Premixed Flame Kinematics in an Axially Decaying, Harmonically Oscillating Vorticity Field Dong-hyuk Shin*, Santosh Shanbhogue*, Tim C. Lieuwen† School of Aerospace Engineering, Georgia Institute of Technology, Atlanta, GA 30332-0150 This paper describes experiments and analysis of the spatio-temporal dynamics of a bluff-body stabilized flame responding to excitation from a harmonic velocity field. The dependence of the flame response is shown to exhibit two fundamentally different dependencies on underlying parameters, referred to here as “interference dominated” and “dissipation dominated” behaviors. Which behavior dominates depends upon the phase velocity and decay rate of the convecting disturbance and the flame angle. The magnitude of the flame sheet wrinkling response initially increases with downstream distance, reaches a maximum and then decays. This basic envelope of flame response can also exhibit short length scale undulations, associated with wave interference effects. The bluff body nearfield response is controlled by linear processes, with a magnitude that grows downstream at a slope proportional to the local normal velocity of excitation. The peaking in response and far-field behavior is influenced by both linear and nonlinear effects, and controlled by phase interference, decay of the excitation, and kinematic restoration processes. Nomenclature v 0 ’ ’peak ’K ’ f0 G K SL u0 uc uf ut,0 u’ ua’ un’ uref’ v0 * † = = = = = = = = = = = = = = = = = = = = = = = = Projection of mean tangential velocity in axial direction Slope of mean flame front position Excitation velocity decay rate Axial velocity fluctuation ratio at the flame base, uref’/u0 Transverse velocity fluctuation ratio at the flame base, v’x=0 /u0 Convective wavelength of harmonically oscillating disturbance, u0 / f0 Mean flame angle Instantaneous flame front position Mean flame front position Fluctuation of flame front Maximum value of ’ Magnitude of ’ at x=xK Magnitude of homogeneous part of ’ Excitation frequency Scalar function representing flame position Ratio of mean axial velocity to disturbance propagation velocity, u0/uc Laminar Flame speed Mean axial velocity Disturbance propagation velocity Axial velocity just upstream of flame Mean velocity tangential to mean flame front Fluctuation of axial velocity Acoustic velocity magnitude of incident disturbance Fluctuation of velocity normal to the mean flame front Magnitude of fluctuating axial velocity at the flame base. Mean transverse velocity Graduate Research Assistant, Student Member AIAA Associate Professor, Associate Fellow AIAA 1 American Institute of Aeronautics and Astronautics 092407 vf v’ xpeak xK x = = = = = Transverse velocity just upstream of flame Fluctuation of Transverse velocity Axial location where ’ is maximum Axial location where interferences are in-phase Axial location where velocity fluctuation decays to e-3 of its initial value I. Introduction T he objective of this research is to understand the flame dynamics of acoustically forced, two dimensional flames. The dynamics of such acoustically forced flames involve complex interactions between gas expansion effects and vorticity dynamics induced by both shear and combustion, as highlighted by, e.g., Poinsot and Veynante1, Schadow et al.2, Rogers & Marble3, Cetegen4, and Coats5. This work is particularly motivated by the problem of combustion instabilities3. In many such instances, it is known that vortical structures excited by harmonic acoustic oscillations interact with the flame, causing its heat release to oscillate. This is illustrated in Figure 1, which depicts instantaneous flame locations and vorticity fields of an acoustically excited flame. a) b) c) d) Figure 1. Instantaneous images of the flame front location and the underlying vorticity field. a) without excitation, b-d) with increasing amplitudes of excitation. (reproduced from Shanbhogue et al.6) A number of prior studies have characterized the interaction of flames with harmonic waves arising due to both acoustic7 and convecting, vortical disturbances8-11. The dynamics of the flame are controlled by flame kinematics, i.e., the propagation of the flame normal to itself at the local burning velocity, and the flow field that the flame is locally propagating into. This is mathematically described by the G equation1: G u G S L | G | t (1) In this equation, the flame position is described by the parametric equation G ( x , t ) 0 . Also, u u ( x , t ) and SL denote the flow field just upstream of the flame and laminar burning velocity, respectively. In the unsteady case, the flame is being continually wrinkled by the unsteady flow field, u . The action of flame propagation normal to itself, 2 American Institute of Aeronautics and Astronautics 092407 the term on the right side of Eq. (1), is to smooth these wrinkles out through “Huygens propagation” / “kinematic restoration”. A wrinkle created at one point of the flame due to a velocity perturbation propagates downstream and diminishes in size due to kinematic restoration. Indeed, the interaction between the excitation (acoustic/vortical flow oscillations) and the