Flamelet structure in turbulent premixed swirling oxy

ARTICLE IN PRESS

JID: PROCI

[m;July 11, 2018;2:3]

Available online at www.sciencedirect.com

Proceedings of the Combustion Institute 000 (2018) 1–8 www.elsevier.com/locate/proci

Flamelet structure in turbulent premixed swirling oxy-combustion of methane N.W. Chakroun∗, S.J. Shanbhogue, Y. Dagan, A.F. Ghoniem Massachusetts Institute of Technology, 77 Massachusetts Ave 3-339Q, Cambridge 02139, USA Received 1 December 2017; accepted 23 June 2018 Available online xxx

Abstract Two key flame macrostructures in swirling flows have been observed in experiments of oxy-combustion (as well as air-combustion); as the equivalence ratio is raised, the flame moves from being stabilized on just the inner shear layer (Flame III) to getting stabilized on both the inner and outer shear layers (Flame IV). We report results of an LES investigation of two different inlet oxy-fuel mixtures, in a turbulent swirling flow at Re = 20,000, that capture these two macrostructures. Previous work on the effects of heat loss have mostly focused on its impact on macro-scale observations. In this paper, we examine how heat loss impacts the flame microstructures as well for these two macrostructures. For both flames, the flamelet structure, as represented by a scatter plot of the normalized fuel concentration against the normalized temperature, depends on whether the combustor walls are adiabatic or non-adiabatic. For the adiabatic case, the flamelets of both macrostructures behave like strained flames. When wall heat transfer is included, Flame III microstructure is more bimodal. Since this flame extends farther downstream and part of it propagates along the walls, heat transfer has a greater impact on it’s microstructure. These results show that heat loss impacts not just the macro properties of the flame such as its shape or interactions with the wall, but also fundamentally changes its internal structure. Scatter plots of the turbulent flames are constructed and compared to different 1D laminar flame profiles (e.g., strained or with heat loss), and comparisons suggest the important role of the wall thermal boundary conditions in the accurate simulations of combustion dynamics and interpretations of experimental data, including data reduction and scaling. © 2018 The Combustion Institute. Published by Elsevier Inc. All rights reserved. Keywords: Swirling flames; Macrostructures; Strained flame; Heat loss; Oxy-combustion

1. Introduction Natural gas oxy-fuel combustion cycles are some of the leading technologies for carbon cap∗

Corresponding author. E-mail address: [email protected] (N.W. Chakroun).

ture and sequestration (CCS) in power plants [1]. These cycles require combustors that can burn fuel and an oxygen stream along with some diluent, typically CO2 , to control flame temperatures. The ability to accurately simulate these mixtures and understand the characteristics of oxy-combustion flames, are important for the design of combustors in these power plants.

https://doi.org/10.1016/j.proci.2018.06.181 1540-7489 © 2018 The Combustion Institute. Published by Elsevier Inc. All rights reserved.

Please cite this article as: N.W. Chakroun et al., Flamelet structure in turbulent premixed swirling oxy-combustion of methane, Proceedings of the Combustion Institute (2018), https://doi.org/10.1016/j.proci.2018.06.181

JID: PROCI

2

ARTICLE IN PRESS

[m;July 11, 2018;2:3]

N.W. Chakroun et al. / Proceedings of the Combustion Institute 000 (2018) 1–8

is laminar, the flamelet data for points upstream do not bear resemblance to any known canonical flame. In this paper, we aim to investigate the flamelet structure at higher Reynolds numbers and the impact of turbulence and heat transfer. Specifically, we propose a couple of questions. Firstly, does a flamelet exist everywhere in turbulent flows for the conditions that we investigate, at Re = 20,000? Secondly, what is the role of wall-heat transfer, along with strain, on the flamelet structure? 1.1. Background on flamelet structures in turbulent flows

Fig. 1. Top: Streamlines (solid lines) and flame position (dotted lines) for the bluff body flame. The top half of this figure plots temperature contours and the bottom half, the fuel mass fraction contours. Bottom Left: Plot of normalized methane mass fraction and temperature for all points upstream of the leading edge, i.e., dx < 3.27. Bottom Right: Plot of normalized methane mass fractions and normalized temperature for all points downstream of the flame leading edge, i.e., dx > 3.27. This data is from simulations reported in Michaels and Ghoniem [7].

