Sunteți pe pagina 1din 16

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/311987094

Reacting Unsteady Reynolds-Averaged Navier–Stokes with the Tabulated


Premixed Conditional Moment Closure Method

Article  in  Journal of Propulsion and Power · December 2016


DOI: 10.2514/1.B35741

CITATIONS READS

2 205

3 authors, including:

Scott Martin
Embry-Riddle Aeronautical University
51 PUBLICATIONS   169 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Effect of mixtures on the S-CO2 power cycle View project

Jet Fuel Kinetic Mechanisms for Turbulent Combustion View project

All content following this page was uploaded by Scott Martin on 14 March 2017.

The user has requested enhancement of the downloaded file.


JOURNAL OF PROPULSION AND POWER

Reacting Unsteady Reynolds-Averaged Navier–Stokes with the


Tabulated Premixed Conditional Moment Closure Method

Carlos A. Velez∗
University of Central Florida, Orlando, Florida 32816
Scott M. Martin†
Embry-Riddle Aeronautical University, Daytona Beach, Florida 32114
and
Subith S. Vasu‡
University of Central Florida, Orlando, Florida 32816
DOI: 10.2514/1.B35741
Downloaded by EMBRY-RIDDLE AERO UNIV. on February 2, 2017 | http://arc.aiaa.org | DOI: 10.2514/1.B35741

The tabulated premixed conditional moment closure model has shown the capability to model turbulent, premixed
methane flames with detailed chemistry and reasonable run times in a Reynolds-averaged Navier–Stokes
formulation. The tabulated premixed conditional moment closure model is a table lookup combustion model that
tabulates species, reaction rates, and thermodynamic data for use by the computational-fluid-dynamics code during
run time. In this work, the tabulated premixed conditional moment closure model is extended to unsteady Reynolds-
averaged Navier–Stokes. The new model is validated against particle image velocimetry and laser Raman
measurements of an enclosed turbulent reacting methane jet from the German Aerospace Center. The flame’s
reaction progress variable, its variance, and the scalar dissipation rate are calculated by the computational fluid
dynamics in three dimensions. These three parameters are used to index detailed species information from the table
for use by the computational-fluid-dynamics code. The scalar dissipation is used to account for the effects of the small-
scale mixing, whereas a presumed shape beta function probability density function is used to account for the effects of
large-scale turbulence on the reaction rates. Velocity, temperature, and major species are compared to the
experimental data. Accurate predictions of the velocity fields were obtained, but accurate predictions of scalar
quantities were limited by the adiabatic assumption of the tabulated premixed conditional moment closure model.

I. Introduction Da < 1). The Reynolds number in GTE combustion chambers is


large [O106 ], with turbulent intensities ranging from 10 to 50% of
A S EMISSION regulations become more stringent and fuel costs
continue to increase, the need for better numerical tools to
predict and improve engine performance are ever more important.
the mean velocity. In these regimes, both the small- and large-scale
turbulent eddies are able to penetrate and alter the flame structure.
This behavior is consistent with a distributed or well-stirred
This is especially true in the field of natural gas turbines, which
combustion regime [4], where mixing is high and the Karlovitz
use lean premixed (LPM) combustion systems to achieve carbon
number Ka is greater than unity. As in well-stirred reactors, by
monoxide (CO) and unburned hydrocarbon emissions in the
constantly mixing incoming unburned fuel with high-temperature
sub-2-ppm level a NOx as low as 9 ppm [1]. With the natural gas partially burned products, stable flames may be sustained under lean
reserves in the United States increasing, and as the emissions from burning conditions and at high flow rates. To accurately predict
coal-fired power plants become less tolerable, LPM gas turbine premixed flame emissions in the well-stirred combustion regime, the
engine (GTE) combustor designs have become important in meeting effect of large- and small-scale turbulent mixing must be well
the energy demands of society [2]. Although there has been success in represented in the combustion model [5], coupled with detailed
the industrial use of ground-based LPM GTEs, these systems are still chemistry. Swaminathan and Bray [6] relate the order of magnitude
largely underdeveloped for use in transport systems such as for of the scalar dissipation rate N to the inverse of the Damkohler
air-based transportation [3]. Further research and model development number. This promotes the need for inclusion of small-scale mixing
are needed before LPM combustion models, in conjunction with effects in premixed combustion models for highly turbulent GTE
computational fluid dynamics (CFD), become suitable for the applications where Da ≤ 1 and scalar dissipation is high [O106 ].
rigorous usage required for the optimization and design of Pitsch [7] includes the fluctuations of scalar dissipation in an Eulerian
succeeding LPM GTEs. unsteady flamelet model for nonpremixed flames and finds better
GTE systems operate under highly turbulent conditions where the predictions of the heat release and species predictions.
turbulent inertial length l 0 is comparable to or smaller than the The largest challenge in combustion models is the accurate closure
laminar flame thickness (i.e., Damkohler number less then unity, of the mean chemical source term ω _ i . Typical closures involve
relating the chemical source term to mixing rates, a reaction progress
Received 26 January 2015; revision received 6 August 2016; accepted for
variable (RPV), the flame position, a turbulent flame speed, or
publication 6 August 2016; published online 30 December 2016. Copyright © reduced reaction rate models. The most common of these models are
2016 by the American Institute of Aeronautics and Astronautics, Inc. All 1) level set approach (G-Eqn) [8], 2) laminar one-dimensional
rights reserved. All requests for copying and permission to reprint should be flamelet (L1DF) models [9–11], 3) mixing limited (Eddy dissipation
submitted to CCC at www.copyright.com; employ the ISSN 0748-4658 model [12,13], Bray, Moss and Libby [14]), or 4) simplified
(print) or 1533-3876 (online) to initiate your request. See also AIAA Rights chemistry models with limited reaction kinetics. The level set
and Permissions www.aiaa.org/randp. approach is unique in that a nonreactive scalar G is used to quantify
*Ph.D. Graduate, Department of Mechanical and Aerospace Engineering, the premixed combustion progress [15]. G is a distance function of
Center for Advanced Turbomachinery and Energy Research, 4000 Central the flame front with no physical interpretation. Although there are no
Florida Boulevard; currently Combustion Research Engineer, GE Global
Research, 1 Research Circle, Niskayuna, NY 12309. Student Member AIAA.
source terms for the nonreacting progress variable, the G-Eqn model

Research Engineer, Eagle Flight Research Center. Member AIAA. depends on an empirically obtained model for the turbulent flame

Assistant Professor, Department of Mechanical and Aerospace Engineer- speed, which is used for closure of the mean reaction rate [16]. L1DF
ing, Center for Advanced Turbomachinery and Energy Research, 4000 models, similar to the premixed tabulated flamelet-generated
Central Florida Boulevard. Member AIAA. manifold (FGM) [10,17], assume that the flow is laminar at the small
Article in Advance / 1
2 Article in Advance / VELEZ, MARTIN, AND VASU

scales within the flame thickness δc and that the time scales of that ∂Qi ∕∂c ≫ ∂Qi ∕∂xi ; ∂Qi ∕∂t. This concept is analogous to
chemical fluctuations are much faster than those of small or large relating a person’s yearly financial income to their natural eyesight.
turbulent scales δt . This effectively decouples the chemical reaction There appear no correlations, and so the behavior of the income when
model from the flowfield conservation equations. Based on this ranging through different levels of eyesight (20∕20, 15∕20,
asymptotic assumption, a set of flamelet equations can be derived, 19.7∕18.3) will be highly nonlinear, and more importantly,
solved offline for a range of conditions, and then tabulated for use by fluctuations from the mean yearly income will be large and even
the CFD code. Only the effect of large-scale turbulence is taken into comparable to the mean. Instead, consider if a person’s yearly income
account in L1DF models by incorporating the variance of the were related to their age. A close to linear relationship would be
progress variable by means of a probability density function (PDF) expected, and most importantly, the fluctuation from the mean would
integration. This PDF is constructed by the RPVand its variance [18], be small. Thus, fluctuations from the conditioned average are much
which act as shape functions for the PDF. This technique is discussed smaller than fluctuations from the unconditioned average. Because
in detail in Sec. II. Mixing limited models assume that the reactions fluctuations from the conditioned mean are reduced, a first-order
are solely dependent on the mixing rate of the RPV (premixed) or the approximation can be used in conditional space with sufficient
fuel and oxidizer (nonpremixed). Although these models allow for accuracy.
strong turbulent and chemical interactions, the chemical kinetics are With this in mind, if the conditional average of the reaction rates
solely dependent on the mixing rate [12,13], instead of the would be taken, smaller fluctuations of the conditioned reaction rates
thermodynamic state and composition of the flame. Reduced reaction would be produced. Because reaction rates are nonlinearly dependent
models use 1–5 step reaction kinetics to model the reaction rates with on stoichiometry, moments conditional on a stoichiometric variable c
Downloaded by EMBRY-RIDDLE AERO UNIV. on February 2, 2017 | http://arc.aiaa.org | DOI: 10.2514/1.B35741

