36th Aerospace Sciences Meeting & Exhibit January 12-15, 1998 / Reno, NV

AIAA 98-0642 Combustion Instability Suppression in Liquid-Fueled Combustors Keith R. McManus, John C. Magill, and Michael F. Miller Physical Sciences Inc. 20 New England Business Center Andover, MA 01810 36th Aerospace Sciences Meeting & Exhibit January 12-15, 1998 / Reno, NV For permission to copy or republish, contact the 1801 Alexander Bell Drive, Suite 500, Reston, VA 22091

COMBUSTION INSTABILITY SUPPRESSION IN LIQUID-FUELED COMBUSTORS Keith R. McManus*, John Magill, and Michael F. Miller Physical Sciences Inc. 20 New England Business Center Andover, MA 01810 AIAA-98-0642 Abstract A study has been performed to evaluate the effectiveness of a closed-loop control system to suppress thermo-acoustic combustion instabilities. The pulsewidth modulation technique was implemented both in a numerical simulation of a dynamic combustion instability and experimentally. The control system was effective in suppressing a simulated thermo-acoustic oscillation based on a simplified system model. The control system is undergoing scale-up design and will be tested on a 200 kw gas turbine combustor rig. The experimental apparatus is described in detail below along with its baseline operating characteristics. The combustor exhibits a strong thermo-acoustic combustion instability which manifests itself in large pressure and flame emission fluctuation amplitudes near a frequency of 140 Hz. The instability tends to grow as the fuel and air flowrates are increased from those corresponding to the lean stability limit. Open-loop control experiments using main fuel flow modulation indicate that the combustion process can be manipulated to produce strong pressure fluctuations at the driving frequency. Closed-loop control experiments using pulsewidth modulation on the main fuel injector are currently underway and will be reported in a future paper. Introduction Thermo-acoustically coupled oscillations occur in many practical combustion systems due to the confinement of a flame inside an enclosure. In many fluid mechanical experiments, it has been observed that dynamic flow instabilities may be sustained and amplified due to feedback from a resonant acoustic mode of the experimental setup. This phenomenon has *Principal Research Scientist, Member AIAA Principal Scientist, Member AIAA Principal Scientist, Member AIAA Copyright 1998 by the American Institute of Aeronautics and Astronautics, Inc. All rights reserved. been shown to be responsible for controlling the frequency of eddy shedding in some cases (see, for example, Ref. 1). In flow devices which are designed to stabilize flames and confine combustion gases, this coupling mechanism can become a strong driver of unwanted pressure oscillations. The amplification factors in these systems can be orders of magnitude larger than those observed in non-reacting flow experiments due to accelerated flame propagation within unsteady flow structures and subsequent oscillations in heat release. In modern combustion systems with design constraints which necessitate very large heat loadings, the problem of strongly coupled combustion oscillations can limit the operating margin and lead to unsatisfactory performance in many respects. Gas turbine engines represent an extreme case where ever increasing performance goals have led to very compact combustors with heat loadings reaching values well in excess of 100 MW/m 3. Thermo-acoustic oscillations in these devices can have a severe impact on their performance with respect to combustion efficiency, pollutant emissions, and service life. Past design procedure has been to reduce the tendency for unstable combustion through costly cut-and-try tests where geometrical and/or secondary flow split variations have been used to change the intrinsic acoustic and fluid dynamic properties of the device. This procedure has become unsatisfactory due to economic and project schedule constraints and the industry is looking for more efficient methods to combat the combustion instability problem. Recently, the idea of actively controlling combustion oscillations has gained much attention. (For reviews of this work, see Refs. 2-4.) The active control systems being studied are designed to suppress unstable combustion oscillations by modulating combustor inputs such as air and fuel feed rates or by directly modulating the pressure field with an acoustic source. When these parameters are modulated at an appropriate 1

