Professional Documents
Culture Documents
net/publication/270591120
CITATIONS READS
7 75
4 authors, including:
Chaitanya Ghodke
Convergent Science Inc., Madison, WI, USA
18 PUBLICATIONS 98 CITATIONS
SEE PROFILE
Some of the authors of this publication are also working on these related projects:
All content following this page was uploaded by Chaitanya Ghodke on 26 May 2015.
I. Introduction
A significant challenge in the design of an operational SCRAMJET propulsion system is to achieve
efficient fuel-air mixing and combustion within a limited combustor length. The parameters responsible for
mixing are directly affected by chemical and thermal changes taking place in the combustor and there exists
close coupling between fluid dynamics and chemical kinetics in the combustor. In addition to efficient fuel-air
mixing, reliable ignition and flame stabilization over a range of operating Mach numbers are necessary to meet
the operational goals. In particular, for hydrocarbon fueled combustors, flame stabilization is especially of
concern due to the lower propagation speed and heat release of such systems.1 Therefore, various flame-holder
concepts that can provide both efficient flame holding and flame stability are being explored. Designs such
as recessed cavity combustor is one of the promising design that appears to be able to achieve flame-holding
by the generation of subsonic recirculation region that ensures sufficient residence time.2, 3 Additional issues
arise if liquid fuels (e.g., JP-8, JP-10 or kerosene are to be employed4 ) but in this study we focus primarily
on gaseous hydrocarbon fuel such as methane mixed with a small amount of hydrogen.
Past experimental investigations of supersonic cavity flameholders2, 5, 6 have shown that the fuel composi-
tion has a significant impact on flame stability and other parameters such as the injection location and cavity
geometry have also been shown to be critical.7 Methane-fueled cavities usually exhibit a smaller stability
domain than hydrogen- or ethylene-fueled cavities, since methane has a slower chemistry.6 For the purpose
of studying flame stability, estimation of time scales has been used to estimate the Damkohler number8
∗ AIAA Student Member, Graduate Student
† AIAA Student Member, Undergraduate Student
‡ AIAA Student Member, Graduate Student
§ AIAA Associate Fellow, Professor
1 of 16
American
Copyright © 2011 by Chaitanya Ghodke. Published by the American Institute
Institute of Aeronautics
of Aeronautics and Astronautics
and Astronautics, Inc., with permission.
and an empirical model for the blowout limit was demonstrated earlier.9 However, there is insufficient data
over the range of operating conditions and geometrical constraints at this time to determine a generalize
scaling rule for blowout conditions. The studies reported here and elsewhere10 have been carried out with
a particular emphasis on finding blowout limits for a combination of fuel mixtures and for conditions that
have not been studied in the past.
Numerical studies of cavity-stabilized combustion are also sparse. Most of the earlier reported studies
have focused on RANS type modeling, for example, Rodriguez et al.11 simulated the SCHOLAR experiment
using VULCAN CFD code. This same code was used also to study cavity-stabilized ethylene-air flames with
various reduced chemical kinetics mechanisms.12 Another numerical study13 focused on varying the cavity
aft wall angle, the offset ratio and the cavity length for hydrogen injection. The cavity type combustor
was found to enhance mixing and combustion efficiency but at the cost of total pressure loss. Recently,
LES methodology has been used to investigate reacting flow and flame ignition/blowout14 in a a cavity
stabilized combustor and some preliminary results for blowout were reported. This study combined the LES
methodology with an advanced subgrid mixing-combustion model based on the Linear Eddy Mixing (LEM)
sub-grid model within the LES framework (LEM-LES) and this approach was also used recently15 to study
cavity and strut stabilized combustors. In the latter approach a more cost-effective subgrid closure for the
reaction rate kinetics was developed using a turbulent artificial neural network (TANN) approach and was
demonstrated that it can correctly capture the flame dynamics in these type of combustors. The TANN
approach combines the ability of the LEM to capture reaction-diffusion-mixing interactions with finite-rate
kinetics in a look up approach that is both cost effective and accurate in the regimes of interest. Earlier
studies16, 17 have demonstrated the ability of this approach in more canonical flows.
