The Role of Hydrodynamic Instabilities on Near-Lean Blowout Flame Shapes in a Swirl-Stabilized Spray Combustor

Swirl-dump combustors are the predominant combustor geometry that is used in gas turbine engines. This design is advantageous due to its improved flame stability, enhanced fuel-air mixing, and its allowance for reductions to the overall combustor length. Despite these advantages, high-swirl flows are also very susceptible to hydrodynamic instabilities that manifest themselves as coherent vortex structures. These coherent vortex structures are known to influence flame stability [1], mixing [2], NO_x emissions [3], the average flame shape [4,5], and thermoacoustic instabilities [6–8]. However, questions remain about whether coherent vortices are desirable for achieving optimal combustor performance.

From a flame stability standpoint, coherent vortex structures are advantageous. For example, Stöhr et al. [2] showed that helical shear layer vortices rapidly mix fuel and air with hot combustion products and this mixing process allows the flame to recover from local extinction instances. Furthermore, vortex cores are ideal locations for flame stabilization because they have low strain rates at typical Damköhler numbers [9,10]. These benefits were demonstrated in additional data taken by Stöhr et al. [1] where lean blowout (LBO) occurred if the flame root remained extinguished for more than one period of the helical vortex cycle.

Coherent vortex structures appear to be undesirable for lowering NO_x emissions. Renaud et al. [11] observed that the PVC can disturb the fuel spray inside of a liquid-fueled combustor's swirler, resulting in a helical spray pattern. Mukherjee et al. [12] argued that such changes to the spray pattern are problematic because they can create a polydisperse spray with unintended high-temperature regions and increased NO_x emissions. Similar conclusions have been reached in gaseous combustion systems. Lückoff et al. [3] showed that coherent vortex structures increase NO_x emissions in both premixed and partially premixed flames.

Studies that have investigated the role of self-excited coherent vortices on thermoacoustic instabilities have shown mixed results. Frederick et al. [6] experimentally demonstrated that global hydrodynamic instabilities were unresponsive to acoustic forcing and may therefore disrupt the coupling processes that are required to close the thermoacoustic feedback loop. Alternatively, Steinberg et al. [7] showed that coherent vortex structures increased the flame surface area and thereby created periodic heat release fluctuations in a thermoacoustically unstable combustor. Hemchandra et al. [8] experimentally studied the interactions between global hydrodynamic instabilities and thermoacoustic instabilities in a backward facing step combustor. They found that self-excited coherent vortices can drive thermoacoustic instabilities if the hydrodynamic frequency is similar to the resonant combustor frequency.

Although multiple hydrodynamic instabilities [13–16] are common in swirl flows, the two that are most consequential to combustor performance are the precessing vortex core (PVC) [14] and the concomitant helical shear layer vortex [13]. The PVC [14] describes a phenomenon where the location of zero azimuthal velocity precesses around the geometrical center of the combustion chamber. Furthermore, the PVC is a global hydrodynamic instability, meaning that it is excited by the flow itself rather than external forcing. Multiple authors [13,15,17] have concluded that the PVC requires the existence of a vortex breakdown bubble, which is a recirculating flow region that helps stabilize the flame. The PVC therefore originates just upstream of the vortex breakdown bubble's leading stagnation point [17] (i.e., the “wavemaker”) and is excited either inside or just downstream of the swirler. Oberleithner et al. [13] further hypothesized that the globally unstable precessing vortex core can force convectively unstable shear layers to respond at the PVC frequency. Therefore, the helical vortex structure that was shown by Oberleithner et al. [13] to wind around the vortex breakdown bubble is ostensibly a result of this PVC forcing. The data of Roy et al. [18] provides support for this hypothesis. They found that the PVC decayed shortly downstream of the swirler exit, while the helical shear layer vortex continued rotating downstream at the original PVC frequency. Manoharan et al. [17] supplemented these findings by demonstrating that helical shear layer instabilities originate at locations where PVC oscillations are the strongest.