Anchored flames, particularly those stabilized via recirculation zones in these practical systems have characteristic macrostructures. Bluffbody flames have a ‘V’ shape, confined Bunsen flames (stabilized in a sudden expansion) look like an inverted-V, and swirling flames (commonly used in gas turbines) can be either conical, bubble + conical, V and tulip shaped [2–5]. Even if the underlying flow is unsteady – whether due to turbulence, combustion instabilities or external excitation – the time or ensemble averaged chemiluminescence images of flames average out to their characteristic macrostructure [6]. A fundamental, yet unresolved question is, what is the flamelet structure (if at all there is one) underneath these macrostructures. To illustrate the complexities, consider a laminar bluff-body flow as shown in Fig. 1. This is data from the DNS study at Red = 500 by Michaels and Ghoniem [7]. There are no unsteady events in this flow at this Reynolds number and dilatation ratio, and the figure shows the steady state solution. As such, this bluff-body flame has an apparently simple macrostructure - two curved flame segments. But, consider the scatter plots of the normalized methane mass fraction versus normalized temperature. The data shows that wherever a flame exists ( dx > 3.27), the relationship between the fuel mass fraction and temperature collapses on to a single line, with very little spread. And this resembles the structure of a strained laminar flame. But upstream, in the recirculation zone close to the bluff body ( dx < 3.27), there is a substantial spread in the scatter plots. In fact, even though the flow

In 1985, Borghi [8] summarized the historical discussion on the structure of premixed turbulent flames, as being either “wrinkled laminar flame, or a set of laminar flames moving randomly” versus “a distributed reaction zone model where no flame fronts are present”. He concluded the discussion by stating that depending on the intensity of turbulence, one could see both. Indeed, the Borghi diagram [9] for turbulent premixed combustion has distinct zones/regimes for wrinkled flames, However, in a comprehensive review in 2008, Driscoll concluded that “evidence for ‘nonflamelet’ behavior is sparse”, and referenced studies at Karlovitz numbers greater than 10 [10]. In a recent 2015 DNS study on n − C7 H16 /air flames, Savard et al. [11] show that even under intense turbulence (Karlovitz numbers >200), the structure of turbulent flames are similar to those of laminar flamelets. Knowledge of the structure of the flamelet and its effective Lewis number also has practical consequences. Bellan [12] provides evidence that the performance of highly reduced mechanisms in reproducing flamelet structures is dependent on the transport model used. Stated differently, some chemical mechanisms work for certain Lewis numbers, but fail for others. This implies more investigations into the flame structure of anchored flames are needed to identify the correct transport models and thus develop accurate, reduced chemistry models. 1.2. Background on Flame IV stabilization and role of heat transfer In our recent study [13], we compared the conditions leading to the stabilization of lean turbulent methane air and oxy-flames along the outer shear layer (Flame IV) of the same premixed swirl combustor as in the current work. We found that the transition in the flame macrostructure from Flame III to Flame IV, scaled according to the calculated extinction strain rate of the inlet mixture. A key conclusion from this work was the importance of performing the laminar flame calculations at realistic thermal conditions in that region, in order to


