SENSITIVITYOFIGNITIONSEQUENCESTOSPARKINGTIMEINA TWO-PHASE FLOW INDUSTRIAL BURNER

A. Eyssartier, L.Y.M. Gicquel and B. Cuenot

CERFACS, 42 Av. G. Coriolis 31057 Toulouse cedex, France

Abstract Ignition sequence of two-phase flows is a recent and on-going research subject of great interest to the aeronautical industry. Indeed, understanding such a transient flow is a critical issue for the next generation of gas turbine engines which need to successfully ignite or re-ignite in very hostile condi- tions. Ensuring full ignition in given conditions is of course a pre-requisite for proper operation of the engine. This is however not mastered by engineers who still rely on expensive test benchmarks in or- der to validate new designs. Numerically, only fully unsteady numerical approaches can provide partial descriptions of such transient phenomena and only Large Eddy Simulation seems accessible. However, two-phase flow modeling is in this case implied and detailed validations are required. In order to demon- strate the capacity of the approach, steady and transient ignition sequences have been proposed in this work. Preliminary results prove the mono-disperse Euler-Euler LES reacting formalism to be adequate for steady reacting conditions. Following on this observation, sensitivity to the flow state at the time of sparking are evidenced in the following by producing eleven distinct LES sequences of an ignition phase.

1. Introduction Ignition, and in particular high altitude re-ignition, are critical processes in aeronautical engines. Ef- ficient and reliable ignition is a crucial point for the certification of engines and innovative systems such as laser ignition devices are actively developed. Ignition is however a complex transient phenomenon, not yet fully understood and controlled. Experiments show that ignition may be successfull or fail in the same operating conditions, and an ignition probability is usually used for engine design [1, 2, 3, 4, 5]. To understand the process of ignition in aeronautical engines, numerical studies have been essentially applied to gaseous ignition [6, 7] and less work address two-phase ignition. Ballal and Lefebvre [8, 9] proposed an extension of their analytical study on gaseous ignition to two-phase flow, leading to the de- termination of a minimum energy criterion. A comparison with experimental data confirms that global tendencies are captured by their analysis and confirms that ignition is strongly influenced by the droplet size. This study is however limited to a simple and ideal case and misses other processes occurring in a real engine. The complexity of ignition in a real engine was first illustrated by Boileau et al. [10] who simulated the ignition of a full annular burner usin Lage Eddy Simulation (LES), which has proven its efficiency to compute combustion chambers [11, 12, 13, 14, 15]. The ignition sequence develops in several steps, from the first flame kernel close to the igniter, to the stabilization of a flame on one injector and its propagation towards the neighboring burning sectors. Based on an similar modeling procedure but applied to an experimentally diagnosed configuration, the first phase of the previous demonstration is investigated here. The primary objective is first to better understand the importance of the local flow state at the location of the energy deposit and to illustrate the capability of the two-phase flow solver to reproduce their effects on the success or failure of the ignition Figure 1: Geometry of the MERCATO burner. Black cross: spark location for ignition sequences. phase. Prior to this aim, a validation in a stationary reacting two-phase flow condition of the burner is produced to assess the modeling capacity in such conditions.

2. Steady state two-phase flow flames The target combustion chamber is a simple rectangular box with optical accesses (Fig. 1) diagnosed in transient and stationary conditions at ONERA. The fuel injector is typical of aeronautical systems and is made of one swirler with a centered liquid atomiser forming a hollow cone spray. The case presented here burns kerosene at a global equivalence ratio of 0.95. Kerosene is modeled in the computations by a surrogate and the 2-step chemical kinetic scheme from Franzelli and Riber (2010) [16]). The fresh gas and liquid are both injected at 285 K at atmospheric conditions. Mass flow rates of air and liquid fuels are respectively 35.5 g/s and 2.26 g/s. LES of the non-reacting two-phase flows were previously performed and validated against measure- ments [17]. The sub-grid scale model is WALE model [18] associated with no-slip adiabatic walls. Characteristic boundary conditions, NSCBC [19], are used for inlet and outlet sections. The two-phase flow is tackled by a mono-disperse Eulerian or a poly-disperse Lagrangian models. Both approaches have been applied to the dispersed phase and gave similar results for the mean flow [20]. Mean velocity profiles obtained numerically and measured for the steady reacting case are presented on Fig. 2. LES matches the measurements, the main differences are for gaseous velocity profiles close to the fuel in- jection. According to the profiles at 10 mm and 26 mm, it seems that the central recirculation zone is slightly upstream in the LES than in the experiments. Excellent agreement is observed for the mean liq- uid phase profiles confirming the potential of the Euler/Euler approach for reacting flows. Note that the lack of data at 116 mm in Fig. 2b) is due to the very low concentration of liquid that is almost evaporated at this position and hence not captured experimentally. Fig. 3 presents the mean reaction rate of the fuel oxidation reaction. The flame is mainly anchored at the combustor end-wall and has a static position for few millimeters; the velocity fluctuations are very low in this zone. On the contrary, the upstream limit of the central recirculation zone (delimited by the black line on Fig. 3) is submitted to quite large velocity (and equivalence ratio) fluctuations which lead to a more diffused flame brush. (a) Gaseous phase (b) Liquid phase

