You are on page 1of 9

482

J. ENERGY

VOL. 7, NO. 6

Aspects of Computer Simulation of Liquid-Fueled Combustors


A. D. Gosman* and E. loannidest Imperial College, London, England
An existing "discrete droplet" model of liquid sprays has been extended to include a stochastic representation of turbulent dispersion effects. Applications to simple test cases, including the dispersion of single particles, produce reasonable agreement. However, two further applications involving volatile and combusting sprays show that the turbulent dispersion effects are small in comparison to those due to uncertainties about the initial conditions of the spray.
Downloaded by Universitats- und Landesbibliothek Dusseldorf on May 5, 2013 | http://arc.aiaa.org | DOI: 10.2514/3.62687

Nomenclature
CD
D

i k K md mfu, mox,mp
MHd MUd,MVd,MWd Nu P Pr r Re
QL

Sc Sh T t tr U,V,W
U,V,W (

u' ,v' ,w x

r
e

Meff P

= drag coefficient = specific heat at constant pressure = droplet diameter = mass transfer coefficient = enthalpy = heat of combustion =; stoichiometric oxygen/fuel ratio = kinetic energy of turbulence = thermal conductivity = droplet mass = mass fraction of fuel vapor, oxygen, and products = droplet enthalpy = droplet U, V, ^momentum = Nusselt number = pressure = Prandtl number = latent heat of vaporization = radial coordinate = Reynolds number = combustion rate = radius of furnace = source term = droplet position vector = Schmidt number = Sherwood number = temperature = time = residence time of a droplet in an eddy = time-averaged velocity components = instantaneous velocity components = fluctuating velocity components = axial coordinate = diffusion coefficient = dissipation rate of turbulence = effective viscosity = density = general variable = droplet relaxation time

fu L m s v

= fuel = liquid - mean = droplet surface vapor

Introduction

Subscripts
d

= droplet = eddy

Presented as Paper 81-0323 at the AIAA 19th Aerospace Sciences Meeting, St. Louis, Mo., Jan. 12-15, 1981; submitted June 29, 1981; revision received July 1, 1983. Copyright American Institute of Aeronautics and Astronautics, Inc., 1983. All rights reserved. * Reader in Fluid Mechanics, Mechanical Engineering Department. tResearcher, Mechanical Engineering Department (presently, Head of Bearing Applications Department, SKF-ERC, Nieuwegein, the Netherlands). Member AIAA.

HE combustion of liquid-fuel sprays has numerous important applications in furnaces, diesel engines, gas turbines, and other equipment. The increasing need for fuel economy and emissions control has generated fresh interest in both experimental and theoretical studies of spray flames. Recent experimental studies include Hiroyashu and Kadota,1 Shearer et al., 2 and Tishkoff et al.,3 on evaporating sprays and Onuma et al., 4 Owen et al.,5 and Found et al., 6 on combusting sprays. On the theoretical side, variants of two models are currently employed, namely, the Williams7 statistical spray model used by Westbrook, 8 Haselman and Westbrook,9 and Cliffe et al.10 and the "discrete droplet" model used by Crowe, 11 Gosman and Johns, 12 El-Banhawy and Whitelaw, 13 Abbas et al.,14 Gosman et al., 15 Dukowicz, 16 and O'Rourke and Bracco.17 The Williams 7 statistical spray model considers a generalized spray distribution function originally defined in an eight-dimensional space of droplet diameter, location, velocity, and time. Conservation principles yield a partial integro differential equation for this function and the solution of this equation, together with the gas conservation equations, provides the required model of the spray. As applied to date, however, this approach is costly in terms of computer storage and time unless simplifications are introduced, such as the assumption of no slip between the droplets or the gases or the representation of the spray by a very limited number of droplet sizes. Furthermore, due to the limited resolution (especially in the vicinity of the atomizer) of the numerical methods generally employed for the solution of the spray equation, this method may introduce substantial spurious numerical diffusion into the calculation. In the "discrete droplet" model, the spray is represented by individual droplets rather than by a continuous distribution function. Because of the large number of actual droplets contained in the spray, the representation is confined to a statistical sample. Therefore, each of these sample droplets characterizes a "parcel" of like numbers, all having the same initial size, velocity, and temperature. The motion, heating, and evaporation of each sample as it traverses the gas are computed by solving numerically the Lagrangian ordinary differential equations governing the mass, momentum, and energy conservation. The effects of the droplets on the gas phase are introduced into the (Eulerian) calculations of the latter by feeding in the local rates of heat, mass, and momentum exchange deduced from the analysis of the droplet