damping (restoration property of the flame) can lead to a range of effects depending upon flame stabilization and the relative values of the flow oscillations and flame speed. Most recent theoretical analyses of flame dynamics in oscillating flows have focused upon predicting the “global” unsteady heat release; i.e., the spatially integrated temporal response of the heat release to the perturbations12. Characterization of the detailed spatio/temporal dynamics of the flame has received considerably less attention. Baillot and co-workers did make such comparisons between experiments and theory (using solutions of the G-equation13) for Bunsen flames, where they showed good agreement between predicted and measured flame shapes. In addition, Shanbhogue et al.6 presented experimental measurements of the spatio/temporal gain and phase response of bluff body stabilized flames to harmonic forcing. A typical set of their results are shown in Figure 2. The convective wavelength of the flame front disturbances, 0=uo/fo, equals the distance a disturbance propagating at the mean flow velocity travels in one acoustic period. The plot on the left shows the spectrum of the flame response at several axial locations. This spectrum was determined from the time variation of the flame position, (x,t) (see Figure 4), at each axial location. The envelope of the flame response at f = fo is also drawn. Figure 3 also plots the gain/phase response at several other conditions at only the forcing frequency. From these plots, we can observe several generic features of the flame response at the forcing frequency: 1) Very low amplitude of flame fluctuation near attachment point, with subsequent growth downstream, i.e.,. |’| ~ x in the bluff body nearfield 2) A peak in amplitude of fluctuation, ’=’peak, at some axial location, x=xpeak 3) Decay in amplitude of flame response farther downstream, i.e., ’(x,f0) ~ 1/x 4) Shorter wavelength modulation in flame response manifested in “ripples” in gain curve, typically downstream of peak location 5) Approximately linear phase-frequency dependence, implying a nearly constant axial convection speed of the flame sheet disturbances As will be shown in this paper, these features reflect the excitation of flame wrinkles by the velocity field, the decay of these wrinkles by kinematic restoration, interference phenomenon between wrinkles on the flame excited by different mechanisms, and the convection of these wrinkles by the velocity tangential to the flame front. Briefly considering the flame response at frequencies other than the forcing frequency in Figure 2a, note that the spectrum also exhibits a monotonic increase in broadband fluctuations with downstream distance. This reflects the random flapping of the flame brush, which increases in magnitude with downstream axial distance, as is clearly seen for the unforced case (see Figure 1a). a) b) Figure 2. a) Dependence of flame front fluctuation spectrum, ’(x, f ) upon axial location (u0 = 4.5 m/s, f0 = 300 Hz). b) Phase dependence upon normalized axial location, where x0 indicates first axial location where data was obtained (ua’/ u0 = 0.05, triangular bluff body, reproduced from Shanbhogue et al.6). 3 American Institute of Aeronautics and Astronautics 092407 a) b) Figure 3. Dependence of flame front fluctuation magnitude (left) and phase (right) at forcing frequency upon axial location for 2D circular bluff-body (x0 indicates the first axial location where data was obtained. Data were reproduced from Shanbhobue et al.14) Some of these flame features were discussed and explained by Shanbogue et al.15, who showed that the flame nearfield response is controlled by flame anchoring and its farfield response by kinematic restoration. However, the velocity model considered in this study had a uniform axial magnitude (i.e., it did not decay downstream) - the constant spatial magnitude velocity disturbance implies that the flame sheet continues to wrinkle and respond indefinitely far downstream, rather than decaying as experimentally observed. The goal of the present work is to further elucidate these flame-sheet dynamics by generalizing the prior model to include a velocity disturbance field that decays downstream. It will be shown that this model can capture the observed spatio/temporal characteristics of the flame sheet, both in the near and farfield. II. Kinematic Model A. Formulation The investigated geometry is a 2-D bluff-body stabilized flame, as shown in Figure 4. The instantaneous location of the flame surface is determined from the G-equation shown in Eq. (1). The principal assumptions made in this analysis are the following: (i) the flame is a thin interface, dividing reactants and products, (ii) the flame base remains fixed to the burner throughout the excitation cycle (i.e., no unsteady liftoff), and (iii) the flame speed is a constant. For small velocity fluctuation, the flame front position is single-valued (i.e., there is only one y axis flame front location value at any axial direction), an approximation that breaks down at high amplitude fluctuations as in Figure 1d). In this case, G(x,y,t) is written as: G x, y, t x, t y (2) By definition, G=0 at the flame front, so that represents the y location of the flame front, as depicted in Figure 4. Substituting Eq. (2) into Eq. (1), and non-dimensionalizing leads to: 2 u v SL 2 t x x (3) where u, v, SL, , x and t are normalized by u0, u0, u0, 0/ 0 and f respectively. The constants, u0, f0, 0=u0/f0, and are mean axial velocity, excitation frequency, mean flow wavelength and the mean flame aspect ratio (in the absence of excitation), respectively. The superscript “tilde” represents non-dimensional quantities. 