phase and magnitude with respect to a combustion oscillation, the oscillation can be suppressed. Most of the active control experiments have been performed in simplified benchtop rigs with relatively low heat release rates (1 to 15 kw) and with gaseous fuel. The present work is focused on extending these active control strategies to suppress combustion oscillations in a 200 kw liquid-fueled model gas turbine combustor. The first part of the paper describes the control methodology adopted in our work and summarizes the results of a proof-of-concept experiment previously reported. 5 Then, the 200 kw experimental facility is described along with detailed measurements of the combustion dynamics associated with it. The results from open-loop frequency response experiments using the full-scale fuel injector will be described. Active control results using this device will be reported in a future paper. Control Technique: Amplitude-Based Pulse Width Modulation (PWM) The controller achieves instability suppression by exploiting the fact that out-of-phase oscillations in heat release and pressure result in damping of a resonantly coupled combustion instability. The controller is designed to track and suppress oscillations by injecting fuel pulses into the combustion chamber to create heat release perturbations which are out-of-phase with those associated with the combustion instability. A block diagram of the controller is shown in Figure 1. The control input is the fluctuating signal from the flame emission sensor (other possible controller inputs we have investigated include the unsteady combustion chamber pressure). The controller tracks this signal and generates a pulse trigger when a zero crossing is detected. At a fixed time delay after the pulse trigger, a control pulse is sent to the auxiliary fuel injector which opens the solenoid valve. Operating in parallel with the Input From Flame Emission Sensor High Pass Filter Amplitude Detector Trigger Generator Valve Pulse Modulator Output to Auxillary Fuel Injector Figure 1. Block diagram for amplitude-based pulse width modulation control. pulse trigger synthesis, a filter is used to detect the peak-to-peak amplitude of the emission oscillation, or the oscillation envelope. The width of the control output pulse (which dictates the amount of fuel injected during the pulse event) is made proportional to the instantaneous magnitude of the signal envelope. The constant of proportionality in calculating pulse width provides an overall gain adjustment to the controller. Feedback Control Results The controller was tested both in computer simulations utilizing a combustion dynamics model previously described 6 as well as experimentally. The simulations show that, without control, the combustor enters a large amplitude oscillation which grows exponentially. This oscillation represents a thermo-acoustic combustion instability and in a real combustor would result in a limit cycle oscillation. Similar behavior is observed in the experimental combustor. When PWM control is applied to the model, the unbounded growth of the oscillation is suppressed; however, a low amplitude oscillation persists. A spectral analysis of the pressure time histories for the cases with and without control indicate a strong suppression of the dominant unstable mode and the persistence of low amplitude modes shifted to slightly higher and lower frequencies (see Figure 2). PSD <p'> 0.025 0.02 0.015 0.01 0.005 0 No Control PWM Control (x 10) 0 50 100 150 200 Frequency (Hz) D-3421z Figure 2. The predicted effect of feedback control on the power spectral density of pressure fluctuations. Feedback control experiments have been performed on a small-scale liquid-fueled combustor. Preliminary results from control experiments using a gaseous-fuel secondary injector were previously reported along with a detailed description of the facility. 5 The results described below were obtained using a liquid-fuel secondary injector for control. 2

The control scheme was identical to that used in the model simulations; however, the control delay parameter was modified to take into account the physical lag associated with the auxiliary fuel injection event. This intrinsic delay is associated with the processes of fuel convection, evaporation, fuel/air mixing, ignition and flame propagation. Feedback control experiments were performed to demonstrate the feasibility of suppressing the thermoacoustic instability near 70 Hz using PWM control. The control scheme was identical to that used in the model simulations; however, the test section pressure was used as the control input rather than the fluctuating heat release parameter, q1. The control algorithm was programmed on a PC-based digital signal processor (National Instruments, DSP2200). Open-loop response experiments were performed to determine the actuator characteristics. The injector demonstrated the ability to produce coherent heat release pulses up to 300 Hz. The relative phase between the valve command signal and the resulting heat release pulse was measured and indicated a time delay of approximately 2 ms. Experiments were performed to determine the effects of varying the control delay on combustor stability. Figure 3 shows the effect of control delay on the rms levels of CH * emission and pressure fluctuations. The results indicate that the pressure and CH * emission fluctuation amplitudes exhibit very similar trends with a strong dependence on control time delay and hence on the phase at which the pulse begins. The amplitude of the fluctuations are suppressed for a broad Figure 3. Effect of varying control delay on rms amplitude of sensor signals. range of delays and are nearly unaffected for delays near 6 ms. The maximum suppression (approximately 50%) occurs with both short(- 2.5 ms) and long (- 12.5 ms) control delays when compared with the instability cycle period (approximately 14 ms). Accounting for the intrinsic actuator delay (approximately 2 ms), these data indicate that maximum instability suppression is achieved when the heat release pulse is added just before the point of minimum pressure during the instability cycle. A spectral analysis of the sensor signals when using feedback control with a delay of 12.5 ms indicates a significant suppression of the peak associated with the thermo-acoustic instability (see Figure 4). Similar to the results from the model simulation, the suppression of the peak is accompanied by a shift in frequency. In the experimental results shown, the shift is to a lower frequency. For other control delays a shift to higher frequencies was also observed. Figure 4. Effects of closed-loop control on pressure fluctuation power spectrum. Facility Description - 200 kw Combustor The high-pressure single nozzle combustion rig at PSI provides a significant capability to perform aerodynamic and combustion testing relevant to gas turbine engine development. The system consists of four main components: 1) a 2400 psi pressure vessel which is charged with ambient air using a compressor capable of delivering breathing quality air, 2) a 250 kw immersion-type electric heater capable of heating the air flow to maximum of 530 K, 3) test section, and 4) exhaust system. The facility is designed to accept a variety of test sections including those appropriate for combustion dynamics experiments and for applying 3