Figure 2: Mean axial velocity. Dotted line: experiments (from ONERA); solid line: LES

Figure 3: Mean reaction rate of the fuel oxidation reaction [mol/m3/s] in the x = 0 plane. Black line denotes 0 m/s axial velocity.

3. LES of ignition sequences Based on the previous validation of the LES modeling strategy, ignition sequences have been per- formed to gauge the numerical approach against the experimental findings of [4]. Several positions of the ignitor are tested. For LES, ignition is initiated by depositing energy [21] at one location: z = 56 mm and y = 57 mm. The relative position of this numerical spark ignitor is given in Fig. 4, where the impact of the central and corner recirculation zones on the droplet distribution is evidenced. The hollow cone spray injects locally mono-disperse droplets (mean diameter 60 µm) in the high shear zone around the central recirculation zone. Droplets go around the recirculation zone and partially evaporate before leaving the chamber through the exit. Note the very high droplet density in the region close to the Figure 4: Liquid volume fraction field in the x = 0 plane (white line denotes 0 m/s axial velocity). injector, Fig. 4.

∗ ∗∗ (a) t0 = 460ms (b) t0 = 462ms (c) t0 = 464ms

Figure 5: Snapshots of 3 characteristic ignition sequences 1ms after the beginning of the depot t0. The light gray isosurface denotes the Tgas = 1500K surface. The slice is the axial median plane colored by the velocity magnitude. *: failed ignition ; **: sharp ignition.

The ignition sequences have been performed with the energy deposition model [21] to mimic the spark, the thickened flame model and reduced chemical scheme for the combustion [10]. The energy deposit has an integrated value of 100 mJ and lasts 50 µs. In order to illustrate the importance of the flow state at ignition, 11 LES’s are produced. Each LES differs from the other ones by the initial flow condition at the sparking time, t0. For the demonstration the location of sparking is always the same. Typical snapshots corresponding to 3 different ignition times, are shown on Fig. 5. Clearly, the flame front is strongly influenced by the surrounding fluid as time proceeds toward full ignition or extinction of the burner. One has already begun to shrink (Fig. 5 a)) when another one has widely spread (Fig. 5 c)). A third one, Fig. 5 b), has an intermediate behavior, and potentially affected by turbulence (highly wrinkled). The temporal evolution of the total consumption rate is shown on Fig. 6 for all 11 tests. Very different evolutions are observed: from fast ignition to immediate failure, with intermediate behaviors where ignition starts more or less slowly and leads to complete ignition or failure. Depending on the time of sparking, the size of the hot gas kernel is very different. The temporal evolution of the liquid mass fraction field, not shown, illustrates that the initial kernel may experience the impact of dense liquid pockets [22]. These pockets come from the interaction between the Precessing Vortex Core (PVC) and the vortex shedding at the shear layer (Kelvin-Helmholtz instabilities). This leads to Figure 6: Ignition tests: time evolution of the total consumption rate (R ωdV˙ ) in the whole combustor sudden high evaporation rate which weakens the flame by both heat lose and decrease of the local flame speed (rich conditions). The last point seems to be of important in determining the success of ignition. With the wrong timing, the initial flame kernel may be too small to consume the evaporated fuel, which can accumulate and exceed the rich flammability limit resulting in local extinction and a failed ignition sequence. This variability underlines the difficulties to predict ignition even if the mean local conditions seem favorable. Likewise one single LES of an ignition sequence is clearly not enough to conclude on the behavior of the flame during this transient phenomenon. This variability seems to be enlarged by the liquid phase through mixing and effective equivalence ratio [23].