4 American Institute of Aeronautics and Astronautics 092407 Figure 4. Schematic of bluff-body stabilized flame with its corresponding coordinates. red solid : instantaneous flame front, red dash : mean flame position, = aspect ratio of mean flame position. The velocity field is written as: u x, y, t u0 x, y u x, y, t v x, y, t v0 x, y v x, y, t (4) where u0 , u , v0 , v , and y are variables normalized by u0, uref’, u0, uref’, and 0/ , respectively and = uref’/u0. We consider both the linear and nonlinear response character of the flame. As such, is written as : x, t 0 x x , t O 2 (5) Note that ’ becomes O( ) when normalized by uref’/( f0 ). By expanding the solution in powers of 16, the zeroth and first order equations for the flame fluctuation are 2 u0 0 S L 2 0 v0 0 x x u0 t u 0 v 0 2 x x 2 0 x SL 0 x (6) (7) Insight into the solution of the first order solution can be obtained from the methods of characteristics17. Note that it is a sub-class of the more general equation: x, t t x, t x f x, t (8) where is assumed to be a constant here (but not in general). Define the function H(x,t) as: H x, t x f , d t x Then, the solution of Eq. (8), subject to boundary condition (0, t ) 0 is: 5 American Institute of Aeronautics and Astronautics 092407 (9) x, t t x 1 H 0, H x, t 1 (10) Particular part Homogeneous part This solution shows that the flame response at each spatial position and time is a superposition of two waves that propagate along the flame front. The first wave, denoted as the “Homogeneous part” is excited at the flame anchoring point and propagates along the flame sheet at an axial velocity of . The second wave propagates at the excitation phase velocity. Note that the particular solution is a convolution of the velocity at all points upstream of x while the homogeneous solution is only influenced by the velocity field near the attachment point. The spatial variation in amplitude and phase of these two solutions leads to interesting interference phenomenon, as will be described later. B. Numerical Method For the fully nonlinear case, Eq. (3) is solved numerically. Spatial derivatives are discretized using a Weighted Essentially Non-Oscillatory (WENO)18 scheme designed specifically for Hamilton-Jacobi equations. This scheme is uniformly fifth order accurate in regions wherein the spatial gradients are smooth and third order accurate in discontinuous regions. Derivatives at the boundary nodes are calculated using fifth order accurate upwinddifferencing schemes so that only the nodes inside the computational domain are utilized. A Total Variation Diminishing (TVD) Runge-Kutta scheme19, up to third order accurate, is used for time integration. The spatial and temporal grid size are 0/1000 and 1/(1000f0), respectively. Sensitivity studies performed with a grid ten times finer demonstrated that the difference between the two grid density results was less than 0.1%. The flame front perturbation at the forcing frequency is determined from the Fourier transform of the computed solution at f=f0. C. Velocity Model Several velocity fields were utilized in this study, ranging from very simple to more complex, experimentally measured fields. A model velocity profile was used for most of these calculations that captures many of the key features this paper is interested in exploring, i.e., the response of a flame to an axially convecting, decaying velocity disturbance: u f 1 e x cos 2 Kx t vf 0 (11) Note that while the full vector velocity field is specified here, the linear flame response is controlled by the scalar velocity field component normal to the flame, un’. The above velocity model is equivalent to the normal fluctuation: un sin e x cos(2 ( Kx t )) (12) III. Results and Discussion This section describes the processing controlling the key flame response features identified in the Introduction. The nearfield behavior is described in section A and the farfield behavior in section B. A. Near Field Behavior 1. Controlling Processes The magnitude of the nearfield flame response is inherently linear. This is due to the =0 boundary condition, which specifies that the flame fluctuation amplitude is very small in the vicinity of the bluff body, so that finite ( / x)2 2 term in Eq. (3) are negligible. Thus, a great deal of insight into this nearfield behavior can be gained from the analytically tractable linearized solutions of Eq. (7). amplitude effects, such as associated with the 6 American Institute of Aeronautics and Astronautics 092407 We first derive an expression for the slope of vs. x dependence in the flame nearfield. Defining