an Arrhenius closure, which is solely dependent on the absolute would eliminate the major source of nonlinearity, and more accurate
temperature of the mixture. closure of the reaction rate terms in the species mass fraction transport
In the preceding discussion of low-run-time premixed combustion equations would be achievable [19]. This statistical feature of
models, closures of the chemical source term are based on relating a conditional averaging is used to close the mean reaction rate term in
distance function, progress variable, mixing rate, or temperature to CMC models, which is known as the first-order CMC approximation.
the mean chemical reaction rate through empirical results (flame By assuming that the mean conditional reaction rate is a function of
speed closure), asymptotic analysis (L1DF, mixing limited), or the individual conditioned species’ mean reaction rates, i.e.,
reduced chemical kinetic mechanisms. Here, a more-fundamental _ i jζi ≃ hρjζiωQ
hρω _ 1 ; Q2 ; : : : ; Qns , all of the species’ thermody-
method, derived from first principles, is proposed (tabulated namic states and individual reaction rates contribute to the closure of
premixed conditional moment closure, T-PCMC) using both detailed the mean chemical reaction rate, which would be a poor assumption
chemical kinetics with the effects of large- and small-scale mixing to in unconditioned space, where fluctuations from the mean are large
close the mean chemical source term through the use of a conditional and can be comparable to the mean. This, in effect, removes the
moment closure (CMC) first-order approximation [19]. The first- nonlinearity of the mean chemical source term hρω _ i jζi and allows
order CMC approximation allows for the individual thermodynamic for a physical closure in terms of the reaction rates and
state and reaction rates of each species to contribute to the mean thermodynamic state of each transported species in the mixture
chemical reaction rate while incorporating the effect of scalar _ 1 ; Q2 ; : : : ; Qns ]. The subscript “ns” is the number of species in
[ωQ
dissipation rate from first principles. A high-level explanation of the the mixture and is equal to 53 species in this work, which uses the
CMC first-order approximation and conditional averaging is GRI3.0 kinetic mechanism [20] to model the thermokinetic
provided next. These concepts will help the reader better understand interactions for the entire combustion process (0 < c < 1).
the more technical aspects of the T-PCMC model discussed in Sec. II. Applications of nonpremixed CMC (NPCMC) have been
The CMC model takes advantage of conditional moments extended for the prediction of soot [21], blowout [22], emissions
[Qi xi ; t; c  hY i x; tjc  ζi]. Qi is the conditional, or condi- [23], lifted flames [24], ignition [22], differential diffusion [25], and
tioned, average of the mass fraction Y i for a species i, where values more. Still, development of the premixed CMC (PCMC) model is
from the entire ensemble are only included in an average hi when the scarce. Amzin [26], in 2012, developed a fully coupled premixed
value satisfies a specific condition (for example, c  1). The species CMC model with encouraging results in a steady Reynolds-
mass fraction Y i is used here, for example, but any thermodynamic averaged Navier–Stokes (RANS) formulation. Recently, in 2013
scalar can also be used. c, the RPV, is the conditioning variable and [27], Thornber et al. validated a high-order fully compressible CMC
ranges from 0 to 1. represents the unburned physical condition, and model for premixed flow. The result of this initial implementation
c  1 is the fully burned physical condition of the scalar quantity Y i provided accurate predictions of species mass transport, as seen in
undergoing the conditioning. The intermediate values of c are then NPCMC, and is the motivation of this work in developing the
used to condition the species mass fraction for the entire combustion PCMC model. Specifically, this work builds upon the development
progress [Qi xi ; t; c  hY i xi ; tjc  0; 0.1; 0.2; : : : ; 1i]. Note of a tabulated PCMC (T-PCMC) model, first presented in [28] and
that, in this example, the conditioned mass fraction Qi xi ; t; c is later developed in [29–31]. The model assumes that conditional
additionally a function of the RPV c, whereas the unconditioned mass moments Qi are uniform in real space under the assumption of a
fraction Y i xi ; t is solely a function of space and time. The solution is well-mixed regime (∂Qi ∕∂xi ; ∂Qi ∕∂t ≡ 0). This allows the
three-dimensional (3-D) in space such that xi  x1 ; x2 ; x3 . reduced PCMC equations to be solved offline. The solution is then
By defining the conditioning variable c with stoichiometrically tabulated, and the resulting species mass fractions and reactive
dependent variables, such as species mass fraction Y i , enthalpy hi, or scalars are accessed by the CFD as a function of progress variable,
temperature T, a strong correlation can be made between conditional variance, and scalar dissipation rate. This model is attractive for its
gradients Qi0 and gradients in the RPV (i.e., changes in Q correspond detailed chemistry, physical first-order closure of the mean
to changes in the RPV more than changes in time and space). This is chemical reaction rate, low run time, and its inclusion of both the
an expected result because physically both the conditioned mass small- and large-scale turbulent mixing effects on the mean reaction
fraction Qi and the RPV c depend on the thermodynamic state and rate and species mass transport.
chemical composition of the mixture. Thus, because Qi and c are The lean premixed flame under investigation reacts in an enclosed
dependent on the same physical quantities, the total fluctuation of combustion chamber with the reacting jet offset from the center of the
Qi c; xi ; t will be much more correlated to changes in the RPV chamber. By offsetting the jet, a recirculation zone emerges above the
(∂Q∕∂c) when compared to fluctuations in space (∂Q∕∂xi ) and time flame, which feeds partially burned hot gases back upstream to mix
(∂Q∕∂t), both of which are not dependent on any physical quantities with cooler unburned gases. In previous numerical work [30] by the
[t; xi ≠ fY i ; Qi ; c]. The conditional mass fraction Qi c; xi ; t is a authors, the T-PCMC model was presented in a steady RANS
function of the RPV, time, and space, whereas the unconditioned formulation and compared to the same experimental measurements
mass fraction Y i xi ; t is solely dependent on time and space. used in this paper. Although the results were encouraging, it is not
Because the conditional mass fraction is correlated more strongly expected that a steady RANS implementation would be able to
with the changes in the RPV than with time and space, it is expected predict transient behavior such as the complex flow considered in this
Article in Advance / VELEZ, MARTIN, AND VASU 3

work. Here, the T-PCMC method is extended to unsteady Reynolds- the CFD. The PDF function takes exact shape based on the values of
averaged Navier–Stokes (URANS) and compared with the same the RPV (c) and its variance (c 002 ). The PDF construction and
experimental data as with the RANS validation case [30]. The intent integration are discussed in detail in Sec. II.B. Once the conditioned
of this work is to develop the T-PCMC model in an unsteady scalars are transformed back to unconditioned space, the 53 species
formulation with URANS turbulence modeling to help predict the mass fractions, density, RPV source term, molecular weight (MW),
transient large-scale turbulent features apparent in this flame. Past and temperature are available for use in the CFD code. The
numerical studies on this specific flame [32] have been performed temperature is used to update the thermodynamic quantities of the
with artificial thickening [17] and FGM models [30]. The past mixture. Sc is used as the source term in the transport of c and c 002 .
numerical studies included the effect of heat loss and show good The transport equations for c and c 002 are solved in transient form by
prediction of mean temperature and velocity fields. The predictions the CFD code and in three spatial dimensions.
of CO2 mole fractions are presented in [17], and CH4 predictions are
presented in [10,33]. This is the first attempt at applying the T-PCMC A. Computational-Fluid-Dynamics Solver in Unconditional Space
to simulate the lean premixed flame under consideration with detailed
The CFD code is required to solve for the flowfield equations
chemical modeling and URANS turbulence modeling.
(continuity and momentum), the RPV transport equation (c), the
The paper is outlined as follows. Section I compares the T-PCMC
variance transport equation (c 002 ), and the scalar dissipation rate N at
model to low-run-time premixed combustion models. Section II
every time step and for every cell in the domain. At each time step,
briefly derives and discusses the governing equations for the
every cell in the domain has available the unconditioned density,
T-PCMC model for both the tabulated results and transport equations
temperature, RPV source term, molecular weight, and the 53 species
Downloaded by EMBRY-RIDDLE AERO UNIV. on February 2, 2017 | http://arc.aiaa.org | DOI: 10.2514/1.B35741

solved in the CFD during run time. Section III discusses the
mass fraction from the T-PCMC table.
validation case and simulation case setup. Section IV presents
To quantify the progress of the flame, a suitable quantity must be
the validation results from the URANS T-PCMC model compared to
chosen to define the RPV. For nonpremixed combustion, the mixture
the measured data. Section V concludes the paper with a summary of
fraction is used as the RPV. In premixed flow, the mixture fraction is
the findings and a description of future work.
constant and has no use as a variable to which remaining variables are
conditioned. In this work, the change in sensible enthalpy is used to
II. Premixed Tabulated Conditional Moment define the progress of the reaction c as in Eq. (1). c  0 and c  1
correspond to the fully unburned and burned conditions, respectively.
Closure Method In this arrangement, c is a monotonically increasing function, and any
To describe the T-PCMC model, a “top-to-bottom” approach is fuel mixture can be used because the RPV does not depend on the
taken. Section II will begin with a description of the T-PCMC model mass fraction of any specific major species:
flowchart, showing the basic inputs, outputs, and shared variables
between the CFD code and the tabulated solutions to the T-PCMC hs − hsu
cx; t ≡ (1)
equations. The model-specific equations solved offline by the hsad − hsu
combustion model and during run time by the CFD code will be
discussed in more detail after a general description of the code is where hs  ∫ TT u Cp dT is the instantaneous sensible enthalpy. The
presented. subscript “ad” represents the adiabatic equilibrium condition, and the
Figure 1 illustrates the T-PCMC model and how the premixed subscript u signifies the unburned state. This formulation of the RPV
CMC table interacts with the CFD solver. The components within the assumes that the total enthalpy is constant and that the there is no heat
dotted box refer to the T-PCMC table, which is generated offline by loss from the high-temperature gases to the surroundings. To develop
solving a T-PCMC equation for each species in the mixture (ns) and a transport equation for the RPV, the RPV is first decomposed into its
for a range of scalar dissipation rate N values. Each species’ T-PCMC mean and fluctuating components using a density-weighted Favre
equation is solved with the species’ respective burned and unburned average c. ~ A density-weighted average is employed because the
conditions. Changing the scalar dissipation alters the rate of diffusion fluctuations in density ρ 0 are not required in Favre averaging. To
of the mixture and the solution of the T-PCMC equation. By solving solve for the RPV, the Favre-averaged transport equation for sensible
the T-PCMC equation for each species and for a range of N values, a enthalpy is substituted into Eq. (1), as in [29], resulting in Eq. (2):
table may be constructed with the tabulated solutions for use by the hP i
CFD code.   ρ
 ω
~ h
∂c
~ ∂u~ i c
~ ∂ ∂c~ i f;i
The value of N computed by the CFD is used to index the correct ρ   ρ Dc − u~ i00 c~ 00 
conditioned scalar hQ~ i i from the table. Once the correct N table is ∂t ∂xi ∂xi ∂xi Δhsad−u
chosen, a set of conditioned values for mass fraction Y i , density ρ,  
μ
RPV source term Sc , and temperature T may be accessed as a function  ∇ t ∇c~  ρ S~c (2)
σc
of c. A PDF function integration is then used to transform the
conditional quantities back to unconditioned space for use by
Here, ρ is the density, u~ i is the velocity in all three directional
components, c~ is the RPV, S~c is the RPV source term, and Dc is the
diffusion rate of c, which requires modeling. The spatial derivatives
of the diffusivity Dc have been negated because, in the T-PCMC
model, conditional species are assumed uniform. If the species are
uniform in conditional space, then there will be no changes in the
mixture diffusion rate in conditional space. Instead, as in previous
CMC work [21,22,27], the diffusion is modeled assuming a Fick’s
law diffusion. Derivatives in density are neglected because a density-
weighted Favre filter is used, which reduces the magnitude of density
derivatives in the governing equations for c and variance. ω~ i is the
reaction rate of species i; hf;i is the heat of formation of species i;
Δhsad−u is the difference in sensible enthalpy between the adiabatic
equilibrium and unburned states; hs is the sensible enthalpy at the
local temperature; and hsu is the sensible enthalpy at the unburned
state. The diffusion term is unclosed due to the second moment term
u~ i00 c~ 00 . The second-to-last term on the right-hand side shows the
diffusion term in closed form, which uses the turbulent viscosity μt
Fig. 1 Flowchart of the T-PCMC model. and turbulent Schmidt number σ t  0.7 to represent the diffusion rate
4 Article in Advance / VELEZ, MARTIN, AND VASU

of the RPV. The last term ρ S~c is the mean reaction rate of the RPVand production term in the variance equation, which is not the case for
is tabulated in the T-PCMC table by nonreacting variance or nonpremixed variance transport equations.
The original values are obtained from nonreacting turbulent theory
hP i
and resulted in large values of the variance, giving a lower flame