optical diagnostics. The air supply system provides the combustor with 0.25 kg/s (0.5 lbm/s) of air preheated up to 500 K (450 F) at up to 2 MPa (300 psig) for approximately 10 minutes. The facility also provides exhaust handling with water-spray cooling, closedcircuit cooling water for cooling of the combustor itself, and a high pressure liquid fuel supply. The combustor test section is mounted on a portable cart, and can be easily inserted into or removed from the facility. Figures 5 and 6 show pictures of the facility and the combustor assembly. Figure 5. Gas turbine combustor test facility. Diffuser Supply Flow Dome Entrance Orifice (Compressor) Fuel Injector Liner Combustor Interior Swirler Liner Passage Exit Orifice (Turbine) Figure 7. Air flow inside combustion chamber. D-7230z enters the diffuser through an inlet orifice intended to provide an acoustic boundary paralleling that provided by the compressor exit in a full engine configuration. From the diffuser, flow is split between the dome swirler and the liner passages. The swirler, as well as the high shear injector, are from a commercial jet engine. Two sides of the combustor are lined, simulating the inner and outer liner walls of an annular combustor. The liner walls are louvered to provide a cooling film, and larger holes in the liners feed dilution air to the combustion zone. Although the PSI combustor rig is designed for lower flowrates than those existing in full-scale aero-engines, it is believed that the dominant thermo-acoustic instabilities will be represented in the PSI system for the purposes of instability control system development. The chamber is fitted with a 4.4 cm diameter window and several instrumentation ports. Many of the ports are aligned with dilution holes to provide access to the interior of the combustor. A spark ignitor serves as the ignition source. Figure 6. Experimental combustor test section. The test section is designed to facilitate combustion instability control experiments and has been scaled to simulate a full-scale gas turbine combustor (see Figure 7). The dome region has been designed to accept a production engine fuel nozzle. In addition, appropriate dilution flows and inlet/exit constrictions have been added to reproduce the fluid dynamic and acoustic properties of an engine. Flow Instrumentation used to evaluate the operating and stability characteristics of the combustor include fast response pressure transducers and a flame emission sensor. The pressure transducers are wall mounted and measure the instantaneous pressure on the combustor wall and in the dilution air passage behind the liner. It was observed that the pressures measured at these two locations were nearly identical in spectral content and for subsequent combustion instability analyses the liner pressure transducer was used. The flame emission sensor consisted of a band pass filter ( max = 430 nm, 10 nm FWHM) placed in front of a photomultiplier tube (PMT) to isolate emission from excited CH radicals in the combustion gases. A focusing lens and aperture were used to provide a narrow field of view 4