the local mean flame angle () with respect to the axial direction (see Figure 4), note that the zeroth order flame front equation is given by: u0 sin v0 cos SL (13) In the same way, the first order perturbation Eq. (7) can be written as: cos u0 cos v0 sin u sin v cos 0 t x cos un ut ,0 (14) cos ut ,0 un 0 t x cos where un and ut ,0 denote a fluctuation of velocity normal to the flame and a mean velocity in the tangential direction of the flame, respectively. Very near the bluff body, the temporal fluctuation of flame response, / t is negligible due to the flame anchoring condition at the bluff body. Thus, the slope of vs. x response is: / x 1 un cos 2 ut ,0 (15) Note that division by occurs because and x are normalized by 0/ and 0, respectively. The cos2 term in Eq. (15) reflects the choice of a coordinate which is rotated with respect to that of the flame; i.e., cos / and x / cos describe the normal and the tangential coordinate, respectively, of a rotated coordinate system that is locally aligned with the mean flame front. Thus: cos / x / cos un ut ,0 (16) This equation shows that the fluctuation of the flame sheet in the direction normal to the mean flame front is equal to the ratio of the fluctuation perturbation velocity normal to the flame front and mean velocity tangential to the flame front. Note also that this equation makes no assumption about the nature of the velocity field. This equation describes the rate of growth of the flame front wrinkling in the very nearfield of the flame sheet. However, axial variations in perturbation velocity magnitude and non-negligible values of the / t term cause a departure from this behavior farther downstream. Flame behaviors in the nearfield of the bluff body that are valid over a larger axial range can be determined by specifying a velocity field and solving the linearized G-equation. 2. Numerical and Experimental results This section compares the results developed in Sec. 1 to data and fully nonlinear computed solutions using a model velocity profile. The basic velocity model shown below in Eq. (17) was used. The objective of this calculation was not to use what is necessarily a physically accurate or realizable velocity field, but only to compare the above described nearfield flame behavior with explicit calculations for various values of the perturbation and mean field parameters (details in Table 1). u f 1 cos 2 Kx t v f v0 v cos 2 Kx t 7 American Institute of Aeronautics and Astronautics 092407 (17) where, v represents the magnitude of fluctuating transverse velocity. The resulting solutions for the magnitude of , normalized by | un | / ut ,0 cos2 , are shown in Figure 5. Note that all of the results converge to the theoretical solution from Eq. (15) in the nearfield, indicated by the line y=x. The solutions diverge downstream due to the nonnegligible temporal fluctuation, / t and nonlinear effects. Figure 5. Dependence of flame response normalized by | un | / ut ,0 cos2 upon axial location. (K=1.2, details are listed in Table 1) Similar results were observed experimentally. Figure 6a plots the magnitude of the flame sheet fluctuations, normalized by ua’/u0. The flame angle( for each experiment is less than 15°, making the cos2 term close to unity, and therefore not included in the normalization. Again, note the collapse of these data to a common curve in the bluff body nearfield, and the divergence of these curves in the farfield, again demonstrating that the nearfield of the flame response is essentially linear. Figure 6. Measured amplitude response normalized by acoustic velocity amplitude upon normalized axial distance, 0=u0/f0 : (o) f0 = 150 Hz, ua’/u0 = 0.028/2.27, () f0 = 150 Hz, ua’/u0 = 0.01/2.27, () f0 = 180 Hz, ua’/u0 = 0.015/2.27, () f0 = 150 Hz, ua’/u0 = 0.021/3.37 (cylindrical bluff body, reproduced from Shanbhogue et al.6) Efforts were also made to quantitatively compare the measured and predicted slope, see also Shanbhogue et al.6 PIV velocity field measurements were used to estimate the velocity field near the flame. Obtaining good velocity field measurements requires excitation of the flame with sufficient magnitude perturbations. This causes the flame sheet to move around, requiring that the conditioned velocity perturbation just upstream of the flame must be used for the un’ specification – i.e., the velocity field cannot be determined at a fixed location, but at a location that moves with the mean flame front. Nonetheless, reasonable quantitative comparisons are possible in the flame nearfield, see Figure 7. The value of RHS of Eq. (15) was determined from the local measured flame angle and velocity field quantities. The results are shown in Figure 7b. Note that this ratio appears to asymptote to a value of | un ( x, f0 ) | / (ut ,0 ( x) cos 2 ) 0.15. This is in good agreement to the slope measured from Mie scattering images of the flame shape, illustrated in Figure 8 American Institute of Aeronautics and Astronautics 092407 | ( x 0, f 0 ) | 0.14. Note that the derivative was not divided by shown in Eq. (15) because these x calculations used dimensional variables. 