 ω~ hf;i
S~c  (3) temperature and larger flame thickness. It was evident from the initial
Δhsad−u values that there was an overproduction of variance; limiting the Cc1
and Cc2 values was performed once and has performed well for
In addition to the governing equation for the RPV [Eq. (2)], a multiple geometries, mesh resolutions, and boundary conditions
transport equation for the variance c 002 of the RPV is solved. [29,31,36]. The URANS model has shown a weaker dependence on
Typically, algebraic models for the variance are used, which assume the values for Cc1 and Cc2 in comparison to the steady RANS version
that the variance is directly proportional to the scalar dissipation rate presented in [29,30].
[10]. This assumption does not allow for a scalar dissipation rate to The unconditioned scalar dissipation N~ c  Dc ∂c∕∂x
~ i ∂c∕∂x
~ i
exist without the presence of variance or a change in the RPV. It is is modeled as
shown in direct numerical simulation (DNS) comparisons of a C2 H4
turbulent jet flame [34] that algebraic models failed to adequately ϵ~
reproduce the DNS results in comparison to results obtained from a N~ c  2Cc2 c~ 002 (6)
k~
transport equation of the variance.
The transport equation for the variance of RPV is derived by where ϵ~ is the turbulent dissipation rate, k~ is the turbulent kinetic
Downloaded by EMBRY-RIDDLE AERO UNIV. on February 2, 2017 | http://arc.aiaa.org | DOI: 10.2514/1.B35741

subtracting the Favre-filtered c~ equation from its instantaneous c energy, and c~ 002 is the variance, which is obtained by solution of
equation. Following previous literature [21,22,27], the derivatives of Eq. (5). An improved model for N~ c, formulated for URANS, has
the diffusivity Dc in space have been negated and are instead modeled been proposed in [26,37], which additionally includes the effects of
assuming Fick’s law diffusion. The mean molecular transport is molecular diffusion. The additional terms, representing the
negated because turbulent transport is the dominant diffusion chemical and molecular diffusion, are inversely proportional to the
mechanism in the high-Reynolds-number flows considered in this Karlovitz number (1∕Ka0.4 ) [26]. Under the well-mixed assumption
work. The transport equation for the variance of the RPV (cVar) is used in this work, the Karlovitz number is large [O10–100];
written next as accordingly, the inclusion of chemical diffusion is considered small
    compared to that of turbulent diffusion. Note that the unconditioned
∂ρc~ 002  ∂ρu~ i c~ 002  ∂ ∂c~ 002 form of N is shown in Eq. (6); in contrast, the T-PCMC table uses a
  ρDc − ρui00 c~ 002
∂t
|{z} ∂xi ∂xi ∂xi |{z} conditional N to index and solve the T-PCMC equation. The
|{z} |{z} cVar4
cVar1 cVar2 cVar3 transformation from unconditioned to conditional N will be
∂c discussed in Sec. II.B.
− 2ρui00 c~ 00 − 2ρN~ c  2c~ 00 ω
_~ c (4)
∂xi |{z} |{z}
|{z} cVar6 cVar7 B. Tabulated Premixed Conditional Moment Closure Solution in
cVar5 Conditional Space
Referring to Fig. 1, the T-PCMC table must take in values for RPV,
The first two terms (cVar1,2) of the left-hand side of Eq. (4)
its variance, and scalar dissipation rate and output the unconditioned
represent the temporal and spatial changes of the variance of c by
species mass fractions, density, temperature, and RPV source term.
means of convection. The term cVar3 represents the diffusive flux of
To generate these values, the T-PCMC method solves a reduced
the variance. Both and ρui00 c 002 require modeling, which introduce a
premixed CMC formulation in conditional space. A brief derivation
model constant Cc1 , whose theoretical value, derived by nonreacting
of the T-PCMC equation is provided later, and interested readers are
turbulent theory, is equal to 2.00. Terms cVar4 and cVar5 both
represent the effects of mean and fluctuating strain fields and the to referred to [31] for a detailed derivation and description.
interaction of turbulence ui00 and scalar fields c 00 . These two terms are To condition the equations of species mass transport Y~ i on the
lumped together because both include a second moment term a 00 b 002 , value of the RPV (c), a Favre-averaged conditional moment of the
which requires modeling; these terms are related to the scalar species mass fraction is performed as
dissipation rate N~ c . Term cVar6 represents the effect of the scalar
dissipation rate on the variance. Because c is a reactive scalar, the hρxi ; tY~ i xi ; tjcx
~ i ; t  ζi
Q~ i c;
~ xi ; t ≡ (7)
production of variance c 002 due to chemical reaction exists and is hρxi ; tjcx~ i ; t  ζi
represented by the last term cVar7 in Eq. (4) [31]. Because models are
required for most of these terms, the implemented form of the where vertical bars indicate that the average is taken over only those
equation is shown in Eq. (5): values of Y, c equals ζ (the condition), and the angle brackets indicate
an average value. ζ is equivalent to c but in conditional space (i.e., ζ is
∂c~ 002  ∂u~ i c~ 002  a set of values in c space representative of the c values in real space).
ρ  ρ  ∇2 αeff c~ 002   Cc1 μt ∇c
~ 2 − ρN~ c
∂t } ∂xi |{z} |{z} |{z} As in c, ζ ranges from 0 to 1, which represent the unburned and fully
|{z |{z} cVar3 cVar4 cVar5
cVar1i cVar2i
i i i burned conditions. The species mass fractions can now be
decomposed into their Reynolds decomposition, representing the
− 10c~ 002 ∕T ~ c~  T u ∕T ad − T u S~c  (5) mean Y~ i and fluctuating component yi as traditionally used [10]. In
|{z
}
cVar6i addition, with the use of Eq. (7), a Favre-averaged conditional
moment decomposition of the instantaneous Y i can be written as
Here, αeff is modeled by the contribution of laminar and turbulent
thermal diffusivities in the URANS formulation. μt is the turbulent Y i x; t  Y~ i x; t  yi x; t (8)
viscosity, which is also modeled in URANS to close the Reynolds
stress tensor. The model constants (Cc1 ; Cc2 ) for the closure of terms
cVar4i and cVar5i in Eq. (4) are 0.715 and 8.00, respectively. Cc2 is Y i x; t  Q~ i cx;
~ t; x; t  qi c; x; t (9)
used in the model for N as in Eq. (6). These values have been adjusted
in comparison to the theoretical values of 2.86 and 2.00. A value of Here, qi is the conditional fluctuation, and Q~ i is the conditional
2.86 is obtained by 2∕σ, where σ is the Schmidt number equal to 0.7. mean of Y i . Notice that, by applying a conditional mean on Y~ i , a new
These values are originally derived from nonreacting turbulence dimension c~ is introduced. With this added dimension, conditional
theory [35]. It is expected that, in the premixed formulation of the fluctuations qi are greatly reduced (yi ≫ qi ), and adequate accuracy
variance equation, these constants must change because there is a can be obtained from a first-order approximation (i.e., Y i ∼ Q~ i ).
Article in Advance / VELEZ, MARTIN, AND VASU 5

With a conditional representation of the species mass fraction the RPV, which resembles convection in RPV space. The third term
available, the derivation of the premixed CMC equation can be represents the conditional reaction rates of species i and is the source
performed as follows. Equation (7) is substituted into Eq. (9), which is term of the Eq. (11).
then used to replace Y i in the reactive species mass fraction transport It is interesting to note that Eq. (11) can also be reached, by
equation. Last, Eq. (2) is substituted into the resulting mass fraction Eq. (10), assuming fast chemistry at a quasi equilibrium, such that
conservation equation and conditionally averaged, resulting in hYjζi  Y e ζ in the CMC equation [38]. This corresponds
mathematically to ∇Q~ i  0 because Q~ i reduces to only a function of
_ ~ Q~ i00
_~ i jζi  hρDi ∇c~ · ∇cjζi c, but the physical assumption of a thin flame is quite different from
hρjζiQ~ i  hρ U~ jζi · ∇Q~ i  hρω
that of a distributed reaction.
− hρS~c jζiQ~ i0  eQ  ey (10) Even though Eq. (11) has a similar form to the steady flamelet
equation, there are three distinct differences in the implementation of
In Eq. (10), the primes indicate derivatives in c space, the gradient Eq. (11) (T-PCMC) to the premixed steady laminar flamelet
operator ∇ is a derivative with respect to physical space, and the dot equations as used in [10,33,39]. First, scalars are conditionally
indicates a time derivative. Equation (10) is similar in form to the averaged, providing reduced fluctuations from the mean, which incur
NPCMC equation, except for the addition of hρS~c jζiQ~ i0 , which less error in gradient diffusion or Fick’s law approximations. Fick’s
appears as a consequence of c being a reactive scalar and represents the law assumes that diffusion occurs at a steady state. For diffusion to
convective velocity in c space. The addition of this convective velocity be theoretically steady, the temporal fluctuations of the mean
term makes the PCMC equations stiffer than the NPCMC formulation. concentration gradients must approach zero. Because temporal
Downloaded by EMBRY-RIDDLE AERO UNIV. on February 2, 2017 | http://arc.aiaa.org | DOI: 10.2514/1.B35741