Although nonreacting swirl flows typically excite a helical vortex, the heat released from combustion can suppress this vortex structure. The primary factors that dictate whether reacting flows will be hydrodynamically unstable are: (1) the reverse flow velocity magnitude, (2) the density gradient magnitude, (3) the location of density gradients relative to the axial velocity gradients, and (4) the location of density gradients relative to the wavemaker. Varying contributions from these factors were at play in the Terhaar et al. [19] data where a range of steam dilution levels were used to produce different flame shapes in their combustor. They observed that a helical vortex was present in the flow fields corresponding to some of the flame shapes but not others. Oberleithner et al. [20] applied a linear hydrodynamic stability analysis to the Terhaar et al. [19] data and found that the flame shapes without a helical vortex were hydrodynamically stable as a result of density gradients that coincided with axial velocity gradients near the wavemaker region. Further research by Oberleithner et al. [5] on two reacting swirl flows, one that had a helical vortex and another that did not, highlighted the importance of the density gradient magnitude on hydrodynamic stability. They found that the case with the helical vortex had smooth radial density gradients whereas the other case had sharp radial density gradients. This same group [21] later performed a model study that synthesized the results from these earlier studies. They concluded that the parameters that govern the formation of helical vortex structures in swirl flows are the reverse flow velocity and the density ratio, where greater reverse flow velocities and lower density ratios will be hydrodynamically destabilizing. These parameters are determined by the combustor geometry [12,17], operating conditions [5], and the average flame shape [19].

Reacting flows that were initially hydrodynamically stable can develop coherent vortices as they approach LBO. This often leads to sudden flame shape changes that are undesirable because they can produce unintended changes in emissions, thermoacoustic oscillations, and thermal stresses [4]. A well-established example of this phenomenon is known to occur in bluff-body stabilized flames [22]. Nonreacting wake flows at large Reynolds numbers experience a sinusoidal flow structure caused by the Bénard-von Kármán instability. However, the heat released by combustion induces density gradients in the flow that usually suppress this hydrodynamic instability such that bluff-body stabilized flames have a symmetric wake [23]. As the equivalence ratio in a combustor is reduced toward the lean blowout limit, flames experience local extinction processes that reduce the overall heat release [22]. Eventually, the density field in the flow approaches the nonreacting condition and vortex shedding [22] intermittently appears. An et al. [24] argued that coherent vortex structures unique to nonreacting swirl flows can similarly manifest themselves as swirl-stabilized flames approach LBO. This conclusion was based on detailed measurements of a bistable flame in their combustor that intermittently lifted-off the burner nozzle near LBO [24]. These lift-off processes were correlated with the appearance of helical vortices that were not present when the flame was attached.

Although research is clear about the significance of coherent vortex structures for combustor performance, the fundamental physical processes that underlie hydrodynamic instabilities are not well-characterized in realistic combustion systems. In particular, there is a lack of research on hydrodynamic instabilities and flame shape changes in aircraft engine relevant combustors. Recent work by the authors [25] on the referee combustor at the Air Force Research Laboratory (AFRL) has shown that the flame fluctuated between two different shapes as LBO was approached. The flow features and flame topology of the original flame shape, FS1, were very similar to those of a stable burning flame. However, a second flame shape, FS2, also appeared near LBO. FS2 filled a larger portion of the combustor volume than FS1, had a higher average CH* emission intensity, and was only observed as LBO was approached. The purpose of this study is to characterize FS2 and determine what processes are responsible for the flame shape change. In addition, the impacts of fuel type on the flame shape will be examined. To the authors' knowledge, this is the first study to demonstrate that the excitation of coherent flow structures can influence the flame behavior in a realistic, liquid-fueled, swirl-stabilized combustor.

This work involved detailed measurements of the flow, flame, and fuel spray in the AFRL referee combustor. This combustor has been used for multiple studies on combustor spray characteristics, LBO, ignition, and combustion instability performance [26]. The combustor design was based on the modern rich-burn, quick-quench, lean-burn (RQL) configuration, and a detailed description is provided in Ref. [26]. High-pressure air (i.e., 207 kPa) at 394 K was routed to the combustion chamber through three sets of swirl vanes, two sets of dilution air jets, and an effusion cooled combustor liner. The liquid fuel was first atomized by a pressure atomizer type fuel injector and was then broken-up further as it was sheared off of a prefilming nozzle surface by the incoming air. Two liquid fuels with very different average LBO equivalence ratios were considered in this study, A-2 (Jet-A with average properties) and C-1 (a highly-branched, low-DCN alternative fuel) [26]. They will be referenced according to the nomenclature used in the National Jet Fuels Combustion Program.