In this paper, we report on experimental and numerical efforts to study cavity stabilized combustion
in a Mach 2.5 cross-stream. Experiments are conducted for several methane-hydrogen fuel blends, while
numerical studies are carried out for only few chosen experimental cases. To ensure proper matching of the
experiments, the entire test facility is modelled in the LES. Comparison with data where available is carried
out and the LES results are used to explain the flow and flame dynamics in the experimental cases for both
stable and blowout conditions.
2 of 16
Figure 2. Test section cut view, L/D = 3.84 shown (dimensions are in mm).
3 of 16
(a) Ignition - Blowout procedure in the mixture space. (b) Stable domain and blowout region in the mixture space:
Mixture trajectories (grey lines) and blowout events (black
symbols).
Figure 3. Construction of the experimental blowout region : (a) depicts a single orbit in the mixture space whereas
(b) is the collection of all orbits (burn data). The end points form the blowout region.
An efficient algorithm is used to post process the data in which the temperature measured in the cavity
is used to detect the ignition and blowout. This process is described in a previous paper.10
4 of 16
past9, 18 and suggests that the flow and the combustion mechanisms depend on the fueling rate. Henceforth,
the physics involved in rich and lean blowout differ and the results reported in this study are specific to the
combustion regime.
(a) Stable combustion (mf = 0.48 g/s). (b) Lean combustion near blowout (mf = 0.31 g/s). The
flame is located near the injector region.
The blowout data points are located on a map with respect to the fuel mixture in Fig. 5(a) for both
Case 1 and 2. The blowout data is represented as a cloud of points which separates the stable combustion
from the non-burning region (close to the ṁCH4 axis). Even in small quantities, addition of hydrogen in
the fuel blend mixture is advantageous to increase the stability of the combustor. The air-methane flame is
stabilized with small hydrogen addition, typically as low as 2 % of the fuel mass flowrate.
As the crossflow static pressure is lowered, the combustion mechanism is maintained in less favorable
conditions: the data dispersion increases and the blowouts become less repeatable. Fuel-lean (ṁf < 0.4 g/s)
and fuel-rich regions (ṁf > 0.7 g/s) are more affected by this phenomenon. Eventually, no combustion is
obtained below 40 kPa of cavity static pressure within the tested conditions, such that the static pressure
draws a lower bound below which no combustion is sustainable. Higher hydrogen concentrations show
that the lack of pressure can be compensated by higher flame speed and mixture diffusivity in the sense
of non-premixed time scale of Driscoll et al.8 Further observations shows that pressure may also shift the
stochiometry of the mixture in the combustor toward the lean limit since the air entrained in the supersonic
shear layer is more dense as predicted by the work of Dimokatis.19, 20
Impact of the supersonic crossflow temperature on the combustion stability is also investigated (Case 2 of
Table 1). The blowout fuel maps conserves the same trends however, the blowout limit trend is shifted down
to lower hydrogen flow rates (as low as 200 parts of methane for 1 part of hydrogen) leading in an increase
of the stable domain. In the preheated case, the amount of energy spent in raising the incoming mixture
to its flammability limits is reduced. The extension of the stable domain may also be due to better flame
anchoring provided by the higher unburnt gas temperature in the lift off region described by Rasmussen and
al.3 In addition for Case 2, the wall heat losses are reduced since the structure is heated and the temperature
gradient may be therefore reduced.
The symptoms of having a low crossflow pressure and temperature are the poor repeatability (high
dispersion of the blowout data points) and a hydrogen increase in the fuel mixture. These results shows that
adjustable fueling composition can play a role in maintaining combustion and extending the stable domain
by offering more flexible heat content.