Moreover, the PCMC equation becomes stiffer as N approaches the fluctuations from the mean are reduced in conditional space, a closer
theoretical limit of zero. For a detailed description of each term, the match to the assumed steady diffusive state is obtained. Second, the
reader is referred to [31]. local scalar dissipation rate is calculated for each cell in the domain,
To make Eq. (10) more usable for tabulation, simplifying assumptions and the effect of small-scale turbulence on the reaction rates is
are made. The following simplifications are made to reach the T-PCMC directly included. Although flamelet models can also achieve this,
equation, which is proposed to be applicable for well-stirred or laminar flamelet models assume that the flow is laminar at the small
distributed reaction regimes. Bilger [19] states that, for stationary scales; therefore, a constant N (N ≠ fc) value is prescribed and is
turbulent flows, as in the discussed flame, conditional averages will volume-averaged across the entire computational domain as in
depend weakly on time (∂Q~ i ∕∂t ≈ 0). Thus, the time derivative of [10,33,39]. The third and most influential distinction is in the
conditional quantities is removed from Eq. (10). Although this physical closure, provided by the first-order CMC approximation, of
assumption will limit the applicability of the model to stable flames, the the mean reaction rate ω~ i . The mean conditional reaction rate from all
temporal fluctuations in c and c 002 are accounted for in Eqs. (2) and (5). n species is closed using a first-order approximation of the
Because Q~ i ∼ fc, the temporal changes of Q, in c space, are implicitly conditioned scalar. Neglecting conditional fluctuations (qi  0)
accounted for in Eqs. (2) and (5). The removal of the time derivative in allows the individual mean reaction rates to be summed up for all ns
the PCMC equation allows the T-PCMC table to be decoupled species and then conditionally averaged. This closure provides a
temporally from the flowfield equations. This is key to solving for the T- physical representation of the mean reaction rate based on the
PCMC equations offline and storing them in a tabulated format. The two chemical composition and thermodynamic state of the entire mixture.
error terms eq , ey are negated. eq is negated because it scales with This first-order closure is written as
Re−1∕2 , making it small for highly turbulent flow. ey is dependent on qi
and is considered negligible because the fluctuation of the conditional _~ i jζi ≃ hρjζiω
hρω _~ Q~ 1 ; Q~ 2 ; : : : ; Q~ ns  (12)
mean qi is small; thus, the gradient of the fluctuations in real space, ∇qi ,
should also be small. Negating eq and ey additionally implies that Note that ω _~ i additionally appears in the closure of the RPV source
differential diffusion effects are neglected. For a detailed derivation and term (Sc), the convective term of Eq. (11).
description of eq and ey , the reader is referred to [31]. In the case of unity Equation (11) is coupled to the flowfield equations by the
Lewis number, such as the methane (CH4 ) flame under consideration, conditional scalar dissipation rate, hN~ c jζi  hρD∇c~ · ∇cjζi∕h
~ ρjζi.
the assumption is appropriate because differential diffusion effects are Accordingly, a scalar dissipation rate must be calculated by the CFD
small when thermal and mass diffusivities are equal. (N~ c ), in unconditioned space, to be able to index the correct N
No major assumptions have been made to restrict the combustion solution set from the table. Note that N~ c must first be converted to a
regime of the premixed flame, except that the flow is turbulent and the conditional scalar dissipation rate hN~ c jζi, as in Eq. (11).
flame is stationary. To adapt the remaining expression to the well- For complete closure of Eq. (11), hN~ c jζi requires modeling.
stirred or distributed reaction regimes, it is conjectured that, in the Experimental measurements of hNjζ ~  ci for premixed flames do
case of a well-stirred reactor, the conditional scalars (not the not exist in the literature, and closures for hNjζ~  ci have solely
unconditioned scalars) are approximately uniform over real space. been developed and validated in nonpremixed CMC formulations. In
This assumption results in small gradients of the conditioned species recent premixed CMC work [27], the nonpremixed closure for
mass fraction in real space, which reduces ∇Q~ i to zero, removing the hNjζ
~  ci is adopted due to a lack of validated models for the
second term from Eq. (10). The same assumption, ∇Q~ i  0, would conditional scalar dissipation rate in premixed flow. Typically, in
also remove the eq term for its additional dependence on ∇Q~ i . The NPCMC models, a bell-shaped PDF is used to statistically weigh the
assumptions that ∇Q~ i  0 and ∇q~ i  0 will limit the range of scalar conditioned N (hN~ c jζi) as a function of its unconditioned value N. ~
dissipation values that are valid in this model, even though other The methodology is best known as the amplitude mapping closure
values may give a solution. To better determine the relative [40]. Similarly, in this work, a parabolic curve is used to model the
magnitude of the assumptions taken to reach Eq. (11), additional shape of hN~ c jζi. The parabolic shape was developed using results
DNS and experimental data are required for the reaction regimes from a premixed flamelet solution set (not shown here), which
under consideration. It is anticipated that this assumption is valid for showed a nearly symmetric relationship for the conditional scalar
low to moderate Damkohler number or flames with characteristics of dissipation as a function of the RPV. hN~ c jζi has its maximum value at
a well-mixed or distributed reaction regime. This final assumption c  0.5 and is formulated as
(∇Q~ i  0) results in the T-PCMC equation written as
hNjζi
~  4N~ c ζ1 − ζ (13)
~ Q~ i00 − hρS~c jζiQ~ i0  hρω_~ i jζi  0
hρDc ∇c~ · ∇cjζi (11)
A constant, equal to 4, is multiplied to ensure that the maximum
The first term in Eq. (11) is the conditioned scalar dissipation rate, value of hNjζi
~ is equal to the value of the unconditioned N (N~ c ) from
which represents scalar diffusion in RPV space. The diffusion Dc is the CFD, i.e., hNjζ
~  0.5i  N~ c .
the diffusivity of the RPV, which is represented in this work as the Now, with a model for hNjζi~ in place, Eq. (11) is fully closed.
thermal diffusivity. The second term is the conditional source term of Equation (11) may now be written into 53 coupled, second-order,
6 Article in Advance / VELEZ, MARTIN, AND VASU

nonlinear ordinary differential equations and solved simultaneously due to the stiffness of the kinetics–turbulence interaction when
for a given value of N. Note that each species’ T-PCMC equation is N < 10. The boundary conditions for Eq. (11) are the unburned mass
solved for the for the entire range of N, resulting in a data set of fractions at c  0 and the adiabatic equilibrium mass fractions at
Q~ i ∼ fN  N min ; : : : ; N max , where i  1; : : : ; ns. c  1. Chemkin along with the full GRI3.0 kinetic mechanism [20] is
To convert the resulting conditional averages Q~ i to ensemble used to provide the species reaction rates ω _ i , sensible enthalpy hs,
averages Y~ i , a PDF integration is performed on the conditional heat of formation hf;i , mixture specific heat Cp , species specific heat
averages [41]. A beta function PDF is used, as in [10,18,36,39], and is Cp;i , mixture MW, and species MW as a function of temperature and
defined as species mass fraction. With these relationships available, the species
reaction rate and diffusion coefficient can be determined for the
1 simultaneous solution of Eq. (10) (Q~ i ) for all ns species. Now, the
Pζ; α; β  ζ α−1 1 − ζβ−1 (14)
Bα; β conditional density, temperature, species mass fractions, and RPV
source term can be computed for the CFD. Thermodynamic
α and β are shape functions of the beta function Bα; β and are quantities were shown to satisfy the ideal gas law in both conditional
defined with respect to the RPV (c) and its variance (c 002 ) as and unconditional space.
    The conditional RPV source term is closed as in Eq. (2) because the
c1 − c c1 − c individual species reaction rates and heat of formations are now
αc − 1 β  1 − c − 1 (15)
c 002 c 002 available. The temperature is determined by iterating through the data
set for a given combination of c and Y i that satisfies energy
Downloaded by EMBRY-RIDDLE AERO UNIV. on February 2, 2017 | http://arc.aiaa.org | DOI: 10.2514/1.B35741

The PDF assumes 0 < ζ < 1 and that the variance stays positive conservation. With the conditional temperature available, fixed
and below its theoretical maximum [c1 − c]. Both of these pressure of 1 atm, and the MW of the mixture known, contributed by
conditions are prescribed in the solution of the T-PCMC equations, all 53 species, the density can be computed via the ideal gas law. The
and the CFD code performs bounding of the c and c 002 fields to ensure conditionally averaged results of density, temperature, species mass
that both criteria are satisfied. Note that this allows for an individual fractions, and the source term for the RPV equation are stored in
PDF shape to be used for each cell in the CFD domain, depending on tabular format for each set of N conditions. At this stage, Eq. (16) is
the RPV and variance at the cell center. Recall that the RPV and used to convert all of the conditioned scalars back to physical space
variance are solved in 3-D and in transient using Eqs. (2) and (5). The for use by the CFD code. This transformation is also performed
beta function PDF is then integrated with the conditioned scalar over offline so that the CFD code receives the unconditioned quantities
RPV space as in Eq. (16). The resulting PDF integration produces the from the T-PCMC table, preventing any added CFD run time from the
unconditioned scalar Y~i : PDF integration. The resulting unconditioned scalars are tabulated
Z1 and linearly interpolated between the tabulated values for both c and
c 002 . In the case of the dissipation rate, the index closest to the value of
Y~i  Q~ i Pζ
~ dζ (16)
0 N is used. The resulting T-PCMC table is a 57 × 100 × 200 × 38 table
where 57 scalars (53 species, density, molecular weight, temperature,
The authors note that this transformation, also used in premixed and source term) are indexed by 100 values of c (0 < c < 1),
FGM [10] and CMC [27] models, is not particularly well suited for 200 values of variance (0 < c 002 < 0.25), and 38 values of
premixed flows. Recall that the RPV is a reactive scalar in premixed N (0.02 < N < 20;000).
combustion, i.e., there exists a production term for the RPV (Sc ) in the To illustrate the output from the T-PCMC table, the resulting
transport equation for c and c 002 . Thus, the total distribution of the density and temperature tables are illustrated in Fig. 2 and shown as
probability of Y~ i taking a value at ζ~ is not fully conserved, yet a functions of c and cVar. Both plots are made at a scalar dissipation
conserved PDF is still used. This is analogous to using a Galton board rate of 200 1∕s. Although these contours are available for the entire
to obtain a Gaussian distribution, except that some of the dropped range of c and cVar values, there are many regions that will never be
balls are removed or added, and still a Gaussian distribution is encountered because the maximum variance is theoretically limited
expected. Further DNS or experimental data are required to quantify by c1 − c. To help limit these contours from nonphysical
the error incurred when using conserved PDF transformations on a combinations of c and cVar, values for which c 002 > c1 − c are
reactive progress variable. It is hypothesized that a corrected removed. It can be seen in Fig. 2 that both density and temperature at
nonconserved PDF can be obtained through information of the c  0 and c  1 have only one value at a variance of zero. At c  0,
current value of the RPV source term (Sc) because the removal of the the temperature is prescribed its unburned value of 573 K, and at
RPV production term would make the RPV a conserved scalar, as in c  1, it is prescribed the fully burned adiabatic value of 2064 K. It
the nonpremixed CMC formulation. can be seen that increasing variance results in a more distributed and
Now, with the details on the governing equations described, uniform variation for density and temperature with c.
Sec. II.C will describe the procedure of numerically solving Eq. (11), Spikes at the extremes of the temperature and density fields
for each species and for a range of N, to develop the T-PCMC table for demonstrate the stiffness of Eq. (11). Although Fig. 2 is useful in
use by the CFD code. illustrating the table output, the role of scalar dissipation rate is
unclear, and still unphysical combinations of c and cVar are
C. Tabulated Premixed Conditional Moment Closure Table encountered (cVar  0 at 0 < c < 1). To help filter out the lower and
To generate the T-PCMC table, Eq. (11) is solved simultaneously upper bounds of the variance, the variance is adjusted based on a
for each ns species for a range of N values. Recall that N is the percentage of the maximum obtainable variance. Mathematically,
diffusive term in a species’ T-PCMC transport equation. Thus, by cVar  c1 − cn%, where n% is the percentage of variance desired.
varying N, the species transport from unburned to fully burned This allows for an average variance to be obtained while preserving
conditions becomes less or more diffusive, while maintaining the local changes in cVar with respect to c instead of holding variance
same unburned (c  0) and fully burned (c  1) conditions. constant for all values of c. In the CFD simulations, it was found that
Equation (11) is a system of second-order, nonlinear, ordinary approximately 10% of the maximum variance is achieved on average
differential equations of the boundary value type and is solved using a within the flame. Accordingly, in Fig. 3, a dynamic variance is used
two-point boundary value problem (TPBVP) solver [42]. The TPBV where all values of c are at a variance of 10% of their own theoretical
solver is a global method to compute numerically the solution of a maximum, c1 − c. Note that this value of average variance is solely
nonlinear TPBVP. Mono-implicit Runga–Kutta numerical schemes used for illustration; the CFD and T-PCMC model are allowed to use
of orders 4, 6, and 8 are solved in a deferred correction framework to the whole range of variance. Now, with an average variance defined
give a solution accurate to a prescribed local tolerance Q10−6  [42]. for each value of c, the scalar dissipation rate can be included as an
Equation (11) is solved over a range of scalar dissipation values axis. This allows for illustration of the tabulated scalars for both major
from 0.02 to 20,000 1/s. Smaller increments of N are used at low N and minor species mass fractions as a function of c, cVar, and N.
Article in Advance / VELEZ, MARTIN, AND VASU 7
Downloaded by EMBRY-RIDDLE AERO UNIV. on February 2, 2017 | http://arc.aiaa.org | DOI: 10.2514/1.B35741

