical reaction on the spectra of the participating species ; Corrsin (1958) inferred
for the case of a unimolecular reaction in an isothermal background fluid that the
Annu. Rev. Fluid Mech. 1976.8:351-376. Downloaded from www.annualreviews.org
Access provided by University of York on 01/27/15. For personal use only.

356

a
LIBBY & WILLIAMS

b
TURBULENT FLOWS INVOLVING CHEMICAL REACTIONS 357

chemical reaction is not spectrally selective, i.e. the shapes of the spectra of Yl will
alter with time as in a pure mixing problem. More generally, the chemical creation
term is a one-point quantity and therefore of itself does not necessitate introduction
of multipoint.correlations, although its high degree of nonlinearity in general can
produce strong spectrai transfers if the dynamics of multipoint correlations are
studied in a transform space.

5 NON-PREMIXED REACTANTS WITH FAST CHEMISTRY

A case of applied interest for which some advance has been achieved is that of
non-premixed reactants in the limit of infinitely fast chemical kinetics. The two­
Annu. Rev. Fluid Mech. 1976.8:351-376. Downloaded from www.annualreviews.org

dimensional mixing layer with one reactant in each stream and the fuel jet
Access provided by University of York on 01/27/15. For personal use only.

discharging into a moving or quiescent oxidizer stream are illustrative of the flow
configurations we consider now. An example in which the fast-chemistry approxi­
mation may be reasonably good is the butane jet discharging into quiescent air, as

shown in Figure la; a spark shadowgraph of the upstream portion of this flow is
given in Figure lb. For conceptual simplicity, we confine attention to isothermal
flows, i.e. to highly dilute reactants in a uniform background fluid. The chemical
system is a simple one, corresponding to equation (3) but with the added provision
that the pressure and temperature are sufficiently high so that the two reactants
do not coexist. Mathematically, Y1(x, t) Y2(x, t) = 0 at all x, t.

5.1 Corifigurations ofReaction Zones


The time history of the concentrations of reactants at a given point in space where
product is formed corresponds to alternating periods during which one reactant
M1 is present and the other absent, followed by a period during which the M1
is absent and M2 present. The times at which the concentrations switch correspond
to crossings of the reaction surface or flame sheet at which reactants are destroyed
and product created. In this physical picture the creation term in equation (4) is
thought of as a pulse train.
The geometry of the reaction surface depends on the flow being considered. In
the two-dimensional mixing layer the flame sheet may be viewed as an osci)lating
surface, which may be multiply connected and which behaves dynamically much
as a turbulent-irrotational interface. In the fuel jet the flame sheet is an oscillating
cylindrical surface that terminates at various downstream positions and occasionally
spawns a free, encapsulated, and fuel-rich eddy, which in due course is consumed
(see Figure la). In the case of grid flows involving non-premixed reactants, the
flame sheet within the reaction zone should perhaps be considered cellular with
alternate cells of fuel-rich and oxidizer-rich fluids, separated by a reacting surface.

+- Figure 1 Photographs of turbulent diffusion flames for butane jets issuing from a tube
of inside diameter 1.02 em into quiescent air (Wohl, Gazley & Kapp 1949). (a) Direct
photograph of short exposure for a Reynolds number of 7000. (b) Spark shadowgraph for
a Reynolds number of 8640, showing the upstream portion, approximately 25%, of the
field of a.
35 8 LIBBY & WILLIAMS

The physics depicted above clearly implies a dominant influence of turbulence


on the effective chemical kinetics. At the molecular level, infinite kinetic rates prevail,
but the effective rate in terms of WI is determined by the frequency with· which
the flame sheet crosses the point in question and by the mean rate of destruction
of reactants associated with each passage.

5.2 Analytical Descriptions


Hawthorne, Weddell & Hottel (1949) appear to have been the first to call attention
to the role of mixing in determining the effective chemical kinetics and to establish
many of the basic ideas involved. Toor (1962) provided an important conceptual
Annu. Rev. Fluid Mech. 1976.8:351-376. Downloaded from www.annualreviews.org

idea in relating the phenomenology of the chemical behavior to that of passive


Access provided by University of York on 01/27/15. For personal use only.

scalars. These early contributions are marred by the assumption of a Gaussian


distribution for the fluctuations of concentration, an assumption not generally
valid. More recent advances have been made by O'Brien (1971), Spalding (1971b,
1975), Lin & O'Brien (1974), Bilger & Kent (1974), Bush & Fendell (1974), Libby
(1974, 1975), Alber & Batt (1975), and others.
To understand the approach from one simple viewpoint, consider the approximate
mean conservation equation for the species i

i= 1, 2,3. (7)

Introduce the element mass fractions


i = 1,2. (8)
The elements are conserved, so
a - - -

",(UkZi+U;'Z;) = 0 i = 1,2. (9)


uXk

We now introduce a linear combination of Zl and Zz as a new variable that we


denote here by C. The exact definition of , or its counterpart differs depending on
the flow considered and from author to author; it is, for example, possible to scale
( from - 1 to 1 or from 0 to 1. In terms of the two-dimensional mixing layer it is
convenient to normalize all concentrations with respect to the concentration of Y1
in the faster moving stream, denoted by Y11; then a concentration parameter
f = Yzz/wY 11 quite naturally arises, where Yzz is the concentration of Yz in the
slower stream, and where w = W2/Wt. In this case
wYll( = \:VY1- Yz = WZI-ZZ (10)

and the range is - f �, � 1. Clearly, is a conserved scalar.

5.3 The Probability-Density Functionfor the Conserved Scalar


Because of the non-coexistence of Y1 and Yz, positive or negative values of , at a
given space-time point can be related directly to either Y1 or Yz, respectively. This
implies that if we had a probability-density function (pdf) for" i.e. P«(, x), then
TURBULENT FLOWS INVOLVING CHEMICAL REACTIONS 359

tl
¥;:/Yll = (P«, x) de.
(11)
o
-(�/Ydf= f (P«(,x)d(.
-y

In other definitions of the characteristic scalar ( similar expectation values arise, but
the limits are changed ; e.g. if 0 � ( � 1, a concentration parameter will determine
the integration limit within the zero-one range.
If P(" x) is known, then at x equations (11) determine the mean concentration
of reactants, and the mean of either of equations (8) determines the mean concen­
Annu. Rev. Fluid Mech. 1976.8:351-376. Downloaded from www.annualreviews.org

tration of the product. Different suggestions for obtaining P(" x) have been made.
Lin & O'Brien (1974) use experimental data on the pdf of passive scalars such as
Access provided by University of York on 01/27/15. For personal use only.

temperature, suitably scaled for the range -f � ( � 1; thus one experimentally


obtained pdf corresponds to a variety of reacting flows depending on f.
For a priori calculations, fluid mechanical information relative to e. e.g. C, ('2,
etc, can be used to estimate P(" x). Various models for P«, x) have been assumed
by Spalding (1971b), Bush & Fendell (1974), Libby (1974), and Alber & Batt (1975).
Libby (1974, 1975) has emphasized the importance of accounting for intermittency
in relating information on the statistical properties of ( to the pdf of (.