Particle image velocimetry (PIV) was used to measure the combustor's flow field. This technique involved pairs of 532 nm laser pulses that were outputted by a Spectra Physics 400 PIV laser with a 15 μs delay between laser pulses. The laser pulse pairs were emitted at a frequency of 10 Hz with 6–7 mJ of energy per laser pulse. Two 12 bit Photron SAZ cameras recorded the Mie scattering from 1.0 μm alumina seeding particles that traced the flow. Each PIV camera was operated in random reset mode with a 50 μs exposure time. The cameras were filtered to detect a 3 nm bandwidth centered at 532 nm and were focused with AT-X M100 Tokina lenses set to f/D = 5.6. Both 2 component PIV and fuel spray Mie scattering measurements were acquired. The Mie scattering measurements used the same PIV data acquisition approach but without seeding the flow with alumina particles. Table 1 outlines the test cases that will be discussed in this paper.

The AFRL referee combustor only has partial optical access. Windows are located on the sides of the combustor but the top and bottom surfaces of the combustor liner are a nickel alloy. Therefore, the laser sheet was formed using a f = +1000 mm spherical lens to focus the laser beam onto a rod lens that was mounted just above the top center dilution air jet hole. The 3 mm diameter rod lens with a focal length of f = +2.2 mm then expanded the beam into a 1.5 mm thick laser sheet along the combustor centerline. This approach is illustrated in Fig. 1. However, limitations in the expansion angle of the laser sheet prevented the flow in the upper-left corner of the combustor from being measured. The laser sheet impinged on several metal surfaces in the combustor, including the liner, inlet dome plate, and the swirler. In order to minimize laser reflections, a red stripe was applied to these surfaces that either absorbed the laser light or shifted it to a wavelength that was blocked by the 532 nm spectral filters [27].

The LaVision DaVis 8.3.1 software was used to calculate vector fields from the recorded PIV images. The raw PIV images were first preprocessed with robust principal component analysis [28] to remove any background features and accentuate the seeding particles. A multipass adaptive PIV algorithm with 50% overlap was then used to calculate the vector fields. The initial interrogation window size was 48 × 48 pixels and the final window size was 24 × 24 pixels. This resulted in a vector spacing of 1.3 mm. As is common in spray flames [29] and flows with coherent vortex structures, certain snapshots included locations with low seeding particle density. In these instances, the gappy-proper orthogonal decomposition method of Saini et al. [30] was used to reconstruct missing vectors. Specifically, 7.2% of the test condition A vectors, 7.4% of the test condition D vectors, and 5.4% of the test condition E vectors were reconstructed.

Since Mie scattering from the fuel spray was also present during the PIV measurements, this raises the question of whether the fuel spray introduced errors in the calculated vector fields. Figure 2 addresses this concern by comparing the accepted average velocity fields, which were calculated by tracking both fuel droplets and seeding particles, with average velocity profiles that are based entirely on the trajectory of liquid fuel droplets. The “fuel only” velocity profiles were calculated by applying the PIV cross-correlation algorithm to the test condition C fuel spray Mie scattering measurements. Figure 2(a) shows that the fuel only axial velocity component does depart somewhat from the accepted (i.e., seed + fuel) velocity component for FS2 in the inner shear layer. On the other hand, there are no major differences between the accepted and fuel only axial velocity component for FS1. These results seem to agree with the conclusions of Emerson and Ozogul [31] who used fuel PLIF measurements to separate liquid and gas velocities in a very similar spray combustor. They found that liquid fuel droplets are reasonably good flow tracers that have little effect on time-averages of axial and azimuthal velocity components. However, they rarely observed fuel droplets with negative axial velocities, which may explain why the average FS2 axial fuel velocity is disproportionately high near the central recirculation zone (CRZ) boundary (i.e., U_x = 0). Nevertheless, fuel droplets are not expected to be present in locations that contain coherent vortex structures because, as will be shown in Sec. 6, coherent vortices effectively centrifuge fuel droplets out of their cores. Furthermore, there is little evidence in Fig. 2(b) that the accepted transverse velocity component for either flame shape has been biased by liquid–gas slip between the liquid fuel droplets and the surrounding gas.