Further studies on cavity blowout lead to the establishment of an empirical model by Rasmussen et al.9
Lean blowout limits with floor injection follows Φ = 0.0028Da−0.8 at best correlation. The Damkohler number
is given by the ratio of characteristic time scale between the flow and the chemical reaction.9
5 of 16
The overall mixture properties such as stoichiometric flame speed and thermal diffusivity are calculated
using Cantera. The equivalence ratio expressed from the characteristic air mass flowrate.3
(a) Fuel mixture composition at blowout: the stable do- (b) Equivalence ratio as a function of Damkohler number
main is located on the right side of the cloud of points. at lean blowout: comparison with blowout data from other
facilities,9 the legend reads as Facility/Fuel/Ramp Type.
6 of 16
∂ρ ∂ρuei
+ =0 (4)
∂t ∂xi
∂ρuei ∂ sgs
+ ρuei uej + pδij − τij + τij =0 (5)
∂t ∂xj
∂ρEe ∂ h e i
+ ρE + p uei + qi − uej τij + Hisgs + σisgs = 0 (6)
∂t ∂xi
∂ρY
fk ∂ h f
sgs sgs
i
+ ρ Yk uei + Y
fk V
g i,k + Yi,k + θi,k = ω˙k k = 1, ..., Ns (7)
∂t ∂xi
where variables with overbar denotes spatially filtered quantities, variables with tilde denote Favre averaged
quantities and variables with superscript sgs represent unclosed subgrid scale terms. In the above equations,
ρ is the density, ui is the Cartesian velocity component, p is the pressure, τ is the shear stress, E is the total
energy, q is the energy flux due to thermal conduction and species diffusion, Yk is the species mass fraction
and ω˙k is the reaction rate. Pressure, p is evaluated using the filtered equation of state, p = ρR̃T̃ + ρT sgs .
2
The viscous stress, τij in Eq. 5 is evaluated from τ ij = 2µSij + µd − 3 µ Skk δij where, δij is the Kronecker
delta function, Sij = 1/2(∂ui /∂xj + ∂uj /∂xi ) is the resolved rate of strain, µ is the dynamic viscosity and
µd is the bulk viscosity evaluated using the formulation given in Ern and Giovangigli.22 The thermal flux,
qi in Eq. 6 is given by q̄i = −κ̄ ∂x∂Te + ρ̄ PNs h̄ Ye Ve where Fick’s law for species diffusion and Fourier’s law
i k=1 k k i,k
for thermal conduction are assumed.
To close the sgs terms, τ sgs , H sgs , σ sgs and Y sgs in Eqs. 5-7, a transport equation for subgrid Kinetic
Energy is solved:
h i
∂ sgs ∂ sgs ∂ ∂ksgs ρνt R
e ∂ Te
∂t ρ k + ∂xi (ρ u
ei k ) = ∂xi (ρνt + µ) ∂xi + P rt∂xi
sgs 2 ρS ksgs
2
sgs ∂ e
uj (ksgs )3/2
(8)
− 1 + αpd Mt e
Dksgs τij ∂xi + ρc ∆
23
The coefficients, cν and c in the above
√ equation are evaluated using a dynamic procedure. Eddy viscosity,
νt computed from k sgs as νt = Cν k sgs ∆ is then used in conjunction with the eddy viscosity hypothesis
and the gradient closure hypothesis to close the subgrid viscous stress, τ sgs and subgrid species flux Y sgs
respectively. In addition, the turbulent Prandtl number, P rt used to close the energy equation, is also com-
puted dynamically and locally.23 The closure procedure and a comprehensive discussion of the formulation
used in this work can be found in Genin and Menon23 and skipped here for brevity. The computation of ω̇
is discussed further below.