0.5 I

AIR + N20 4 -0- DATA BATT


PDF THEORY
[N02] [N204]
004 0.74 0.00268
[N2J
= -- =

[N2J
UI 25 FTISEC
=

TI 252°K
=

0.1

o
-2 2
<T = 12y/K
Figure 2 Profiles of intensities of concentration fluctuations for nitrogen dioxide, measured
and calculated for the two-dimensional mixing of air at room temperature with cooled air
containing nitrogen tetroxide (Alber & Batt 1975). The theoretical curve is obtained from
the probability-density function of temperature fluctuations by use of the assumption of
fast chemistry.
360 LIBBY & WILLIAMS

If the reaction zone is embedded in a fully turbulent region and the intensity of
the fluctuations of ( is sufficiently small so that the bounds on ( are ineffective,
then a Gaussian form of P«(. x) may be reasonable, and estimates from equations (11 )
are readily made on the basis of and ('2. This has been the approach to
"(
calculations in the past but without adequate acknowledgment of the limitations.
Recently Alber & Batt (1975) have presented such a calculation for a two­
dimensional mixing layer in which a cooled air stream containing added N204 is
mixed with room-temperature air. The reaction

(12)
Annu. Rev. Fluid Mech. 1976.8:351-376. Downloaded from www.annualreviews.org

is fast and has a known equilibrium constant which favors N204 at the lower
Access provided by University of York on 01/27/15. For personal use only.

temperature and N02 at room temperature. Here temperature serves as the


conserved scalar, and its pdf was measured and fitted to a Gaussian density at
various positions in the flow. From this, the equilibrium relation and element
conservation, the pdf for the concentration [N02] can be calculated. In particular,
the concentration fluctuation intensity [N02]'2 can be computed at the points

0.6

0.5

0.4

�I�
0.3



Q. 0.2

0.1

0.0 I I
-4.0 -2·0 o 2.0 4.0 6.0

a (8-9)/�
Figure 3 Normalized probability-density function for temperature in the wake of a heated
cylinder, 400 diameters downstream, at a Reynolds number of 2800 (LaRue & Libby 1974).
The smooth line represents a Gaussian density for comparison and the vertical dashed
line corresponds to the threshold level used for intermittency discrimination. The ratio of
the transverse distance to the geometric mean of the cylinder diameter and the distance
downstream is 0.0275 for a, 0.209 for b, and 0.257 for c. Corresponding fractions of the
time that the wake is turbulent are 1.000,0.951, and 0.882, respectively.
TURBULENT FLOWS INVOLVING CHEMICAL REACTIONS 361

0.6

0.5

0.4

�-...
0.3
IQ;
I
Annu. Rev. Fluid Mech. 1976.8:351-376. Downloaded from www.annualreviews.org



Access provided by University of York on 01/27/15. For personal use only.

0.2
Q.

0.1

0 I I I
-4.0 0 2.0 4.0 6.0

(8-8)/ ..f?
b

3.0

2.5

2.0


-...

ICD
1.5
I


Q. 1. 0

0·5

0
-4.0 -2.0 0 2.0 4.0 6.0

(8-8)r/�
c
362 LIBBY & WILLIAMS

where the pdf for temperature was measured. Alber & Batt (1975) performed this
calculation and thereby obtained theoretical profiles for the intensity of NOr
concentration fluctuations, shown in Figure 2. This result then was compared with
measured profiles of the same quantity. The agreement is reasonably good,
differences in the wings being attributable to departure of the temperature pdf
from normality.
That substantial departures from Gaussian densities can occur is verified by the
data of LaRue & Libby (1974) shown in Figure 3. Near the centerline of the
turbulent wake of a heated cylinder, no intermittency is observed, and the pdf is
nearly Gaussian (Figure 3a). At points farther off axis (Figures 3b and 3c),
intermittency distorts the pdf appreciably, placing a spike at the temperature of
Annu. Rev. Fluid Mech. 1976.8:351-376. Downloaded from www.annualreviews.org

the external fluid. Data in Figure 3 can be used in equation (11) to calculate mean
Access provided by University of York on 01/27/15. For personal use only.

reactant concentrations in a turbulent wake after Lin & O'Brien (1974), for example.

5.4 Dynamics of the Flame Sheet


The location of the reaction zone within the mlXlng layer is determined by
stoichiometry and by the relative concentrations of reactants in the two streams,
i.e. by a parameter equivalent to Y. Libby (1974) has conjectured that if either
Y � 1 or Y � 1 the flame sheet can be made contiguous with the turbulent­
irrotational interface on the high- or low-speed side of the mixing layer, respectively.
If this conjecture is true, information on the statistical geometry of turbulent­
irrotational interfaces can be interpreted in terms of flame-sheet behavior, at least
for these special cases. Although there is no direct experimental evidence to
support this conjecture, it is difficult to imagine how it could be in error.
There has been considerable study of the structure of the flame sheet in this fast­
chemistry limit to determine the dependence of its thickness on the reaction rate
alld on the flow in its neighborhood. The point of view has been that, on a scale
appropriate for examination of its structure, the flame sheet is a flat, quasi-steady
laminar flame embedded in a strain field such that reactants diffuse into the sheet
and product diffuses away. These studies are not discussed here; Williams (1975a)
provides a current entry into the literature of this matter.
There are various approaches to the difficult problem of calculating mean
consumption rates from the dynamics of flame sheets. At one extreme there are
attempts to solve equations for the evolution of an appropriate pdf and at the
other direct modeling of �i in terms of mean flow properties. The physical basis
for a fundamental approach is described at the end of Section 5.1. Spalding (1971a)
in connection with premixed flames assumed that �1 is proportional to aatlaxz,
an assumption clearly invalid if the mean strain vanishes; recently Spalding (1975)
developed improved modeling for both premixed and non-premixed systems,
replacing the mean strain rate by the average of a fluctuating strain rate. Marble
(1974, unpublished) incorporates unsteady-laminar-flame theory into a strain theory
of turbulent reacting flows. For the passive scalar, the joint pdf of concentration
and of the magnitude of the concentration gradient can be shown to be relevant
to �l (Williams 1975a, Bilger 1975); under the assumption of statistical independence
for these two variables, the mean scalar dissipation rate is found to be proportional
TURBULENT FLOWS INVOLVING CHEMICAL REACTIONS 363

to Wi (Bilger 1975). Detailed comparisons of predictions obtained by the various


methods have not yet been performed.