NOV.-DEC. 1983

LIQUID-FUELED COMBUSTORS

483

parcel trajectories. The overall solution is obtained by iterating between the calculations of the two phases. An important component of any theoretical model is the manner in which the effects of turbulence on droplet dispersion are represented. In many studies (e.g., Westbrook, 8 Haselman and Westbrook,9 Crowe,11 Crowe et al.,18 El-Banhawy and Whitelaw, 13 and Gosman et al. 15 ) this effect is ignored. Some researchers, such as Abbas et al.,14 introduce it in a deterministic way by working out from some phenomenological model a "diffusion" velocity, which is added to that obtained from the original Lagrangian equations of motion, thus modifying the parcel trajectory. A third, and probably more correct, approach is the stochastic one developed by Dukowicz16 and later elaborated for thick spray applications by O'Rourke and Bracco,17 in which the gas turbulence is randomly sampled during each droplet's flight and allowed to influence its motion, the gross spray behavior being obtained by averaging over a statistically significant sample of droplets. The method to be described in this paper is a stochastic "discrete droplet" method in the spirit of Dukowicz. However, it differs in details of the treatment of the gas-phase turbulence and the manner in which it is thought to interact with the droplet motion. In particular, this interaction is held to occur over a time interval that is the minimum of two time scales, one being a typical turbulent eddy lifetime and the other the residence time of the droplet in the eddy. These time scales, as well as the local turbulence intensity, are obtained in the present study by solving for the time-averaged turbulence energy k and its dissipation rate e, using a version of the wellknown "k-e" turbulence model of Launder and Spalding.19 The turbulence model also forms part of a conventional Eulerian finite-volume calculation procedure for the gas phase, the interactions between the droplets and gas being dealt with in an iterative fashion as outlined earlier. The accuracy of the overall scheme has been checked by reference to the known analytical solution for turbulence diffusion of inertialess particles from a point source in a homogeneous isotropic turbulent flow (Hinze 20 ) and by comparison with the measured averaged dispersion rates of individual particles introduced into the turbulent gas flow downstream of a grid, as reported by Snyder and Lumley. 21 In addition to these two cases, where the boundary conditions on both particles and gas are well-known and detailed information is available on the behavior of both phases, thereby allowing reasonably definitive assessment of the modeling, further calculations are presented for the evaporating spray reported by Tishkoff et al.3 and the combusting swirling spray reported by Found et al.6 In both of the latter cases information on the boundary conditions and interior behavior is not sufficiently complete to allow definitive assessment. They are used here instead to explore the sensitivity of the calculations of such flows to the turbulent dispersion modeling.

where // is the velocity component in direction xi9 the implied summation being restricted to the axial and radial components; 4> represents any of the variables just mentioned apart from pressure. The terms S and Feff are, respectively, the "source" and the effective diffusion coefficient for entity, while Sd represents the particular sources due to the presence of the droplets. The continuity equation is obtained by setting 3> = 1 and Feff = 1. The particular expressions for 5, F, and Sd pertaining to each variable are given in Table 1. In deriving Eq. (1), certain terms involving correlations of the fluctuating components arise that must be modeled. The approach employed here is similar to that described by Jones22 for single-phase combusting flows, although it should be said that at this stage that correlations involving fluctuating properties of the droplet field have been ignored for want of _a_better method. Thus, the turbulent Reynolds stresses pu-uj and scalar fluxes pw/c/>' are modeled by
dU,

Downloaded by Universitats- und Landesbibliothek Dusseldorf on May 5, 2013 | http://arc.aiaa.org | DOI: 10.2514/3.62687

(2)

(3)

where \it and F, are the turbulent viscosity and diffusivity, respectively, and k the time-mean kinetic energy of turbulence. The turbulent viscosity is obtained from
(4)

where e is the time-mean dissipation rate of k, CM is a constant, and Tt is calculated from


U

t,<t>

(5)

Here at>(t) is the turbulent Prandtl/Schmidt number, also taken as a constant. Finally, the local values of k and e are determined by solving two additional transport equations for them of the general form of Eq. (1). The associated transport coefficients and source terms are shown in Table 1, together with the values of the various empirical constants. For the furnace application, a combustion model is employed that assumes the droplets evaporate sufficiently rapidly to form a cloud of vapor burning as a gaseous flame. However, it is necessary to recognize that as a result of the variety of conditions occurring in furnaces this flame can be of a diffusion and/or premixed character (Faeth 27 ) and therefore the combustion model should be capable of handling both types. In the present study, the Magnussen and Hjertager 28 "eddy mixing control" model was used, which possesses this capability. This model assumes that the fuel and oxygen combine irreversibly in a single global reaction to form products, the time-averaged fuel consumption rate Rfu and species concentrations being linked by
-

The Mathematical Formulation


The Gas Field

(6)

The dependent variables characterizing the time-averaged axisymmetric gas flow are the axial, radial, and circumferential velocity components, denoted by U, V, and W, respectively, the static pressure P, the total enthalpy h, and the mass fractions ra, of the chemical constituents. The partial differential conservation equations that govern these are, for the case of a dilute spray in which the effects of displacement of the gas by the droplets may be ignored, are of the following form:
d ^