7 of 16
V. Numerical Approach
A block structured finite-volume methodology23 that is second order accurate in space and time is used
to solve the LES governing equations described in Section IV. The evaluation of the fluxes in finite-volume
algorithm is performed using a hybrid methodology developed for the resolution of high speed turbulent flow
environment characterized by shock-turbulence interactions. The algorithm has been validated and applied
for supersonic applications elsewhere23 and is implemented in a multiblock parallel framework using the
Message Passing Interface (MPI) library. The time-step being used for the time integration of the governing
equations is determined from stability consideration for the advection and the diffusion equations.
8 of 16
where M is a radical. The TANN is trained on this kinetics for the range of turbulent conditions in the
combustor.
Figure 6. Time-averaged flowfield at the cavity centerplane for stable combustion case
Figure 6 shows time averaged temperature and H2 O field in the cavity for the stable combustion case.
In this case, there exists an efficient fuel/oxygen mixing region that sustains stable combustion with hot
products recirculating in the cavity, maintaining high temperature. Hence, overall cavity shows region
of high temperature and significant amount of products. Figure 7(a) and Fig. 7(b) show instantaneous
temperature field at cavity centerplane and at 3 spanwise locations (x = 0.18, 0.23, 0.26 m), respectively.
Combustion takes place in most of the cavity region where both fuels are being mixed with the entrapped
oxidizer. Peak temperature occurs at the corner upstream of the injectors but is also well distributed in the
entire combustor. Spanwise temperature field in Fig. 7(b) shows 3D wrinkled structures with local regions
of high temperature. When compared to LEMLES approach,14 turbulence flame interactions appear to be
well captured by TANN-LES approach and in computationally much affordable way.
To understand flow features inside the cavity in stable combustion case, velocity vectors are plotted and
are shown in Fig. 8. It can be clearly seen that a large vortical structure forms and spans over almost entire
cavity length (shown by red box). This vortical flow helps hot products of combustion to recirculate inside
the cavity and preheat oncoming fresh fuel and oxygen. This subsonic recirculation region provides more
residence time for fuel/air to get well mixed and react with each other. This appears to be the primary
mechanism of flame holding for the stable combustion case.
Although fuel is injected into the cavity and therefore, this combustion problem could be classified to
be non-premixed combustion, due to product recirculation and variable mixedness, the combustion process
is much more complex than a simple diffusion flame. To delineate between premixed and non-premixed
∇YF ·∇YO
flame fronts and analyze the flame structure, the flame index26 given by, Flame Index (F.I.) = |∇Y F ·∇YO |
is calculated; where F denotes the fuel and O being the oxidizer. F.I. is positive for premixed flame and
negative for diffusion flame. In premixed region, the gradients of fuel and oxidizer are aligned. So the positive
values of F.I. are seen due to rapid consumption of the reactants across the flame. In non-premixed flames
the gradients of fuel and oxidizer oppose each other, and hence F.I. is a negative value. Fig. 9(a) shows
flame index at stream-wise location for stable combustion case indicating both premixed and nonpremixed
burning occurs inside the cavity. Negative values of F.I. are mostly observed near the shear layer and near
the injectors where diffusion flame is present, while positive values of F.I. showing premixed burning are
observed mostly in the aft cavity portion. As discussed earlier, presence of a large vortical structure inside
the cavity increases local flow time and enhances fuel/air mixing process and this is confirmed by the positive
values of F.I. in this region. In this premixed zone, CH4 / air could burn by itself even there is very little
or no H2 present.
9 of 16
(b) Temperature field at 3 spanwise locations: x = 0.18, 0.23, 0.26 m from left to right
Figure 8. Velocity vectors showing recirculation region inside cavity in stable combustion case (zoomed in view of
cavity)
(a) Flame index at centerplane (b) Stoichiometric mixture fraction overlaid on tempera-
ture field
10 of 16
Figure 11 shows zoomed in view of the time-averaged velocity vectors inside the cavity in near blowout
case. When compared to the stable case there are substantial differences. There is no single large recirculation
region present inside the cavity (as opposed to stable combustion case). Instead, there are small local regions
present (shown by red box) in the aft cavity. As a result, there is no efficient feedback mechanism present,
which can recirculate hot products of combustion inside the cavity and preheat oncoming fuel/air mixture.