(approximately 5 mm diameter) across the combustor at an axial location of x/h = 1.0 (measured from the fuel injector nozzle face). The output from the PMT was taken as a measure of the instantaneous energy release from gases contained in the measurement volume. The output from the sensors were acquired and analyzed using a laboratory PC. Results - Non-Reacting Flow Experiments Prior to installation in the combustor, the injector was tested to determine if, when coupled to a pulsed valve, the injector could modulate the fuel at the frequencies necessary to control instabilities. The tests furthermore provided a pressure vs. mass flow calibration plot for the nozzle. The injector actually contains two injection components - a pressure atomizing pilot nozzle for low flow rates and a set of periphery jets which feed fuel onto a surface on the swirler, constituting a high-shear air-blast atomizer. The nozzle contains an internal valve to turn on the periphery jets only at high pressures. Only the primary nozzle was used for these experiments. The test arrangement is shown in Figure 8. The frequency response was measured by applying a train of pulses at varying frequency to the pulsed valve, and observing the flow characteristics of fluid in the spray cone. These tests were conducted using a water spray for safety reasons. Cone flow rate fluctuations were detected optically by passing a HeNe laser beam through the spray and onto a silicon photodiode. Variation in the spray density along the beam propagation path causes variation of the extinction of the laser beam and this appears in the photodetector output as a signal modulation. The entire flow to the injector was modulated using a pulsed valve (General Valve, Series 9) Compressed Air Regulator Water Storage Vessel Adjustable Spring Check Valve Pulsed Valve Filter Laser Fuel Injector Photodetector Valve Driver Amp & Filter Figure 8. Fuel injector test configuration. Function Generator To Data Acquisition coupled to a modulation amplifier (General Valve, Model Iota One Driver). The frequency of the flow pulses was swept from 10 Hz to 500 Hz during a sweep time of 8.1 sec. The valve command and photodetector outputs were sampled at 1 khz. The photodetector was amplified and low-pass filtered at 500 Hz to prevent aliasing. The frequency response curve was constructed by computing the cross-spectra of the valve command and detector output. Four sweeps were averaged to compute the curves. The frequency response of the injector for a fuel pressure of 80 psig is shown in Figure 9. The logarithmic magnitude scale is referenced to the gain at 10 Hz. In all test cases, the coherence between valve and sensor is strong over nearly the entire range from 10 to 500 Hz. Phase(deg) Magnitude (db) 60 40 20 0-20 -40-60 Freq (Hz) D-7418z (a) 300 200 100 0-100 -200 Freq (Hz) D-7419z (b) Figure 9. Frequency response of injector with 80 psig water (a) amplitude; (b) phase. The magnitude plot contains interesting structure which appears at first to be a series of harmonics. A more careful investigation revealed that this structure in the frequency response plot is due to an oscillation (about 2 Hz) in the fuel supply. Since this oscillation is 5

present during the frequency sweep, it appears as a frequency-dependent phenomenon. However, repeated runs conducted under the same conditions exhibit the peaks and valleys at different frequencies, indicating that the oscillation is not correlated with the frequency sweep. This problem appears to be an artifact of the experimental setup, which contains two spring-loaded valves. Since the nozzle will be operated in the rig using an entirely different fuel supply system, more experiments will be needed later to accurately assess frequency response. These experiments were, however, sufficient to show that the injector has adequate bandwidth for controlling combustion instabilities in the range of frequencies where they are expected (approximately 100 to 500 Hz). Phase plots for 80 and 120 psig supply pressures have different slopes. Some of the phase can be attributed to transport delay since the phase varies with pressure and hence with flow rate. The slopes do not vary with the square root of the pressure, as might be expected if the phase delay were due entirely to transport lag. Other delays, such as in the valve response, must play a part in producing the overall phase lag. Results - Reacting Flow Experiments The operating characteristics of the combustor were evaluated for a range of fuel and air flowrates. The minimum allowable fuel flowrate was dictated by the fuel injector s ability to produce a well atomized spray at a low pressure drop and corresponded to a flowrate of approximately 3 gm/s of heptane. In the experiments described below, the fuel flowrate was varied between 3 and 4 gm/s corresponding to thermal outputs between 150 and 200 kw. Airflows were adjusted over a range from approximately 180 to 250 gm/s and the overall combustor equivalence ratio was fuel-lean with φ between 0.25 and 0.3. The combustor test section pressure was kept relatively constant in the range from 1.7 to 2.0 bar. Figure 10 shows a comparison of two power spectra of the fluctuating test section pressure when operating the combustor at 150 kw and 175kW heat output. The test section pressure for these cases was 1.7 and 2.0 bar, respectively. The spectrum exhibits a broad feature at low frequencies (f < 30 Hz) and a sharper peak centered near 140 Hz. The 140 Hz peak 160 140 120 100 80 60 P(f) = 40 psig, p(air) = 106 psig P(f) = 55 psig, p(air) = 142 psig Test section T(air) = 220 F 40 Frequency (Hz) Figure 10. Pressure fluctuation power spectra for cases with 150 and 175 kw heat output indicating combustion instability near 140 Hz. represents a thermo-acoustic combustion instability and in this example shows an increased intensity with increasing heat output and test section pressure. This trend was observed in other experimental runs as well. Although the broad spectral feature at lower frequencies indicates relatively large pressure amplitudes, further analysis indicated that it was not associated with a thermo-acoustically driven instability. In addition, its amplitude remains relatively constant as the fuel and air supply flowrates are varied. Figure 11 shows corresponding CH * emission spectra for the same experi- Figure 11. Power spectrum of fluctuating energy release for cases with 150 and 175 kw heat output indicating combustion instability near 140 Hz. 6

