OH- PLIF studies of acoustically forced swirl flames

Lean premixed combustors are highly susceptible to combustion instabilities, caused by coupling of heat release fluctuations with combustor acoustics. In order to predict the conditions under which these instabilities occur and their limit cycle amplitudes, understanding of the amplitude dependent response of the flame to acoustic excitation is required. This paper describes an experimental investigation of the amplitude dependent processes controlling the nonlinear heat release response of a swirling flame to harmonic excitation. In depth OH-PLIF studies were carried out at representative conditions to visualize the spatial dynamics of the flame and hence identify the key controlling physical processes and qualitatively discuss their key characteristics. These results illustrate that the flame response is not controlled by any single physical process but, rather, by several simultaneously occurring processes which are potentially competing, and whose relative significance depends upon forcing frequency, amplitude of excitation, and flame stabilization dynamics.

O H - P L IF S t u d i e s o f A c o u s t i c a l l y F o r c e d S w ir l F la m e s Sai K u m a r Thumuluru, Tim Lieuwen School of Aerospace Engineering, Georgia Institute of Technology, Atlanta, GA, USA 30332-0150 ABSTRACT Lean premixed combustors are highly susceptible to combustion instabilities, caused by coupling of heat release fluctuations with combustor acoustics. In order to predict the conditions under which these instabilities occur and their limit cycle amplitudes, understanding of the amplitude dependent response of the flame to acoustic excitation is required. This paper describes an experimental investigation of the amplitude dependent processes controlling the nonlinear heat release response of a swirling flame to harmonic excitation. In depth OH-PLIF studies were carried out at representative conditions to visualize the spatial dynamics of the flame and hence identify the key controlling physical processes and qualitatively discuss their key characteristics. These results illustrate that the flame response is not controlled by any single physical process but, rather, by several simultaneously occurring processes which are potentially competing, and whose relative significance depends upon forcing frequency, amplitude of excitation, and flame stabilization dynamics. INTRODUCTION This paper describes an investigation of the physical processes that control the dynamics of acoustically excited swirl flames. These include fluid mechanic instabilities, flame stabilization dynamics, and turbulent flame brush development. This work is motivated by the propensity of lean, premixed combustors to instabilities [1,2]. These instabilities arise as a direct consequence of the coupling between the combustion processes and the acoustic modes of the chamber. A key unsolved problem lies in understanding how the flame responds to flow perturbations, and how this response varies with perturbation amplitude. Such an understanding is needed in order to predict the conditions under which instabilities occur and their limit cycle amplitude. In general the flame response increases linearly with perturbation amplitude over a certain range of low amplitudes, before exhibiting nonlinear behavior at high amplitudes [3,4], e.g., see Figure 1. Further discussion of this issue can be found in Refs. [5,6,7, 8,9,10,11,12]. Measurements have shown that the flame response to excitation can exhibit a variety of complex, nonlinear behaviors. These include the flame's heat release response exhibiting multiple saturating behaviors or a non-monotonic dependence upon amplitude - several representative results are illustrated in Figure 1. Figure 1 (a) describes the variations in the flame response at the flow velocity corresponding to Re = 21,000. Note the initial linear increase in flame response with flow forcing, followed by saturation at chemiluminescence fluctuation levels ranging from 40-100% of the mean. In some cases, the saturation plateau ends with further increases in amplitude and the flame response increases again. Presumably, this is followed by saturation