ARTICLE IN PRESS

JID: PROCI

[m;July 11, 2018;2:3]


properly model the experimental turbulent flame behavior. The importance of laminar flame characteristics, e.g., extinction strain rate, on the turbulent flame behavior has also been seen in other recent studies [2,14]. The dependence of these macrostructures on strain and heat loss effects was investigated by several researchers [15–17]. They all show that inadequately representing the thermal boundary conditions of the combustor wall results in incorrect predictions of the flame macrostructure found in experiments. Our present paper serves to better understand the roles of strain and heat transfer for methane oxy-combustion mixtures which has not been sufficiently investigated yet. As we will show later in this study, heat loss, besides strain, also significantly impacts the microstructures of flames. The rest of the paper is organized as follows. We first present details of the LES computational model and then detail the reduced order models that we use to compute laminar flame profiles to compare with the LES data. We then describe two macrostructures that we study, namely Flame III and Flame IV, then analyze their flamelet structures and assess the role of heat-transfer. Finally, we conclude with the implications of these findings.

considered constant, and to simplify the equations, Favre or density-weighted filtering is applied. We use the ATF model allowing the use of an Arrhenius type reaction rate [20]. This involves artificially thickening the flame front so that it can be resolved on the LES grid, in a DNS-like approach, while maintaining the same laminar flame speed and turbulence-flame interaction. In the ATF model, the molecular diffusivity (D) and reaction rate (ω˙ ) are modified accordingly using a thickening factor, F, (FD & ω˙ /F) to maintain the same flame speed. When the flame is thickened, however, this leads to a modified turbulence-chemistry interaction and the wrinkling of the flame front is reduced. To account for the wrinkling effect of the unresolved features on the thickened flame front, an efficiency function, E, is introduced. This efficiency function is implemented following the algebraic expression proposed by Colin et al. [20]. A dynamic thickening approach is implemented [22] wherein the thickening factor and the diffusivity are represented locally (Floc & Dloc ). The filtered species conservation equation then takes the form: k k k ∂ ρY ∂ ρ ui Y ∂ ∂Y + = ρFloc EDeff ,loc,k ∂t ∂xi ∂ xi ∂xi E ω˙ k Floc

+ 2. Computational methods 2.1. LES model The LES code utilized in this work was described in detail in the work of Kewlani et al. [18] and Nemitallah et al. [19] and it has been shown to reasonably predict the velocity profiles, flow structures and flame shapes observed in experiments for various configurations, including the swirl combustor. This code is implemented using OpenFOAM’s C++ libraries based on finite volume spatial discretization. Implicit second order temporal schemes are used along with a combination of first and second order schemes for spatial discretization. For the turbulent combustion model, the artificially thickened flame (ATF) approach is used [20]. LES resolves the large scales of the flow but models the smallest (and most expensive) scales of the solution, rather than resolving them as DNS does. The governing equations for LES are obtained by applying a spatial filter to the conservation equations of mass, momentum and energy, and the species transport equations. This low pass filter operation eliminates the small scales of the solution. The sub-grid-scale stresses resulting from the filtering operation are unknown, and require modeling. These terms can be modeled using different approaches but in our code the one-equation eddy viscosity model [21] is implemented. Also for these reacting flows, since the density cannot be

3

(1)

where, Floc = 1 + (F − 1)(c )

Deff ,loc,k =

(2)

μ μsgs EFloc + (1 − (c )) Sc Scsgs

(3)

with (c ) = 16[c(1 − c )]2 is a locally defined function based on the reaction progress variable: c = fuel (1 − YYinlet ). μ and Sc are the dynamic viscosity and fuel

the Schmidt number respectively. Following the expression proposed by Colin et al. [20] for modeling the efficiency function in terms of the dimensionless wrinkling factor, , local filter width, , unstrained laminar flame speed, S0L , thickness of the laminar (δL0 ) and thickened flames (δL1 ), and local sub-grid scale velocity fluctuation, u : E =

(δ = δL0 ) 1 (δ = δL1 )

(4)

u u

, SL0 δL0 3Cms (Ret1/2 − 1) SL0 2 / 3 −0.3 u u

0 , 0 = 0.75 0 exp −1.2 0 S L δL δL SL 2ln2

= 1+α

(5) (6)

u

The term α S0 expresses the increase in flame L

wrinkling due to the turbulent stretch. While the


ARTICLE IN PRESS

JID: PROCI

4

[m;July 11, 2018;2:3]