The flame shape was identified throughout the experimental campaign from chemiluminescence imaging. During the 2 component PIV and droplet Mie scattering measurements, a separate camera was dedicated to simultaneously record CH* chemiluminescence images at 10 Hz. The chemiluminescence camera had a f/D = 2.8 lens setting and was filtered using a 427 nm spectral filter with a 10 nm bandwidth. High-speed OH*, CH*, and C2* chemiluminescence measurements that were acquired as part of a previous study by Monfort et al. [32] will also be discussed.

Coherent flow structures were identified in the experimental data using proper orthogonal decomposition (POD). POD is an analysis technique [33] that extracts orthogonal modes from a series of snapshots. The modes are ordered according to the amount of kinetic energy that they contribute to the flow, where the first mode contributes the most kinetic energy to the flow and the last mode contributes the least. Coherent flow structures are highly energetic flow features and, if they are present, usually appear in the POD modes with the highest kinetic energy fraction. The mathematical details that underlie POD can be found in the overview by Taira et al. [33] and in the authors' previous work [25].

As will be discussed later in the paper, POD was applied to both the velocity field data and the fuel spray Mie scattering images. Depending on the test case, each recorded series of instantaneous velocity fields was comprised of 308-439 image pairs. Test condition C similarly includes 483 fuel Mie scattering images for FS1 and 499 fuel Mie scattering images for FS2. In order to ensure that the number of available snapshots is sufficient for convergence of the POD modes, the authors repeated the analysis with 250 snapshots and observed only minor differences in the POD mode shapes. In a study designed to determine the minimum number of image pairs that are required to characterize coherent flow structures, Lacarelle et al. [34] also concluded that converged POD results can be achieved with 250 snapshots.

Flames are considered to be bistable when they have a fixed equivalence ratio but are susceptible to sudden bifurcations in the flame shape. This section will show that the AFRL referee combustor exhibited clear bistable flame behavior when it operated near LBO. For example, Fig. 3(a) shows a time series of spatially integrated chemiluminescence intensity that was recorded during an interval of approximately 100 s. Test condition C is shown in this figure where the flame burned on the threshold of LBO. Despite maintaining a constant equivalence ratio throughout the recording, it is clear that the integrated chemiluminescence signal bifurcates between low and high values throughout this run. These bifurcations represent a sudden, temporary change in the flame shape that only occurred when the equivalence ratio was reduced toward the LBO limit. The average CH* chemiluminescence images corresponding to the low and high spatially integrated chemiluminescence values in Fig. 3(a) are also shown in this figure. The image in Fig. 3(b) resembles the flame shape that was observed at the stable burning test condition B and corresponds to the lower spatially integrated chemiluminescence values shown in this plot. This flame shape will be called “FS1” hereafter. The image in Fig. 3(c) shows the new flame shape, “FS2,” which only appeared near LBO and fluctuated wildly in chemiluminescence intensity. It is clear that FS2 occupied a larger portion of the combustor volume and had more intense CH* emission on average than FS1.

Although data are only shown for fuel A-2 in Fig. 3, the flame shape change also occurred for fuel C-1. The two liquid fuels that were considered in this work, A-2 and C-1, were selected because they have very different average LBO equivalence ratios at the studied operating condition. Previous work concluded that this is due to the higher DCN of fuel A-2 [26], which is an inverted measure of the ignition delay time of a liquid fuel. However, the recorded chemiluminescence images suggest that FS2 persisted longer for A-2 than C-1. The A-2 flames burning on the threshold of LBO were on average in FS2 46.3% of the time compared to 31.9% of the time for C-1. Furthermore, A-2 remained in FS2 for a 10 s average duration while C-1 remained in FS2 for 7 s on average.