at even higher amplitudes which was generally not attainable in our system - although this is apparently what is happening in the 210 Hz case. These results illustrate that the forcing frequency strongly impacts both the linear gain, as well as initial saturation amplitude. The non-monotonic behavior of the flame response was also observed at higher flow velocities (Figure 1 (b)). For example, in the Re=44,000, 140 Hz case the saturation plateau ends with a 1 further increase in amplitude. More complex results were obtained at other conditions, such as the 210 Hz and the 270 Hz (Re=44,000 case), where the flame response first increases, and then decreases before increasing once again. This decreasing response is interesting as it shows that increasing forcing amplitude causes a decreasing response at the forcing frequency. 0.5 1 r__________ 140 H 210 H 270 H 420 H 08 ■ 0.6 o o uo □ 0 * o 0 * o o 0.4 □ 0.3 o 0 o O 0.4 0.2 a o R°° 0- 0 0.2 0 0.4 0.6 0.8 1 1.2 0 O 0OooOo O nO. #□ □ ® ++ VJ+ 0.1 02 140 Hz 210 Hz 270 Hz 420 Hz 0.1 □ * * * 0.2 0.3 0.4 0.5 U '/ U U /U (a) (b) Figure 1 : Dependence of fluctuating CH* chemiluminescence upon acoustic excitation amplitude at flow conditions of (a) Re =21,000 and (b) Re =44,000 [13,14]. These data suggest that the flame response is not controlled by any single key physical process, but rather several processes occurring simultaneously which are potentially competing, and whose relative significance depends upon forcing frequency, amplitude of excitation, and flame stabilization dynamics. This latter point is the key motivator for the present work - identifying and cataloguing the key physical processes is required before rational sense can be made of these results. The objective of the current study is to present an analysis of phase locked OH PLIF images to identify these key physical processes, qualitatively discuss their key characteristics, and illustrate their role in controlling the flame's amplitude dependent response characteristics. EXPERIMENTAL FACILITY A schematic of the experimental setup is shown in Figure 2 and Figure 3 and consists of an annular flow passage (4.5 mm ID and 15.9 mm OD) centerbody, a 400 axial swirler (swirl number, S ~ 0.65), followed by rapid expansion into the combustor (70 mm diameter). The Reynolds number (Re) is based on the incoming cold flow, the mean nozzle exit velocity and the center body diameter. Fuel (natural gas) and air were premixed (^ = 0.8) upstream of a choke point to prevent the occurrence of fuel/air ratio oscillations. Acoustic oscillations were excited using two drivers mounted upstream of the combustor. Velocity oscillations were calculated using the two microphone method. Heat release fluctuations shown in Figure 1 were characterized from measurements of the global CH* and OH* chemiluminescence with photomultipliers (PMT) fitted with a 10 nm bandwidth filter centered at 430 nm and 310 nm, respectively. The field of view of the photomultipliers was such that it covered the entire combustion region. As the relative fluctuations in CH* and OH* were virtually identical [14], only the CH* results are shown. 2 Computer ICCD camera for PLIF UV Beam : 281.4 nm R1(9) line : OH (1,0) Band Pulsed Nd:YAG Laser a t 5 3 2 n m Pressure Taps Loudspeakers T u n a b le SHG Dye Laser O u tp u t at 57 0 nm _r -Air + Natural Gas Figure 2: Schematic of the combustor along with the OH-PLIF setup. Phase locked OH Planar Laser Induced Fluorescence (PLIF) was used to visualize the spatial dynamics of the flame. This system is detailed in Thumuluru et al. [13] and consists of a cluster of an Nd: YAG laser, a dye laser, and a high-resolution ICCD camera. The frequency-doubled output from the dye laser was tuned near 281.4 nm to pump the R1 (9) transition of the A 1Z X2U (1, 0) band. OH fluorescence integrated from 300-380 nm is captured by the ICCD camera through a WG-305 and UG-11 Schott glass filter. Since the frequency of imaging was limited by the Nd: YAG laser to 10 Hz, successive phase-locked images were actually obtained several cycles apart from each other. Fifty flame images per phase were obtained at six equally spaced phase angles, at a resolution of about 270 ^m / pixel. The captured