Fig. 2. Computational domain dimensions for the swirl case studied in this work. Table 1 Operating conditions for each flame. Flame type Mixture composition Re Teq (K) SL (cm/s) κ ext (s−1 ) Le Ka

XCH4 XO2 XCO2

III

IV

0.08 0.27 0.65 20,000 1590 7.75 370 0.75 29

0.10 0.27 0.63 20,000 1838 13.40 989 0.75 17

function represents the dimensionless stretch of a flame with velocity SL0 and thickness δL0 impacted by a range of vortices. The average cell size imposed a choice of a thickening factor F used by the combustion model. As the filter size is taken to be the cell size, thickening factors of F = 1.5 and 3 were chosen for Flame III and Flame IV, respectively, allowing approximately five grid points across the thickened flame to resolve it. Sensitivity tests for the static (initial) thickeningfactor were conducted and produced very similar flamelet structures, suggesting that the statistical representation of the phenomenon we are investigating, i.e., the effect of chemical kinetics on largescale flame response would not be affected by the choice of initial thickening factor coefficient. Reactions are modeled using the multi-step chemical mechanism developed by Frassoldati et al. [23]. This mechanism was developed for oxycombustion mixtures by tuning the widely used Jones and Lindstedt [24] mechanism for methaneair. It involves 6 reactions and 9 species. It also gave acceptable approximations of adiabatic flame temperature, laminar burning velocity and extinction strain rate [25] when compared to the detailed mechanism proposed by Mendiara and Glarborg [26], which has been widely used for modeling CO2 diluted methane-oxy mixtures [13,27–29]. The swirl combustor configuration is shown in Fig. 2 and is based on the same combustor geometry as in previous simulations [18] and similar to the experimental setup [13,28]. The inlet conditions tested are shown in Table 1 where CH4 /O2 /CO2 mixtures are introduced at two different equivalence ratios (0.56 and 0.70) at atmospheric conditions (Tu = 300 K, P0 = 1 atm). Karlovitz num-

bers (Ka) were computed based on the definition in [11] as the ratio of the flame time scale to the Kolmogorov time scale. The code and mesh for the swirl combustor has been tested and validated before for CH4 /air mixtures and was also shown to be capable of resolving the global dynamics of the flame [18]; in this study we change the inlet mixture to study methane oxy-combustion and understand flame structures. The mesh consists of 0.65 million cells and the cells get coarser downstream. The average cell sizes in the whole domain are 1 mm with maximum and minimum cell sizes of 1.6 mm and 0.32 mm. The mesh is also non-uniform with a mix of tetrahedral and hexahedral cells. Tetrahedral cells are used around the swirler to better model its geometry. The swirler has 8 blades each inclined at 45o to the cylinder cross-section, with an estimated swirl number of 0.7. Grid sensitivity tests were performed up to a 2 million cells grid [30–32]. The computational time increased by a factor of three and became impractical for use in LES but more importantly, the results did not show significant improvements, when compared to experiments, in order to justify this increased time. Dirichlet conditions are used for all variables at the inlet except the pressure for which zero-gradient conditions are specified. The inlet velocity is uniform with turbulent fluctuations imposed. No-slip conditions are applied along all walls while zerogradient conditions are used for other variables. Similarly at the exit, zero-gradient conditions are specified for all variables except pressure for which wave-transmissive conditions are used. Convective heat transfer at the walls is considered for two cases as will be shown later, while radiation within the gases and with the wall is neglected. The thermal boundary condition used at the wall is a Robin (mixed) type boundary condition that takes into account an effective external heat transfer coefficient. 2.2. Flamelet model Laminar flame profiles were computed by running various CHEMKIN modules [33]. For comparisons with an unstretched flame, we used the PREMIX module, which solves for laminar burning velocities to compute unstretched flame profiles (referred to as κ = 0 in the figures). For comparisons with strained flames, we used the OPPDIF module to generate the flame profiles. In this module, the code models two opposed jets. We use two variants of this configuration. In one, both jets are identical mixtures of premixed CH4 /O2 /CO2 and the result is identical twin flames. In the second, one jet contains a premixed unreacted mixture, and the other contains equilibrium products of the mixture, but at various product temperatures. We use this to compute flame profiles under different levels of heat loss with product