One question that naturally arises from the previous discussion is whether the flame shape change is caused by the combustor undergoing a thermoacoustic instability. Thermoacoustic instabilities can cause clear differences in the average flame shape and could potentially explain the FS2 heat release fluctuations. To answer this question, begin by considering test conditions D and E, which had the same equivalence ratio but corresponded to FS1 and FS2, respectively. Test condition E had a higher average sound pressure level (SPL) than either test condition D (i.e., 145.3 dB for FS2 and 142.0 dB for FS1) or the stable burning test condition B (i.e., 143.2 dB). However, the FS2 SPL (i.e., 145.3 dB) is still much lower than the 155 dB value [32,35] corresponding to the ϕ = 0.11 condition where previous work found the lowest amplitude thermoacoustic instabilities were activated [32]. Furthermore, Fig. 4(b) plots the PSD of spatially integrated 10 kHz chemiluminescence measurements for an FS1 case, an FS2 case, and a thermoacoustically unstable case. A narrowband heat release peak is clearly evident in the spectrum of the thermoacoustically unstable case but no such features are observed in the FS2 spectrum. The FS2 spectrum is dominated by broadband, low-frequency structures that likely represent incoherent fluctuations in heat release. Although these fluctuations may explain the relatively higher SPL for FS2 compared to FS1, there is no evidence that they were coupled with the natural acoustic modes of the combustor. Taken together, these results suggest that a transition to thermoacoustic instability was not the cause of the flame shape change.

As was discussed in Sec. 1, the sudden excitation or suppression of coherent flow structures often underlies flame shape changes. This is especially true near LBO where the reduced heat release is less effective at suppressing the hydrodynamic instabilities that are excited in nonreacting flows. Therefore, the measurements that were taken in the nonreacting flow field will now be compared with those from the FS1 and FS2 flow fields.

Figure 5 shows the average velocity fields and Fig. 6 shows the first two POD modes for the nonreacting test condition A, the FS1 test condition D, and the FS2 test condition E. It should be noted that test conditions D (i.e., FS1) and E (i.e., FS2) had the same equivalence ratio. Results are only shown for x ≤ 40 mm to emphasize the combustor primary zone. However, the full field of view can be seen in the authors' previous work [25].

Consider first the flow features for the nonreacting test condition A. Beginning with the average velocity field in Fig. 5(a), it can be seen that this flow has a large CRZ and that the annular jet of air that exits the swirler experiences rapid radial expansion as the flow progresses downstream. Furthermore, the POD modes in Fig. 6(a) demonstrate that a majority of the nonreacting kinetic energy is concentrated in large shear layer vortices that appear immediately downstream of the swirler exit. The fact that multiple vortices appear sequentially suggests that these structures are convected by the flow after being initiated in the swirler. The first two nonreacting POD modes depict the same vortical structures but they are shifted by a quarter wavelength relative to each other. This provides evidence of a mode pair that has captured a periodic phenomenon. Furthermore, Oberleithner et al. [13] performed a three-dimensional reconstruction of similarly shaped POD modes and showed that they represent a helical vortex that winds around the vortex breakdown bubble.

The FS1 data demonstrate that combustion had a significant effect on the flow field in this combustor. Consider, for example, the differences between the nonreacting and FS1 time-averaged velocity fields. Figure 5(b) shows that the annular jet exiting the swirler had a narrower spreading angle for FS1 than for the nonreacting flow, as evidenced by the 11% difference on average between the nonreacting jet core location and the FS1 jet core location. The FS1 CRZ was also smaller than the nonreacting CRZ. Assuming that the 0.5 progress variable contour in Fig. 5(a) is an accurate estimate of the time-averaged flame location, the heat released by FS1 in the inner shear layer may contribute to these differences in the time-averaged velocity fields. The progress variable field was calculated by time-averaging raw PIV images that were binarized based on locations with sharp gradients in seeding particle density [8,38]. Furthermore, vortices are only observed in the FS1 POD modes around x = 30 mm in Fig. 6(b), which is far downstream of the swirler exit. Although the physical meaning behind the FS1 POD mode shapes is unclear, it is safe to conclude that they do not represent a periodic vortex structure similar to the nonreacting POD modes in Fig. 6(a). These results are consistent with the commonly observed phenomenon that the heat released from combustion tends to suppress the formation of coherent vortex structures [5,20,21]. No significant differences were observed between the FS1 POD modes for fuels A-2 and C-1 [25], and both resemble the dominant POD modes for the stable burning test condition B case [25].