7a, a) b) Figure 7. a) Dependence of flame front fluctuation spectrum, | ( x, f0 ) | / 0 upon axial location. b) Dependence of normal velocity fluctuation amplitude, | un ( x, f0 ) | / (ut ,0 cos 2 ) upon axial direction. (u0 = 4.5 m/s, f0 = 300 Hz, reproduced from Shanbhogue et al.6) B. Farfield Response We next turn to the flame response farther downstream, in order to consider the peaking, decay, and modulation of the flame front amplitude. As suggested by the data and computational results above, this farfield region is inherently nonlinear and, thus, less amenable to analytical insight. However, some insight into various features of the flame response can be gained by first considering the linear response. This analysis highlights some of the controlling physics which nonlinearity acts upon – specifically, it provides insight into the roles of interference of the waves propagating along the flame (see Eq. (11)), and the role of decay of the velocity field downstream. 1. Linear Response The first order flame response for the velocity model in Eq. (11) is x x x i 2 K t i 2 t i i e e e 2 K 1 i 2 K 1 i (18) where the parameter, 2 /(1 2 ) physically represents the projection of the tangential mean velocity in the axial direction. Note that the solution consists of two parts, as also shown in Eq. (10). The first term is the particular solution that is excited by the unsteady velocity field at all positions upstream of the position, x. The second term is the homogeneous solution, which is determined by the zero response of the flame at the anchoring point to the nonzero velocity disturbance at that point. Mirroring the underlying perturbation velocity field, the particular solution decays downstream and propagates with the same phase velocity as the excitation‡. In contrast, the homogenous solution has a constant axial magnitude and propagates at a phase velocity equal to the projection of the mean flow on the mean flame front. Note that these two disturbances propagate at the same phase velocity when K=1. This equation shows that the flame response in the axial coordinate x / is controlled by two parameters, K and whose influence is discussed next. Figure 8 illustrates the influence of the decay parameter, . First, notice that all five solutions converge to a common solution in the very nearfield, as described in the previous section. Next, note that all the results exhibit flame responses that grow, reach a local maximum, and then oscillates with constant local maximum value (for the ‡ Note that the particular solution is actually a superposition of infinitesimal, non-decaying waves (within the constant flame speed assumption, decay of flame wrinkling occurs at O(2) and higher – i.e., it cannot be captured with a linear analysis), whose superposition sums to a decaying field due to interference effects. 9 American Institute of Aeronautics and Astronautics 092407 no decay case, =0) or with decaying local maximum (for >0) until settling at a constant magnitude given by the homogenous solution. The highest case shows the lowest response due to the lower spatially integrated disturbance field. For the slowest decaying cases, such as = 0 and 0.5, the flame response has local minimum that are even lower than this =1 case, due to the negative interference between the waves propagating along the front. As the decay coefficient increases, this interference phenomenon plays a monotonically lesser role in the flame response, until the flame response is primarily controlled by the homogeneous solution (referred to as “dissipation dominated” behavior in this paper), as shown in Figure 8b. a) b) Figure 8. a) Dependence of | | upon axial coordinate for different values of velocity decay rate parameter, . B) Dependence of maximum value of | | upon velocity decay rate parameter, (K= 1.28). The above results illustrate the important role of interference phenomenon between the waves propagating along the flame sheet. Such interference occurs when K≠1, due to differing propagation speeds of the two waves along the flame sheet. As such, the value of K has important influences upon the solution. Figure 9 plots typical solutions of Eq. (18) for various values of K. As Ka increases from zero, the overall response increases, reaches a maximum at K = 1, and then decreases for the higher Ka. This can be seen in Figure 9b, which plots the , upon K. These plots clearly show dependence of the magnitude of the local maximum in flame response, peak the role of K and interference processes in controlling the spatial character of the flame response. Note also that when K=1, the flame response is not oscillatory, showing the role of interference phenomenon in inducing an oscillatory gain in flame response. The linear flame response is symmetric about |K-1|. This shows that the magnitude of the flame’s spatial response is controlled by the magnitude of the difference in phase velocity, and not their individual values. This symmetry does not extend into the non-linear regime. a) b) Figure 9. a) Dependence of | | upon axial coordinate for different values of disturbance convection velocity parameter, K b) Dependence of maximum value of | | upon K ( = 1.87). 