In the near blowout case, as less hydrogen is blended with methane, there is no enough heat release for
thermal expansion of flow inside the cavity. The reduced temperature and thermal expansion effect seems to
breakup the recirculation in the cavity. Small local vortical structures can recirculate only locally and thus,
seem to be inefficient in capturing the overall mixing process.
Figure 11. Velocity vectors showing recirculation regions inside cavity - near blowout case (zommed in view of cavity)
Flame index in Fig. 12(a) shows diffusion type burning (black color) near the injector location. Due to
absence of strong recirculation current, there is hardly any premixed region seen in this case. Stoichiometric
mixture fraction contours overlaid on temperature field in Fig. 12(b) show primarily diffusion flame surface.
It can be clearly seen that, flame is mostly seen near injector location. In effect, reaction zone is shrunk and
this maybe the cause of the blowout event. Figure 13 shows CH4 field in near blowout case. As it can be
seen, lot of CH4 is left and is escaping out of cavity without getting mixed with oxidizer inside cavity. This
could be due to absence of strong recirculation current in the near blowout case.
Further analysis of the thermal expansion effect and the sensitivity of the combustion process will be
carried out in the future when more cases are simulated and when other fuel-air mixtures are considered.
11 of 16
For now, the current studies agree with experimental observations that the flame structure and the hot
temperature region changes significantly from stable to unstable combustion. Finally, an estimate of the
Damkohler number based on local properties is carried out (note that this is a local estimate as opposed to
a global estimate in the experimental case). Equivalent to Fig. 5(b), scatter plot of Damkohler number as
a function of local equivalence ratio is plotted in Fig. 14 along with relevant global data. There are some
clear differences between the local Da variation and the global estimate. Further research is still needed to
understand this discrepancy.
Figure 15 shows the top wall pressure comparison with available experimental data. It can be seen that,
locations of shocks and peak pressure inside the combustor are predicted reasonably well. Location of leading
edge shock (x ≈ 0.22 m), expansion (x ≈ 0.26 m) followed by shock at (x ≈ 0.32 m) are well predicted.
Multiple reflections off the wall (x ≈ 0.34 m, 0.37 m and 0.42 m) are also captured well.
12 of 16
VII. Conclusions
This paper discusses a combined experimental and numerical study of ignition and blow-out limits of
methane-hydrogen fuel blend in high speed flow environment. Experiments are conducted to study combus-
tion ignition and blow-out limit in a Mach 2.5 crossflow in a facility developed at Georgia Tech. Results
show that addition of even a small amount of hydrogen greatly influences the reaction time, altering the
combustion mechanism in a more favorable manner. The blowout data collected during experiments is com-
pared to the data from other facilities. Subsequently, LES of one stable and one blowout case guided by
experiments are carried out. ANN based filtered chemical rates modeling (Turbulent-ANN) approach for
turbulent combustion is employed. The flow structures observed in stable combustion and blow-out cases
show distinct differences. In stable combustion, a large vortical flow is formed in the aft region of the cavity
13 of 16
VIII. Appendix
VIII.A. Combustion Regimes
In order to determine different combustion regimes and understand the turbulent flame structure a Borghi
diagram modified by Pitsch27 for LES applications is constructed based on premixed F.I. values. The analysis
requires total RMS velocity (u0∆ ), LES filter size (∆), local laminar flame thickness (LF ) and local laminar
flame speed (SL ). The RMS velocity and the LES filter width (same as grid size) can be evaluated using
the data obtained from the simulations. However, the laminar flame speed and the flame thickness will
also depend on amount of products present at particular location and they need to be estimated locally
everywhere inside the combustor. Using the flame speed and the flame thickness, Karlovitz number (Ka)
defined for LES27 is evaluated. Figure 16(a) shows different flame regimes seen in the present study. Strong
variation of Ka number from corrugated flamelet to broken reaction zone is observed. For Ka < 1 turbulent
motions wrinkle the flame front causing convolution leading to formation of pockets of fresh and burnt
gases. For 1 > Ka > 100, turbulent motions enter and modify flame preheat zone. For Ka > 100, both
diffusion and reaction zone are affected by turbulent motions. Figure 16 shows probability distribution of
flame regimes inside the cavity combustor. Injector plane location is at x ≈ 0.006 m (x = 0 is at cavity LE).