The PIV measurements revealed that the FS2 flow field had more in common with the nonreacting flow than the FS1 flow. For example, Fig. 5(b) indicates that the spreading angle of the annular swirler jet is very similar between the average FS2 and nonreacting flow fields with an average percent difference of only 2% between their jet core locations. Furthermore, the FS2 POD modes in Fig. 6(c) parallel those of the nonreacting flow field in Fig. 6(a). Large vortices are observed immediately downstream of the swirler exit and they extend throughout the combustor primary zone in a staggered pattern, although with more noise than what is observed in Fig. 6(a). The fact that the first two FS2 POD modes capture a pair of vortices that have been shifted by a quarter wavelength suggests that these large-scale structures potentially represent a helical vortex that is intermittently activated as the flame approaches LBO.

Many of the features that were unique to the FS1 and FS2 flow fields also manifested themselves in the respective spray patterns for these flame shapes. Consider, for example, the fuel spray probability contours shown in Fig. 5, which were calculated by averaging binarized spray Mie scattering images from test condition C. The FS1 fuel spray was entirely confined within the annular jet of air exiting the swirler, whereas more of the FS2 fuel spray extended into the CRZ. Furthermore, FS2 generally had thicker spray probability contours than FS1. Figure 7 shows several instantaneous spray Mie scattering images from test condition C. The image representing the FS1 fuel spray, which is shown in Fig. 7(a), denotes a set of finely atomized droplets that spread throughout the combustor primary zone in an orderly fashion. In contrast, the FS2 spray images in Fig. 7(b) show clusters of droplets near the swirler exit that often developed into pronounced hook-like features further downstream. As described by Menon [39], these are notable characteristics of vortex-spray interactions because the high vorticity in vortex cores centrifuges the droplets outward to where they then accumulate around the vortex circumference. This sequence of droplet clusters upstream followed by hook-like features downstream is also consistent with the vortex-spray interactions observed in the LES simulation of Patel and Menon [40]. They observed that the droplets cluster around the PVC shortly after the fuel is injected and then spread radially outward as the vortex progresses downstream.

Consider next the patterns that were extracted by POD from the entire series of FS1 and FS2 spray images in test condition C. Figure 8 shows the first two spray POD modes for FS1 in plot (a) and FS2 in plot (b). It can be seen that the FS1 spray may be stronger on either the top or bottom of the prefilming surface, but in general the FS1 spray pattern does not materially deviate from the mean. In contrast, the first two FS2 spray POD modes display an alternating staggered pattern and were much more energetic than the dominant FS1 spray POD modes (i.e., 7.2% and 4.6% for FS2 compared to 3.2% and 2.4% for FS1). Similar results were observed by Renaud et al. [11] in their Mie scattering measurements of vortex-spray interactions both inside and downstream of the swirler. By applying dynamic mode decomposition to these data, they also observed mode shapes with an alternating pattern and concluded that these were evidence of a helical vortex disturbing the fuel spray. Furthermore, the time coefficients corresponding to the FS2 spray POD modes in Fig. 8(b) make it possible to phase-average the Mie scattering images using the method of Stöhr et al. [41]. This phase-average, which is shown in Figs. 9(a), 9(c), 9(e), and 9(g) at phase angles of 0 deg ± 22.5 deg, 90 deg ± 22.5 deg, 180 deg ± 22.5 deg, and 270 deg ± 22.5 deg, illustrates an alternating, asymmetric spray pattern where a disturbance convects through one branch of the fuel spray and then this process repeats itself in the opposite branch. Again, this plot seems to capture the progression of a helical spray pattern through a planar slice in the flow. No such asymmetries were observed in a phase-average of the FS1 spray images (not shown).