images were corrected for background noise and for beam profile in-homogeneities and filtered using a 3-pixel width Gaussian filter. The flame edges were captured manually. Determination of the flame front location was straightforward in most cases, except in regions where the flame is close to the combustor walls and/or where there is back mixing of OH laden products. In the latter case, determination of whether the interface between regions of high and low OH levels corresponded to a flame or a non-reacting, product-reactant interface was determined from the OH gradient. Gradient values that fell substantially below typical gradient levels for regions with known flames were not marked as flame fronts. FLOW AND FLAME CHARACTERISTICS In order to provide some context for interpreting the results of this study, a brief overview of the key flame and fluid mechanical characteristics of this system are provided below. In a time 3 averaged sense, the flow field consists of four main regions*, see Figure 3: (1) the outer recirculation zone (ORZ), a toroidal recirculating regime generated by the rapid expansion of the nozzle into the combustor, (2) the inner recirculation zone (IRZ, also referred to here as the vortex breakdown bubble), due to vortex breakdown accompanying the swirling flow, (3) the high velocity, annular fluid jet that divides these regions, and (4) two annular shear layers that divide the ORZ and annular jet, and the IRZ and the annular jet. Visualizations of these features can also be seen in Huang and Yang [15] and Santavicca et al. [16]. Each of these flow features have their associated fluid mechanic instabilities which are highlighted in the next section. Velocity perturbations most visibly affect the flame by modulating the velocity of the annular jet, causing the flame to surge back and forth. In addition, they apparently cause the IRZ recirculation strength to fluctuate and to move axially back and forth in location. The perturbations exhibit a similar effect upon the ORZ, apparently causing its strength and axial length to fluctuate. Finally, they influence the shear layer by synchronizing its rollup due to the Kelvin-Helmholtz instability. (d ) (e) Figure 3: Time averaged base flow field (a) and flame structures (b-e). (dimensions are in mm.) The flame itself can be spatially distributed in four basic configurations, depending upon fuel/air ratio and flow velocities, see Figure 3(b)-(e). As shown, it can be stabilized in the shear region at the inner centerbody or outer centerbody or downstream by the vortex breakdown bubble. How the flame is stabilized plays a crucial role in determining which physical processes control its dynamics. For example, the configuration shown in (d) is clearly affected by the dynamics of both shear layers, while that in (c) is presumably lesser affected by the shear layer from the *In addition, there is a small recirculation zone downstream of the centerbody. 4 centerbody. Similarly, the dynamics of the central portion of the flame is strongly influenced by vortex breakdown bubble dynamics in configurations (b) and (c), while configurations (d) and (e) are presumably less affected. As will be shown below, the time averaged stabilization location can vary with perturbation amplitude, implying that one set of fluid mechanic processes are important at low amplitudes, and another at higher amplitudes. FORCED FLAME RESULTS From the discussion in the earlier section, it can be seen that the key physical processes controlling the response of the flame to acoustic oscillations are jointly controlled by fluid mechanics and the flame configuration. Based upon analysis of several thousand instantaneous OH PLIF images, four key processes have been identified: (1) annular jet fluctuations, (2) turbulent flame brush development, (3) flame stabilization, and (4) excitation of and flame interactions with wake mode, jet column, swirl flow, and Kelvin-Helmholtz fluid mechanic instabilities. These are separately addressed in the sub-sections below. Some of these processes appear in isolation, but they usually appear in combination with others. For example, oscillations of the jet and inner recirculation zone usually occur