5.5 Experimental Aspects


It is worth noting that the situation described here is so idealized that it provides
a useful vehicle for the experimental verification of ideas as to the primitive effect
of turbulence on chemical reaction. A passive scalar, such as temperature or
concentration of a dilute contaminant, must be measured and related to our ( or
its counterpart by appropriate scaling; the concentrations of the two "reactants" are
given by the values of ( � 0, while the concentration of product is obtained from
equation (8). The statistical geometry of and the velocities at the flame sheet are
Annu. Rev. Fluid Mech. 1976.8:351-376. Downloaded from www.annualreviews.org

given by the behavior of a surface of constant (. Such measurements involve the


Access provided by University of York on 01/27/15. For personal use only.

technique of conditioned sampling based on a specified· level of the passive scalar


(cf Libby 1974). Measurements of the sort suggested here can be made with a precision
not generally possible in more realistic reacting flows. While the experimental
verification of our ideas of turbulent reacting flows as provided by such
experiments on passive scalars is valuable, it will clearly be necessary in the longer
term to obtain quality measurements in turbulent flows with actual chemical
reactions, perhaps at first in simple chemical systems (cf Alber & Batt 1975).

6 THE EFFECTS OF HEAT RELEASE

The situations we have described so far relate largely to idealized and simplified
flows in which the reactants are sufficiently dilute so that chemical reaction does
not alter the isothermal background fluid. In liquids this idealization may be
reasonably valid even in practically interesting flows, but in most of the cases that
motivate the study of turbulent reacting flows, e.g. those involving combustion,
significant temperature and density variations occur.
Even if reactants are too dilute to influence the temperature field, preexisting
temperature fluctuations in the turbulent fluid can be significant in terms of the
chemical kinetic behavior of the reactants, especially if the activation te�peratures
of the chemical steps are high. This might be the situation for a trace species, such
as a pollutant from a combustor, which would be fluid-mechanically and
energetically unimportant but whose behavior would be determined by the tempera­
ture fluctuations in the background fluid. In this case density fluctuations may or
may not be appreciably large, depending on the extent of the prevailing
temperature fluctuations.
The ideas on non-coexistence of reactants in cases involving fast chemistry carry
over to flows involving significant heat release. In particular, relating the chemical
behavior to that of a passive scalar in the nonisothermal flow remains valid if
molecular Lewis numbers are unity, Mach numbers are low, temperature and
composition fluctuations are absent in the approaching streams, and the chemistry
is rapid. Conditions on chemical kinetics needed for the occurrence of this limit
have been discussed by Williams (1975b). When the kinetic rates are taken to be
infinite, the effect of temperature on chemical kinetics of the primary reaction is
364 LIBBY & WILLIAMS

negligible. However, the variability of density leads to dynamic coupling among


the velocity components and the state variables, pressure, temperature, and mass
fractions; the density plays a role in the conservation of all dynamical quantities
and in the creation term as well. In addition, exothermicity is expected to generate
pressure disturbances that have both local and long range influence. Thus, there
arises a variety of effects that are inadequately understood at present.
A fundamental question concerns the influence of exothermicity on the turbulent
kinetic energy and its spectrum. Although Eschenroeder (1964, 1965) postulated an
effect of local, spectrally dependent power sources on evolution of turbulence in
wave-number space, no physical basis for the occurrence of such an effect was
presented. Many investigator2 appear to believe that influences of heat release on
Annu. Rev. Fluid Mech. 1976.8:351-376. Downloaded from www.annualreviews.org

turbulence, such as this, are not likely to be large, . possibly because estimates
Access provided by University of York on 01/27/15. For personal use only.

suggest the coupling coefficient to be of the order of the Mach number of the
turbulence. In this regard, it may be relevant that the extensive literature on
predictive methods for turbulent boundary layers in high-speed flows is largely based
on straightforward extensions of the corresponding methods of low-speed flows,
with the explicit effect of density fluctuations neglected on the basis of Morkovin's
hypothesis (Morkovin 1964) that density fluctuations should not directly influence
turbulence if fluctuations in Mach number are small (although mean-density
variations must be taken into account). The situation is less clear when the density
variations arise from local chemical heat release. There appears to be some
experimental evidence (Gunther 1970) that turbulence levels in non-premixed flames
are lower than those in corresponding isothermal flows and that the opposite is
true for premixed flames, but the degree of generality of these observations is
uncertain at present. This entire topic needs more thorough attention, directed
toward developing understanding and predictive methods that are more firmly
based.
It is interesting that there is little agreement concerning an optimum averaging
method. The straightforward decomposition of density into its mean and fluctuating
parts, p(x, t) = p(x)+ p'(x, t), results in proliferation of terms involving p', all of
which must be modeled or neglected (cf Bray 1974). The alternative of Favre
averaging, mentioned in Section 3, in which all quantities except pressure are mass
Uj(x, t) PUj(x)/p(x)+ u;'(x, t), produces describing equations in which
---

averaged, e.g. =

all terms except those of molecular transport are formally close to those of
constant-density flows, e.g. the flux terms p u; uj become pui'u'j. Accordingly, the
modeling needed to provide closure becomes easier to rationalize, perhaps
deceptively so. The resulting advantages are not uniformly accepted; Bray (1974)
has argued, for example, that if development of a phenomenology of turbulent
reacting flows is to parallel an experimental effort that will provide detailed data
on the velocity components, Uj, i = 1,2,3, and on the state variables, p, p, T; Yj,
i = 1,2, ... N, then conventional averaging must be pursued despite the clutter of
the resulting equations and despite the resulting extensive modeling of terms needed
to handle the equations. An alternative view might be that the experimental data
suitable for direct comparison with prediction may be relatively limited compared
with the variety of quantities appearing in the describing equations, and may in fact
TURBULENT FLOWS INVOLVING CHEMICAL REACTIONS 365

be more closely related to Favre quantities, in which case conventional averaging


is required only for certain particular terms subject to comparison.
Another consideration relative to the choice of conventional versus Favre
averaging concerns the molecular terms. Strict Favre averaging of terms such as
the transport term, (O/OX2)(Jl. OUt/OX2). or such as the terms leading to the dissipation
of turbulent kinetic energy, Ui(OjOX,J(}J. ou;jox,J. introduces considerable compli­
cation. However, these terms and their counterparts in conventional averaging
generally require modeling to achieve closure. Thus the complication in the
molecular terms associated with Favre averaging does not appear to be crucial, and
the choice probably should be made on other grounds.
Annu. Rev. Fluid Mech. 1976.8:351-376. Downloaded from www.annualreviews.org

Experimentally, there currently exists growing recognition of the occurrence and


Access provided by University of York on 01/27/15. For personal use only.

potential importance of low-frequency, coherent structures in turbulent shear flows


of variable density (Brown & Roshko 1974). The occurrence in diffusion flames of
low-frequency oscillations has recently been documented experimentally (Kumar
1974, Grant & Jones 1975). Theoretical studies of these phenomena and of the ways
in which they can be of practical importance are just beginning.
Sufficient research on the phenomenology of turbulent flows with variable density
has not been carried out to establish a sound basis for the treatment of such flows.
Until such a basis exists progress on the treatment of flows involving significant
heat release may be slow.

