COMPARISON OF CENTER NOZZLE STAGING TO OUTER NOZZLE ...

Proceedings of ASME Turbo Expo 2018:Turbomachinery Technical Conference and Exposition

GT2018June 11-15, 2018, Lillestrøm, Norway

GT2018-75423

COMPARISON OF CENTER NOZZLE STAGING TO OUTER NOZZLE STAGING IN AMULTI-FLAME COMBUSTOR

Wyatt Culler, Xiaoling Chen, Stephen Peluso,Domenic Santavicca, Jacqueline O’Connor∗

Center for Combustion, Power, and PropulsionDepartment of Mechanical and Nuclear Engineering

The Pennsylvania State UniversityUniversity Park, Pennsylvania

David Noble

Senior Technical LeaderElectric Power Research Institute

Charlotte, North Carolina

ABSTRACTCombustion instability in gas turbines is often mitigated us-

ing fuel staging, a strategy where the fuel is split unevenly be-tween different nozzles of a multiple-nozzle combustor. This workexamines the efficacy of different fuel staging configurations bycomparing axisymmetric and non-axisymmetric fuel staging ina four-around-one model gas turbine combustor. Fuel stag-ing is accomplished by increasing the equivalence ratio of thecenter nozzle (axisymmetric staging) or an outer nozzle (non-axisymmetric staging). When the global equivalence ratio isφ = 0.70 and all nozzles are fueled equally, the combustor un-dergoes longitudinal, self-excited oscillations. These oscillationsare suppressed when the center nozzle equivalence ratio is in-creased above φStaging = 0.79. This bifurcation equivalence ratiovaries between φStaging = 0.86 and φStaging = 0.76 for the outernozzles, and is attributed to minor hardware differences betweeneach nozzle. High speed CH* chemiluminescence images in com-bination with dynamic pressure measurements are used to deter-mine the instantaneous phase difference between the heat releaserate fluctuation and the combustor pressure fluctuation through-out the combustor. This analysis shows that the staged flamehas similar phase relationships for all staging configurations. Itis found that axisymmetric staging can be as effective as non-axisymmetric staging; however, the aforementioned hardware

∗Address all correspondence to [email protected]

variations can impact both the bifurcation equivalence ratio andthe effectiveness of staging.

NOMENCLATUREIQR Inner quartile rangep′ Combustor pressure fluctuationp′RMS Root-mean-square amplitude of the combustor pres-

sure fluctuationΦ Phase angleφ Equivalence ratioφStaging Equivalence ratio of the staged flameq Heat release rateζ Stochastic drivingν Damping rateτ Characteristic instability decay or onset time

INTRODUCTIONCombustion instability is a potential issue for gas turbines oper-ating at lean-premixed conditions. Combustion instability occurswhen heat release rate oscillations couple with acoustic oscilla-tions in a feedback loop [1], and is undesirable in engines be-cause it reduces engine operability, increases emissions, and, inrare cases, results in catastrophic hardware failure [1,2]. It can besuppressed by passive techniques such as Helmholtz resonatorsor active techniques such as fuel flow modulation, but passivetechniques are generally preferred in industry for their robust-

1

ness [2]. Fuel staging, which is when the fuel is distributed un-evenly between different nozzles of a multiple nozzle combustor,is a common passive method of suppressing instabilities [1–6].

The primary effect of fuel staging is the redistribution ofheat release, and Li et al. [7] recently conducted a computa-tional study that examined the effect of non-axisymmetric heatrelease rate distributions on self-excited instabilities in a Rijketube. They modeled the circumferential non-uniformity usinga sine function and a parameter, α , to control the “strength”of the non-uniformity. They found that the magnitudes of theinstability growth rates decreased with increasing α , indicatingthat increasing the amount of non-uniformity in the heat releasehelps to stabilize the combustor. They suggested that the for-mation of vortical waves may be partially responsible for thestabilizing effect they noted. Recent work by Samarasinghe etal. [8] proposed a mechanism for how axisymmetric staging sup-presses self-excited instability in more realistic combustor con-figurations. Their study utilized a five nozzle, four-around-oneconfiguration can combustor, the same configuration used in thepresent work. Fuel staging was accomplished by increasing thecenter nozzle equivalence ratio. They found that fuel staging al-tered the phase relationships between the staged flames and theouter flames, resulting in destructive interference that suppressedthe instability.

Fuel staging in gas turbines may be conducted in an axisym-metric [3] or non-axisymmetric manner [4, 5]. Bulat et al. [3]used a control algorithm to vary the fuel splits between main fueland pilot fuel in an axisymmetric manner to suppress longitudi-nal instabilities in a Siemens gas turbine. They found that fuelstaging successfully suppressed instability at different operatingconditions, which increased the operability of the engine whilesimultaneously decreasing emissions. Cohen et al. [5] used non-axisymmetric fuel splits in order to suppress an azimuthal insta-bility in an annular combustor. They used a reduced-order modelto determine the optimal fuel staging distribution for instabilitysuppression. They noted that the heat-release coupling, which iscontrolled by the fuel staging, was skew-symmetric, where in-creasing the damping of one mode decreased the damping of theother mode. They devised an optimal non-axisymmetric stagingstrategy by decreasing the damping of the quieter mode to in-crease the damping of the louder mode. These previous studiesshow that both axisymmetric and non-axisymmetric fuel stagingcan be used to suppress instabilities, depending on the nature ofthe instability.

The studies by Bulat [3] et al., Cohen [5] et al., and Davisand Black [6] indicate that the fuel staging splits in actual en-gines are varied in a fundamentally unsteady manner. Dependingon the timescale and direction of this unsteady change, nonlin-ear behaviors such as triggering [9] or hysteresis [10] may occur.Recent work by Culler et al. [11] considered the effect of time-varying axisymmetric staging on self-excited instabilities usingthe same multiple nozzle combustor in the present work. Their

study specifically examined the effect of impulse (very fast) tran-sients in staging equivalence ratio on the onset and decay of com-bustion instability. They found that the end states of the transient,quantified by the RMS of the combustor pressure fluctuations,only depended on the staging equivalence ratio. However, thetime it took for the combustor to become stable or unstable de-pended on both the staging amount and the transient direction(fuel addition or removal). It is unclear if these transient endstates are dependent on the symmetry (or lack of thereof) of thefuel staging injection location.