It can be seen that the probability of broken zone and thin reaction zone is more near the injector location.
This could be the effect of high level of turbulence due to floor injection. Thin reaction zone dominates in
the overall cavity, whereas flamelet type burning is mostly seen at aft portion of the cavity. It is emphasized
that this analysis only looks that the premixed regions based on the F.I. values.
(a) Borghi diagram for LES (b) PDF of flame regimes inside the combustor
Figure 16. Flame regimes and their probability inside the combustor
14 of 16
Figure 17. Correlation between the target value and TANN prediction for all species
IX. Acknowledgements
This work is supported in part by the Air Force Office of Scientific Research. The support at the Depart-
ment of Defence (DOD) High Performance Computing Center (HPC) of Army Research Laboratory(ARL),
NAVY and Engineer Research and Development Center (ERDC) for simulations is acknowledged.
15 of 16
stabilized flames in supersonic flow. Proceedings of the Combustion Institute, 30:2825–2833, 2005.
4 V. A. Vinogradov, S. A. Kobigsky, and M. D. Petrov. Experimental investigation of kerosene fuel combustion in supersonic
experiments with a cavity-based fuel injector. Journal of Propulsion and Power, 17(6):1305–1312, 2001.
6 M. R. Gruber, J. M. Donbar, C. D. Carter, and K. Y. Hsu. Mixing and combustion studies using cavity-based flameholders
2008-4571, 44th AIAA/ASME/SAE/ASEE Joint Propulsion Conference and Exhibit, 21 - 23 July 2008, Hartford, CT, 2008.
10 Kovitch S. B. Verma S. Retaureau, G. J. and S. Menon. Experimental studies of cavity flame-holding in a Mach 2.5 cross
flow. AIAA-2009-0810, 47th AIAA Aerospace Sciences Meeting, 5 - 8 Jan 2009, Orlando, Florida, 2009.
11 C. G. Rodriguez and A. D. Cutler. Computational simulation of supersonic-combustion benchmark experiment. AIAA
2005-4424, 41st AIAA/ASME/SAE/ASEE Joint Propulsion Conference and Exhibit, July 10 - 13, 2005, Tucson, Arizona,
2005.
12 J. Liu, C.-J. Tam, T. Lu, and C. K. Law. Simulations of cavity-stabilized flames in supersonic flows using reduced
chemical kinetics mechanisms. AIAA 2006-4863, 42nd AIAA/ASME/SAE/ASEE Joint Propulsion Conference & Exhibit,
Sacramento, California, 2006.
13 Kyung Moo Kim, Seung Wook Baek, and Cho Young Han. Numerical study on supersonic combustion with cavity-based
AIAA 2011-323, 49th AIAA Aerospace Science Meeting, Orlando, FL, 2011.
16 B. Sen and S. Menon. Artificial neural networks based chemistry-mixing subgrid model for les. AIAA-2009-241, 47th
cavity-stabilized flames. AIAA-2005-3660, 40th AIAA/ASME/SAE/ASEE Joint Propulsion Conference and Exhibit, 2004.
19 M. D. Slessor, M. Zhuang, and P. E. Dimotakis. Turbulent shear-layer mixing: growth-rate compressibility scaling. J.
48(3):526–539, 2010.
24 B. Sen and S. Menon. Turbulent premixed flame modeling using artificial neural networks based chemical kinetics.
16 of 16