JID: PROCI

ARTICLE IN PRESS

[m;July 11, 2018;2:3]


Fig. 3. Instantaneous stream lines and heat-release contours for the methane-oxy flames at two different inlet conditions. We refer to these flames as Flame III and Flame IV, respectively.

temperatures extracted from LES simulations. Results from these configurations are shown in Fig. 6. While the OPPDIFF module provides flame profiles of strained flames, we can’t use it to compute profiles close to extinction. For this, we use the EXTINCTION module. This uses a numerical continuation procedure to compute profiles of flamelets close to extinction (referred to as κ = κext in the figures). In all these modules, we can explicitly specify the transport model to be used. We use mixture averaged transport properties in most of the laminar flame calculations except for the results in Fig. 6 where we model heat loss. In these simulations we use a fixed unity Lewis number transport model to generate some of the profiles.

5

swirler center body and two locations where the flame touches the expansion plane.) On the other hand in Flame III, we see that despite starting at almost the same location near the swirler centerbody, and spreading along the innershear layer; the flame does not wrap back around the outer recirculation zone, but proceeds to spread further downstream, very close to the walls. In fact almost half of the flame, measured lengthwise, borders the walls. The two mixtures that resulted in Flame III and Flame IV had adiabatic flame temperatures of 1590 K and 1838 K. There are several other macrostructures possible in the swirling flow especially for mixtures with lower flame temperatures, but these are long flames and are less desirable in practical combustors. Also in our previous study, we have shown that when combustor acoustic conditions are favorable, the transition from Flame III to Flame IV is concomitant with the transition of the combustor from stable to unstable [4]. Hence we focus on these two macrostructures in this study. 3.2. Laminar flamelet comparisons

3.1. Flame III and Flame IV structures

It is common to use a phase diagram (compositional space) or scatter plots in which the fuel mass fraction is plotted against the normalized temperature to compare turbulent and laminar flame structures [35–37]. The LES data is constructed by extracting methane mass fraction and temperatures for all points in the 3D domain at a single time step. The results presented do not depend on the time instant chosen. The fuel mass fraction begins at the reactant composition and temperature, and drops all the way to temperatures close to their equilibrium values. The laminar flame data falls essentially on a single line while turbulent flame data exhibits a scatter corresponding to how turbulence modifies the flame structure at different points in the domain. The comparison helps determine how turbulence affects combustion under the conditions being examined.

We begin by plotting the macrostructure of the flame for two different inlet fuel-oxygen ratios (see Table 1) with wall convective heat loss, as shown in Fig. 3. These plots show instantaneous images of heat-release contours, superimposed on the streamlines. These are 3D simulations, but shown as 2D slices. The flow is rich in dynamical events and has been documented extensively in previous work [34]. What we refer to as Flame IV is a flame that begins by anchoring itself at the swirler upstream, and spreads beyond the expansion plane in the shearlayer enveloping the inner recirculation zone (IRZ), but then wraps back around the outer recirculation zone (ORZ). In the literature, they are also referred to as “M-flames” or “tulip flames” [3]. Notice also that in this case, the flame barely interacts with the wall, except near the tips (two locations near the

3.2.1. Adiabatic case We plot the scatter plots of the LES of Flame III and Flame IV assuming adiabatic boundary conditions first, in Figs. 4 and 5. For reference, results for an unstretched flame and a strained flame very near extinction, are added. For Flame III, notice that the data from the LES data agrees well with the laminar flame case especially at higher temperatures. Now consider the data in Flame IV (Fig. 5). In this figure, we also display unstretched flame data (κ = 0) and scatter plot of the same very near to the extinction strain rate (κ = κext ). The latter forms the upper bound of the LES data, while the unstretched flame forms the lower bound. This is also seen for Flame III near the highest temperatures. Both figures show that the scatter plot of the LES data are bounded

3. Results


JID: PROCI

6

ARTICLE IN PRESS

[m;July 11, 2018;2:3]


Fig. 4. Scatter plots for Flame III, with adiabatic boundary conditions, and the corresponding laminar flames at zero strain and very close to extinction.