simultaneously however, the complication arises in that they may not have the same phase or amplitude dependence. Fluctuating annular jet velocity To understand the physics underlying the nonlinearities present in the flame response and to trace the changes in the flame structure, OH PLIF images were analyzed. It should be noted that the interpretation of these images is limited by the fact that the flow is actually highly three dimensional with the flame coming in and going out of the images. Also, the consecutive phase images are actually taken several cycles apart. Because of sufficient swirl in the flow, vortex breakdown occurs, leading to flow reversal which, in a time-averaged sense, is equivalent to placing a blockage in the flow. This blockage requires that the axial flow remain in an annular, high speed fluid column, between the IRZ and ORZ. Acoustic excitation causes an oscillatory flow velocity in the jet between the IRZ and ORZ, causing a fluctuation in flame length and, therefore, heat release due to the expanding and contracting of the flame surface area. A typical set of PLIF images of a flame at six phases of an acoustic cycle and forced at various forcing amplitudes is shown in Figure 4. The forcing frequency here was 140 Hz at a Reynolds number of 44,000, (the corresponding flame response is shown in Figure 1 (b)). The phase angles correspond to the difference in phases between the heat release and the nozzle exit velocity fluctuations. In each of these images, the red lines represent the location of the flame front (marked by the steepest change in the gradient). The nozzle exit is located at the bottom of the image - note that the nozzle image shown in the picture is translated downward from its actual position. Also, we have only highlighted the main flame features and have not outlined many of the smaller islands that are also present. At this condition, the flame is attached on the outer annulus but not to the inner one, similar to what is shown pictorially in Figure 3 (c). Figure 4 (a) and (b) describe the flame structure at low forcing amplitudes; i.e. in the linear regime of flame response. Figure 4 (c) & (d) are images of the flame under large forcing amplitudes, and correspond to the cases, where the flame response saturates, and when the flame response starts rising once again after the saturation regime, respectively. In general, from Figure 4, it can be seen that the flame is surging back and forth axially, due to the oscillating flow velocity in the annular jet region shown in Figure 3 (a). This phenomenon is well illustrated in Figure 4 (d) - at various phase angles, the flame in the ORZ and "Jet" region is moving up and down - hence resulting in fluctuations of the overall flame length. The overall level of fluctuation of the flame length is seen to grow with forcing amplitude. Moreover, the size of the ORZ is also modulated, as is evident from the oscillating angle of the 5 flame front facing the ORZ (e.g., compare the images at 180 and 300 degrees). Further, at higher forcing amplitudes, the location of the flame at the centerline is oscillating axially, suggesting a corresponding fluctuation in location of the vortex breakdown bubble. Figure 4: OH PLIF images showing flame structure at the forcing frequency of 140 Hz, Re=44,000 and amplitudes of (a) u '/u0 = 0.07, (b) u '/u0 = 0.1, (c) u '/u0 =0.17 & (d) u '/u0 =0.24. Turbulent Flame Brush Development Referring back to the flame configurations shown in Figure 3, it can be seen that three of them are attached in the shear layer at the centerbody and/or nozzle exit. Because the flame moves very little at this point, the turbulent flame brush is quite thin in the immediate vicinity of the nozzle exit, but grows in thickness downstream. Turbulent flame parameters, such as turbulent flame speed and brush thickness consequently vary with downstream distance [17]. Moreover, the flame length varies throughout a forcing cycle. A related point was noted by Sathiah and Lipatnikov [18], who modeled the flame response to 6 low amplitude perturbations while incorporating the changes in the turbulent flame speed along the developing flame. This effect may be more significant during large