10 American Institute of Aeronautics and Astronautics 092407 From these results we can see that the magnitude of the flame’s spatial response is controlled by interference and dissipation. In general, both of these processes control the flame response, but for certain parameter combinations one or the other is controlling. We will next exploit this point to derive insight into the scaling for the axial location in the maximum of flame response, x peak . As will be shown next, the flame response at approximate axial locations of 1/2(K – 1) and 3/is close to the maximum for the interference and dissipation dominant regimes, respectively. Note that in the dissipation dominated case, there is no strong peak as seen in Figure 7a and Figure 8a. Rather, x peak describes how fast the flame response approaches the maximum. Which regime is more dominant in controlling x peak is determined by which length is shorter. This implies the following criterion for dissipation vs. interference dominance: Dissipation dominant : 6 | K 1| (19) Interference dominant : 6 | K 1| (20) In the interference dominated regime, the maximum and minimum flame magnitude response occurs when the relative phase between the two flame front waves is 0º or 180º, respectively (the phase between these two waves is 180º at x =0, due to the =0 boundary condition). In contrast, the dissipation dominant regime response exhibits a far less distinct local maximum, because the particular solution magnitude already has negligible values at the spatial point where it is in phase with the homogeneous solution. In fact, a key difference between the interference dominant shape and the decaying dominant shape is the relative magnitudes of the two waves at the spatial location where they are in phase. Analytic approximations of each regime are listed in appendix. The two waves are first in phase at the location, xK , indicated in Eq. (21). Figure 10a plots the linearized flame response, the magnitude of the sum of the two waves, and the relative phase between the two waves. It also indicates the actual maximum in flame response and the estimated response based upon Eq. (22). Note that | | parti | at the axial locations where the respective phase of the two terms is zero. This result shows | | | homo that the estimated and actual peak locations are quite close for the chosen parameter values. They differ due to the non-zero value which causes the actual peak to occur earlier, because of the decaying amplitude of the particular solution. xK / |x xK | parti | x xK | K | | homo 1 (21) | 2 K 1 | 1 4 2 K 1 2 2 x K 1 e (22) The dissipation dominant regime is next considered. In this parameter regime, the axial location of xpeak is controlled by how fast the particular solution decays, because the particular and homogeneous solutions are out of phase in the very nearfield. This decay length scale is 1/. If the criterion is set to the point where the excitation amplitude is lower that e-3 (about 5%) of its initial amplitude, this leads to the estimate for an axial location in Eq. (23) and its corresponding estimated response in Eq. (24). A typical solution is shown in Figure 10b. Note also that defining a “local maximum” for this case is often not a good representation of the results, as the curves really do not have a peak (see Figure 10), but are better described as asymptotically (but with some small level of fluctuation) approaching their maximum, given by | |max = 1/. Similarly, x is also better interpreted as the length scale describing the location where | | asymptotes to its maximum value. x / 3 11 American Institute of Aeronautics and Astronautics 092407 (23) | | | | homo a) Figure 10. 1 4 2 K 1 2 2 (24) b) Dependence of | | upon axial coordinate for (a) interference dominant case (K=1.28, = 0.75, estimation of peak is calculated by Eq. (21) and (22)) and b) dissipation dominant case (K=0.89, = 0.75, estimation of max is calculated by Eq. (23) and (24)) This interference and dissipation dominant behavior can also be used to understand the “ripples” in flame response discussed in the Introduction section. These ripples are due to interference effects and, thus, can be expected to be more prominent as K departs from unity. This is clearly seen in our computations, such as Figure 9a and Figure 11a. We are currently evaluating these K values for experimental data as well to assess this prediction about when ripples will and will not appear experimentally. 