It is evident from the literature that both axisymmetric andnon-axisymmetric fuel staging can be effective in suppressingcombustion instability. However, it is still an open question asto whether or not one staging strategy is more effective than theother in realistic combustor configurations. The present workseeks to answer this question by comparing the effectiveness ofaxisymmetric and non-axisymmetric staging on suppressing self-excited longitudinal instabilities in a can combustor. The “effi-cacy” is quantified using three metrics: the RMS amplitude ofthe combustor after fuel staging is applied, the amount of timeit takes for the stability transition to occur, and the damping rateof the combustor when fuel staging is applied, which is a mea-sure of the stability margin at a given staging condition. Therest of the paper is organized as follows. First, the experimentalsetup of the multi-nozzle can combustor is described. Next, thedata analysis techniques are described. Finally, the characteristictimes, damping rates, and instantaneous phase difference imagesare compared between the staging configurations.

EXPERIMENTAL SETUP AND METHODSA schematic of the multi-nozzle combustor is shown in Fig. 1a;the experimental details are more extensively described in previ-ous work [11, 12]. The combustor burns a preheated, premixedmixture of natural gas and air at atmospheric pressure. The com-bustor liner is a quartz tube that permits full optical access. Airis supplied from an in-house compressor and preheated usinga 50kW process air heater. The air flow rate is metered usinga Sierra Instruments 780S mass flow meter and controlled us-ing needle valves. The main and staging natural gas flowratesare measured using Telledyne-Hastings HFM-301 and HFM-201flowmeters, respectively. The combustor contains five fuel noz-zles based on industrial hardware; a simplified nozzle cutaway isshown in Figure 1b. Each nozzle consists of an annulus for thepremixed fuel and air, an annulus for the staging fuel, a swirler,and a centerbody. The bulk flow velocity variation in each nozzleis less than 8%; this is due to small variations in the manufactureof each nozzle. Fuel staging is accomplished by injecting a smallamount of extra fuel in the fuel staging annulus while simultane-ously decreasing the premixed fuel flowrate. This staging fuelmixes with the main flow through upstream-facing holes in theswirler, and previous work [13] indicates this extra fuel is wellmixed by the nozzle exit. While there is a slight increase (no

2

(a) (b)

FIGURE 1: MULTI-NOZZLE COMBUSTOR (A) AND FUELSTAGING FLOW PATH (B)

more than 4.2%) in the global equivalence ratio when stagingfuel is added, the primary effect of fuel staging is a redistributionof fuel rather than an increase in thermal power.

The staging fuel is controlled using a Humphrey ProCon-trol PC3 proportional control solenoid valve, which in turn iscontrolled using LabView and a constant-current analog circuit.This system allows the amount of staging fuel (staging ampli-tude), the timescale of fuel change (transient duration), and di-rection (fuel increase or decrease) to be controlled. In this study,we consider changes in staging amplitude and direction, wherethe commanded valve opening time is held constant at 1 ms.

Axisymmetric (center nozzle) and non-axisymmetric (outernozzle) configurations are considered in this work. These config-urations are shown in Fig. 2. The outer nozzles are labeled 1-4(red text). For axisymmetric staging, the center nozzle equiva-lence ratio is varied during the transient, as shown Fig. 2a. Forthe non-axisymmetric staging case, the equivalence ratio of oneof the four outer nozzles is varied. It was found that Nozzles 2,3, and 4 had transition timescales and damping rates similar tothe center nozzle for a given φStaging, while Nozzle 1 had longertransition timescales and lower damping rates. The discussionfocuses on the comparing and contrasting the behavior of non-axisymmetric staging in Nozzles 1 (outlier case) and 2 (repre-sentative case) to the axisymmetric staging in the center nozzle.We consider staging amplitudes of φStaging = .80,φStaging = .85and φStaging = 0.90.

DiagnosticsK-type thermocouples monitor centerbody and dump plane tem-peratures. A water-cooled, recess-mounted PCB dynamic pres-sure transducer mounted on the dump plane measures the com-bustor pressure fluctuation amplitudes at a data acquisition rateof 16,384 Hz. The combustor pressure data are high-pass filteredto retain all frequencies above 10 Hz. Table 1 summarizes otherexperimental parameters.

CH* chemiluminescence is used as a qualitative marker of

FIGURE 2: SCHEMATIC OF NOZZLE STAGING CONFIG-URATION FOR (A) AXISYMMETRIC STAGING AND (B)NON-AXISYMMETRIC STAGING

TABLE 1: EXPERIMENTAL PARAMETERS

Parameter Value

Inlet Temperature 200 C

Inlet Velocity 26 m/s

Inlet Reynolds Number Red 17,000

Nozzle Swirl Number 0.7

Air Flow Rate 0.142 kg/s

heat release rate; this technique is widely used to estimate heatrelease rates from hydrocarbon-air premixed flames [14, 15].High-speed CH* chemiluminescence images are obtained usinga Photron SA4 high speed camera coupled with an Invisible Vi-sion UVi 1850-10 intensifier, a Nikon AF Micro-Nikkor 60mmf/2.8 lens, and a 432±5 nm bandpass filter. Images are capturedat 4000 frames per second for one or two seconds.

Data Screening

The instability onset and decay process is stochastic, and a smallpercentage of cases do not show clear stability transitions. Welimit the characteristic times and damping rates calculations tothe cases that show a clear transition. In this work, unstable op-eration is defined as having an RMS pressure amplitude greaterthan 0.07 PSI (0.483 kPa), which is 0.5% of the mean combustorpressure. The combustor pressure must additionally have a peakpower spectral density amplitude 30 times greater than the aver-age amplitude of all other frequencies to ensure the instability istonal. These criteria remove 14% of onset transitions and 16%of decay transitions. A fuel-staging amplitude is considered in-

3

effective when more than 50% of the tests at a given equivalenceratio fail to suppress the instability using these two criteria.