(1)

where A and B are constants, the subscripts fu, ox, and pr refer to fuel, oxygen, and product, respectively, / is the stoichiometric combination ratio, and k/e is the time scale of the turbulent eddies, which is assumed to be much larger than the chemical kinetic time scales of the hydrocarbon reaction. The species mass fractions are therefore determined by solving transport equations for each possessing the form of Eq. (1), in which, as indicated in Table 1, the reaction rates are determined from Eq. (6) or an appropriate multiple thereof. Finally, radiation is calculated by way of a "four-flux" model similar to that employed by Gosman and Lockwood 29 ;

484

A. D. GOSMAN ANDE. IOANNIDES

J. ENERGY

however, at this stage droplet and soot radiation are not modeled. The local density is obtained from the equation of state for an ideal mixture.
The Droplet Field

trajectories may be obtained by a further integration of the equation for the position vector sd,
ds^

The model employed for the droplet calculations in the present study is outlined in this section. The Lagrangian forms of the governing equations for the instantaneous motion and rates of heating and evaporation used in the present study are as given below.
Droplet Motion and Trajectory

dt

(10)

The drag coefficient CD is obtained from the following expressions, taken from Wallis30:
CD = 0.44 for Red>1000 (11)

CD = (l + 0.15Red687)/(Red/24)

for Red<100Q

(12)

The axial, radial, and tangential momentum equations are, respectively, as follows:
(7)

where the droplet Reynolds number Red is defined by

Red = p\u-ud\D/n
Downloaded by Universitats- und Landesbibliothek Dusseldorf on May 5, 2013 | http://arc.aiaa.org | DOI: 10.2514/3.62687

(13)

dt

Droplet Heating

dt

(8)

The balance equation is (from Borman and Johnson31): (14)

dwd 3 p ^ i CD(w-wd)^ \\u~dt ~ ~4 PdD


where u and ud are the local gas and droplet velocity vectors, respectively. Integration of these equations provides the velocity components of the droplets, from which the where

-7

dt

(15)

Table 1 Transport coefficients and source terms for the variable </>

Variable

Source term S

Droplet source term 5^

d (
Meff

dU\
dx '

1d (
r dr \ 1d ( r dr \

dV\
dx /

dP
IT

1 ^
~ 77

IT Ueff ) + - Ueff r lT /

dx \ d (

dx

Ns

k=]

LJ ( MUd0 ~ MUdj ) ,

'

dU\ dx / pW
2

dV\ dr /

2^V r

1 ^
~ 77

Meff

T~ Ueff 7~ J + ~ 7 ~ U e f f / ' ~ ) ~ ~T

dx \
'

Ns

2-f ( MK ^ ~ M K rf/^
k=]

'

dP dr
V J
N

r2

w
k
e

\ 2
^eff
(Tgff A,

r dr /

7V5 ^

-^a

(e/*)(C 7 G*-C 2 pe)


7
N

0
V fM//
^5 A:=7 N V^

eff, e Meff A///

h
a

eff,/2

rrifu

^eff ^eff./w
N

s k=l

//?
a

n
0

eff , ox
Meff

mpr

(7 I /)/>

^eff , pr

Notes: 1) 5^ expressions are presented in the forms employed in the numerical calculations. 2) Turbulence model constants are assigned the following values: CM = 0.09,C/ = 1.44,C2 = 1.92, ak --

NOV.-DEC. 1983

LIQUID-FUELED COMBUSTORS
where r is the droplet "relaxation" time, defined as

485

and the Nusselt number Nu is calculated from the Ranz and Marshall23 correlation (16)
Droplet Evaporation

T=(4/3)PdD/(pCD\u-ud\)

(23)

The evaporation rate is obtained from


^ = -TTDPvDvtn(l+By)Sh

(17)

where By = (mfUjS -mfuoo)/(\ -mfUtS). The Sherwood number Sh is obtained from Eq. (16), with Nu and Pr replaced by Sh and Sc, respectively.
Turbulent Dispersion of Droplets
Downloaded by Universitats- und Landesbibliothek Dusseldorf on May 5, 2013 | http://arc.aiaa.org | DOI: 10.2514/3.62687

In circumstances where Ie/(r\u ud\)>l9 Eq. (22) has no solution. This can be interpreted as implying that the droplet has been "captured" by the eddy, in which case t-mt = te. The above procedure may be repeated for as many interaction times as are required for the droplet to traverse the required distance. Clearly, if a statistically significant number of droplet samples is tracked in this way, the ensembleaveraged behavior should represent the turbulent dispersion induced by the prevailing gas field.
Boundary Conditions

As mentioned in the Introduction, the effect of the turbulence on the droplet motion is simulated here by a stochastic approach, one element of which is the evaluation of the instantaneous gas velocity u in the droplet equations of motion (14-16) from the time-averaged gas velocity U and turbulence energy k fields. For this purpose, the turbulence is assumed to be isotropic and to possess a Gaussian probability distribution in the fluctuating velocity, whose standard deviation an is given by
(18)

The boundary conditions for the gas-phase calculations must be specified on all surfaces enclosing the region of interest; details will be provided when the individual cases are presented. As for the droplet calculation, this is essentially an initial-value problem, but the necessary information regarding the initial velocity, size, and temperature of the droplets is rarely available, due to the complexity and the irregular character of the atomization process that produces them. As a consequence, a strong element of guesswork and empiricism enters into spray calculations, which inevitably hinders their evaluation. The particular practices employed in the present study will be given separately for each case.