2. Nonlinear response The farfield flame response is intrinsically nonlinear and, thus, the above linear treatment provides some guidelines on the role of interference and dissipation, but does not capture the critical nonlinear effect – kinematic restoration. Kinematic restoration is flame propagation normal to itself which smoothes out wrinkles7 and acts as a source of flame wrinkling dissipation. Note that the dissipation referred to in the linear discussion was in the excitation velocity – in the linear, constant flame speed regime, flame wrinkles propagate with constant amplitude. The rate of flame sheet dissipation increases with amplitude and is inversely proportional to flame front wrinkling length scale. This nonlinear effect exerts two key influences upon the farfield – it causes the location of peak flame response to be amplitude dependent and the flame disturbance magnitude to decay to zero (as opposed to the constant values shown in Figure 10). We begin first by illustrating results showing this latter effect. Figure 11a) and b) depict typical fully nonlinear computations showing the flame response at different K and values. The black line is associated with dissipation dominant parameter values, whereas blue and red cases with interference dominated. In general, the farfield non-linear flame response is smaller than the linear solution. Moreover, the reduction in interference generated oscillations in flame response, as well as the monotonic reduction in magnitude of flame response with downstream distance is clearly evident. This dissipation in flame wrinkling is amplitude dependent, as can be seen by comparing the scaled linear solution with the =0.1 and 0.3 cases. 12 American Institute of Aeronautics and Astronautics 092407 a) Figure 11. b) a) Dependence of | | upon axial distance for at different excitation amplitude, solid : linear solution, dash : =0.1, dot : =0.3 (=0.93). b) Dependence of | | upon axial distance at different excitation amplitudes, solid : linear solution, dash : =0.1, dot : =0.3 (=1.17). Figure 12 plots experimental flame response measurements, illustrating similar points – i.e., the faster decay of the flame wrinkle with increasing amplitude. Note that these results are dimensional, hence the growing peak response and nearfield slope with amplitude of excitation. In contrast, the results in Figure 11 are dimensionless and scaled by the magnitude of excitation. Figure 12. Measured dependence of flame front fluctuation magnitude upon normalized axial distance, 0=u0/f0. (o) ua’= 0.028, (+) ua’ = 0.016, () ua’ = 0.010. (u0 = 2.27 m/s, f0 = 150 Hz, cylindrical bluff body, reproduced from Shanbhogue et al.6) The amplitude dependence of x peak is illustrated in Figure 13. For the interference dominant parameter sets, depicted on the left, the peak location of the linear solution is a good indicator for the non-linear solution, as shown in Figure 13a. This is also shown in Figure 14a) and c), which compares the location and magnitude of the peak flame response for different perturbation amplitudes over a range of different parametric conditions (detailed in Table 2). Increasing deviation between the two solutions occurs with increasing amplitude of excitation, as expected. in the nonlinear case is higher and lower than the linear solution when K>1 and K<1, The value of peak respectively. In the dissipation dominant regime, the peak in the linear solution is not a good indicator for x peak for the nonlinear solution. This is because the linear solution does not necessarily have a strong peak, while the nonlinear solution has a very well defined peak due to kinematic restoration. However, x in Eq. (23) is a good estimate for the location of x peak . Note also that as excitation amplitude increases, nonlinear x peak shifts upstream because nonlinearity starts to smooth out wrinkles earlier. Furthermore, as shown in Figure 14d, the linear value of | | . provides a good estimation of nonlinear peak 13 American Institute of Aeronautics and Astronautics 092407 a) Figure 13. b) Dependence of | | upon axial distance illustrating variation in location of peak response upon excitation amplitude for (a) interference dominant regime (K=2, =0.47,) and (b) dissipation dominant regime ( K=0.95, =2.8). a) b) c) d) Figure 14. Comparison of location and magnitude of peak flame response in interference ((a) and (c)) and dissipation ((b) and (d)) dominated regimes. Conditions are listed in Table 2 IV. Conclusion This paper has described the features and parameters that control the spatio-temporal dynamics of flame front fluctuations of a harmonically excited flame. It has been shown that the magnitude of the flame response grows, reaches peak, and then, decays with axial distance. There may also be short wavelength fluctuations superposed upon this behavior. In the nearfield, the flame response grows linearly with downstream distance, at least for well anchored flames that do not fluctuate in position at the attachment point. The slope of this growth region is proportional to the ratio of the fluctuating normal and mean tangential velocity at the flame. 