amplitude oscillations, i.e., in the nonlinear flame response regime. During low amplitude oscillations, the flame length and therefore brush thickness and flame speed change very little over the cycle. During high amplitude oscillations, the flame length can oscillate considerably, implying that the maximum flame brush thickness varies considerably throughout the cycle. To illustrate, a number of instantaneous realizations of the flame sheet at two different phases of the forcing cycle are overlaid in Figure 5. Figure 5: Overlay of 10 instantaneous flame realizations at u '/u0 = 0.83 showing the variation in flame lengths and brush thickness at two phases of (1) 135o and (2) 180o (Re=21,000, f= 130 Hz) If the turbulent flame speed grows with downstream distance, then this introduces a nonlinear saturation mechanism. As the flow velocity increases, the flame lengthens, but not as much as it would have, if the flame speed remained constant. In addition, as the flow velocity drops, the flame shortens, but also less than if the flame speed were constant. Amplitude Dependent Flame Stabilization In the results discussed in the prior sections, the flame stabilization locations essentially remained fixed throughout the forcing cycle and across the range of amplitudes. However, the flame configurations shown in Figure 3 need not remain constant at a given operating condition, but can change with disturbance amplitude. This mechanism was noted in a previous study [13] and shown to be responsible for saturation of the flame response. At other conditions, the flame's attachment point can exhibit amplitude dependence. This can result in a situation where the flame resembles that shown in Figure 3 (d) for low amplitudes and Figure 3 (c) for high amplitudes. Alternatively, the attachment point can oscillate between these configurations throughout the cycle. An example of the latter scenario is shown in Figure 6. At low forcing amplitudes (Figure 6(a)), the flame remains attached to the centerbody through the acoustic cycle, while at high forcing amplitudes (Figure 6 (b)) the flame blows off the inner centerbody and moves downstream. However, it does not blow out of the combustor, but restabilizes at what is presumably the leading edge of the IRZ. Since the flame is lifted off, the two flames that are present at low amplitudes are merged together into one flame at high amplitudes - the flame area is lower in this case than it would be if the flame were attached. This unsteady liftoff, and consequent reduction in flame area, is a very significant and abrupt source of nonlinearity [10]. 7 (a) (b) Figure 6: PLIF images showing oscillation in flame stabilization point and unsteady liftoff at the forcing frequency of 410 Hz, Re=21,000 and u '/u0= 0.6. Flame interactions with fluid mechanic instabilities The fluid mechanics of a swirl combustor is complex, due to several inherent flow instabilities which are simultaneously present and described below. Swirling Annular Jet Column and IRZ Dynamics The swirling jet column possesses a range of instabilities ranging from helical disturbances to strong spiral and axisymmetric vortex breakdown states (19,20,21). While the influence of these helical and spiral modes is difficult to elucidate from these planar images, the axisymmetric rollup of the flame into the IRZ is clearly evident under certain conditions, such as shown in Figure 7 (a). In prior publications [10], it was shown that this rollup was strongly amplitude dependent, leading to saturation of the flame response at high disturbance amplitudes. (a) (b) Figure 7: PLIF images showing (a) vortex rollup in IRZ at 130 Hz, Re=21,000, u '/u0 = 0.9 and (b) coherent structures in the flame (1) Re=21,000 and (b) Re=44,000 8 In addition, the location and width of the IRZ oscillates axially. This has important influences upon the flame dynamics for situations where the flame is IRZ stabilized, as in the two configurations shown in Figure 3 (b) and (c). In general, the dynamics of the bubble itself are not well understood (even in the non-reacting case); and our analysis indicate that it has a complex frequency-amplitude dependence. Shear Layer Dynamics The shear layers between the annular jet