Random sampling of this distribution at appropriate points in the trajectory calculation then yields the estimated prevailing fluctuating velocity field u' and hence the instantaneous field u = U+ u'. A second important element is the manner of determining the time interval t-mt over which the droplet interacts with the randomly sampled velocity field. Here, it is convenient to envisage the latter as being associated with a turbulent eddy, in which case the interaction time is determined by one or the other of the following possible events: 1) the droplet moves sufficiently slowly relative to the gas to remain within the eddy during the whole of its lifetime te or 2) the relative or "slip" velocity between the gas and droplet is sufficient to allow it to traverse the eddy in a transit time tR shorter than te. The interaction time scale will therefore be the minimum of the above (see Ref. 24), i.e.,
t-mi=min(te,tR)

Numerical Analysis
The Gas Field

(19)

The solution method used for the gas equations is that described by Caretto et al.25 and embodied in the TEACH-T code that was adapted for the present circumstances. Inasmuch as details of both the method and code have been published elsewhere (see, e.g., Hutchinson et al. 26 ), only an outline will be provided here. The method is of the finite-volume variety, in which the differential equations are cast into an algebraic form that preserves their conservation and boundedness properties, on a computing mesh formed from cylindrical-polar coordinate surfaces. The resulting algebraic system is solved iteratively in a sequence that extracts the velocities from the momentum equations and then employs a continuity-based equation to compute the pressures. The calculation of the other gas field variables is embedded in this sequence, during which the "sources" associated with the droplets are held constant.
The Droplet Field

Estimates of the eddy and transit time scales are made under the further assumption that the characteristic size of the randomly sampled eddy is the dissipation length scale le, given

by*
le =Ck3/2/e (20)

The eddy lifetime is then estimated as

te = le/ l i / ' l

(21)

The ordinary differential equations governing the behavior of each droplet are solved by forward numerical integration, starting from the injection location, at which the initial conditions are (stochastically) prescribed, and proceeding until it either leaves the calculation domain or evaporates to a negligible size. During this process, the gas field properties appearing in the equations are interpolated from the prevailing values at the nearest nodes of the computational grid.
Overall Solution Procedure

The transit time scale tR is estimated from the following solution of a simplified and linearized form of the equation of motion of the droplet:
tR = -rHn[LO-le/(r\u-ud I ) ]

(22)

$In an earlier version of this paper a CM exponent of 3/4 was erroneously deduced (the authors are grateful to Prof. G. M. Faeth for drawing their attention to this). The results here are based on the more appropriate value of 1/2. The change has not altered the conclusions.

Each iteration cycle of the two-phase calculation involves two stages. First, a chosen number of droplets (18 in the present case) is introduced and their trajectories are calculated according to the above procedure. At each computational cell traversed by the droplets, the mass, momentum, and energy extracted or deposited are accumulated and averaged via the "droplet source" expressions appearing in Table 1. These are of the form for, e.g., the mass source,
(24)

486

A. D. GOSMAN ANDE. IOANNIDES

J. ENERGY

where Ns is the cumulative total number of computational droplets introduced over all iterations performed, TV the cumulative number which traverse the cell in question, and Mdi and Mdo are the mass fluxes of the kth parcel entering and leaving, respectively, i.e.,
= (>K/6)pdD3knk
(25)

Downloaded by Universitats- und Landesbibliothek Dusseldorf on May 5, 2013 | http://arc.aiaa.org | DOI: 10.2514/3.62687

Here nk is the number of "real" droplets issuing from the spray nozzle per unit time and per computational droplet. Similar interpretations apply to the remaining droplet source entries in Table 1 . Second, the droplet sources are inserted into the gas-phase equations and one iteration is performed. This cyclic process is repeated until the gas-phase calculation converges, by which stage it follows that the sources derived from the stochastic droplet treatment have attained statistically stationary values. It may, of course, be possible to accelerate this process by introducing a more or less computational droplets per cycle, but this has not yet been explored.

- 5 - 4 - 3 - 2 - 1 0 1 2 3 4 5
Fig. 1 Analytical and numerical solutions for the transverse distributions from a point source at 2.6 m downstream.

Applications and Assessment


Diffusion from a Point Source in a Homogeneous Flow

It is well known that the spatial dispersion of "marked" fluid particles introduced at a constant rate from a point source into a uniform turbulent flow of the same fluid is amenable to exact analysis when the turbulence is homogeneous and isotropic and long diffusion times are considered. The distribution of the concentration C(xlrx2,x3) of these particles at a point (x},x2,x3) is, according to Hinze,20
C(x/,x2,x3)=S Q\p[-U(x22+x23)/(4Tt \
t

o hollow glass,D=46.5um,p=0.26g/cc^ --A corn pollen,D-87 ym, P =2.5g/ccX" glass,D=87Mm sP =lg/cc _-A copper,D=46.5pm,p=8."