7 PREMIXED TURBULENT REACTING FLOWS

As indicated earlier, phenomena may arise in premixed turbulent reacting flows


that differ from those in non-premixed systems. The most striking new quantity is
the turbulent-flame speed. Situations occur in which just as in the case of laminar
flames (Williams 1971), a reaction front propagates into the flow of reactants at a
definite velocity. For example, in the premixed grid flow described in Section 4. the
grid may be unheated and thus be incapable of providing temperatures sufficiently
high for reaction to occur but the downstream flow may be ignited by an external
heat source. It may then be found that the reaction region remains spatially
stationary in the mean only if the average upstream velocity is fixed at a defiilite
value, the turbulent-flame speed.
Laminar-flame speeds depend on transport and chemical-kinetic properties of the
combustible mixture, but turbulent-flame speeds depend in addition on characteris­
tics of the turbulence, e.g. those produced by the grid. Statements sometimes found
in the literature to the effect that turbulent-flame speeds do not exist are based
either on the occurrence (in some experiments) of transient conditions rather than
conditions that are steady in the mean or on the reluctance to accept a flame speed
dependent on the fluid dynamical configuration. There are many situations in which
the concept of a turbulent-flame speed is useful.
One example of practical importance is provided by the stabilization of a flame
in a premixed turbulent flow by means of an array of rods placed transverse to
the mean flow direction, such as is found in turbojet afterburners. Provided that the
mean velocity of the approach flow exceeds the turbulent-flame speed, in the mean
366 LIBBY & WILLIAMS

the flames spread obliquely from the stabilizer rods at angles simply calculable from
the flame speed. This flow has been employed in numerous experiments, ranging
from design of power-producing equipment to fundamental measurements of
turbulent-flame speeds and structures; it affords a convenient vehicle for discussion
of the various regimes of premixed turbulent flame propagation that may occur.

7.1 Wrinkled Laminar Flames


There is wide agreement that turbulence of sufficiently low intensity and large scale
wrinkles a laminar flame without disrupting its integrity. This idea was first expressed
by Damk6hler (1940) and subsequently has been verified in a number of experiments
Annu. Rev. Fluid Mech. 1976.8:351-376. Downloaded from www.annualreviews.org

on rod-stabilized flames. Figures 4a and 4b show one such flame in a turbulent


flow of low intensity ; the wrinkled structure is seen in the short exposure but is
Access provided by University of York on 01/27/15. For personal use only.

washed out in the normal exposure. In the limit of extremely low intensity, the
structure of the wrinkled flame will be the same as that of a normal, unstrained
laminar flame,provided that the integral scale of the turbulence is large compared
with the laminar-flame thickness. However, as soon as the turbulent intensity
becomes large enough (at a fixed integral scale, say) for the Kolmogorov length
scale to be no longer infinitely large in comparison with the laminar-flame thickness,
the strain of the turbulence modifies the laminar-flame structure and flame speed.
The correct theoretical analysis of this "flame-stretch" effect was first given by
Klimov (1963) and has been reviewed previously (Williams 1975a). Klimov has also
indicated how flame stretch modifies the structure of diffusion flames (see Williams
1975b).
There has been a common belief that if the turbulent flame consists of a
wrinkled laminar flame, then it is a simple matter to calculate any desired property.
This belief is false ; for example not even the turbulent-flame speed can be calculated
immediately, since it will depend on the geometry of the wrinkling. Many authors
assign a shape to the wrinkles essentially arbitrarily, e.g. a saw tooth with angles
dependent on the ratio of the rms turbulent-velocity fluctuation to the laminar-flame
speed, thereby determining the desired properties of the turbulent flame. There is
little justification for such procedures,and it is better to develop a statistical analysis
of the interaction of the flame discontinuity with a turbulent field, as Tucker (1956)
tried to do. Unfortunately,any such undertaking is beset with fundamental difficulties
from the outset, stemming from laminar-flame instability. Landau ( 1944) showed
that the plane laminar flame, treated as a discontinuity whose propagation speed
is independent of curvature, is unconditionally unstable to two-dimensional dis­
turbances,and much subsequent theoretical and experimental work was devoted to
resolving the question of why steady laminar flames are observed to exist at all (e.g.
Istratov & Librovich 1966). The explanation lies in the effect of flame curvature
on flame speed, calculable approximately when proper account is taken of transport
effects within the flame; it is now well accepted that laminar flames are stable to
two-dimensional disturbances of sufficiently small wavelength.
To build these results into a proper wrinkled-laminar-flame theory, accounting
for statistical aspects of the turbulent field, is a formidable but not insurmountable
task that apparently never has been tackled. Experimentally, for reasons that are
TURBULENT FLOWS INVOLVING CHEMICAL REACTIONS 367
Annu. Rev. Fluid Mech. 1976.8:351-376. Downloaded from www.annualreviews.org
Access provided by University of York on 01/27/15. For personal use only.

a b

c d

Figure 4 Photographs of premixed turbulent flames. a and b are stoichiometric city gas/air
flames stabilized on a 7.6 cm diameter ring in a flow having an integral scale of 1.5 mm
and an rms velocity fluctuation equal to 2.8% times the mean velocity of 420 cm sec-1
(Williams, Hottel & Gurnitz 1965). Exposure times are 1 msec in a and 40 msec in b.
Prints c and d are city gas/air flames at a fUel/air equivalence ratio of 0.71, stabilized at
the 7.5 cm diameter mouth of a short contraction section placed above a perforated plate
having hole diameters of 6 mm, producing at the port an rms velocity fluctuation lying
between 17% and 20% of the mean velocity of 430 cm sec-1 (Yamazaki & Tsuji 1962).
c is a visible long-time exposure and d a spark schlieren.
368 LIBBY & WILLIAMS

now understood theoretically, the wrinkled laminar flame in the turbulent flow
assumes a cusped shape or cellular structure, with the cusps typically pointing
toward the burnt gas. This can be seen not only in Figure 4a but also in Figure
5, where the sharp ridges of the spherically expanding flame lie on the inside. This
aspect of the structure must be included in a proper analysis, the cusps demanding
special treatment.
These difficulties of analyzing the dynamics of a wrinkled laminar flame in a
turbulent flow suggest that alternative theoretical approaches also are worthy of
pursuit. One such approach is to average the original conservation equations and
work with moments. The principal difficulty that this method encounters for real
flames is caused by the strong temperature dependence of the overall rate of heat
Annu. Rev. Fluid Mech. 1976.8:351-376. Downloaded from www.annualreviews.org

release. The high nonlinearity of the reaction-rate term dictates that the intensity of
Access provided by University of York on 01/27/15. For personal use only.

the temperature fluctuation must be low if closure is to be achieved justifiably (i.e.


without ad hoc modeling assumptions) at a manageably small number or moments.
Recently Williams (1975a) completed an analysis of this type for grid turbulence of
low intensity, in the final stage of decay. The limit of a large overall activation
energy for the reaction was considered, and to facilitate solving the mean equations
spectral decompositions in two space dimensions and time were introduced. It was
shown that just as in the case of the laminar flame (Williams 1971) convective­
diffusive and reactive-diffusive zones exist. Scalar fluctuations, e.g. of temperature
and concentrations of species, are generated in the convective-diffusive zone by the
action of velocity fluctuations on the mean scalar gradient. Persistence of the
fluctuations so generated into the reactive-diffusive zone increases the mean-square
scalar fluctuation level there and in so doing modifies the flame speed. A formula