Fig. 3 Unconditioned hydrogen mass fraction Y H2 (top) as a function of


RPV (c) and scalar dissipation rate N at a variance equal to
1∕10Cvar;Max colored by methane mass fraction Y CH4 . Unconditioned
carbon monoxide mass fraction Y CO (bottom) as a function of RPV (c)
and scalar dissipation rate N at a variance equal to 1∕10Cvar;Max
colored by carbon dioxide mass fraction Y CO2 .

(∂c∕∂xi  → 0), producing more intermediates and radicals. In


contrast, at high N, the flame is thin (∂c∕∂xi  → ∞), and the
majority of CH4 is directly converted to CO2 , similar to the behavior
Fig. 2 Density (top) and temperature (bottom) as a function of RPV and of single-step kinetic mechanisms.
variance at a scalar dissipation rate of 200 1/s. The limiting cases of N in Eq. (11) result in analogous solutions, in
conditional space, for a perfectly stirred reactor and the fast chemistry
approximation. These limiting cases are solely mentioned in this
Figure 3 shows two 3-D comparisons. On the top, hydrogen (H2 ) work, and the interested reader is referred to [31] for a more detailed
mass fraction is plotted and colored by methane (CH4 ) mass fraction. discussion and derivation of the limiting cases of N. This promotes
CO mass fraction is plotted on the bottom and colored by the CO2 the idea that the validity of Eq. (11) may be larger than expected and
mass fraction; both plots are shown as functions of c and scalar that the assumed regime in conditional space, where ∂Q~ i ∕∂xi   0,
dissipation rate. Here, the strong dependence of major and minor is applicable for a wider range of combustion regimes than originally
species on the scalar dissipation rate is evident. It can be seen that, at assumed.
low N, where H2 formation is largest, more CH4 has been burned,
resulting in lower CH4 with decreasing N. From a kinetics
perspective, because there is less CH4 , there must be more CO2 or
more intermediates (CO, H2 ) being formed. Looking at CO2 on the III. Experimental Comparison
right side of Fig. 3 shows that, in fact, less CO2 is formed at low N. A. Experimental Data
Thus, if at low N there is less CO2 and less CH4 , there must be more To compare the T-PCMC CFD results to experimental work, the
intermediate species, which is in accordance with the increase of H2 DLR, German Aerospace Center [32,33] single-jet premixed
and CO at low N values in Fig. 3. methane flame is chosen for simulation. Methane and air are
For both plots in Fig. 3, the mass fractions become independent of premixed at an equivalence ratio of 0.71. The mixture is preheated to
any changes in N at approximately N  6000. The CFD results 573 K with an inlet velocity V jet of 90 m∕s and a Reynolds number of
predicted N to range between 0.5 and 400 1∕s within the flame approximately 18,000. The Damkohler and Karlovitz values for the
(0 < c < 1). This encourages the notion that the effect of fluctuating measured flame are not provided in the experimental work. Previous
N must be taken into account in combustion regimes where Da < 1 work by the authors [36] goes into detail to numerically predict the
and changes in N are large within the flame. turbulent and chemical time scales of the flame to determine the
The results from the CMC model show directly the effect large appropriate combustion regime of the flame. Based on the findings of
(variance) and small (N) scale turbulence have on the reaction rates this work, the average Damkohler and Karlovitz values within the
and ultimately the species mass fractions. Moreover, the small- flame are approximately equal to 0.22 and 2.4, respectively. This
scale mixing, N  D∂c∕∂xi ∂c∕∂xi , has been shown to directly would place the DLR flame in between the well-stirred and
alter the reaction pathways, where at low N, the flame is thick distributed reaction regimes.
8 Article in Advance / VELEZ, MARTIN, AND VASU
Downloaded by EMBRY-RIDDLE AERO UNIV. on February 2, 2017 | http://arc.aiaa.org | DOI: 10.2514/1.B35741

Fig. 4 DLR combustion chamber dimensions (left) as well as PIV mean and instantaneous velocity field (right).

The jet is offset from the center of the chamber such that a geometry, a two-dimensional slice along the plane of symmetry is
recirculation zone may stabilize the flame and redirect burned shown in Fig. 5. Figure 5 is cut at a distance of 7.5d downstream of the
products downstream of the nozzle. The rectangular rig had optical nozzle for illustration. The outlet of the domain is located 38d
access for particle image velocimetry (PIV) and laser Raman downstream of the nozzle exit. Grading of the mesh cell size is used to
measurements of velocity and scalars, respectively. refine the mesh resolution near the jet shear layer and along the
On the left side of Fig. 4, the dimensions of the test geometry are interior walls of the nozzle.
labeled for the single jet case. The inlet pipe diameter d is 1 cm A three-dimensional image of the mesh in the nozzle region is
(d  0.01 m) and is used to scale the dimensions of the domain on shown in Fig. 6. The top and front patches of the domain are removed
the left side of Fig. 4. On the right-hand side of Fig. 4 are mean and in Fig. 6 to offer a view of the mesh progression from the outer
instantaneous velocity fields captured by the experimental PIV [33]. boundaries to the jet region. The mesh consists of a main core with
The PIV data provide a 2.2 × 2.2 mm2 spatial resolution. An five surrounding blocks added to place the jet core in the correct offset
uncertainty of 2.3 m∕s in both axial and vertical velocity location as the experimental apparatus. The geometric center of the
measurements was reported, and the Raman measurements of enclosure is illustrated in Fig. 6, which shows that the nozzle is
temperature and species concentration report an average 5% error in symmetric with respect to the axial (X axis) and spanwise directions
the flames used for calibration. Though not specified, it appears that (Z axis) but offset in the vertical direction (Y axis).
the data error of 2.3 m∕s corresponds to the instantaneous velocity To determine the independence of the solution on the
measurements; it is not specified how this would affect the error of the computational domain, a URANS simulation is run with identical
averaged results. conditions, except that the mesh is replaced. The URANS solution for
The inside diameter of the inlet nozzle is 10 mm. Measurements each mesh is initialized with the same steady RANS solution. The
were performed at atmospheric pressure with the walls air-cooled. simulation is set to run for five flow passages. Based on quantitative
The heat loss from the rig and wall temperatures were not reported, comparisons of the mean velocity and temperature fields at one, two,
and the measured peak flame temperature was approximately 200 K three, four, and five flow passages, the average difference in mean
below the adiabatic flame temperature computed by the T-PCMC velocity fields was below the 2.3 m∕s experimental uncertainty, and
model. A recent numerical study [17] estimates the wall temperature the difference in mean temperature was below the 5% experimental
to be approximately 1000 K. Comparing these estimates to the uncertainty by the third flow passage. Because the change in mean
adiabatic fully burned temperature (2064 K) demonstrates the velocity, with additional run time, was below the experimental
presence of heat loss. uncertainty, the field is considered to have reached a statistically
converged solution by the third flow-through time. An additional two
flow passages were included to secure the converged solution.
B. Computational-Fluid-Dynamics Simulation
To quantify the effects of spatial discretization and the mesh
The CMC table lookup combustion model described in Sec. II.A resolution in the computational domain, the solution of the mean
was added to the Open-FOAM CFD code as described in Sec. II.B temperature is compared. The temperature is chosen as a comparison
and used to model the experiments described in [32]. The 3-D low- variable because it is a good indicator of when the solution and
Mach-number version of OpenFOAM [43,44] is employed to solve combustion progress are statistically converged. Temperature is
the coupled pressure–velocity equations from. Although the dependent on the velocity, RPV, variance, and scalar dissipation
flowfield experiences large density variations, the Mach number is fields. Comparisons of the mean temperature fields are plotted next to
low enough that compressibility effects are considered negligible, one another in Fig. 7. The solutions are similar except for the small
and the velocity is large enough that buoyancy effects are negligible. changes in temperature along the shear layer of the reacting jet. To
A pressure implicit split operator algorithm [45] is used to solve the better quantify the difference in mean temperature among the three
discretized Navier–Stokes equations in time and space. Backward meshes, line plots of the mean temperature are compared at a distance
Euler implicit time marching scheme is used for all variables, and a of 1d and 15d downstream of the nozzle along the plane of symmetry.
second-order cubic discretization scheme is used for the divergence, As seen in the temperate contours, differences between the three
Laplacian, gradient, and interpolation of the discretized equations solutions are greatest along the shear layer. The largest percent
and field values in physical space [46]. difference in temperature, weighted by the total change in
The computational domain, illustrated in Fig. 4, is discretized temperature (T ad − T u ), among the three solutoins was equal to 4% at
into approximately 800,000, 1,200,000, and 2,000,000 cells for the 1d location and 2.3% at the 15d location. Because the differences
comparison. To illustrate the block distribution developed for this between the three solutions are smaller than the experimental
Article in Advance / VELEZ, MARTIN, AND VASU 9
Downloaded by EMBRY-RIDDLE AERO UNIV. on February 2, 2017 | http://arc.aiaa.org | DOI: 10.2514/1.B35741

Fig. 5 Two-dimensional slice of the computational domain along the plane of symmetry.