\x, I

(26)

where F, is the (uniform) turbulent diffusivity and S the volumetric source strength. A comparison of concentration profiles extracted from Eq. (26) and predictions from the present dispersion model for corresponding circumstances (i.e., very small particles of the same density as the fluid) is shown in Fig. 1 . In the numerical calculations, fixed values were ascribed to (7, k, and e and the trajectories of some 800 particles were computed and then processed to obtain C. The analytical solution was obtained with F / evaluated as C^pk2 /e, which is consistent with the diffusion model on which Eq. (26) is based. Both profiles are normalized by the analytically derived Cmax . Agreement between the two sets of results is reasonable and could, of course, have been improved by introducing more particles. The irregularities in the calculated curve are indicative of the stochastic nature of the model and of the differences that occur when a different random number sequence is used.
Dispersion of Single Particles

100 200 300 400 Time from the first station (ms) Fig. 2 Measurements21 (, , - - - ) and calculation ( o , A , , A ) of particle dispersion.

Snyder and Lumley 21 performed an interesting experiment in which single spherical solid particles of various densities and sizes were isokinetically introduced into the uniform turbulent flow downstream of a grid and their trajectories were photographed. The present dispersion model was applied to this problem, making use of the fact that gridgenerated turbulence has been well studied in the past20 and the decay rates of both the turbulence kinetic energy and its dissipation rate are known. Moreover, it was (reasonably) assumed that both the turbulence and the mean flow are unlikely to have been appreciably disturbed by the single particles. As in the experimental study, the trajectories of
The calculated curves in this and the next application represent averages over radial intervals of 0.005 m.

-5 -4-3-2-10

1 2 3 4 5

r(cm)

Fig. 3 Measurements21 and calculations for copper particle distributions at 1.95 m downstream.

some 800 examples of each particle type were calculated and then processed to yield the dispersion profiles displayed in Fig. 2, along with the data. The predictions for the heavier particles (i.e., copper and glass) are reasonably good, but agreement deteriorates in the case of the light ones (hollow glass and corn pollen). (The diameters of the copper and glass particles are such that they possess nearly identical relaxation times, which is why they exhibit nearly identical dispersion rates.)

NOV.-DEC. 1983

LIQUID-FUELED COMBUSTORS

487

Downloaded by Universitats- und Landesbibliothek Dusseldorf on May 5, 2013 | http://arc.aiaa.org | DOI: 10.2514/3.62687

Figures 3 and 4 show examples of the predicted and measured number density probability distributions in a crossstream plane for heavy (copper) and light (hollow glass) particles, respectively. According to Ref. 21, the measurements are well represented by the Gaussian distributions shown. (The profiles are normalized in each instance by the predicted Cmax values.) These graphs reinforce the conclusions drawn from Fig. 2, i.e., the numerical prediction agrees fairly well with the heavyparticle distribution of Fig. 3, but less with the light-particle case of Fig. 4, where the calculated distribution has about the right width, but is more concentrated at the edges, due in part to the smaller samples of particles that reach these positions. The reason for the larger errors observed for light particles is not entirely understood, although it is probably connected with the fact that their interaction time is invariably the eddy lifetime, whereas with the heavier particles it is the transit time. The fact that good agreement was obtained for the diffusing particles of the previous example, whose interaction time was also the eddy lifetime, suggests that the inhomogeneity of the turbulence prevailing in the present case may be inadequately allowed for in the model, although there may be other explanations.

30

2Q
Q_ -M I O 13 U-

10

10

20 30 40 r (mm) .

50

40

30
>o
a,^ 20
OLO

Station at 50 mm

experiment

10

10

b)

20 30 r(mm)

40

50

-5-4-3-2-10

4 5

Fig. 4 Measurements21 and calculations for hollow glass particle distributions at 2.6 m downstream.

c)

10

20 30 40 r(mm) -

50

Experiment photograph -- shadowgraph - scattering

40

20-

x ( mm )
Fig. 5 Measured3 and computed spray boundaries (computations: A no random entry, no droplet dispersion; o no random entry, droplet dispersion; A random entry, no droplet dispersion; random entry, droplet dispersion).

10

d)

20 r (mm) -

Fig. 6 Measured3 and calculated radial variation of liquid volume fraction at two stations: a) and b) o calculation without droplet dispersion, calculation with droplet dispersion; c) and d) o calculation without random entry, calculation with random entry.