Figure 5 Flash schlieren of a spherically expanding turbulent flame, taken 44 msec after
ignition, in a rich propane-air mixture baving an equivalence ratio of 1.768 (Palm-Leis &
Strehlow 1969). The rms velocity fluctuation is 2.3% of the mean velocity of 418 cm sec-!,
and the integral scale is 1.5 mm.
TURBULENT FLOWS INVOLVING CHEMICAL REACTIONS 369

for the turbulent-flame speed was derived that in the first approximation involves
both the variance and the skewness of the scalar fluctuations at the downstream
side of the convective-diffusive zone.
When the analysis of Williams (1975a) is specialized to the case in which the
ratio of the laminar-flame thickness to the integral scale of turbulence is small, the
flow conditions correspond precisely to those under which the wrinkled laminar
flame is expected to exist. Under these conditions, the relevant variance and
skewness in the flame-speed formula become those of the time integral of the sum
of the nondimensional velocity and temperature fluctuations in the approach flow.
The ratio of the turbulent- to laminar-flame speed then is predicted to increase as
Annu. Rev. Fluid Mech. 1976.8:351-376. Downloaded from www.annualreviews.org

the turbulence intensity, turbulence integral scale, or reciprocal of the overall


activation energy increases. Secondary predicted effects are an increase in flame speed
Access provided by University of York on 01/27/15. For personal use only.

with decreasing overall reaction order and with increasing ratio of the initial thermal
energy to the heat of reaction. The first two of these dependences can be tested
qualitatively against experiment. The ratio of turbulent- to laminar-flame speed has
indeed been found to increase with increasing turbulence intensity at constant scale
and with increasing turbulence scale at constant intensity (Ballal & Lefebvre 1974),
in the large-scale, low-intensity regime. The latter effect, the increase in flame speed
with increasing scale, has not been predicted before by any wrinkled-laminar-flame
theory. There must be implications of this dependence concerning the variation of
instantaneous laminar-flame shapes with turbulence scale, but such implications
are difficult to extract from the averaged equations that appear in the statistical
treatment. Further studies would be of basic interest, directed toward calculating
most probable wrinkled-laminar-flame shapes from statistical theory.

7.2 Distributed Reaction Zones


In the same paper in which Damkohler enunciated the wrinkled-laminar-flame
view (Damkohler 1940), he reasoned that the effect of turbulence of sufficiently
small scale should be to increase transport coefficients within the laminar flame.
The idea is that, under conditions such that the laminar flame is thick in
comparison with the integral scale of turbulence, the propagation mechanism of
the turbulent flame will be the same as that of the laminar flame, but the laminar
diffusivity must be replaced by the turbulent diffusivity. A conclusion from known
laminar-flame structure then is that the ratio of turbulent- to laminar-flame speeds
becomes ST/SL Jrt.T/rt.L. wherert.T and rt.L are the turbulent and laminar diffusivities.
=

A further idea is that under these small-scaie conditions the turbulence convects
the reaction regions, producing a distributed reaction zone (Summerfield, Reiter,
Kebely & Mascolo 1955).
The view has a certain appeal and also some merit. At low-turbulence intensities,
the two-zone structure of the laminar flame persists (Williams 1975a). However,
turbulent influences in the reactive-diffusive zone can be of overriding importance,
and a more thorough look does not entirely support the simple flame-speed formula.
At higher intensities, the tendency for developing the distributed reaction zone may
occur. The limit in which the laminar flame is thick in comparison with the
turbulence scale has received relatively little attention because it is seldom en­
countered in practice. Laminar flames normally are less than one millimeter thick
370 LIBBY & WILLIAMS

and can be made to approach thicknesses on the order of one centimeter only by
burning at pressures well below one tenth of an atmosphere.

7.3 Open Flames in High-Intensity Turbulence


Under conditions of greatest practical importance, the rms-velocity fluctuation
exceeds the laminar-flame speed, and it becomes more important to distinguish
between open flames with little or no confinement and enclosed flames stabilized
on a rod within a duct. For the enclosed flame, the low-density reaction products
behind the flame often have a sufficiently high velocity with respect to the cool
reactants for turbulence levels in the reaction zone to be controlled largely by
Annu. Rev. Fluid Mech. 1976.8:351-376. Downloaded from www.annualreviews.org

generation of intense turbulence in the relatively strong shear layer that follows the
obliquely spreading flame. Open flames are more complex in the sense that in
Access provided by University of York on 01/27/15. For personal use only.

addition to possible influences of local shear-layer instabilities, turbulence charac­


teristics of the approach flow always have an important effect on the premixed­
flame structure.
It has become clear on theoretical grounds that the wrinkled-laminar-flame model
is untenable when the turbulence intensity is high. Specifically, Klimov (1963)
showed that if the product of a representative strain rate of the turbulent field with
a characteristic residence time in the laminar flame becomes large, then significant
strain-induced velocity gradients must occur in the interior of the flame. The
definition of a burning velocity for the laminar flame becomes imprecise ; if it is
defined as the gas velocity at the heat-release plane, measured with respect to the
motion of that plane, then it is negative throughout most of the flame. Consequently,
back-to-back flames tend to move toward each other on the average, and they can
meet, leading to local extinction, with a portion of a hot spot being replaced by
cool reactants. Thus, even if initially a wrinkled-laminar-flame sheet existed, it would
be twisted, develop holes, and soon would be totally unrecognizable. Figures 4b
and 4c show a premixed flame in turbulence of high intensity ; its structure is quite
different from that of the flame in Figures 4a and 4b. where the intensity of
turbulence is low.
Structures of premixed flames at high-turbulence intensities are not understood
today, although various qualitative ideas have been advanced [see, for example,
Shetinkov (1959), Schelkin & Troshin (1965), and Chomiak (1970)] . The topic is
complex and is in need of further fundamental elucidation. There have been a
number of attempts at modeling. Bray & Moss (1974) have embarked on a
promising line in which the well-known Shvab-Zel'dovich variables are used to
reduce to one the number of independent variables that influence the reaction rate
for a one-step overall process, and then possible shapes of the pdf for this
independent variable are inferred for limiting types of premixed-flame structure and
are used in estimating possible limiting forms for the reaction-rate term in the mean
equations.

7.4 Enclosed Flames in High-Intensity Turbulence


The classic case of a flame stabilized on a rod in a duct has received a considerable
amount of experimental attention because of its closeness to afterburner configura-
TURBULENT FLOWS INVOLVING CHEMICAL REACTIONS 371

tions (e.g. Cushing, Faucher, Grandbhir & Shipman 1967). There exist conditions
(Spalding 1959, 1967, 1971a) in which the flame-spread rate is controlled not at all
by chemical kinetics but rather by the rate of turbulent engulfment of fresh mixture
downstream from the stabilizer, and fluid mechanics rather than chemistry deter­
mines the combustion efficiency of a duct of a given length.