Fig. 6. Scatter plots for Flame III, with heat loss, and for the corresponding laminar flames with different adiabatic flame temperatures and different strains.

Fig. 7. Instantaneous normalized temperature ( = (T − Tu )/(Teq − Tu )) contours for Flame III with heat loss. Fig. 5. Scatter plots for Flame IV, with adiabatic boundary conditions, and the corresponding laminar flames at zero strain and very close to extinction.

by laminar flame profiles at different strain rates. These adiabatic LES results indicate that turbulence impacts these flames primarily by subjecting them to different strain rates. Furthermore, we see that overall, the turbulent flamelet structures are similar to those of laminar strained flames. The slight deviations from the laminar profiles could be because of the tendency of turbulent flames to survive strains higher than κext under unsteady conditions [38,39]. But further investigations are needed to confirm this conclusion. 3.2.2. Heat loss effects We perform a similar analysis in this section, to try and explain the non-adiabatic turbulent flame scatter data using 1-D strained laminar flames. Clear differences are observed between the two wall boundary conditions for Flame III in Figs. 4 and 6. For Flame IV, the majority of the

scatter data lie along a curve extending from top left to bottom right. Flame III scatter plot, and its underlying flamelet structure, is more complex. We observe a large cluster of points, located further downstream in the flow, below the scatter observed in Flame IV. The location within the flow of these low temperature data was found along the outer walls of the combustor; suggesting that the corresponding flames have a different structure, perhaps determined by heat loss along the wall. Therefore we attempt to reproduce this turbulent flame scatter data and the local flame structures they correspond to using laminar flame simulations but with varying product gas temperatures as shown in the figure, hence modeling the impact of heat loss on the flame. Note that b = (Tb − Tu )/(Teq − Tu ), where Tb is the product gas temperature. These laminar flamelets simulate the different temperatures present along the wall of the combustor (see Fig. 7). Fig. 6 shows that we are able to successfully capture the spread in the LES data using data from non-adiabatic strained flames. In order to reproduce the impact of heat loss, a


JID: PROCI

ARTICLE IN PRESS N.W. Chakroun et al. / Proceedings of the Combustion Institute 000 (2018) 1–8

Fig. 8. Normalized instantaneous temperatures and heat release rates along the combustor wall for Flame III with and without heat loss.

[m;July 11, 2018;2:3]

7

( = 0.5–1.0), resulting in a larger spread in the scatter plot. On the other hand, the effect of heat loss on the Flame IV structure is pretty minimal. From Fig. 3 it was evident that the Flame IV macrostructure is more compact and has less interaction with the combustor wall. We do not observe the same effects we see in Fig. 8 for Flame IV where fuel is being consumed at low temperatures. We added data from a laminar flame calculation at a very low b in order to show that heat loss does not impact the structure of this flame. The additional slight scatter below the zero strain curve is also not explained by strain or heat loss effects but perhaps could be due to the unsteady effect mentioned above (or is an artifact of the artificially thickened flame model). Adiabatic unstrained and strained laminar flames, can thus reasonably reproduce the microstructure of this flame. 4. Concluding remarks

Fig. 9. Scatter plots for Flame IV, with heat loss, and for the corresponding laminar flames at zero strain and very close to extinction.