DATA PROCESSINGCharacteristic Time CalculationA system transitioning from a limit cycle to a stable fixed pointor vice-versa shows a two-asymptote structure, and the transitiontimescale between asymptotes is an important characterizationmetric. For acoustic systems that exhibit oscillatory behavior ata narrow-band, dominant frequency, the envelope of the pres-sure fluctuations is the best way to isolate the behavior of theasymptotes and the transition time between them. The pressureenvelope is obtained by taking the magnitude of the analytic sig-nal, z(t), in the same method described in previous work [11].The analytic signal is shown in Eq. 1, which has been shown byGabor [16] to be equivalent to Eq. 2 as long as the timescale ofthe phase variations is shorter than the timescale of the amplitudefluctuations [17].

z(t) = s(t)+ jH[s(t)] (1)

z(t) = a(t)e jΦ(t) (2)

The pressure signal envelope is modeled using the logistic equa-tion shown in Eq. 3. The logistic equation is an unambiguousway of modeling a two-asymptote transition.

p′(t) =A−B

1+ ek(t−t0)+B (3)

In Eq. 3, A is the limit cycle amplitude, B the staged-stable am-plitude, t0 the curve center, and k, the exponential rise or de-cay factor; the curve is symmetric relative to t0. There are anumber of features that make Eq. 3 a good model for this sys-tem. First, the sigmoid nature of the curve captures the dual-asymptote behavior of a combustor transitioning to or from sta-bility. Second, the initial growth or decay rate is approximatelyexponential, which is consistent with the behavior of a dampedacoustic oscillator. Finally, Eq. 3 can be used to define anunambiguous characteristic transition time. The characteristictime equation is shown by Eq. 4, and is obtained by solvingA−B

2e +B = A−B1+ek(t−t0)

+B for t− t0. The fractional decay ampli-tude is 2e, and using this value is similar to approximating thelogistic fit as the piecewise fit of two exponentials and determin-ing the resulting time constant.

τ = | ln(2e−1)k

| (4)

The value of k depends on transient direction; therefore, the ab-solute value in Eq. 4 ensures the characteristic time is alwayspositive. Figure 3 provides a graphic illustration of the logisticmodel applied to an axisymmetric staging instability decay tran-sition in (a) and an axisymmetric staging instability onset transi-tion in (b). The combustor pressure trace is in blue and the modelfit, obtained from least-squares nonlinear regression of the pres-

FIGURE 3: LOGISTIC FIT APPLIED TO (A) INSTABILITYDECAY AND (B) INSTABILITY ONSET TRANSITIONS

sure fluctuation envelop, is shown by the black dotted line. Theblack solid lines show t0± τ . In all cases the model is fit to thedata for 4 seconds, or approximately 65,000 data points. The fitparameters do not have a strong dependence on fit start time orinitial parameter guesses as long as the fit algorithm converges.However, to reduce any statistical bias, the start time of the fit isadjusted on a case-by-case basis to ensure the instability transi-tion happens near the middle of the fit time. The built-in functionnlinfit() in MATLAB R2015a is used for the fitting.

Combustor Damping CalculationIn addition to understanding the timescale of stability transitionsbetween the staging configurations, we also compare the rela-tive stability margins using the combustor damping rates [18,19].The physical effect of damping is to de-correlate fluctuations atone time with those at a later time. As such, systems that exhibitlarger overall damping will have fluctuations that are correlatedfor shorter lengths of time. This oscillation correlation time isproportional to the inverse of the damping rate, and it is this pro-portionality that allows the total system damping to be related tothe system damping rates. Combustor damping rates are calcu-lated using the method described by Stadlmaier et al. [19], andis only briefly summarized here. This method, which is based onthe method proposed by Lieuwen [18], assumes that the acousticpressure inside of a combustor is due to the superposition of Inonlinearly-interacting oscillators [18–20]. The motion of theseoscillators is governed by a second order system in η , which is

4

written as Eq. 5 for a linearly stable system [19].I

∑i=1

ηi +2νiηi +ω2i ηi = ζ (t) (5)

In Eq. 5, i is the current mode, ηi the modal amplitude, νi thenet damping rate, ωi the angular frequency, and finally ζ (t) thestochastic driving. Using Eq. 5, it can be shown that the autocor-relation of the pressure is related to the damping rate as shown inEq. 6 [19, 21].

ACF(τLag) =I

∑i=1

e−νiτLag cos(ωiτLag) (6)

In Eq. 6, ACF is the autocorrelation function and τLag is thecurrent autocorrelation lag of the pressure fluctuation. The righthand side of Eq. 6 is fit to the autocorrelation of the pressurefluctuation in the time domain in order to extract the damp-ing rate. Bayesian network analysis, which uses Markov ChainMonte Carlo methods to numerically simulate probability distri-butions for each parameter based on the given data, is a robustway of fitting nonlinear equations to data. The software packageJAGS [22] is used for the fitting. A single oscillator frequencyis used in this work as this combustor exhibits a single dominantinstability frequency around 530 Hz. The net damping rate is ob-tained from ν from the fit of Eq. 6 to a 1 second autocorrelationof the pressure during the staged-stable portion of the transientas shown in 3. Tests with synthetic data indicate a fit length of10 cycles is optimal for the dominant instability frequency anddamping rates examined in this work.

Instantaneous Phase Difference Images

The Rayleigh index, described by Eq. 7, is an important met-ric that characterizes the propensity of combustion systems toundergo instability. In Eq. 7, T is the period of an acoustic os-cillation, Ω the flame integration volume, p′(~x, t) the combustorpressure as a function of time and space, and q′(~x, t) the heat-release rate as a function of time and space. Positive Rayleighindex values indicate that energy is added to the acoustic field,while negative Rayleigh index values indicate energy is removedfrom the acoustic field. The sign of the Rayleigh index is de-termined by the relative phase difference between p′(~x, t) andq′(~x, t). A drawback of the Rayleigh index approach is that it re-quires accurate relative measurements of the heat release rates.While chemiluminescence intensity is proportional to heat re-lease rate, the constant of proportionality depends on the localequivalence ratio [14], and this dependence makes it difficult todetermine relative heat release rates in configurations where theequivalence ratio is spatially non-uniform. However, since thesign of the Rayleigh index is determined by the phase relation-ships between p′ and q′, and the relative phase relationships areinsensitive to local variations in equivalence ratio [23], instan-taneous phase difference images may be used to determine the

regions of the combustor that drive and damp the instability.