7.5 The Stirred Reactor


There is a related limiting case that ,in some respects is similar and in others
antithetical to that just mentioned. Imagine not one but many staggered rods
holding a flame in the duct in a high-speed flow. The rods are viewed as being
arranged to produce rapid mixing. It is then tempting to think of the reactants as
Annu. Rev. Fluid Mech. 1976.8:351-376. Downloaded from www.annualreviews.org

having mixed , completely with the hot gases before their combustion , begins. Such
Access provided by University of York on 01/27/15. For personal use only.

a limiting situation has been termed a well-stirred reactor. There is a residence time
(or distribution thereof) for fuel elements in the well-stirred reactor, and combustion
proceeds at a finite rate governed by the chemical-kinetic rate of the combustion
reactions at the flame temperature. The combustion efficiency is determined by this
rate and the residence time. The similarity with the rod-stabilized flame in the duct
is the strong mixing, but the difference is chemical instead of fluid-mechanical
control of the reaction rate.
In practice, intense mixing is obtained by use of a spherical reactor with multiple
fuel and oxidizer jets, instead of by the rod array indicated above (Longwell &
Weiss 1955). This reactor has received extensive study and use over the years, often
with devout faith that experiments with it yielded fundamental chemical-kinetic
data, totally unaffected by fluid mechanics, except through the residence time. Later
studies (Hottel, Williams & Miles 1967, lenkins, Yumlu & Spalding 1967) explored
experimentally the degree of mixing actually achieved in stirred reactors, showing
that the theoretical concept is good in some cases but not in others.
There are tendencies for investigators either to believe completely in stirred
reactors, to the extent that every rod-stabilized flame is thought to carry a stirred
reactor in the recirculation zone behind the rod, or to feel secretly that they cannot
possibly exist, on the grounds that mixing rates must at least influence, if not control
completely, the overall chemical-reaction rates. Actually, both views are too extreme.
While there should be many cases in which the Spalding (1971a) model is good,
there also are many situations, e.g. for relatively slowly reacting mixtures with low
flame temperatures, in which the stirred reactor model should be good. This serves
to emphasize the large number of different limiting cases that can be encountered
in premixed turbulent combustion.

7.6 Experimental Developments


For many years, experimenters apparently held the view that the only aspect of
turbulence relevant to turbulent-flame propagation was the intensity. The only
turbulent quantity reported in the experiments was the rms velocity [see, for
example, the experiments summarized by Williams (1965)] . This prompted
Kovasznay ( 1956) to chide the community for not at least reporting a length scale
as well. Since that time, the combustion experimenter's understanding of turbulence
372 LIBBY & WILLIAMS

has improved markedly. Not only have length scales been reported, but moreover
Ballal & Lefebvre (1974, 1975b) went to considerable effort to vary intensity and
scale independently in measuring turbulent-flame speeds of pilot-stabilized flames.
In so doing, they clearly demonstrated that the dependence of flame speed on scale
at high intensity is opposite from the dependence at low intensity. Even when
intensity and scale are not controlled separately, data are reported and interpreted
in forms that take both into account. For example, Andrews, Bradley &
Lwakabamba (1975), who used-four fans to establish in the central region of a large
chamber a nearly homogeneous turbulent field, in which propagation of a spark­
ignited flame was studied, report their data in terms of a turbulent Reynolds number
based on the Taylor scale, although fan speed was the only quantity varied. The
Annu. Rev. Fluid Mech. 1976.8:351-376. Downloaded from www.annualreviews.org

knowledge of turbulence appears not only in flame-propagation studies but also in


Access provided by University of York on 01/27/15. For personal use only.

other aspects of experimental studies of turbulent combustion. Thus, Ballal &


Lefebvre (1975a) use concepts of turbulence scale in interpreting spark-ignition
experiments.
New experimental techniques continue to be developed for studying premixed­
turbulent-flame propagation. Chomiak (1974) used the Townsend approach for
establishing a uniform, constant mean strain rate in grid turbulence to study the
influence of strain in premixed flames. He found that the strain can increase the
overall average intensity of combustion markedly, as would be expected from
theory (Klimov 1963). Kumar (1974) developed acoustic-probing methods for
ascertaining significant differences between turbulent structures of premixed and
diffusion flames stabilized in identical jets. The experimental information that is
now being generated should stimulate theoretical discussion and progress.

8 ENGINEERING ANALYSES

Because of the importance of making predictions of the properties of turbulent flows


with chemical reactions of a realistic nature despite the absence of a sound theory,
a number of engineering analyses of such flows have appeared. It is appropriate to
mention at the outset the extensive and important contribution to the methodology
of turbulent flows of a variety of types by Spalding and his co-workers. Although
some of the details of the present phenomenology are undoubtedly unsatisfactory,
the general framework for the treatment of various turbulent flows with and without
chemical reaction is provided by this work (cf Spalding 1974).
There are two types of engineering approaches. In one, conservation equations
for the various second-order correlations among the several species, the temperature
and in cases with significant heat release the density as well are developed according
to the general strategy of the so-called "second-order closure" methods, which have
become so highly evolved for the prediction of turbulent shear flows (cf Bradshaw
1972, Mellor & Herring 1973). In the second, one or more equations giving the
distribution functions for the velocity and/or the species concentration are developed
in a heuristic manner and solved ; from these solutions the various statistical
properties of the species of interest are obtained by calculating moments of the
distribution function.
TURBULENT FLOWS INVOLVING CHEMICAL REACTIONS 373

With respect to the first general approach we review the work of Borghi (1974)
first. Borghi considered the oxidation of carbon monoxide in the exhaust of a jet
engine according to a simple chemical reaction
(13 )

so that only one creation term needs to be treated. The molecular weight of all
species is considered the same and pressure fluctuations are neglected so that the
simple equation of state p pRT applies, The chemical kinetics of the reaction are
==

assumed to be described by an Arrhenius form that is expanded in terms of the


mean and fluctuating temperature as suggested by
Annu. Rev. Fluid Mech. 1976.8:351-376. Downloaded from www.annualreviews.org

exp ( - T.fT) = exp (- T.fT) { 1 + (T.fT2)T'


Access provided by University of York on 01/27/15. For personal use only.