Up to this point, the discussion surrounding vortex-spray interactions has not involved velocity field data. This limitation is overcome here using the test condition E Mie scattering images, which were the basis for calculating the FS2 velocity field. By applying POD to the region in the raw FS2 Mie scattering images where most of the fuel spray is distributed (i.e., x ≤ 15 mm and −26 mm ≤y ≤ 26 mm), POD modes shapes were obtained (not shown) that are nearly identical to those shown in Fig. 8(b) for the FS2 spray. This demonstrates that the same fuel spray dynamics can be extracted from both the FS2 seed+fuel Mie scattering images (i.e., test condition E) and the FS2 fuel only Mie scattering images (i.e., test condition C). Therefore, POD and the method of Stöhr et al. [41] were again used to phase-average the test condition E Mie scattering images. These phase-averages are shown in Fig. 9(b), 9(d), 9(f), and 9(h) at phase angles of 45 deg ± 22.5 deg, 135 deg ± 22.5 deg, 225 deg ± 22.5 deg, and 315 deg ± 22.5 deg, respectively. To help distinguish the fuel spray from the seeding particles, the 10% spray probability contours from the fuel only (i.e., test condition C) phase-averages were added to Figs. 9(b), 9(d), 9(f), and 9(h) at the appropriate phase angles. Furthermore, since the test condition E Mie scattering images are coincident with velocity data, phase-averages of the FS2 velocity field and the swirling strength corresponding to the individual FS2 snapshots are included in Figs. 9(b), 9(d), 9(f), and 9(h). The swirling strength quantifies the swirl intensity in vortex cores and is commonly used to identify vortices in high-shear flows [42].

The combination of fuel only and fuel + velocity field phase-averages in Fig. 9 illustrates the mechanism underlying the unique progression of the FS2 fuel spray. Begin by considering the cluster of droplets located around x = 8 mm, y = 10 mm in Fig. 9(b). It is noteworthy that this droplet cluster is positioned adjacent to an inner shear layer (ISL) vortex, as demonstrated by the rotating vectors and concentration of swirling strength at x = 5 mm, y = 7 mm. As this droplet cluster advects downstream, it ostensibly transitions to the hook-like spray feature that appears in the upper half of the flow in Figs. 9(c), 9(d), and 9(f). Consistent with what was observed in Fig. 9(b), an ISL vortex is also located adjacent to the hook-like spray feature in Figs. 9(d) and 9(f). A similar pattern can be seen to commence in the lower half of the flow in Fig. 9(f). Here, the droplet cluster at x = 6 mm, y = −8 mm appears to be linked with an ISL vortex emerging from the swirler exit. This lower ISL vortex then tracks the hook-like spray feature in Figs. 9(h) and 9(b). Interestingly, the hook-like spray feature is no longer observed once its neighboring ISL vortex has dissipated. This can be seen in the lower half of the flow in Fig. 9(d) and the upper half of the flow in Fig. 9(h). Taken together, these findings support the hypothesis that coherent vortices interact with the FS2 fuel spray and cause the unique spray dynamics that are shown in Figs. 7(b) and 8(b). Furthermore, Figs. 9(b) and 9(f) provide evidence that the FS2 vortex structure is helical in nature, as the upper and lower ISL vortices are shifted axially relative to each other.

In addition to bordering the fuel spray disturbances, the ISL vortices in Fig. 9 share very similar axial locations with highly concentrated regions of swirling strength in the outer shear layer (OSL). The OSL vortices in the lower half of the flow in Figs. 9(b), 9(f), and 9(h) and the upper half of the flow in Figs. 9(d) and 9(f), for example, appear to be synchronized with corresponding ISL vortices. This is an important finding because the ISL and OSL vortices could be synchronized by a common source, such as the PVC, during a global hydrodynamic instability [43]. However, it should be noted that the ISL and OSL vortices only remain synchronized up to a point. Using the upper half of the flow as an example, the ISL vortex achieves its maximum phase-averaged swirling strength just downstream of the swirler exit in Fig. 9(b). As the flow proceeds downstream in Figs. 9(d) and 9(f), the ISL vortex's swirling strength progressively decreases, whereas an OSL vortex enters the field of view. Eventually, the ISL vortex disappears completely in Fig. 9(h) and only the OSL vortex remains. Figure 9(d) shows that a similar phenomenon takes place in the lower half of the flow. In this 135 deg phase-average, the OSL vortex has advected throughout the extent of the field of view but the ISL vortex that was located at x = 12 mm, y = −12 mm in the preceding phase-average (i.e., Fig. 9(b)) is no longer detectable. These observations seem to be consistent with the data of Roy et al. [18] who found that the PVC decayed shortly downstream of the swirler exit, while a helical OSL vortex continued rotating downstream at the original PVC frequency.