Fig. 6 Three-dimensional image of the computational domain near the jet nozzle.

uncertainty in the temperature field (5%), the solution is deemed


independent of the three mesh computational domains. Moreover,
this same conclusion was found in a previous numerical study [17],
and for this reason, the 800,000 cell mesh is used in this study. Each of
the three mesh cases is prescribed different time steps of the order of a
microsecond to determine the solution dependence on the temporal
resolution. The 800,000 mesh uses a dynamic time step such that the
Courant number would not exceed unity. For the case of the
1,200,000 and 2,000,000 cell mesh, the maximum Courant number
allowed is 0.5 and 0.8, respectively. The same model constants (Cc1
and Cc2 ) are used for each case, which suggests that the two model
constants are effectively independent of the grid resolution. Figure 8
compares the temperature at 1d and 15d for the three meshes.
The mesh consists mostly of hexahedral (hex) cells, with the
addition of a few tetrahedral (tet) cells introduced at the interface
between course and fine cell levels. Inserting tet cells above the height
of the boundary layer allows for twice the cell refinement of the
boundary layer. In conjunction with an expansion ratio of 1.2 near the
wall, the mesh is able to sustain Y values of O10 along the nozzle
interior wall. The walls of the combustion chamber have a Y  ≈300
and employ wall functions for the turbulent kinetic energy and
Fig. 7 Comparisons of the mean temperature fields from the 800,000 dissipation rate profiles within the boundary layer.
cell mesh (top), 1,200,000 cell mesh (middle), and 2,000,000 cell mesh Three external boundaries are considered in this simulation: the
(bottom). inlet, which is located 4d upstream; the outlet, which is 38d
10 Article in Advance / VELEZ, MARTIN, AND VASU

Fig. 9 Volume-weighted average RPV vs time.

under the same flow conditions, making it attractive for design


purposes where low run times are desired with the inclusion of
detailed chemistry in the combustion model.

IV. Results
Fig. 8 Comparisons of the mean temperature at locations 1d and 15d
away from the nozzle.
To validate the T-PCMC implementation in OpenFOAM, both
CFD and measured data sets are compared against one another. The
Downloaded by EMBRY-RIDDLE AERO UNIV. on February 2, 2017 | http://arc.aiaa.org | DOI: 10.2514/1.B35741

measured PIV results provided radial line plots of the mean and rms
downstream of the nozzle exit; and the remaining walls of the velocity fields and Raman laser-measured contour plots of the scalar
enclosure, which include the pipe walls inside the nozzle. The quantities along the center plane of the combustion chamber, as in
simulation occurs at atmospheric pressure and is prescribed this Fig. 4. Before comparing experimental and predicted results,
pressure at the outlet. All other scalar and vector quantities demonstration of the unsteady nature of the simulation is illustrated in
are imposed a zero-gradient condition at the outlet boundary. The Fig. 10. The reaction of the URANS algorithm to the intial RANS
pressure on the remaining walls and inlet is allowed to change and is steady state solution causes perturbations at the nozzle exit (at time
imposed the pressure value from the interior cells adjacent to the 0.608 ms), which are dissipated down the flame causing the tip of the
boundaries (i.e., a zero gradient). The velocity profile at the inlet is flame to shed off (at time 1.216 ms). At this point, the flame is
made to be parabolic, and it was confirmed that the velocity profile shortened and allowed to grow (at time 1.825 ms) until a relative
matched the measurements at the nozzle exit and maintained the steady state is once again obtained (at time 3.041 ms), after which
same mass flow rate. Random turbulent fluctuations are imposed on point the only transient behaviors that exist are small oscillations near
the inlet velocity profile at 10% about its mean velocity of 90 m∕s. the flame tip, which is most evident in the last frame of Fig. 10.
This turbulence intensity has been reported in previous numerical Analysis of the velocity rms verifies that the largest and sustained
turbulent fluctuations occur in the regions directly above and below
studies to provide excellent correspondence between measured and
the flame tip. Although the flame tip fluctuates moderately, the
simulated velocity rms profiles at the nozzle exit (4d downstream of
recirculation length remains relatively constant. This evidently stable
the inlet). The inlet mixture equivalence ratio, Reynolds number,
behavior, of the flame and recirculation zone, is a consequence of
density, and mass flow rate are equal to 0.7; 1.8×103 , 0.5793 kg∕m3 ,
the time averaging used in the k − ω SST turbulence model. By time
and 4.2 g∕s, respectively.
The k − ω shear-stress transport (SST) RANS turbulence model
was used to model the Reynolds stress tensor in the momentum
equation. The k − ω SST model is based on the assumption that the
principal shear stress is proportional to the turbulent kinetic energy k
in the definition of the eddy viscosity. In addition, the SST version has
the ability to account for the transport of shear stress in adverse
pressure gradient boundary layers. The model was prescribed
k  0.304 m2 ∕s2 and ω  5511∕s at the inlet boundary based on the
relevant length and velocity fluctuation scales. The length of the
nozzle was sufficient enough (see Fig. 4) to allow the inlet values to
adjust to the experimental profiles by the end of the nozzle. N, RPV,
and its variance were set to zero at the inlet, and a zero gradient
condition was set for the remaining walls and outlet. The inlet
temperature is prescribed a fixed value of 573 K; the remaining
boundaries impose a zero gradient with temperature.
The simulation flowfield is initialized by a converged RANS solution
and then runs “unsteady” for five flow passages through the entire
domain. A single flow passages lasts 0.7 ms. A quasi-steady state is
again reached after 3 ms of simulation time, at which point there still
exist turbulent fluctuations, but the flame length remains approximately
constant. To quantify the steadiness of the flowfield, the RPV from every
cell in the domain is included in a weighted average, where the RPV is
weighted by its respective cell volume. The results for the volume-
weighted RPV are plotted against time for each mesh size in Fig. 9.
When the weighted average of the RPV reaches a constant value, the
flame can be considered to have reached steady burning conditions even
in the presence of sustained turbulent fluctuations both in the jet and
recirculation zones. This calculation is performed at every time step, and
the transient results are plotted in Fig. 4.
Statistics are then taken, for the mean and rms, for an additional
35 ms. The simulation is run in parallel on 16 Xeon E5650 hex-core
processors for a total of 22 computational hours. The T-PCMC run Fig. 10 Sequential contours of the progress variance c during
time is approximately 10% higher than nonreactive simulations initialization of the flowfield.
Article in Advance / VELEZ, MARTIN, AND VASU 11

averaging velocity fluctuations, as in URANS, a dissipative solution field history. The rms of the axial and vertical velocities resulted in
is obtained, void of any chaotic or incoherent structures. It is expected overestimation of the measured rms results by more than double. This
that large-eddy simulations (LES) can predict sustained turbulent was an expected result because there is no loss in total enthalpy and
fluctuations throughout the jet and recirculation region better than the kinetic energy in the flow. When assuming axial and vertical velocity
URANS implementation performed in this work. fluctuations are equivalent, the velocity rms can be approximated via
To access the accuracy of the T-PCMC model, in obtaining k  3∕2U2rms in a URANS formulation. Thus, the turbulent kinetic
predictions of the mean and rms velocities, comparisons are made to energy k is used to determine the velocity rms for comparison with
the exact velocity PIV measurements from the DLR data set [32]. measured PIV data. Figure 12 compares this modeled velocity rms
Figure 11 compares radially the predicted velocity components to the (in meters per second) to the measured axial velocity rms at four
PIV measured data at four planes 1, 5, 10, and 15 diameters planes 1, 5, 10, and 15 diameters downstream of the nozzle exit as
downstream of the nozzle exit, as illustrated in Fig. 4. The vertical in Fig. 11.
axis in Fig. 11 is the velocity (in meters per second), and the The rms profiles show reasonable comparison with increasing
horizontal axis is the radial distance from the bottom wall scaled by error downstream of the nozzle. The profile shape and distribution is
the nozzle diameter y∕D. well predicted, although an underprediction in the rms values is
The top of Fig. 11 shows the x-component velocity. The overall apparent with a maximum deviation from the measured results of
match is apparent, but the T-PCMC model predicts a slightly faster jet 36%. This underprediction can be attributed to the fact that URANS
that also spreads slightly farther downstream. The predicted jet modeling produces a dissipative solution, specifically in wake
maximum velocity is higher than the measured data by approximately regions, due to its time averaging of the turbulent fluctuations. It is
Downloaded by EMBRY-RIDDLE AERO UNIV. on February 2, 2017 | http://arc.aiaa.org | DOI: 10.2514/1.B35741

15%, mostly due to a lack of any heat loss in the T-PCMC model. expected that LES simulations will reduce this error in rms by
Because there is no loss in heat and total enthalpy, the kinetic energy of preserving the small-scale turbulent fluctuations, which contribute
the flow will also be overpredicted under the assumption of an largely to the total rms.
adiabatic system. Even with the apparent jet velocity overprediction, Figure 13 compares the Raman laser-measured and T-PCMC-
the recirculation zone is well predicted, as is evident by correct predicted temperature at the jet centerline, relative to the 4d side of
prediction of the jet recirculation length (Ux < 0) at axial locations 10d the base in Fig. 4. Results from the Raman laser measurements have
and 15d at y∕D  5 (top wall). The right-hand side of Fig. 11 coarser resolution than PIV measurements, and consequently Raman
compares the vertical velocities. Although there are some differences, laser measurements of species moles fractions are illustrated by
the main features of the recirculation zone are well recovered, and an contour plots. Measurements are taken 0.5d − 15d downstream of
adequate match between the two data sets is found. At the 1d location, the jet exit. The vertical axis is the 5d side of the base shown in Fig. 4
the sharp gradient in velocity, from the shear layer of the jet, was with 0.5d clipped from the top and bottom to match the range of the
difficult to accurately resolve and showed the largest variations among recorded data.
tested turbulence models. Unlike the axial velocity comparisons, the Comparison of the measured and predicted temperature field
match between the predicted and measured vertical velocity profiles clearly shows the effect of the adiabatic assumption. At c  1,
does not change with increasing downstream direction. outside of the flame, the temperature is uniform, whereas the
To compare the performance of the T-PCMC model in predicting measured results show heat loss outside of the flame structure and in
the unsteady features of the flowfield, the rms is taken of the velocity the recirculation region. The majority of heat loss is expected to occur

Fig. 11 Axial (left) and vertical (right) velocity comparisons between predicted (URANS) and measured PIV (data) results.
12 Article in Advance / VELEZ, MARTIN, AND VASU

Fig. 14 Dry CH4 mole fraction (percent) contours from CFD (top)
compared to Raman laser-measured data (bottom).
Downloaded by EMBRY-RIDDLE AERO UNIV. on February 2, 2017 | http://arc.aiaa.org | DOI: 10.2514/1.B35741

Fig. 12 Velocity rms computed from k and compared to the measured


PIV (data) axial velocity rms. Fig. 15 Dry CO2 mole fraction (percent) contours from CFD (top)
compared to Raman laser-measured data (bottom).