RI(T ) =∫ T

0

∫Ω

p′(~x, t)q′(~x, t)dΩdt (7)

Instantaneous phase difference images are obtained using amethod similar to Kheirkhah et al. [23] and described in moredetail in previous work [11]. A ±25Hz bandpass filter centeredat the dominant instability frequency is applied to the combustorpressure and each pixel of the CH* chemiluminescence imagein the frequency domain. Note that this band-pass fluctuationimage is also used to calculate the RMS and phase images pre-sented in Fig. 8. The CH* image is downsampled to half ofits original resolution to increase the signal-to-noise ratio anddecrease processing time before the bandpass filter is applied.Next, the analytic signal is computed for the filtered data and theinstantaneous phase angle Φ(t) obtained by taking the angle be-tween real and imaginary parts of the analytic signal. Finally, themagnitude of the instantaneous phase difference between p’ andq′, |Φ(p′)−Φ(q′)|, is obtained by subtracting the instantaneousphase angle of p′ from each downsampled pixel in the chemilu-minescence image.

RESULTSNozzle CharacterizationWhile the nozzles are nominally identical, there are small differ-ences in the bulk velocity and fuel convection time due to manu-facturing differences. These differences are characterized in thissection so they do not confound the comparison of axisymmet-ric to non-axisymmetric staging shown later. The bulk velocityvariations between nozzles is shown in Table 2. The differen-tial pressure is measured across the swirler and empirically cal-ibrated to yield bulk velocity. Table 2 reports the percent differ-ence in bulk velocities relative to the center nozzle and the actual∆P value. Nozzle 2 has the largest difference in velocity relativeto the center nozzle at 8%. This difference does not appear tobe significant, however, as analysis of the staged-stable pressureamplitudes, characteristic decay timescales, and damping ratesshow the behavior of nozzle 2 is not statistically significantly dif-ferent from the behavior of the center nozzle. Table 3 shows thepercent difference in characteristic fuel convection time for eachnozzle relative to the center nozzle. The characteristic convec-

TABLE 2: NOZZLE BULK VELOCITY MEASUREMENTS

Nozzle V (m/s) ∆V Relative to Center (%) ∆PPMean

(%)

Center 26.6 0 0.569

Nozzle 1 25.2 5.4 0.597

Nozzle 2 24.4 8.0 0.516

Nozzle 3 27.0 0.8 0.609

Nozzle 4 24.5 7.9 0.644

5

TABLE 3: FUEL CONVECTION TIMESCALE VARIATIONRELATIVE TO CENTER NOZZLE

φStaging Nozzle 1 Nozzle 2 Nozzle 3 Nozzle 4

φStaging = 0.80 6.77 % 9.56 % 4.78 % 9.96 %

φStaging = 0.85 4.85 % 1.21 % 24.24 % 1.82 %

φStaging = 0.90 12.7 % 9.4 % 20.8 % 5.37 %

tion time is obtained using an infrared absorption measurementconducted at equivalent non-reacting conditions. An equation ofthe same form as Eq. 3 is fit to the infrared signal, and a char-acteristic time obtained using an analogous form of Eq. 4. Forreference, the characteristic fuel convection times range from 14to 25 ms for the test conditions shown. Nozzle 3 tends to havethe greatest difference in fuel convection timescale relative to thecenter nozzle. Despite this difference, Nozzle 3 has very simi-lar stability transition timescales and damping rates to the centernozzle. This similarity suggests these differences in convectiontimescales are not significant.

Bifurcation Equivalence RatioThe bifurcation equivalence ratio is defined as the staging equiva-lence ratio where at least 90% of the staging attempts result in thecombustor transitioning from an unstable limit cycle to a staged-stable condition. As a result, staging equivalence ratios greaterthan the bifurcation equivalence ratio will almost always stabilizethe combustor. The bifurcation equivalence ratio was determinedby conducting unstable-to-stable transients using staging equiva-lence ratios ranging from 0.75 to 0.86 in 0.01 increments, where0.01 is the uncertainty of our measurement in equivalence ratio.At least four ensembles of the equivalence ratio sweep were con-ducted for each nozzle. Although the five injectors are nominallyidentical, we find variations in the bifurcation equivalence ratiofor each injector. Table 4 contains a summary of the bifurcationequivalence ratio and shows that it differs for each nozzle. Noz-zles 2 and 3 exhibit bifurcation equivalence ratios similar to thecenter nozzle. This similarity suggests that non-axisymmetricstaging can be as effective as axisymmetric staging. The differ-ence in performance of Nozzles 1 and 2 is likely driven by minorvariations in the staging fuel mixing or staging fuel circuit dy-namic response between the nozzles, as the characterization sug-gests that the nozzle-to-nozzle differences in bulk velocity andfuel convection timescales are not significant.

Comparison of p’RMS AmplitudesThe p′RMS amplitudes before and after the transient are comparedto determine how well different staging strategies reduce the in-stability amplitude. Figure 4a shows the p′RMS amplitudes be-fore and after the transient for axisymmetric staging, Fig. 4bfor non-axisymmetric Nozzle 1 staging, and Fig. 4c for non-axisymmetric Nozzle 2 staging. In these and all subsequent box-