488

A. D. GOSMAN AND E. IOANNIDES

J. ENERGY

a)
Downloaded by Universitats- und Landesbibliothek Dusseldorf on May 5, 2013 | http://arc.aiaa.org | DOI: 10.2514/3.62687

measured at 25 mm

measured at 50mm

10
b)

20

30 40

50

r (mm) *-

Fig. 7 Measured and calculated vapor concentrations: a) n calculation without droplet dispersion, station 25 mm; calculation with droplet dispersion, station 25 mm; o calculation without droplet dispersion, station 50 mm; calculation with droplet dispersion, station 50 mm; b) n calculation without random entry, station 25 mm; calculation with random entry, station 25 mm; o calculation without droplet dispersion, station 50 mm; calculation with droplet dispersion, station 50 mm.

Steady Conical Evaporating Spray

The n-heptane spray measurements reported by Tishkoff et al.3 form the first testing ground for the fully coupled gas motion and droplet model. The evaporating solid-cone spray was injected into a coflowing airstream of almost uniform velocity and detailed measurements were made at two planes 25 and 50 mm downstream of the nozzle of the following: the droplet-size distributions of the axis, edge, and middle of the spray; the liquid-phase volume fraction; and vapor volume concentration. In addition, photographs of the spray boundary were taken from which such global characteristics as the loci could be deduced. Predictions for one of the several injection pressures examined (1.48 MPa) were generated using estimated initial conditions for the spray. The practice adopted for this was to subdivide the measured initial spray angle of about 50 deg into three equal-angle bands and then estimate the initial conditions for each (in an admittedly crude manner) from the available measurements in the plane nearest to the injector by assuming that all droplets evaporated in a vapor-free environment according to the so-called d2 law during their travel from the injector to this plane. Different assumptions were made for the initial axial velocities and temperatures of the droplets, which were assigned uniform values calculated from the nozzle exit conditions. Within each band, the initial proportion of the total mass flow and the droplet size distribution were estimated in this way. The latter was randomly sampled during the calculations, as was the

radial velocity compdnent of the droplet between upper and lower limits corresponding to the included angle of the particular band considered. In order to assess the separate influence of the various methods employed in the droplet modeling, three additional calculations were performed in which, independently: 1) the random radial entry velocity treatment was suppressed and all droplets in a particular band were ascribed the median entry angle; 2) the turbulent dispersion model was suppressed; and 3) both of the foregoing practices were invoked simultaneously, which effectively makes the approach a deterministic one in the mold of the earlier discrete-droplet models of this kind mentioned in the Introduction. The spray boundaries obtained from each method are shown in Fig. 5 and indicate that droplet dispersion has only a marginal effect on the spray angle, especially in comparison with that of the random radial entry velocity treatment. As expected, the latter causes some droplets to reach larger radii. The spray boundary demarcated by these is in good agreement with the outer band of the experimental measurements. Further details of the predictions are shown in Fig. 6, which shows the radial variations in the liquid volume fraction at the two axial measurement locations. Figures 6a and 6b contain results using the stochastic entry treatment with and without dispersion modeling, while Figs. 6c and 6d show the effect of retaining the latter and dispensing with the former. From Figs. 6a and 6b it can be seen that the influence of turbulent dispersion in these circumstances is predicted to be small, with the most pronounced effects occurring near the axis where, as would be expected, dispersion gives rise to smaller radial gradients. By contrast, the influence of the initial conditions ascribed to the droplets is very pronounced, as is illustrated in Figs. 6c and 6d, where the stochastic treatment tends to substantially reduce the radial gradients. It is also clear that none of the predictions is in acceptable agreement with the measurements, for in all cases the calculated liquid volume fraction distribution diminishes rapidly away from the axis and increases toward the edge of the spray to only a fraction of the measured values. These errors are probably the consequence of inaccuracies in estimating the initial conditions of the spray (further calculations are in progress that point in this direction) and suggest the need for measurements in, and/or a model of, the atomization zone. Figures 7a and 7b show similar plots of the radial variations of the vapor concentration in the two measurement planes. The two sets of predictions in Fig. 7a relating to the different dispersion treatments are again close, showing that here too the effect is small at all but a few locations where the concentration is itself low. According to Fig. 7b the vapor distribution is less sensitive to the different entry treatments than was the liquid, but the considerable differences between all of the predictions and the experimental data near the outer periphery of the spray at the downstream plane (which are consistent with the underprediction of the liquid concentrations there) reinforce the above comments about the need for a better inlet treatment. These calculations were performed on a 20x20 computational grid and required 35.4
and 43.5 s of CDC 7600 CPU time without and with the

dispersion time, respectively.