laminar flame model with opposed reactants and products stream at different temperatures, as was described in Section 2.2, was used. As indicated in the figure, there is a difference in the results using the two transport models, described in Section 2.2. The mixture averaged points don’t seem to fit the data in the region of Y/Yin close to 1. We also add a strained flamelet at an intermediate strain to capture the upper bound of the data. This strain rate was chosen by taking an average value in the region of the flow near the combustor walls where heat loss effects are important. Points with b = 0.5–0.6 are present in Flame III, but at much lower density. The data points in the second branch lie exactly along the combustor wall, where the majority of the data points are at = 0.4–0.5. Figure 8 depicts the normalized instantaneous temperatures and heat release rates along the wall for Flame III with and without heat loss. It clearly shows that fuel is being consumed but at lower temperatures ( = 0.4–0.6) for the case with heat loss compared to the adiabatic flame

In this study, we use LES to investigate the local flamelet structure of two oxy-fuel mixtures at different fuel-oxygen ratios that resulted in two flame macrostructures: Flame III and Flame IV. While previous studies focused on examining the macroscale properties of these flames, this work examines how heat loss impacted their flame microstructures. The combustor wall boundary conditions had a significant impact on the flamelet structure. For the adiabatic cases, the flamelet structure of both macrostructures resembled strained laminar flames subjected to varying levels of strain. When wall heat transfer is included, Flame III exhibits larger scatter data because the fuel is consumed at lower temperatures along the combustor wall. The correspondence of the 3D turbulent flames to different 1D laminar flame profiles (e.g., strained or with heat loss), suggests that the thermal boundary conditions play an important role in determining the burning characteristics and dynamics in oxy-combustion [28]. Acknowledgments Financial support from the King Abdullah University of Science and Technology (KAUST) (KUS-110-010-01), the King Fahd University of Petroleum and Minerals (KFUPM) (R12-CE-10), the TATA Center for Technology and Design, and MIT-Technion fellowships, is gratefully acknowledged. References [1] N. Chakroun, A. Ghoniem, Int. J. Greenh. Gas Control 41 (2015) 163–173, doi:10.1016/j.ijggc.2015.06. 025.


JID: PROCI

8

ARTICLE IN PRESS

[m;July 11, 2018;2:3]


[2] S. Shanbhogue, Y. Sanusi, S. Taamallah, M. Habib, E. Mokheimer, A. Ghoniem, Combust. Flame 163 (2016) 494–507. [3] I. Chterev, C. Foley, D. Foti, et al., Combust. Sci. Technol. 186 (8) (2014) 1041–1074. [4] S. Taamallah, S.J. Shanbhogue, Y.S. Sanusi, E.M. Mokhiemer, A.F. Ghoniem, in: ASME Turbo Expo 2015, Montreal, Canada. GT2015-43998, ASME, 2015. [5] C. Foley, I. Chterev, B. Noble, J. Seitzman, T. Lieuwen, Int. J. Spray Combust. Dyn. 9 (1) (2017) 3–18. [6] S. Taamallah, Z.A. LaBry, S.J. Shanbhogue, A.F. Ghoniem, Proc. Combust. Inst. 35 (3) (2015) 3273–3282. [7] D. Michaels, A.F. Ghoniem, Combust. Flame 172 (2016) 62–78. [8] R. Borghi, in: Recent Advances in the Aerospace Sciences, Springer, 1985, pp. 117–138. [9] R. Borghi, Prog. Energy Combust. Sci. 14 (4) (1988) 245–292. [10] J.F. Driscoll, Prog. Energy Combust. Sci. 24 (1) (2008) 91–134, doi:10.1016/j.pecs.2007.04.002. [11] B. Savard, B. Bobbitt, G. Blanquart, Proc. Combust. Inst. 35 (2) (2015) 1377–1384. [12] J. Bellan, Combust. Flame 184 (2017) 286–296. [13] S. Taamallah, N.W. Chakroun, H. Watanabe, S.J. Shanbhogue, A.F. Ghoniem, Proc. Combust. Inst. 36 (3) (2017) 3799–3807, doi:10.1016/j.proci.2016.07.022. [14] N.W. Chakroun, S.J. Shanbhogue, G. Kewlani, S. Taamallah, D. Michaels, A. Ghoniem, in: 55th AIAA Aerospace Sciences Meeting, 2017, p. 1572. [15] L. Tay-Wo-Chong, M. Zellhuber, T. Komarek, H.G. Im, W. Polifke, Flow Turbul. Combust. 97 (1) (2016) 263–294. [16] F. Proch, A. Kempf, Proc. Combust. Inst. 35 (3) (2015) 3337–3345. [17] R. Mercier, T. Guiberti, A. Chatelier, et al., Combust. Flame 171 (2016) 42–58. [18] G. Kewlani, S. Shanbhogue, A. Ghoniem, Energy Fuels 30 (4) (2016) 3451–3462. [19] M.A. Nemitallah, G. Kewlani, S. Hong, S.J. Shanbhogue, M.A. Habib, A.F. Ghoniem, Energy 95 (2016) 211–222. [20] O. Colin, F. Ducros, D. Veynante, T. Poinsot, Phys. Fluids 12 (7) (2000) 1843–1863. [21] S. Huang, Q. Li, Int. J. Numer. Methods Eng. 81 (7) (2010) 835–865.