4. Conclusions LES of reactive two-phase flow in an ONERA configuration has been produced. First, stationary two-phase reacting LES predictions are compared with measurements, showing very good agreement for the mean velocity profiles for both phases. Second, several ignition sequences have been performed numerically. All mean operating conditions as well as the location of sparking being identical, all simu- lations differ by the time of sparking. Based on eleven different initial conditions at sparking, transients appear to be subject to large variabilities of the process. Sensitivity to the initial flow conditions at spark- ing time is here evidenced and initial investigations point to the variation of the equivalence ratio seen by the flame as a key parameter since it can reach the rich flammability limit. References [1] E. Mastorakos, Progress in Energy and Combustion Science 35 (2009) 57–97. [2] S. F. Ahmed, R. Balachandran, T. Marchione, E. Mastorakos, Combust. Flame 151 (2007) 366– 385. [3] T. Marchione, S. F. Ahmed, E. Mastorakos, Combustion and Flame 156 (2008) 166–180. [4] R. Lecourt, Ignition and Extinction results at ambient conditions, Technical Report, Timecop Re- port No. 2.2.1a, 2008. [5] A. Neophytou, E. Mastorakos, R. Cant, Combustion and Flame 157 (2010) 1071–1086. [6] P. Boudier, S. Henriot, T. Poinsot, T. Baritaud, in: T. C. Institute (Ed.), Twenty-Fourth Symposium (International) on Combustion, pp. 503–510. [7] T. Kravchik, E. Sher, Combust. Flame 99 (1994) 635–643. [8] D. R. Ballal, A. H. Lefebvre, Proc. R. Soc. Lond. A 364 (1978) 277–294. [9] D. R. Ballal, A. H. Lefebvre, in: T. C. Institute (Ed.), Eighteenth Symposium (International) on Combustion, pp. 1737–1747. [10] M. Boileau, G. Staffelbach, B. Cuenot, T. Poinsot, C. Berat,´ Combust. Flame 154 (2008) 2–22. [11] S. Roux, G. Lartigue, T. Poinsot, U. Meier, C. Berat,´ Combust. Flame 141 (2005) 40–54. [12] L. Selle, Simulation aux grandes echelles´ des interactions flamme-acoustique dans un ecoulement´ vrille,´ Phd thesis, INP Toulouse, 2004. [13] H. Forkel, J. Janicka, Flow, Turb. and Combustion 65 (2000) 163–175. [14] H. Pitsch, Ann. Rev. Fluid Mech. 38 (2006) 453–482. [15] V. K. Chakravarthy, S. Menon, Flow, Turb. and Combustion 65 (2000) 133–161. [16] B. Franzelli, E. Riber, M. Sanjose,´ T. Poinsot, Combust. Flame 157 (2010) 1364–1373. [17] R. Lecourt, Injection system two-phase flow caracterization (LDA-PDA), Technical Report, Time- cop Report No. 2.2.1c, 2009. [18] F. Ducros, F. Nicoud, T. Poinsot, in: ICFD, Baines M. J., 1998, pp. 293–300. [19] T. Poinsot, S. Lele, J. Comput. Phys. 101 (1992) 104–129. [20] J. Senoner, M. Sanjose,´ T. Lederlin, F. Jaegle, M. Garc´ıa, E. Riber, B. Cuenot, L. Gicquel, H. Pitsch, T. Poinsot., C. R. Acad. Sci. Mecanique´ 337 (2009) 528–538. [21] G. Lacaze, E. Richardson, T. J. Poinsot, Combust. Flame 156 (2009) 1993–2009. [22] A. Eyssartier, L. Gicquel, T. Poinsot, Assessment of the ignition/extinction modeling process for two-phase flows: FAUGA experiment, Technical Report, Timecop Report No. D3.2.7c, 2010. [23] A. Neophytou, Spark Ignition and Flame Propagation in Sprays, Ph.D. thesis, University of Cam- bridge, 2010.