Combusting Spray

The final application of the method described here is to the liquid-fueled furnace experiments reported by Founti et al.,6 in which a hollow-cone swirling kerosene spray was axially injected into a cylindrical combustion chamber. The initial conditions of the spray were estimated from knowledge of the atomizer geometry, low-resolution photographs of the nearby region under cold and burning conditions, and measurements of the cold-flow droplet size distribution a short distance downstream. Details of the estimated conditions are given in an earlier publication by the present authors. 15 Reference 15 also presents calculations based on the deterministic version

NOV.-DEC. 1983 No droplet dispersion

LIQUID-FUELED COMBUSTORS

489

Downloaded by Universitats- und Landesbibliothek Dusseldorf on May 5, 2013 | http://arc.aiaa.org | DOI: 10.2514/3.62687

W i t h droplet dispersion Fig. 8a Streamlines in a plane through the axis of the furnace.

of the method, which were found to produce poor agreement with the measurements. In the further calculations reported here, the deterministic specification of the initial conditions was retained but the stochastic turbulent dispersion treatment was introduced, the objective being to determine the sensitivity of the predictions to this facet of the modeling. An overall impression of the sensitivity may be obtained from Figs. 8a-8c, which show, respectively, the streamlines, fuel concentration contours, and isotherms, in each case with and without the dispersion treatment. Evidently the effect is small in both cases, and it may therefore be concluded that the absence of turbulent dispersion modeling is unlikely to have been responsible for the poor agreement obtained in our earlier calculations. Although other sources of error are known to exist in the present version of the method, notably in the treatment of combustion and radiation, it is believed that the overriding factor is once again inadequate knowledge of the initial conditions of the spray. In this instance, the CPU times on a 20 x 20 grid were 300 s with the dispersion modeling and some 50% less without.

Conclusions
No droplet dispersion The present study has demonstrated that it is possible to calculate turbulent dispersion effects on droplets and particles to a tolerable degree of accuracy using a stochastic discrete droplet approach, provided that the initial conditions are well defined, as they were in the simple test cases. It is true that the accuracy deteriorated for the lighter particles, whose interaction times were invariably the eddy lifetime, suggesting that refinement of the approach is required for these circumstances. Nevertheless, the current performance is believed to be adequate enough for the purposes of evaluating the importance of turbulent dispersion in the much more complex circumstances of real sprays. The initial impression gained from the two such evaluations performed here is that the dispersion effects are small as compared with the major source of uncertainty, which appears to be inadequate knowledge of the initial size distributions, mean velocities, and other properties of the droplets. This impression should be tempered by the acknowledgment that there are other dispersion-dependent phenomena occurring in sprays that were not considered in the present study, notably droplet proximity and collision/coalescence effects. Some progress in these has been made recently by O'Rourke and Bracco.17 However, in the context of furnace and gas turbine applications, these effects are likely to be confined to the immediate vicinity of the atomizer and may therefore be regarded as part and parcel of the initial-condition problem. It should also be noted that the extent of dispersion is dependent on the ratio of the length of the droplet trajectory to the average eddy size, which tended to be small in the cases examined here, due to either the limited region considered (evaporating case) or the rapid disappearance of the volatile droplets (combusting case). In other circumstances, notably those of pulverized coal combustion, dispersion would be expected to play a much more important role.

W i t h dropTet 'dispersion Fig. 8b Calculated vapor mass fraction contours in a plane through the axis of the furnace.

1930
1750 1570

1390
1210 1030 850 670

References
^iroyasu, H. and Kadota, T., "Fuel Droplet Size Distribution in Diesel Combustion Chamber," Bulletin of the Japan Society of Mechanical Engineers, Vol. 19, 1976, pp. 1064-1072. 2 Shearer, A. J., Tamura, H., and Faeth, G. M., "Evaluation of a Locally Homogeneous Flow Model of Spray Evaporation," Journal of 3Energy, Vol. 3, Sept.-Oct. 1979, pp. 271-278. Tishkoff, J. M., Hammon, D. C. Jr., and Chraplyvy, A. R., "Diagnostic Measurements of a Fuel Spray Dispersion," ASME Paper 80-WA/HT-35, 1980. Onuma, Y., Ogasawara, M., and Inoue, T., "Further Experiments on the Structure of a Spray Combustion Flame," 16th Symposium (International} on Combustion, The Combustion Institute, Pittsburgh, Pa., 1977, pp. 561-568.

490

W i t h droplet dispersion Fig. 8c Calculated temperature contours in a plane through the axis of the furnace.