Flame shape characteristics were investigated in this study as a liquid-fueled, swirl-stabilized combustor was operated on the threshold of LBO. In particular, the flame abruptly bifurcated between two different shapes when the equivalence ratio was held constant near the LBO limit. A new flame shape that only appeared near LBO (i.e., FS2) filled more of the combustor volume and had stronger CH* emission than the original flame topology (i.e., FS1). Although this flame shape change was observed to happen for both a conventional jet fuel (A-2) and an alcohol-to-jet (ATJ) alternative jet fuel (C-1), it was more likely to happen when fuel A-2 was burned near LBO. Advanced diagnostic techniques were used to study the spray, flame, and flow field corresponding to each flame shape.

Clear differences in coherent vortex structures and spray patterns were evident between the flame shapes. These conclusions were reached by reducing the spray and flow field data using proper orthogonal decomposition (POD). Although large vortices developed immediately downstream of the swirler exit in the nonreacting flow, the FS1 combustion process apparently suppressed these vortex structures. Furthermore, the FS1 fuel spray was symmetrical and deviated very little from its average spray pattern. In contrast, POD identified large vortices in the FS2 flow field that resembled those of the nonreacting flow. The FS2 spray POD modes similarly displayed an alternating staggered pattern, which other studies [11] have linked to a coherent vortex periodically disturbing the fuel spray. Phase-averages were then used to link the FS2 fuel spray dynamics with the velocity field. They showed an asymmetric vortex structure disturbing the fuel spray and progressing through the measurement plane in a presumably helical pattern. Near the swirler exit, this helical vortex structure involved both outer and inner shear layer vortices that appeared to be synchronized with each other. However, the inner shear layer vortices decayed as the flow progressed downstream and only the outer shear layer vortices remained throughout the measurements' field of view. Taken together, these results suggest that the new flame shape is related to a helical vortex that develops in the flow near LBO.

Multiple studies have shown that the excitation of hydrodynamic instabilities can induce flame shape changes in gaseous premixed combustors. Although the instability mechanisms vary, these phenomena have been observed to affect both gaseous bluff-body and swirl-stabilized flames. However, to the authors' knowledge, this is the first study to demonstrate that similar physical processes can influence the flame behavior in a realistic, liquid-fueled, swirl-stabilized combustor. Due to their ability to disrupt the fuel spray, coherent vortex structures may be more consequential to the performance of liquid-fueled combustors than the gaseous systems where most previous research efforts have been concentrated.

The authors would like to thank Jerry Greiselhuber, Jeff Gross, and Alan Forlines for their contributions to the measurements and facility operation. This material is based on research sponsored by the Air Force Research Laboratory, under agreement numbers FA8650-19-F-2049, FA8650-15-D-2505, and FA8650-16-2-2605. The U.S. Government is authorized to reproduce and distribute reprints for Governmental purposes notwithstanding any copyright notation thereon. The views and conclusions contained herein are those of the authors and should not be interpreted as necessarily representing the official policies or endorsements, either expressed or implied, of the Air Force Research Laboratory or the U.S. Government. A portion of this work was also funded by the U.S. Federal Aviation Administration (FAA) Office of Environment and Energy as a part of ASCENT Project 30 under FAA Award Number: 13-C-AJFE-UD. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the authors and do not necessarily reflect the views of the FAA or other ASCENT Sponsors. Distribution Statement A: Approved for Public Release; Distribution is Unlimited. PA# AFRL-2022-0467.

• Air Force Research Laboratory (Award ID: FA8650-16-2-2605; Funder ID: 10.13039/100006602).

• Federal Aviation Administration (Award ID: 13-C-AJFE-UD; Funder ID: 10.13039/100006282).

The datasets generated and supporting the findings of this article are obtainable from the corresponding author upon reasonable request.

AFRL =

Air Force Research Laboratory

$c¯$ =

time-averaged progress variable contour

CRZ =

central recirculation zone

DCN =

derived cetane number

FS1 =

flame shape 1

FS2 =

flame shape 2

ISL =

inner shear layer

LBO =

lean blowout

OSL =

outer shear layer

PIV =

particle image velocimetry

POD =

proper orthogonal decomposition

PVC =

precessing vortex core

ΔP/P =

pressure drop across the swirler represented as a percentage of the mean combustor pressure

SPL =

sound pressure level

U_x =

axial velocity component

U_y =

transverse velocity component

η =

spatial POD modes

ϕ =

equivalence ratio