TABLE 4: BIFURCATION EQUIVALENCE RATIOS

Nozzle Bifurcation Equivalence Ratio

Center 0.79

Nozzle 1 0.85

Nozzle 2 0.83

Nozzle 3 0.80

Nozzle 4 0.78

plots, D indicates decay transients, O indicates onset transients,U means un-staged, and S means staged. The upward and down-ward facing triangles show the estimated 95% confidence inter-val [24]. This conservative confidence interval includes the un-certainty involved in comparing different sample sizes [24]. Redcrosses denote outliers, which are defined using the standard con-vention of 1.5 times the inner quartile range (IQR). The num-bers at the top show the number of ensembles for each condition.The staging effectiveness is evaluated by comparing the unsta-ble combustor amplitudes to the staged-stable amplitudes. Thered dotted line shows the delineation between stable and unsta-ble pressure amplitudes. The RMS pressure is calculated usingtwo seconds of pressure data both before and after the transient.When a logistic fit to the pressure time series was possible, thebefore transient RMS is taken starting at t0−3τ−2 seconds andthe after transient RMS was taken starting at t0 + 3τ . When alogistic fit was not possible, the before transient region was de-fined as starting 2.75 seconds before the valve actuation and theafter transient region defined starting 0.75 seconds after the valveactuation time. These offsets prevent averaging over the pressurechange during the transient. Figure 4a indicates that all threeequivalence ratios are effective at suppressing the instability, asall of the staged pressure amplitudes ((D,S),(O,S)) are stable andall un-staged pressure amplitudes ((D,U),(O,U)) are unstable. Incontrast, φStaging = 0.80 for non-axisymmetric staging cases inFigs. 4b and 4c fail to reliably suppress the instability as themedian p′RMS amplitude is unstable during the staged part of thetransient ((D,S),(O,S)). Although Nozzle 2 in Fig. 4c fails to sup-press the instability at φStaging = 0.80, the median staged p′RMSamplitude is lower than the median staged p′RMS amplitude forNozzle 1 staging in Fig. 4b. Furthermore, when the Nozzle 2equivalence ratio is greater than the bifurcation equivalence ra-tio, the staged-stable amplitude is not significantly different fromaxisymmetric staging at the same equivalence ratio.

Comparison of Characteristic TimesThe previous section focused on the asymptotic behavior of thetransients by comparing the differences in p′RMS amplitudes be-fore and after staging. The current section examines the char-acteristic decay and rise timescales for different staging config-urations and equivalence ratios. Figure 5a shows boxplots of

6

(a) (b) (c)

FIGURE 4: BOXPLOTS OF P’RMS FOR AXISYMMETRIC STAGING (A) AXISYMMETRIC STAGING, (B) NON-AXISYMMETRIC NOZZLE 1 STAGING, AND (C) NON-AXISYMMETRIC NOZZLE 2 STAGING

the characteristic decay time for axisymmetric staging, Fig. 5bshows boxplots of non-axisymmetric staging (Nozzle 1), and Fig.5c shows boxplots of non-axisymmetric staging (right). The y-axis shows the characteristic time, computed using Eq. 4, andthe x-axis shows the staging equivalence ratio. The y-limits inFig. 5 have been scaled to highlight the differences in the box-plots; this scaling hides one outlier for φStaging = 0.80 in Fig.5a and two outliers for φStaging = 0.85 in Fig. 5c. Figure ??shows that the non-axisymmetric Nozzle 1 staging case in Fig.5b has a longer median decay time at a given equivalence ratiothan axisymmetric staging in Fig. 5a or non-axisymmetric rightstaging in Fig. 5c. In contrast, the median decay timescales foraxisymmetric center staging in Fig. 5a and non-axisymmetricNozzle 2 staging in Fig. 5c are nearly the same. This suggeststhat non-axisymmetric and axisymmetric staging behave simi-larly during a transient, which is congruent with the fact thatthey can be equally effective at suppresing the instability. Addi-tionally, non-axisymmetric Nozzle 1 staging in Fig. 5b exhibitsmore variability than the other configurations. This higher vari-ation is consistent with the higher p′RMS variation noted in Fig.4b, and is likely due to a difference in the mixing of the stagingfuel or the response of the fuel circuit within the nozzle, as men-tioned previously. Figure 6 shows boxplots of the characteristicrise time for axisymmetric staging in Fig 6a, non-axisymmetricNozzle 1 staging in Fig. 6b, and non-axisymmetric Nozzle 2staging in Fig. 6c. The y-axis shows the characteristic rise time,in milliseconds, computed using Eq. 4, and the x-axis showsthe staging amplitude. Figure 6 does not show a clear relation-ship between characteristic onset time and staging amplitude forthe configurations, as the confidence intervals on the medians

(a) (b) (c)

FIGURE 5: CHARACTERISTIC DECAY TIMES FOR (A) AX-ISYMMETRIC STAGING, (B) NON-AXISYMMETRIC NOZ-ZLE 1 STAGING, AND (C) NON-AXISYMMETRIC NOZZLE2 STAGING

overlap. However, comparing the onset time in Fig. 6 to thedecay time in Fig. 5 shows the characteristic rise times are sig-nificantly longer than the characteristic decay times at a givenstaging equivalence ratio and configuration. The instability risecases also have larger IQR’s, indicating that the instability risecases are more variable than the instability decay cases at a givenequivalence ratio and configuration. These results are consistentwith rise rates obtained in a previous work [11], and suggest that

7

(a) (b) (c)

FIGURE 6: CHARACTERISTIC RISE TIMES FOR (A) AX-ISYMMETRIC STAGING, (B) NON-AXISYMMETRIC NOZ-ZLE 1 STAGING, AND (C) NON-AXISYMMETRIC NOZZLE2 STAGING

the instability growth process is largely independent of the initialstaging condition. The longer, more variable timescales for theinstability rise cases are consistent with the idea that combustioninstabilities grow from stochastic perturbations to the flame thateventually couple with the combustor acoustics.

Comparison of Damping RatesIn addition to knowing the staging effectiveness, quantified bythe reduction in instability amplitude, and the timescale of insta-bility transition, quantified by the characteristic transition time,it is important to know the relative stability margin of the com-bustor for each staging strategy. A good way to quantify thestability margin is to compute the combustor damping rates, ashigher damping rates indicate larger stability margins [18, 19].Figure 7a shows boxplots of the damping rate for axisymmetricstaging, Fig. 7b shows boxplots of the damping rate for non-axisymmetric Nozzle 1 staging, and Fig. 7c shows boxplots ofthe damping rate for non-axisymmetric Nozzle 2 staging. They-axis shows the damping rate, in radians/s, while the x-axisshows the staging equivalence ratio and direction. The medianstaged-stable damping rates for axisymmetric staging in Fig. 7aare not statistically significantly different from the median damp-ing rates for non-axisymmetric Nozzle 2 staging in Fig. 7c. Thismirrors the results in the pressure amplitude section, and sug-gests that in general, axisymmetric staging can be as effective asnon-axisymmetric staging.