and IRZ, and the IRZ and ORZ are unstable and subject to rollup, due to the Kelvin-Helmholtz instability. From non-reacting flow studies, this rollup phenomenon is known to organize itself spatially and temporally when periodically forced [22]. As shown in Figure 7 (b), this rollup phenomenon can occur in the shear layers between the annular jet and IRZ, the annular jet and ORZ, or both. Images of these vortical structures in the shear layer can also be seen in the computations of Yang [15] and Najm and Ghoneim [23], who also present analyses of their temporal characteristics. ORZ Dynamics In addition, the rapid expansion at the dump plane causes wake type mode instabilities, as discussed by Najm and Ghoneim [24]. As shown in their simulations, the recirculating flow in the ORZ periodically detaches and travels downstream. This strongly recirculating flow entrains the annular jet under certain circumstances, causing it to roll-up in the outward direction, see Figure 8. As in the related case for roll-up into the IRZ, this process is amplitude dependent and grows in prominence with increasing disturbance amplitude. Even in cases where the ORZ does not cause flame-rollup, its size is apparently modulated through the cycle. This is evident in many images at high flow velocities that show a fluctuating angle of the outer rim stabilized flame, which is riding along the shear layer between the annular jet and ORZ, e.g., see Figure 4. Figure 8: PLIF images showing (a) vortex rollup in ORZ at 130 Hz, Re=21,000, u '/u0 = 0.9 and (b) coherent structures in the flame at 130 Hz, Re=21,000, (1) u'/u0 = 0.2 and at 270 Hz, Re=44,000, (2) u'/uo = 0.05. 9 Concluding Remarks These results show that the flow moves the flame around in complex ways, involving the annular jet, IRZ, and ORZ. In addition, these results also illustrate that the flame response is not controlled by any single physical process but, rather, by several simultaneously occurring processes which are potentially competing, and whose relative significance depends upon forcing frequency, amplitude of excitation, and flame stabilization dynamics. Our ongoing work is aimed toward using this improved understanding of the flow-flame phenomenology to develop better predictive capabilities. References [1] [2] [3] [4] [5] [6] Lee, J., Santavicca, D., J. Prop. Power, 19(5) 2003 735-750. Richards, G., Straub, D., Robey, E., J. Prop. Power, 19(5) (2003) 795-810. Lieuwen, T., J. Prop. Power, 18(1), (2002) 61-67. Dowling, A.P., J. Fluid Mech., 346 (1997) 271-290. Dowling, A.P., J. Fluid Mech., 394, (1999) 51-72. Peracchio, A.A., Proscia, W.M., Journal of Engineering for Gas Turbines and Power, 121 (1999) 415-421. [7] Baillot, F., Durox, D., Prud'homme, R., Combust. Flame, 88 (1992) 149-168,. [8] Durox, D., Baillot, F., Searby, G., Boyer, L., J. Fluid Mech. 350 (1997)295-310. [9] Bourehla, A., Baillot, F., Combust. Flame, 114 (1998) 303-318. [10] Bellows, B.D., Bobba, M. K.,. Seitzman J.M , Lieuwen. T., Proc. Combust. Inst. 31(2) (2007) 3181-3188. [11] Kulsheimer, C., Buchner, H., Combust. Flame, 131 (2002) 70-84. [12] Balachandran, R., Ayoola, B.O., Kaminski, C.F., Dowling, A.P., Mastorakos, E., Combust. Flame, 143 (2005) 37-55. [13] Thumuluru, S.K., Bobba M.K., Lieuwen, T., ASME paper, GT 2007-27932, 2007. [14] Thumuluru, S.K., Ma, H., Lieuwen, T., AIAA 2007-845, 2007. [15]Huang, Y and Yang, V.., Combust. Flame, 136 (3) (2004) 383-389. [16]Venkataraman, K.K., Preston, L.H., Simons, D.W., Lee, B.J., Lee, J.G. and D.A. Santavicca, Journal o f Propulsion and Power, 15 (6) (1999) 909-918. [17] Lipatnikov, A.N. and Chomiak, J, Prog. in Energy and Combustion Science, 28 (1) (2002) 1­ 73. [18] Lipatnikov, A.N. and Sathiah, P. Combust. Flame, 142 ( 2005) 130-139,. [19] Sarpkaya T. Phys. Fluids 7(10) (1995) 2301-2303. [20] S. Leibovich and K. Stewartson, J Fluid Mech 126 (1983) 335-356. [21] Althaus, W., Bruecker, Ch. & Weimer, M. 1995. in Fluid Vortices (ed. S. I. Green), pp. 373­ 426. Kluwer. [22] C Ho, P Huerre, Annual Review of Fluid Mechanics, (1984) (16) 365-422. [23]Najm H.M., and Ghoniem A.F., AIAA Journal 29 (6) (1991) 911-919. [24]Najm H.M., and Ghoniem A.F., AIAA Journal of Propulsion and Power, 10 (6) (1994) 769­ 776. 10