+[(Ta�/2T4) - (T./P)] T'2+ . . . }. (14)

All third- and higher-order correlations are neglected so that attention is focused on
the second-order correlations of the forms Y/ YJ, Y;'T', and T'2. By introducing
Shvab-Zel'dovich relations among the temperature, the element mass fractions for
carbon and oxygen, and the concentration of the two reactants, CO and O2, Borghi
finds he can express all the needed second-order correlations in terms of YiT'
and T,2, where i corresponds to oxygen. Conservation equations for these two
correlations (including the various contributions thereto from the creation term)
and for the intensity of the fluctuations in the elements are closed by modeling
assumptions suggested by the second-order closure methods for nonreactive flows.
There results no effect of chemical reaction on turbulent kinetic energy. The
transport properties of the flow are described in terms of appropriate turbulent
Prandtl and Schmidt numbers and an eddy viscosity developed from the well-known
algebraic relation between the mean shear stress and the turbulent kinetic energy.
This analysis produces twelve equations that are of the generalized convection­
diffusion type, and which are solved by the Patankar-Spalding routine. This brief
description of the approach gives a picture of the complexities involved and of the
various assumptions necessary to treat this relatively simple flow. Nevertheless, the
results from a calculation based on this method and given by Borghi are
interesting, even if only qualitatively correct. The calculations are initiated at the
exit plane of a jet discharging into quiescent air. Chemically the air should be
considered an additional source to oxidize the CO in the jet. Initial values for the
mean and intensities of the temperature and species concentrations are assumed.
The flow that develops in the downstream exhaust jet has an identifiable structure ;
in the core region of the jet, the temperature fluctuations on top of a high mean
temperature lead to rapid oxidation of the CO. Near the edges of the jet where
mixing with the cool external air occurs, reaction is inhibited because of the
reduced mean temperature. Moreover, dilution of the CO due to mixing is found
not to be sufficiently rapid to reduce the CO concentrations in the outer portions
of the jet below that in the central region, so that the lowest CO concentrations
are in the middle of the jet.
Donaldson in a series of papers (cf Donaldson & Hilst 1972) has extended his
374 LIBBY & WILLIAMS

second-order closure methodology to chemically reacting flows. In published work


to date he has considered two reactants undergoing a unidirectional reaction with
finite chemical kinetics. Thus in the creation terms various second- and third-order
correlations of the forms Y/YJ. k'Y/. ¥;,2 arise. The conservative equations for these
quantities lead to correlations with the velocity components, e.g. in the form ui YjY':.
which are expressed in terms of a gradient model. In recent work Donaldson &
Varma (1975) have discussed the implications of the various correlations and the
influence of turbulence on chemical behavior in terms of an idealized turbulent
flow consisting of a repeated sequence of eddies with various assumed temperatures
and concentrations of reactants and products. Donaldson concludes on the basis
Annu. Rev. Fluid Mech. 1976.8:351-376. Downloaded from www.annualreviews.org

of several calculations that the second-order closure methodology gives promise of


permitting useful calculations to be carried out.
Access provided by University of York on 01/27/15. For personal use only.

We mention here that there are many other contributions to the engineering
literature of turbulent reacting flows. In the early 1960s when there was considerable
interest in the chemical behavior of hypersonic wakes and in the supersonic com­
bustion of hydrogen, there were many papers published on the subject. In these
papers the fluid mechanics of the turbulence was described by the best
phenomenology available at the time, but the chemical effects embodied in the
creation term were treated without any turbulent effects whatsoever ; Libby (1962)
is representative. In recent years this shortcoming has been attacked by several
investigators. We consider the Borghi and Donaldson contributions to be indicative
of the current status of work on more practical flows.
The second engineering approach to turbulent-reacting flows involves develop­
ment of one or more equations for a distribution function from which any
desired statistical quantity can be computed. Chung in a series of papers beginning
in 1967 (cf Chung 1975) adopts a Fokker-Planck type of equation for the
distribution function for the velocity feu ; x) and then assumes that the joint
distribution functions of species concentration and velocity can be expressed as a
delta function in concentration multiplied by that velocity distribution function.
There seems little physical basis for this assumption, but it may provide approxi­
mate solutions.
Although not completely compatible with the spirit of this section, we mention
here the approach to chemically reacting turbulent flows based on development of
an equation either for the probability-density function of the concentration of a
passive scalar, such as our C. or for the joint probability-density function of a
scalar such as a species concentration, and temperature. Lundgren (1967, 1969) has
explored such an approach for nonreacting flows. Such direct attacks face a variety
of critical assumptions in order to achieve progress ; their contribution to the
understanding and prediction of turbulent reacting flows of even the simplest type
is unclear at the present time. An advantage of this procedure is that by starting
with correct functional equations, the character of all approximations introduced
can be made explicitly clear. Dopazo & O'Brien (1974, 1975) have reviewed this
approach and have indicated the importance ofexperimental data on several joint and
conditioned probability-density functions in order to guide theoretical developments.
TURBULENT FLOWS INVOLVING CHEMICAL REACTIONS 375

9 CONCLUDING REMARKS

Our essential message in this limited review of turbulent flows with chemical
reactions is that notwithstanding the basic difficulties involved the field is important
and rich in variety and complexity. Although we have not emphasized it sufficiently,
we recognize the promise of combined theoretical and experimental efforts in
attacking the difficulties involved. With respect to the latter we note that the
combination of various laser diagnostic techniques with the digital techniques of
modern experimental turbulence research leads to the hope that, in the near future,
Annu. Rev. Fluid Mech. 1976.8:351-376. Downloaded from www.annualreviews.org

time series of the velocity components and the state variables of composition and
temperature at well-resolved space-time points will be obtained in some turbulent
Access provided by University of York on 01/27/15. For personal use only.

flows with simple chemical re,actions. Such series provide a means for obtaining not
only the mean values, the first few moments of the fluctuations, and the correlations
of these flow variables, but the various probability densities, single, joint, and
conditioned, which as we have tried to suggest here play an important role in
turbulent reacting flows.

ACKNOWLEDGMENTS

The present work was supported by the Office of Naval Research under Contract
NOOO14-67-0226-005 (Subcontract No. 4965-26) as part of Project SQUID and by
the Air Force Office of Scientific Research, Office of Aerospace Research, US Air
Force under Grant AFOSR-72-2333.

Literature Cited

Alber, I. E., Batt, R. O. 1975. AIAA J. To turbulence in flames. AGARD Con! Proc.,
appear 1 64, 112-20
Andrews, G. E., Bradley, D., Lwakabamba, Bray, K. N. C., Moss, J. B. 1 974. A unified
S. B. 1975. Symp. Int. Combust., 15th, statistical model of the premixed turbulent
655---64. Pittsburgh : Combust. Inst. flame. AASU Rep. No. 335, Vniv.
Ballal, D. R., Lefebvre, A. H. 1 974. Acta Southampton, Dep. Aeronautics Astro­
Astron. 1 : 471-83 nautics
Ballal, D. R., Lefebvre, A. H. 1975a. Symp. Brown, G., Roshko, A. 1974. J. Fluid Mech.
Int. Combust., 15th, 1473-8 1 . Pittsburgh : 64 : 775-8 16
Combust. Inst. Bush, W. B., Fendell, F. E. 1974. Acta
Ballal, D. R., Lefebvre, A. H. 1975b. Proc. Astron. 1 : 645-66
R. Soc. London Ser. A. To appear Chomiak, J. 1970. Combust. Flame 1 5 : 3 19-
Bilger, R. W. 1975. Combust. Sci. Technol. 21
To appear Chomiak, J. 1974. Combust. Flame 22 : 99-
Bilger, R. W., Kent, J. H. 1974. Combust. 104
Sci. Technol. 9 : 25-29 Chung, P. 1975. Combust. Sci. Technol. To
Borghi, R. 1974. Methode analytique de appear
prevision des taux de reaction chImique Corrsin, S. 1958. Phys. Fluids 1 : 42-47
en presence d'une turbulence non homo­ Cushing, B. S., Faucher, J. E., Gandbhir, S.,
gene. (Application a la combustion Shipman, C. W. 1 967. Symp. Int. Combust.,
turbulente.) AGARD Con! Proc., 164, 1 1 th, 8 1 7-24. Pittsburgh : Combust. Inst.
114-26 Damk6hler, G. 1940. Z. Elektrochem. 46 :
Bradshaw, P. 1972. Aeronaut. J. 76 : 40,3-18 601-26
Bray, K. N. C. 1974. Kinetic energy of Donaldson, C. duP., Hilst, G. R. 1 972.
376 LIBBY & WILLIAMS