However, there is a significant difference between mediandamping rates for axisymmetric staging in Fig. 7a and non-axisymmetric Nozzle 1 nozzle staging in Fig. 7b, where Nozzle

(a) (b) (c)

FIGURE 7: DAMPING RATES FOR (A) AXISYMMETRICSTAGING, (B) NON-AXISYMMETRIC NOZZLE 1 STAG-ING, AND (C) NON-AXISYMMETRIC NOZZLE 2 STAGING

1 staging is less effective than center nozzle staging. The anal-ysis in the nozzle characterization section suggests the reducedeffectiveness is not due to different nozzle bulk velocity or fuelconvection times, as nozzles that have larger differences in bulkflow velocity and convective time still behave similarly to thecenter nozzle. We suspect the difference is in the mixing of thestaging fuel, the response of the staging fuel line to the instabil-ity, or some combination of the two in Nozzle 2. This highlightsthe importance of considering nozzle-to-nozzle hardware varia-tions when determining the efficacy of different staging strate-gies. Finally, Fig. 7 suggests the damping rate for φStaging = 0.90does depend on transient direction, where decay cases show lessvariable and larger median damping rates than onset cases. Thegreater variability in the initial, staged-stable p′RMS noted in Fig.4b for the axisymmetric φStaging = 0.90 case is likely due to theincreased variability in damping rate, as the staged-stable pres-sure amplitude is dependent on the damping rate [20]. At thispoint it is not clear why this dependence of damping rate on tran-sient direction is only evident for the highest staging equivalenceratio. It is possible the φStaging = 0.90 is rich enough that it beginsto destabilize the combustor, as the stability map shows increas-ing instability amplitudes with increasing equivalence ratios atthis inlet velocity and preheat temperature.

Time-Averaged Flame Structure Comparison

The flame structures are analyzed in the following sections. Fig-ure 8 shows time-averaged flame structure for each staging con-figuration. Following the approach of Samarasinghe et al. [8],the flame structure is decomposed into mean, RMS, and phaseimages filtered at the instability frequency. The first column

8

shows the time-averaged flame structure of an unstable flamewith φ = 0.70 and no staging, the second column shows ax-isymmetric staging at φStaging = 0.85, the third column showsnon-axisymmetric Nozzle 1 staging with φStaging = 0.90, and thefourth column shows non-axisymmetric Nozzle 2 staging withφStaging = 0.85. The number at the bottom of the image corre-sponds to the outer nozzle number as shown in Fig. 2. The firstrow shows the time-averaged CH* chemiluminescence signal,the second row shows the RMS of the fluctuations normalized bythe time-averaged CH* chemiluminescence for each pixel, andthe third row shows the phase of the fluctuations at the dominantinstability frequency. The colorbar for the unstable RMS imageranges from 0 to 50 % of the mean, while the colorbars for thestaged-stable RMS images range from 0 to 25 % of the mean;these ranges are chosen to highlight differences in the structureof the RMS images.

Figures 8a,b,c, and d show the highest heat release is con-centrated in the flame interaction regions. When the flame isunstable in Figure 8a, the high heat release region is fairly broad.This region becomes slightly narrower when axisymmetric fuelstaging is applied in Fig. 8b, suggesting less flame motion. Thetime-averaged heat release is biased towards the interaction re-gion between the staged flame and the center flame for the non-axisymmetric staging cases in Figs. 8c and d. This bias is to-wards the right side of the image for Nozzle 1 and the left sideof the image for Nozzle 2. Figure 8d shows that when the flameis unstable, the largest heat release rate fluctuations occur in theflame/wall interaction region and towards the base of the flame-flame interaction region. There are additional fluctuations nearthe flame tips. The majority of these fluctuations are suppressedwhen the flame is staged, as indicated by the lower intensity col-ors in Figs. 8f, g and h. All three staging configurations showthe largest amplitude fluctuations are in the flame-wall region.The non-axisymmetric staged cases in Figs. 8g and h also showthe largest amplitude flame-wall fluctuations occur in the stagedflame; for Nozzle 1 this is the right flame and for Nozzle 2 thisis the left flame. However, the general similarity in structure be-tween RMS images for staged-stable conditions is remarkablegiven the differences in time-averaged flame structures betweenthe configurations.

The final comparison is made for the phase images in thethird row. When the combustor is unstable in Fig. 8g, horizontalbands of constant phase are visible, indicating a convective insta-bility as has been reported previously [8]. The horizontal bandsof constant phase are significantly disrupted for axisymmetricstaging in Fig. 8h and non-axisymmetric Nozzle 2 staging in8l, where the structure of the phase image is largely incoherent.Much of this apparent incoherency is due to low signal-to-noiseratios caused by low fluctuation amplitudes, as the instantaneousphase comparisons in Fig. 10 show more clear structure. The leftside of the non-axisymmetric Nozzle 1 staging phase in Fig. 8ishows structure similar to the unstable phase in Fig. 8i while the

right side shows a discontinuity in the bands of constant phaseabove the staged flame. The clearer phase structure in Fig. 8k islikely due to the larger pressure fluctuation amplitude.

The time-averaged flame structure is examined in more de-tail using slices from a tomographic reconstruction of the flame,where the grey squares at the bottom of the image indicate thelocation of the dump plane. The reconstructions are obtainedby calculating the inverse Radon transform using a filtered backprojection algorithm as described in Refs. [12] and [25]. Fig-ure 9 shows slices of CH* chemiluminescence taken through thecenter of the center nozzle and nozzles 3 and 1. For the cen-ter staging case in Fig. 9a, the flame stabilizes in both the innerand outer shear layers. In contrast, the non-axisymmetric stag-ing case in Fig. 9b shows the flame stabilizes in the inner shearlayer only. It is surprising that non-axisymmetric and axisym-metric staging can be equally effective yet have different flamestabilization mechanisms. Future work will examine these dif-ferences using OH-PLIF measurements.