14 American Institute of Aeronautics and Astronautics 092407 Farther downstream, the key flame features are controlled by interference phenomenon and the dissipation in magnitude of the velocity excitation. Interference between waves propagating along the flame sheet is controlled by the parameter |K-1|, which represent the difference between mean flow and excitation phase velocities. Dissipation is controlled by the rate of decay of the velocity fluctuations, . Depending upon the relative values of these two parameters, the farfield response can be “Interference dominated” or “Dissipation Dominated”. In the interference dominant regime, the flame response is oscillatory and possesses local maxima and minima. In the dissipation dominant regime, the flame response monotonically grows or exhibits no recognizable peak. Nonlinearity plays an important role downstream due to kinematic restoration effects, which are responsible for dissipation of flame front wrinkles. This effect causes the relative decay rate of flame wrinkles to increase with disturbance amplitude. The computational and analytical results presented above were obtained with relatively simple model velocity fields. Although not shown here, additional results mirroring those presented above were obtained with a more complex velocity field model, obtained from fits to data described in Shanbhogue et al.6 and given by: u f 1 (1 x)e x cos 2 Kx t v f v0 2 c1 x e x cos 2 ( Kx t ) (25) The above results on the nearfield and farfield behavior, role of wave interference and dissipation, and peak location estimates were reproduced with similar results. Appendix Table 1. Lists of Numerical simulation conditions depicted in Figure 5 for the velocity Model, Eq. (17). Case No. u0 v0 v 1 1 0.1 0.1 0.01 15 2 1 0.1 0.1 0.01 30 3 1 0.2 0.05 0.02 30 4 1 0.5 0.1 0.05 45 5 1 1 0.1 0.1 60 Table 2. Tabulation of conditions used for parameter sweep results depicted in Figure 14 for the velocity model described in Eq.(11). Interference dominant regime Dissipation dominant regime K K 0 - 0.5 0.47 0.9 - 1.1 2.80 1.5 - 2 0.47 0.9 - 1.1 3.73 0 - 0.5 0.93 1.5 - 2 0.93 (1) Approximation of Interference dominant Regime (Taylor Series expansion of Eq. (18) about (=0) x 2 i Kax t 2 i t i e e 2 K 1 2 i K 1 x e x 2 i Ka x t 2 i t e e 2 O 4 2 K 1 K 1 K 1 x 2 i Ka t 15 American Institute of Aeronautics and Astronautics 092407 (26) (2) Approximation of Dissipation dominant Regime (Taylor Series expansion of Eq. (18) about (=0) e a x 1 e x i 2 t x x a x a 2 i e e 1 x K 1 2 i 2 t K 1 e O (27) Acknowledgments This research was supported by the US-DOE & NSF under contracts DE-FG26-07NT43069 and CBET0651045; contract monitors Rondle Harp and Dr. Phil Westmoreland, respectively. References 1 Poinsot, T. and D. Veynante, Theoretical and Numerical Combustion. 2001, Flourtown PA: RT Edwards Inc. 2 Schadow, K.C., et al., Large-Scale Coherent Structures as Drivers of Combustion Instability. Combustion Science and Technology, 1989. 64(4): p. 167-186. 3 Rogers, D.E. and F.E. Marble, A Mechanism for High Frequency Oscillations in Ramjet Combustors and Afterburners. Jet Propulsion, 1956. 26(1): p. 456–462. 4 Chaudhuri, S. and B.M. Cetegen, Blowoff characteristics of bluff-body stabilized conical premixed flames with upstream spatial mixture gradients and velocity oscillations. Combustion and Flame, 2008. 153: p. 616-633. 5 Coats, C.M., Coherent Structures in Combustion. Progress in Energy and Combustion Science, 1996. 22: p. 427-509. 6 Shanbhogue, S., et al., Flame Sheet Dynamics of Bluff-Body Stabilized Flames during Longidudinal Acoustic Forcing, in Combustion Symposium. 2008: Montreal. 7 Lieuwen, T., Nonlinear kinematic response of premixed flames to harmonic velocity disturbances. Proceedings of the Combustion Institute, 2005. 30(2): p. 1725-1732. 8 Preetham and T. Lieuwen, Nonlinear Flame-Flow Transfer Function Calculations: Flow Disturbance Celerity Effects Part 2, in 43rd AIAA Aerospace Sciences Meeting & Exhibit. 2005. 9 Chaparro, A., E. Landry, and B.M. Cetegen, Transfer function characteristics of bluff-body stabilized, conical V-shaped premixed turbulent propane–air flames. Combustion And Flame, 2006. 145(1-2): p. 290-299. 10 Balachandran, R., et al., Experimental investigation of the nonlinear response of turbulent premixed flames to imposed inlet velocity oscillations. Combustion And Flame, 2005. 143(1-2): p. 37-55. 11 Nottin, C., et al., Large Eddy Simulations of an Acoustically Excited Turbulent Premixed Flame. Proceedings of the Combustion Institute, 2000. 28: p. 67-73. 12 Preetham, S. H., and T. Lieuwen, Dynamics of Laminar Premixed Flames Forced by Harmonic Velocity Disturbances. Journal of Propulsion and Power, 2007. 13 Baillot, F., D. Durox, and R. Prudhomme, Experimental and Theoretical-Study of a Premixed Vibrating Flame. Combustion and Flame, 1992. 88(2): p. 149-168. 14 Shanbhogue, S.J. and T.C. Lieuwen, Response of a rod stablized, premixed flame to longitudinal acoustic forcing, in ASME Turbo Expo. 2006: Barcelona, Spain. 15 Shanbhogue, S.J., et al., Response of Rod Stabilized Flames to Harmonic Excitation: Shear Layer Rollup and Flame Kinematics, in Joint Propulsion Conference. 2006: Sacramento, CA. 16 Holmes, M.H., Introduction to perturbation methods. Texts in applied mathematics. 1995, New York: Springer-Verlag. xiii, 337 p. 17 Zachmanoglou, E.C. and D.W. Thoe, Introduction to partial differential equations with applications. 1986, New York: Dover Publications. x, 405 p. 18 Jiang, G.S. and D.P. Peng, Weighted ENO schemes for Hamilton-Jacobi equations. Siam Journal on Scientific Computing, 2000. 21(6): p. 2126-2143. 19 Gottlieb, S. and C.W. Shu, Total Variation Diminishing Runge-Kutta Schemes. Mathematics of Computation, 1998. 67(221): p. 73-85. 16 American Institute of Aeronautics and Astronautics 092407