Overprediction of the mean CH4 and CO2 mole fractions is expected


to be less severe in the vicinity of the nozzle exit where heat loss is less
intense. This is in accordance with the improved prediction of the first
contour lengths for both CH4 and CO2 . As contour levels approach
their fully burned values, deviation from the measured results is
found. This encourages the notion that discrepancies in the results are
due to the adiabatic assumption and that a nonadiabatic formulation
would result in better predictions of the flame length and species mass
fractions.
Using identical inlet conditions as in [32], the inlet dry CH4 mole
percentage is 6.959%, and the maximum measured xCH4 is 6.548%.
This is an error of approximately 6% and slightly larger than the 5%
estimated error stated in [32]. The predicted data also show small
amounts of H2 O, CO2 , and H2 in the inlet stream that are within the
Fig. 13 Temperature contours (in Kelvins) from CFD (top) compared to 5% error range of the measurements.
Raman laser-measured data (bottom). To compare the predicted scalar fluctuations with Raman laser
experimental results, rms contours are compared for the predicted
and measured dry CH4 mole fraction.
downstream of the step in the recirculation region where mixing As seen in Fig. 16, CH4 fluctuations are well predicted in
is fast (high convection) and downstream of the step where magnitude but not in the distribution. The shorter adiabatic flame
temperatures are largest (high radiation). Near the jet exit, predicted by the T-PCMC model burns all the available CH4 within
where heat-loss effects are lowest, a better match of the temperature the flame, whereas the measured Raman laser data shows the largest
contours is found. The peak measured temperature is about 200 K CH4 fluctuations near the exit of the sampled region (x  15d), i.e.,
lower than the adiabatic flame temperature and is due to heat loss pockets of unburden CH4 appear intermittently downstream of the
from the hot burned gases to the enclosure walls. flame tip. As seen in the time-averaged results, the rms predictions
To compare the predictions of species mole fractions, contours of near the nozzle are best predicted where heat-loss effects are
the dry (xH2 O removed) CH4 and CO2 mole percentages are plotted relatively small.
with their measured values in Figs. 14 and 15, respectively. To compare the predictive ability of the T-PCMC model in
It can be seen for both the CH4 and CO2 predictions that the predicting turbulent fluctuations, which are of the same time scale as
adiabatic assumption has significantly shortened the species mole the bulk flow, instantaneous velocity streamlines from experimental
fraction distributions in comparison to the measured flame. and simulated results are compared in Fig. 17. In Fig. 17, the velocity
Article in Advance / VELEZ, MARTIN, AND VASU 13

The k − ω shear-stress transport turbulence model was used along


with adiabatic walls. The table lookup method was used with the
tabulted premixed CMC (T-PCMC) combustion model using three
controlling variables: the reaction progress variable (RPV), its
variance, and the scalar dissipation. This combustion model uses
detailed kinetics information from the GRI3.0 kinetic mechanism to
give a detailed representation of the chemical state of the flame. The
detailed kinetics information is stored in a table and accessed by three
controlling variables. By recasting the 53 species in the kinetics
model to the three control variables, the computational effort is
significantly reduced. The effect of large-scale turbulence on the
reaction rates is accounted for by using a presumed shape probability
density function, similar to other table lookup models. The effect of
small-scale turbulence is directly accounted for by the scalar
Fig. 16 Dry rms CH4 mole fraction (percent) contours from CFD (top)
dissipation in the T-PCMC governing equation. Additionally, a
compared to Raman laser-measured data (bottom). chemical and thermodynamic representation of the mean reaction
rate term is obtained using a CMC first-order closure for the mean
chemical reaction rate. The URANS version of the CMC model
Downloaded by EMBRY-RIDDLE AERO UNIV. on February 2, 2017 | http://arc.aiaa.org | DOI: 10.2514/1.B35741

provided good prediction of the velocity field and was able to


predict the main features the temperature and major species. The
main limitation of the model was the assumption of adiabatic walls,
whereas the experiment had significant heat loss. The next step is to
adapt the premixed CMC model to large-eddy simulations and add a
heat-loss model, which will add a fourth dimension to the lookup
table (total enthalpy). The results of this work show the promise
of the premixed CMC model in URANS to predict the velocity and
scalars of an idealized lean premixed gas turbine combustor with
detailed chemistry and low run times, which warrants further
development.

Acknowledgments
Carlos Velez would like to thank the McKnight Fellowship by the
Florida Education Fund. Research at the University of Central
Florida was supported by financial assistance from the Department of
Fig. 17 Instantaneous velocity streamlines of the measured (top) and
Mechanical and Aerospace Engineering.
predicted (bottom) velocity field.

magnitudes range from 122 m∕s (shown in red) to 0 m∕s (shown in References
dark blue). It is evident from the comparison that the URANS [1] “National Ambient Air Quality Standard,” U.S. Environmental
solution is able to capture the small-scale eddies, which are generated Protection Agency, Washington, D.C., 2008, https://www.epa.gov/
predominantly along the shear layer of the jet, while preserving the criteria-air-pollutants [retrieved 01 Aug. 2015].
large-scale eddies driving the recirculation region above the jet. The [2] Lieuwen, T., Torres, H., Johnson, C., and Zinn, B. T., “A Mechanism of
simulated streamlines follow the experimental trend of moving Combustion Instability in Lean Premixed Gas Turbine Combustors,”
Journal of Engineering for Gas Turbines and Power, Vol. 123, No. 1,
downstream from the inlet and then back upstream toward the nozzle,
2001, pp. 182–189; also American Soc. of Mechanical Engineers Paper
with vorticies being shed off from the tip of the flame and then either 99-GT-3, New York, June 1999.
recirculated or dissipated downstream. Because this turbulent doi:10.1115/1.1339002
shedding time scale is comparable to the time it takes the fluid to enter [3] Steele, R., “NOx and N2 O Formation in Lean Premixed Jet Stirred
and exit the domain, it is expected that a URANS model will preserve Reactors Operated from 1 to 7 atm,” Ph.D. Dissertation, Univ. of
the history of the recirculation zone. Correct prediction of these shed Washington, Seattle, WA, 1995.
vorticies is essential in the accurate prediction of emissions. [4] Tanahashi, M., Nada, Y., Ito, Y., and Miauchi, T., “Local Flame
Although this detail is lost when comparing averaged quantities, it is Structure in the Well-Stirred Reactor Regime,” Proceedings of the
apparent from Fig. 17 that the URANS simulation detects the same Combustion Institute, Vol. 29, No. 2, 2002, pp. 2041–2049.
doi:10.1016/S1540-748980249-8
turbulent phenomena as those measured by PIV imaging.
[5] Chakraboty, N., and Swaminathan, N., “Influence of the Damkohler
Though the comparison has demonstrated that heat-loss effects are Number on Turbulence-Scalar Interaction in Premixed Flames,” Physics
strong in the measured flame and that the adiabatic nature of this of Fluids, Vol. 19, No. 4, April 2007, Paper 045103.
formulation is not well suited for this flame, accurate prediction of doi:10.1063/1.2714076
the velocity mean and rms quantities is obtained. Additionally, the [6] Swaminathan, N., and Bray, K. N. C., “Effect of Dilatation on Scalar
coupling of unsteady RANS, with detailed chemistry and the Dissipation in Turbulent Premixed Flames,” Combustion and Flame,
inclusion of the small-scale mixing effects, has been demonstrated Vol. 143, No. 4, 2005, pp. 549–565.
and shows encouraging results, which warrant further development. doi:10.1016/j.combustame.2005.08.020
[7] Pitsch, H., “Improved Pollutant Predictions in Large-Eddy Simulations
of Turbulent Non-Premixed Combustion by Considering Scalar
Dissipation Rate Fluctuations,” Proceedings of the Combustion
V. Conclusions Institute, Vol. 29, No. 2, 2002, pp. 1971–1978.
The tabulated premixed conditional moment closure (T-PCMC) doi:10.1016/S1540-748980240-1
[8] Wang, P., and Bai, X. S., “Large Eddy Simulation of Turbulent Premixed
turbulent combustion model was coupled to the open-source Flames Using Level-Set G-Equation,” Proceedings of the Combustion
computational fluid dynamics program OpenFOAM and used to Institute, Vol. 30, No. 1, 2005, pp. 583–591.
model an enclosed, reacting jet from DLR experimental data [32]. doi:10.1016/j.proci.2004.08.218
This work expands on earlier work by using unsteady Reynolds- [9] Benim, A. C., and Syed, K. J., “Laminar Flamelet Modelling of
averaged Navier–Stokes (URANS) instead of steady RANS. Turbulent Premixed Combustion,” Applied Mathematical Modelling,
14 Article in Advance / VELEZ, MARTIN, AND VASU