Comparison of Instantaneous Phase ImagesFigure 10 shows snapshots of the |Φ(p′)−Φ(q′)|, the magni-tude of the instantaneous phase difference between pressure andheat release rate fluctuations, taken at different non-dimensionaltimes during the transient for representative axisymmetric stag-ing in Fig. 10a, non-axisymmetric Nozzle 1 staging in Fig. 10b,and non-axisymmetric Nozzle 2 staging in 10c. The combustorpressure time trace is shown at the bottom of each figure, wherethe black lines show the instant of each |Φ(p′)−Φ(q′)| snapshotrelative to the pressure. The evolution of the instantaneous phaseimage for axisymmetric staging in Fig. 10a has been discussedin more detail in a previous publication [11]; only the main fea-tures are highlighted here. Long before the start of the transientin Fig. 10a(i) the heat release rates and pressure fluctuations arein-phase throughout much of the center of the combustor, as in-dicated by the large dark blue region. As the transient progressesto the beginning of the pressure decay in Fig. 10a(ii), out-of-phase structures begin to appear in the center of the combustor inthe flame interaction region. These out-of-phase structures growthroughout the transient until they displace much of the regionthat was formerly in-phase, as illustrated by the large yellow re-gions visible at the end of the transient in Fig. 10a(vi). Thisdestructive interference between pressure and heat release is re-sponsible for suppressing the instabilities during axisymmetricstaging.

Figure 10b shows snapshots of the instantaneous phase be-tween p′ and q′ for a representative φStaging = 0.90 Nozzle 1 stag-ing. This equivalence ratio is chosen because it is above the bi-furcation equivalence ratio and has a more similar damping rateand final p′RMS amplitude to the other staging cases, and bettershows the changes in instantaneous phase relationships duringthe transient. The initial phase distribution before the transientin Fig. 10b(i) very much resembles the phase distribution before

9

FIGURE 8: MEAN, RMS, AND PHASE IMAGES OF DIFFERENT STAGING CONFIGURATIONS

(a) (b)

FIGURE 9: TOMOGRAPHIC CH* SLICES THROUGH THESTAGED NOZZLE FOR (A) AXISYMMETRIC STAGING ATφStaging = 0.85 AND (B) NOZZLE 1 STAGING AT φStaging =0.85

the axisymmetric staging transient in Fig. 10a(i). This similaritymakes sense, as the initial self-excited instability state is simi-lar for all staging configurations. The phase distribution changeslittle until Fig. 10b(iii), the halfway point of the transient de-cay. At this point, the left-side of the combustor has begun togo out-of-phase while the center region and the right, staged side

of the combustor remain predominantly in-phase. The generalstructure of the in- and out-of-phase regions closely resemble thephase image previously shown in Fig. 8(i).

The difference between the phase relationships in the rightnozzle and the rest of the combustor is most striking at the end ofthe transient in Fig. 10b(v). The right, staged nozzle shows alter-nating bands of in- and out-of-phase oscillations, while the cen-ter and left side of the combustor do not. The alternating patternin the staged nozzle resembles the pattern in the axisymmetricstaged nozzle at the same non-dimensional time in Fig. 10a(v),suggesting that effect of staging on flame structure is similar forboth nozzles. It should also be noted that the non-axisymmetricNozzle 1 staging decay occurs over a longer timescale than theaxisymmetric staging decay and exhibits more intermittency, asthe pressure amplitude increases again by the end of the tran-sient in Fig. 10b(vi). This increase is caused by small, stochasticvariations of the phase relationships in Nozzle 1, which causeweaker destructive interference and lead to larger pressure fluc-tuation oscillations. Figure 10c shows instantaneous phase dif-ference snapshots for Nozzle 2 staging. In this view, the stagedflame is on the left of the image. The evolution of the instanta-neous phase difference structure in Nozzle 2 closely resemblesthe structure in Nozzle 1, suggesting that non-axisymmetric fuelstaging has similar effects on each flame. Future work will in-

10

(a) (b) (c)

FIGURE 10: INSTANTANEOUS PHASE DIFFERENCE BETWEEN P’ AND Q′ FOR (A) AXISYMMETRIC ΦStaging = 0.85, (B)NON-AXISYMMETRIC (NOZZLE 1) ΦStaging = 0.90, AND (C) NON-AXISYMMETRIC (NOZZLE 2) ΦStaging = 0.85

vestigate these dynamics in more detail using OH-LIF.We suspect the efficacy of different fuel staging configura-

tions is strongly dependent on the instability mode shape. Thelongitudinal instability is nominally axisymmetric at this con-dition, and this axisymmetry is likely why axisymmetric stag-ing can be as effective as non-axisymmetric staging. When theinstability mode is not axisymmetric, it is likely that a non-axsymmetric staging strategy can be designed to better suppressthe instability, as Cohen et al. [5] did. This work also high-lights the importance of considering nozzle-to-nozzle variations,as these variations can confound effects. In particular, we findthe bifurcation equivalence ratio is very sensitive to these poten-tial differences. However, when staging is sufficient to suppressthe instability (greater than the bifurcation equivalence ratio), wefind the staged-stable p′RMS and damping rates in general do nothave a statistically-significant dependence on the staging con-figuration. Nozzle 1 staging behaves as an outlier case, and itsunique behavior is likely due to a difference in the mixing ofthe staging fuel or the response of the staging fuel circuit in thatnozzle.

CONCLUSIONSThis work has compared the effect of fuel staging configurationon longitudinal, self-excited instabilities in a multiple-nozzle can

combustor. In general, non-axisymmetric staging can be as ef-fective as axisymmetric staging. However, small hardware vari-ations between nozzles can affect the bifurcation equivalence ra-tio. These changes in the bifurcation equivalence ratio can makestaging configurations appear less effective; however, both ax-isymmetric and non-axisymmetric staging configurations exhibitsimilar staged-stable p′RMS and damping rates for a given equiva-lence ratio greater than the bifurcation equivalence ratio. Thisresult is surprising given that the time-averaged flame shapesare very different between axisymmetric and non-axisymmetricstaging configurations. Future work will investigate these differ-ent flame shapes using OH-LIF. Finally, this work highlights theimportance of considering hardware variations among nominallyidentical nozzles in a multiple nozzle combustor.