490

A. D. GOSMAN AND E. IOANNIDES

J. ENERGY

5 Owen, F. K., Spadaccini, L. J., Kennedy, J. B., and Bowman, C. T., "Effects of Inlet Air Swirl and Fuel Volatility on the Structure of Confined Spray Flames," 17th Symposium (International) on Combustion, The Combustion Institute, Pittsburgh, Pa., 1978, pp. 467,473. 6 Founti, M., Hutchinson, P., and Whitelaw, J. H., "Measurements and Calculations of a Kerosene-Fueled Flow in a Model Furnace," Mechanical Engineering Dept., Imperial College, London, Rept. FS/79/19, 1979. 7 Williams, A., Combustion of Sprays of Liquid Fuels, Elek Science, London, England, 1976. 8 Westbrook, C. K., "Three-Dimensional Numerical Modeling of Liquid Fuel Sprays," 16th Symposium (International) on Combustion, The Combustion Institute, Pittsburgh, Pa., 1976, pp. 15IT1525. 9 Haselman, L. C. and Westbrook, C. K., "A Theoretical Model for Two-Phase Fuel Injection in Stratified Charge Engines," SAE Paper 780138, 1978. f Cliffe, K. A., Lever, D. A., and Winters, K. H., "A FiniteDifference Calculation of Spray Combustion in Turbulent, Swirling Flow," Paper presented at International Conference on Numerical Methods in Thermal Problems, Swansea, U.K., July 1979. H Crowe, C. T., "A Computational Model for the Gas-Droplet Flow Field in the Vicinity of an Atomiser," Proceedings of the llth JANNAFSymposium, 1974, Paper 74-23. . 12Gosman, A. D. and Johns, R., "Computer Analysis of Fuel-Air Mixing in Direct-Injection Engines," SAE Paper 800091, 1980. 13 El-Banhawy, Y. and Whitelaw, J. H., "The Calculation of the Flow Properties of a Confined Kerosene-Spray Flame," AIAA Paper 79-7020, 1979. 14 Abbas, A. S., Koussa, S. S., and Lockwood, F. C., "The Prediction of the Particle Laden Gas Flows," Proceedings of the 18th Symposium on Combustion, Combustion Institute, Waterloo, Canada, 1981, pp. 1427-1438. 15 Gosman, A. D., loannides, E., Lever, D. A., and Cliffe, K. A., "A Comparison of Continuum and Discrete Droplet FiniteDifference Models Used in the Calculation of Spray Combustion in Swirling Turbulent Flows," AERE Harwell Rept. TP 865, 1980. 16 Dukowicz, J. K., "A Particle-Fluid Numerical Model for Liquid Sprays," Journal of Computational Physics, Vol. 35, 1980, pp. 229253. 17 O'Rourke, P. J. and Bracco, F. V., "Modeling of Drop Interaction in Thick Sprays and a Comparison with Experiments," Paper presented at IME Stratified Charge Automotive Engines Conference, London, 1980.

18 Crowe, C. T., Sharma, M. P., and Stock, D. E., "The ParticleSource-in-Cell (PSI-cell) Model for Gas-Droplet Flows, ASME Paper No. 75-WA/HT-25, 1975. 19 Launder, B. E. and Spalding, D. B., Mathematical Models of Turbulence, Academic Press, London, 1972. 20 Hinze, J. O., Turbulence, McGraw-Hill Book Co., New York, 1975. 21 Snyder, W. H. and Lumley, J. L., "Some Measurements of Particle Velocity Autocorrelation Functions in a Turbulent Flow," Journal of Fluid Mechanics, Vol. 48, Pt. 1, 1971, pp. 41-47. 22 Jones, W. P., " Prediction Methods for Turbulent Flows," Lecture Series 1979-2, Von Karman Institute for Fluid Dynamics, Brussels, 1979. 23 Ranz, W. E. and Marshall, W. R. Jr., "Evaporation from Drops," Chemical Engineering Progress, Vol. 48, 1952, p. 173. 24 Brown, D. J. and Hutchinson, P., "The Interaction of Solid or Liquid and Turbulent Fluid Flow FieldsA Numerical Simulation," Journal of Fluids Engineering, Vol. 101, 1979, pp. 265-269. 25 Caretto, L. S., Gosman, A. D., Patankar, S. V., and Spalding, D. B., "Two Calculation Procedures for Steady Three-Dimensional Flows with Recirculation," Proceedings of the 3rd International Conference on Numerical Methods in Fluid Mechanics, Vol. 2, 1972, p. 60. 26 Hutchinson, P., Khalil, E. E., and Whitelaw, J. H., "Experimental Investigation of Flows and Combustion in Axisymmetric Furnaces," Journal of Energy, Vol. 1, 1977, pp. 212-219. 27 Faeth, G. M., "Current States of Droplet and Liquid Combustion," Progress in Energy Combustion Science, Vol. 3, 1977, pp. 191-224. 28 Magnussen, B. F. and Hjertager, B. H., "On Mathematical Modeling of Turbulent Combustion with Special Emphasis on Soot Formation and Combustion," 16th Symposium (International) on Combustion, The Combustion Institute, Pittsburgh, Pa., 1976, pp. 719-728. 29 Gosman, A. D. and Lockwood, F. C., "Prediction of Influence of Turbulent Fluctuation of Flow, Heat Transfer Surfaces," 14th Symposium (International) on Combustion, The Combustion Institute, Pittsburgh, Pa., 1973, p. 661. 30 Wallis, G. B., One Dimensional and Two Phase Flow, McGrawHill Book Co., New York, 1969. 31 Borman, G. L. and Johnson, J. H., "Unsteady Vaporization Histories and Trajectories of Fuel Drops into Swirling Air," SAE Paper 598C, Oct. 1962.

Downloaded by Universitats- und Landesbibliothek Dusseldorf on May 5, 2013 | http://arc.aiaa.org | DOI: 10.2514/3.62687

You might also like