[22] L. Durand, W. Polifke, in: ASME Turbo Expo 2007: Power for Land, Sea, and Air, American Society of Mechanical Engineers, 2007, pp. 869–878. [23] A. Frassoldati, A. Cuoci, T. Faravelli, E. Ranzi, C. Candusso, D. Tolazzi, in: 1st International Conference on Sustainable Fossil Fuels for Future Energy, 2009, pp. 6–10. [24] W. Jones, R. Lindstedt, Combust. Flame 73 (3) (1988) 233–249. [25] N. Chakroun, Dynamics, Stability and Scaling of Turbulent Methane Oxy-Combustion, Massachusetts Institute of Technology, USA, 2018 Ph.D. thesis. [26] T. Mendiara, P. Glarborg, Combust. Flame 156 (10) (2009) 1937–1949. [27] H. Watanabe, S.J. Shanbhogue, A.F. Ghoniem, ASME Turbo Expo GT2015-43224, 2015. [28] H. Watanabe, S.J. Shanbhogue, S. Taamallah, N.W. Chakroun, A.F. Ghoniem, Combust. Flame 174 (2016) 111–119. [29] N.W. Chakroun, D. Michaels, S.J. Shanbhogue, A.F. Ghoniem, J . Propuls. Power 34 (4) (2017) 825–835. [30] S. Taamallah, Impact of Fuel and Oxidizer Composition on Premixed Flame Stabilization in Turbulent Swirling Flows: Dynamics and Scaling, Massachusetts Institute of Technology, 2016 Ph.D. thesis. [31] G. Kewlani, Z. Labry, N. Abani, S. Shanbhogue, A. Ghoniem, in: 50th AIAA Aerospace Sciences Meeting, American Institute of Aeronautics and Astronautics, 2012. [32] G. Kewlani, K. Vogiatzaki, S. Shanbhogue, A.F. Ghoniem, in: 51st AIAA Aerospace Sciences Meeting, 2013, p. 169. [33] CHEMKIN-PRO 15101Reaction Design Inc., San Diego, CA (2010). [34] S. Taamallah, S.J. Shanbhogue, A.F. Ghoniem, Combust. Flame 166 (2016) 19–33. [35] A. Aspden, M. Day, J. Bell, J. Fluid Mech. 680 (2011) 287–320. [36] J. Kwon, Y. Park, K.Y. Huh, Combust. Flame 164 (2016) 85–98. [37] S. Lapointe, B. Savard, G. Blanquart, Combust. Flame 162 (9) (2015) 3341–3355. [38] Y. Marzouk, A. Ghoniem, H. Najm, Proc. Combust. Inst. 28 (2) (2000) 1859–1866. [39] Y. Marzouk, A. Ghoniem, H. Najm 41 (4) (2003) 641–652.