Environ. Sci. Technol. 6 : 8 1 2-16 1 1 : 590-99


Donaldson, C. duP., Varma, A. K. 1975. Morkovin, M. V. 1964. The Mechanics of
Combust. Sci. Technol. To appear Turbulence, ed. A. Favre, 367-80. New
Dopazo, c., O'Brien, E. E. 1974. Acta York : Gordon & Breach
Astron. I : 1 239-66 O'Brien, E. E. 1966. Phys. Fluids 9 : 2 1 5-16
Dopazo, c., O'Brien, E. E. 1975. Combust. O'Brien, E. E. 1 9 7 1 . Phys. Fluids 1 4 : 1804-6
Sci. Technol. To appear Palm-Leis, A., Strehlow, R. A. 1969. Combust.
Eschenroeder, A. Q. 1964. Phys. Fluids 7 : Flame 1 3 : 1 1 1-29
1 735-43 Schelkin, K. I., Troshin, Ya. K. 1965. Gas­
Eschenroeder, A. Q. 1 965. AIAA J. 3 : 1839- dynamics of Combustion. Baltimore :
46 Mono Book Corp.
Favre, A. 1 969. Statistical equations of Shetinkov, Eo S. 1959. Symp. Int. Combust.,
turbulent gases. Problems of Hydro­ 7th, 583-89. London : Butterworth
dynamics and Continuum Mechanics, 231- Spalding, D. B. 1959. Symp. Int. Combust.,
Annu. Rev. Fluid Mech. 1976.8:351-376. Downloaded from www.annualreviews.org

66. Philadelphia : SIAM 7th, 595-603. London : Butterworth


Grant, A. J., Jones, J. M. 1975. Combust. Spalding, D. B. 1967. Symp. Int. Combust.,
Access provided by University of York on 01/27/15. For personal use only.

Flame. To appear 1 1 th, 807':"15. Pittsburgh : Combust. Inst.


Gunther, R. 1970. J. Inst. Fuel 43 : 1 87-92 Spalding, D. B. 1971a. Symp. Int. Combust.,
Hawthorne, W. R., Weddell, D. S., Hottel, 13th, 649-57. Pittsburgh : Combust. Inst.
H. C. 1 949. Symp. Combust., Flame Spalding, D. B. 1971b. Chem. Eng. Sci. 26 :
Explosion Phenomena, 3rd,,266--8 8 95
Hottel, H. c., Williams, G. c., Miles, G. A. Spalding, D. B. 1974. See Libby 1974,
1 967. Symp. Int. Combust., 1 1 th, 77 1-78. 1 1 -24
Pittsburgh : Combust. Inst. Spalding, D. B. 1 975. Combust. Sci. Techno/.
Istratov, A. G., Librovich, V. B. 1966. Prikl. To appear
Mat. Mekh. 30 : 45 1-66 Summerfield, M., Reiter, S. H., Kebely, V.,
Jenkins, D. R., Yumlu, V. S., Spalding, Mascolo, R. 1955. Jet Propu/. 25 : 377-84
D. B. 1 967. Symp. Int. Combust., 1 1 th, Toor, H. L. 1962. AIChE J. 8 : 70-78
779-90. Pittsburgh : Combust. Inst. Tsuge, S. 1974. Phys. Fluids 1 7 : 22-33
Klimov, A. M. 1963. Zh. Prikl. Mekh. Tekh. Tucker, M. 1956. Interaction of a free flame
Fiz. 3 : 49-58 front with a turbulence field. NACA Rep.
Kovasznay, L. S. G. 1956. Jet Propul. 26 : No. 1277
485 Williams, F. A. 1 965. Combustion Theory,
Kumar, R. N. 1974. Some experiments on Chap. 7. Reading, Mass : Addison-Wesley
structure and acoustics of turbulent jet Williams, F. A. 1971. Ann. Rev. Fluid Mech.
flames. WSS/CI Pap. pp. .74-25 3 : 1 71-88
Landau, L. D. 1944. Zh. Eksp. Teor. Fiz. Williams, F. A. 1 975a. A review of some
1 4 : 1 1 2-18 theoretical considerations of turbulent
LaRue, J. c., Libby, P. A. 1974. Phys. Fluids flame structure. See Libby 1974, II I 25 -

1 7 : 1956-67 Williams, F. A. 1975b. Recent advances in


Libby, P. A. 1962. ARS J. 32 : 388-96 theoretical descriptions of turbulent
Libby, P. A. 1974. Analytical and numerical diffusion flames. In Turbulent Mixing in
methods for investigation of flow fields Nonreactiveand Reactive Flows, ed. S. N. B.
with chemical reactions, especially related Murthy. New York : Plenum. (Proceedings
to combustion. AGARD Con! Proc., 164, of the Workshop on Turbulent Mixing
IIS-8 held at Purdue Univ., May 20 and 21,
Libby, P. A. 1975. Combust. Sci. Technol. 1 974)
To appear Williams, G. C., Hottel, H. C., Guniitz,
Lin, C. H., O'Brien, E. E. 1974. J. Fluid R. N. 1 965. Symp. Int. Combust., 1 2th,
Mech. 64 : 195-206 1081-92. Pittsburgh : Com bust. lost.
Longwell, J. P., Weiss, M. A. 1955. Ind. Wohl, K., Gazley, c., Kapp, N. 1949. Symp.
Eng. Chem. 47 : 1 634-39 Int. Combust., 3rd, 288-300. Baltimore :
Lundgren, T. S. 1967. Phys. Fluids 1 0 : 969- Williams & Wilkins
75 Yamazaki, K., Tsuji, H. 1 962. Symp. Int.
Lundgren, T. S. 1 969. Phys. F/uids 12 : 485- Combust., 8th, 543--5 3. Baltimore :
97 Williams & Wilkins
Mellor, G. L., Herring, H. J. 1973. AIAA J.

You might also like