Vol. 22, No. 1, 1998, pp. 113–136. [26] Amzin, S., “Conditional Moment Closure Method for Turbulent
doi:10.1016/S0307-904X00012-2 Premixed Flames,” Ph.D. Dissertation, Cambridge Univ., Cambridge,
[10] Donnini, A., Martin, S. M., Bastiaans, R. J. M., van Oijen, J. A., and de England, U.K., 2012.
Goey, L. P. H., “Numerical Simulations of a Premixed Turbulent [27] Thornber, B., Bilger, R. W., Masri, A. R., and Hawkes, E. R., “An
Confined Jet Flame Using the Flamelet Generated Manifold Approach Algorithm for LES of Premixed Compressible Flows Using the
with Heat Loss Inclusion,” Proceedings of the ASME Turbo Expo Conditional Moment Closure Model,” Journal of Computational
2013: Turbine Technical Conference and Exposition, American Soc. Physics, Vol. 230, No. 20, 2011, pp. 7687–7705.
of Mechanical Engineers Paper GT2013-94363, New York, doi:10.1016/j.jcp.2011.06.024
June 2013. [28] Martin, S. M., Kramlich, J. C., Kosaly, G., and Riley, J. J., “The
doi:10.1115/GT2013-94363 Conditional Moment Closure Method for Premixed Turbulent
[11] Veynante, D., and Vervisch, L., “Turbulent Combustion Modeling,” Combustion,” 2001 Eastern States Section Fall Technical Meeting of
Progress in Energy and Combustion Science, Vol. 28, No. 3, 2002, the Combustion Institute, Paper B-3-53, 2001, pp. 245–248.
pp. 193–266. [29] Martin, S. M., Kramlich, J. C., Kosaly, G., and Riley, J. J., “The
doi:10.1016/S0360-12850017-X Premixed Conditional Moment Closure Method Applied to
[12] Magnussen, B. F., “On the Structure of Turbulence and a Generalized Idealized Lean Premixed Gas Turbine Combustors,” Journal of
Eddy Dissipation Concept for Chemical Reaction in Turbulent Flow,” Engineering for Gas Turbines and Power, Vol. 125, No. 4, 2003,
19th AIAA Aerospace Sciences Meeting, AIAA Paper 1981-0042, pp. 895–900.
Jan. 1981. doi:10.1115/1.1587740
doi:10.2514/6.1981-42 [30] Martin, S. M., Jemcov, A., and de Ruijter, B., “Modeling an
[13] Spalding, D. B., “Mixing and Chemical Reaction in Steady Confined Enclosed, Turbulent Reacting Methane Jet with the Premixed
Turbulent Flames,” Proceedings of the Combustion Institute, Vol. 13, Conditional Moment Closure Method,” Proceedings of ASME
Downloaded by EMBRY-RIDDLE AERO UNIV. on February 2, 2017 | http://arc.aiaa.org | DOI: 10.2514/1.B35741

No. 1, 1970, pp. 649–657. Turbo Expo 2013: Turbine Technical Conference and Exposition,
doi:10.1016/S0082-078480067-X American Soc. of Mechanical Engineers Paper GT2013-95092,
[14] Bray, K. N. C., Libby, P. A., and Moss, J. B., “Unified Modeling New York, June 2013.
Approach for Premixed Turbulent Combustion—Part 1: General doi:10.1115/GT2013-95092
Formulation,” Combustion and Flame, Vol. 61, No. 1, 1985, [31] Martin, S. M., “The Conditional Moment Closure Method for Modeling
pp. 87–102. Lean Premixed Turbulent Combustion,” Ph.D. Dissertation, Univ. of
doi:10.1016/0010-218090075-6 Washington, Seattle, WA, 2003.
[15] Pitsch, H., “A Consistent Level Set Formulation for Large-Eddy [32] Lammel, O., Stohr, M., Kutne, P., Dem, C., Meier, W., and Aigner, M.,
Simulation of Premixed Turbulent Combustion,” Combustion and “Experimental Analysis of Confined Jet Flames by Laser Measurement
Flame, Vol. 143, No. 4, 2005, pp. 587–598. Techniques,” Journal of Engineering for Gas Turbines and Power,
doi:10.1016/j.combustflame.2005.08.031 Vol. 134, No. 4, 2012, Paper 041506.
[16] Dusing, M., Kempf, A., Flemming, F., Sadiki, A., and Janicka, doi:10.1115/1.4004733
J., “Combustion LES for Premixed and Diffusion Flames,” [33] Di Domenico, M., Beck, C. H., Lammel, O., Krebs, W., and Noll, B. E.,
Progress in Computational Fluid Dynamics, Vol. 5, No. 7, 2005, “Experimental and Numerical Investigation of Turbulent, Lean, High-
pp. 363–374. Strained, Confined, Jet Flames,” 49th AIAA Aerospace Sciences
doi:10.1504/PCFD.2005.007423 Meeting, AIAA Paper 2011-0238, Jan. 2011.
[17] Proch, F., and Kempf, A. M., “Modeling Heat Loss Effects in the Large doi:10.2514/6.2011-238
Eddy Simulation of a Model Gas Turbine Combustor with Premixed [34] Knudsen, E., Richardson, E. S., Doran, E. M., Pitsch, H., and Chen, J.
Flamelet Generated Manifolds,” Proceedings of the Combustion H., “Modeling Scalar Dissipation and Scalar Variance in Large Eddy
Institute, Vol. 35, No. 3, 2015, pp. 3337–3345. Simulation: Algebraic and Transport Equation Closures,” Physics of
doi:10.1016/j.proci.2014.07.036 Fluids, Vol. 24, No. 5, 2012, Paper 055103.
[18] Stankovic, I., Triantafyllidis, A., Mastorakos, E., Lacor, C., and Merci, doi:10.1063/1.4711369
B., “Simulation of Hydrogen Auto-Ignition in a Turbulent Co-Flow of [35] Menter, F. R., “Improved Two-Equation k-Omega Turbulence Models
Heated Air with LES and CMC Approach,” Flow Turbulence and for Aerodynamic Flows,” NASA TM-103975, Oct. 1992.
Combustion, Vol. 86, No. 3, 2011, pp. 689–710. [36] Velez, C. A., Martin, S. M., Jemcov, A., and Vasu, S., “LES Simulation
doi:10.1007/s10494-010-9293-0 of an Enclosed Turbulent Reacting Methane Jet with the Tabulated
[19] Bilger, R. W., “Conditional Moment Closure for Turbulent Reacting Premixed CMC Method,” Journal of Engineering for Gas Turbines
Flow,” Physics of Fluids A: Fluid Dynamics, Vol. 5, No. 2, 1993, and Power, Vol. 138, No. 10, 2016, Paper 101501; also American Soc.
pp. 436–444. of Mechanical Engineers Paper GT2015-43788, New York, June 2015.
doi:10.1063/1.858867 doi:10.1115/1.4032846
[20] Smith, G. P., , Golden, D. M., Frenklach, M., Moriarty, N. W., Eiteneer, [37] Amzin, S., Swaminathan, N., Rogerson, J. W., and Kent, J. H.,
B., Goldenberg, M., Bowman, C. T., Hanson, R. K., Song, S., and “Conditional Moment Closure for Turbulent Premixed Flames,”
Gardiner, W. C., et al., “GRI 3.0 Kinetic Mechanism,” http://www.me. Combustion Science and Technology, Vol. 184, Nos. 10–11, 2012,
berkeley.edu/gri_mech/ [retrieved April 2006]. pp. 1743–1767.
[21] Bolla, M., Wright, Y., Boulouchos, K., Borghesi, G., and Mastorakos, doi:10.1080/00102202.2012.690629
E., “Soot Formation Modeling of n-Heptane Sprays Under Diesel [38] Klimenko, A., and Bilger, R. W., “Conditional Moment Closure for
Engine Conditions Using the Conditional Moment Closure Approach,” Turbulent Combustion,” Progress in Energy and Combustion Science,
Combustion Science and Technology, Vol. 185, No. 5, 2013, Vol. 25, No. 6, 1999, pp. 595–687.
pp. 766–793. doi:10.1016/S0360-128500006-4
doi:10.1080/00102202.2012.752362 [39] Fiorina, B., Gicquel, O., Vervisch, L., Carpentier, S., and
[22] Garmory, A., and Mastorakos, E., “Capturing Localised Extinction in Darabiha, N., “Approximating the Chemical Structure of Partially
Sandia Flame f with LES–CMC,” Proceedings of the Combustion Premixed and Diffusion Counterflow Flames Using FPI Flamelet
Institute, Vol. 33, No. 1, 2011, pp. 1673–1680. Tabulation,” Combustion and Flame, Vol. 140, No. 3, 2005,
doi:10.1016/j.proci.2010.06.065 pp. 147–160.
[23] Bottone, F., Kronenburg, A., Gosman, D., and Marquis, A., “The doi:10.1016/j.combustflame.2014.07.008
Numerical Simulation of Diesel Spray Combustion with LES- [40] Devaud, C. B., Bilger, R. W., and Liu, T., “A New Method of Modeling
CMC,” Flow, Turbulence and Combustion, Vol. 89, No. 4, 2012, the Conditional Scalar Dissipation Rate,” Physics of Fluids, Vol. 16,
pp. 651–673. No. 6, 2004.
doi:10.1007/s10494-012-9415-y doi:10.1063/1.1699108
[24] Navarro-Martinez, S., and Kronenburg, A., “LES–CMC Simulations of [41] Klimenko, A., “On the Relation Between the Conditional Moment
a Lifted Methane Flame,” Proceedings of the Combustion Institute, Closure and Unsteady Flamelets,” Combustion Theory and Modelling,
Vol. 32, No. 1, 2009, pp. 1509–1516. Vol. 5, No. 3, pp. 275–294.
doi:10.1016/j.proci.2008.06.178 doi:10.1088/1364-7830/5/3/302
[25] Ma, M., and Devaud, C. B., “A Conditional Moment Closure (CMC) [42] Cash, J. R., and Mazzia, F., “A New Mesh Selection Algorithm, Based
Formulation Including Differential Diffusion Applied to a Non- on Conditioning, for Two-Point Boundary Value Codes,” Journal of
Premixed Hydrogen–Air Flame,” Combustion and Flame, Vol. 162, Computational and Applied Mathematics, Vol. 184, No. 2, 2005,
No. 1, 2014, pp. 144–158. pp. 362–381.
doi:10.1016/j.combustflame.2014.07.008 doi:10.1016/j.cam.2005.01.016
Article in Advance / VELEZ, MARTIN, AND VASU 15

[43] Weller, H. G., Tabor, G., Jasak, H., and Fureby, C., “A Tensorial [45] Issa, R., “Solution of the Implicitly Discretised Fluid Flow Equations by
Approach to Computational Continuum Mechanics Using Object Operator-Splitting,” Journal of Computational Physics, Vol. 62, No. 1,
Oriented Techniques,” Computers in Physics, Vol. 12, No. 6, 1998, 1986, pp. 40–65.
pp. 620–631. doi:10.1016/0021-999190099-9
doi:10.1063/1.168744 [46] Ferziger, J. H., and Peric, M., Computational Methods for Fluid
[44] Jasak, H., Jemcov, A., and Tukovic, Z., “OpenFOAM: A C++ Dynamics, Springer, Berlin, 2002, pp. 157–216, Chap. 7.
Library for Complex Physics Simulation,” The International Workshop
on Coupled Methods in Numerical Dynamics, IUC, Dubrovnik, J. M. Powers
Sept. 2007, pp. 1–20. Associate Editor
Downloaded by EMBRY-RIDDLE AERO UNIV. on February 2, 2017 | http://arc.aiaa.org | DOI: 10.2514/1.B35741

View publication stats

S-ar putea să vă placă și