ACKNOWLEDGMENT

The authors would like to thank Dr. Keith McManus and Dr.Janith Samarasinghe from G.E. Global Research, and Dan Dolei-den from Penn State for their contributions to this work. Theauthors also want to thank the U.S. Department of Energy forfunding this work under award number DE-FE0025495 and con-tract monitor Mark Freeman.

11

REFERENCES[1] Lieuwen, T. C., and Yang, V., eds., 2005. Combustion In-

stabilites in Gas Turbine Engines, Vol. 210 of Progress inAstronautics and Aeronautics. American Institute of Aero-nautics and Astronautics.

[2] Lieuwen, T. C., and Yang, V., 2013. Gas Turbine Emis-sions. Cambridge University Press, New York.

[3] Bulat, G., Skipper, D., McMillan, R., and Syed, K. “Ac-tive control of fuel splits in gas turbine dle combustion sys-tems”. In ASME Turbo Expo 2007: Power for Land, Sea,and Air, no. GT2007-27266.

[4] Hermann, J., Orthmann, A., Hoffmann, S., and Berenbrink,P., 2000. “Combination of active instability control andpassive measures to prevent combustion instabilities in a260MW heavy duty gas turbine”. RTO AVT Symposium.

[5] Cohen, J., Hagen, G., Banaszuk, A., Becz, S., and Mehta,P., 2011. “Attenuation of combustor pressure oscillationsusing symmetry breaking”. In 49th AIAA Aerospace Sci-ences Meeting, no. AIAA 2011-60.

[6] Davis, L. B., and Black, S. H. Dry low NOx combus-tion systems for ge heavy-duty gas turbines. Report GER-3568G, GE Power Systems.

[7] Li, L., Sun, X., Lioi, C., and Yang, V., 2017. “Effect of az-imuthally nonuniform heat release on longitudinal combus-tion instabilities”. Journal of Propulsion and Power, 33(1),pp. 193–203.

[8] Samarasinghe, J., Culler, W., Quay, B. D., Santavicca,D. A., and O’Connor, J., 2017. “The effect of fuel stag-ing on the structure and instability characteristics of swirl-stabilized flames in a lean premixed multi-nozzle can com-bustor”. Journal of Engineering for Gas Turbines andPower, 139, pp. 121504–121504–10.

[9] Kim, K. T., and Hochgreb, S., 2012. “Measurements oftriggering and transient growth in a model lean-premixedgas turbine combustor”. Combustion and Flame, 159(3),pp. 1215–1227.

[10] Kabiraj, L., Sujith, R. I., and Wahi, P., 2011. “Bifurcationsof self-excited ducted laminar premixed flames”. Jour-nal of Engineering for Gas Turbines and Power, 134(3),pp. 031502–1–031502–7.

[11] Culler, W., Samarasinghe, J., Quay, B. D., Santavicca,D. A., and O’Connor, J. “The effect of transient fuel stag-ing on self-excited instabilities in a multi-nozzle model gasturbine combustor”. In ASME Turbo Expo 2017, Charlotte,North Carolina, no. GT-2017-63479.

[12] Samarasinghe, J., Peluso, S., Szedlmayer, M., De Rosa, A.,Quay, B., and Santavicca, D., 2013. “Three-dimensionalchemiluminescence imaging of unforced and forced swirl-stabilized flames in a lean premixed multi-nozzle can com-bustor”. Journal of Engineering for Gas Turbines andPower, 135(10), p. 101503.

[13] Orawannukul, P., 2014. “An experimental study of forced

flame response in technically premixed flames in a lean pre-mixed gas turbine combustor”. Dissertation, The Pennsyl-vania State University.

[14] Nori, V., and Seitzman, J., 2008. “Evaluation of chemilu-minescence as a combustion diagnostic under varying oper-ating conditions”. In 46th AIAA Aerospace Sciences Meet-ing and Exhibit, Reno, Nevada, no. AIAA 2008-953.

[15] Lee, J. G., and Santavicca, D. A., 2003. “Experimentaldiagnostics for the study of combustion instabilities in leanpremixed combustors”. Journal of Propulsion and Power,19(5), pp. 735–750.

[16] Gabor, D., 1945. “Theory of communication”. Journal ofthe Institution of Electrical Engineers - Part III: Radio andCommunication Engineering, 93(26), pp. 429–441.

[17] Boashash, B., 1992. “Estimating and interpreting the in-stantaneous frequency of a signal. i. fundamentals”. Pro-ceedings of the IEEE, 80(4), pp. 520–538.

[18] Lieuwen, T., 2005. “Online combustor stability margin as-sessment using dynamic pressure data”. Journal of engi-neering for gas turbines and power, 127(3), pp. 478–482.

[19] Stadlmair, N. V., Hummel, T., and Sattelmayer, T., 2017.“Thermoacoustic damping rate determination from com-bustion noise using bayesian statistics”.

[20] Culick, F., and Kuentzmann, P., 2006. Unsteady mo-tions in combustion chambers for propulsion systems.No. RTO-AG-AVT-039. NATO Research and TechnologyOrganization,Neuilly-Sur-Seine.

[21] Noiray, N., 2017. “Linear growth rate estimation from dy-namics and statistics of acoustic signal envelope in turbu-lent combustors”. Journal of Engineering for Gas Turbinesand Power, 139(4), p. 041503.

[22] Plummer, M., 2015. JAGS Version 4.0.0 User Manual.[23] Kheirkhah, S., Geraedts, B., Saini, P., Venkatesan, K., and

Steinberg, A., 2017. “Non-stationary local thermoacousticphase relationships in a gas turbine combustor”. Proceed-ings of the Combustion Institute, 36(3), pp. 3873–3880.

[24] McGill, R., Tukey, J. W., and Larsen, W. A., 1978. “Vari-ations of box plots”. The American Statistician, 32(1),pp. 12–16.

[25] Samarasinghe, J., 2016. “The structure and velocity-forcedresponse of swirl-stabilized flames in a lean-premixedmulti-nozzle can combustor”. Ph.D. Thesis, The Pennsyl-vania State University.

12