mental runs. These spectral plots clearly indicate that a strong flame oscillation occurs near 140 Hz and the magnitude of oscillation increases as the fuel and air flowrates are increased. The Rayleigh index can be computed from the same data used to compute the fluctuation power spectra for these cases and is shown in Figure 12. This spectral plot indicates that there is a strong positive coupling between the flame energy release and the acoustic pressure at the 140 Hz mode and the lower frequency oscillations (10 < f < 40 Hz) tend to be damped locally. Figure 12. Rayleigh index for 150 and 175 kw cases indicating flame driving at 140 Hz. Open-loop experiments were performed to determine the effects of fuel flow modulation on the stability characteristics of the combustor. In these experiments, the main fuel injector was modulated at a fixed frequency and the effects of unsteady fuel addition were investigated. Figure 13 shows an example of the resulting pressure fluctuation power spectrum when modulating the main fuel flow at 200 Hz and with a mass flow modulation of approximately 10 percent. The operating condition under which these data were acquired corresponds to a nominally stable regime, where the 140 Hz oscillation is relatively weak (rms pressure amplitude < 16 mbar). The pressure spectrum with forcing shows strong peaks at the forcing frequency and the first harmonic at 400 Hz. In addition, there are other peaks of lower amplitude which appear in the forced spectrum. It is believed that these lower amplitude peaks are the result of a fuel system resonance which is being excited by the action of the 200 Hz fuel modulation. It is also 160 140 120 100 80 60 40 Unforced Forced: f = 200Hz Test section P(air) = 110 psig T(air) = 240 F P(f) = 40 psig Frequency (Hz) Figure 13. Pressure fluctuation power spectrum for case wit open-loop fuel modulation. possible to drive the 140 Hz instability through openloop forcing. This is displayed in the pressure spectra of Figure 14. In this example the fuel flow was modulated at approximately 140 Hz with a modulation amplitude of 10 percent. Similar to figure 13, strong peaks are seen at the driving frequency and its harmonics. The oscillation at 140 Hz is driven to an amplitude over 20 db higher than that observed in the unforced case. This result demonstrates the control authority which can be achieved through main fuel flow modulation. This actuation technique will be used in future closed-loop experiments for suppression of the 140 Hz unstable oscillation. 160 140 120 100 80 60 40 Unforced Forced: f = 130Hz Test section P(air) = 110 psig T(air) = 240 F P(f) = 40 psig Frequency (Hz) Figure 14. Pressure fluctuation power spectrum with open-loop fuel modulation at instability frequency. 7

Summary A study has been performed to evaluate the effectiveness of a closed-loop control system to suppress thermo-acoustic combustion instabilities. The pulse-width modulation technique was implemented both in a numerical simulation of a dynamic combustion instability and experimentally. The control approach is to use fuel injection modulation as a control actuator and a flame emission or pressure sensor to detect unstable combustion oscillations. A high-speed microprocessor is used to implement an amplitudebased pulse width modulation control scheme. It was demonstrated that the control system was effective in suppressing a simulated thermo-acoustic oscillation. The system model representing the combustor dynamics was based on simplified conservation equations and described the coupling between the linear acoustics of the system and the heat release dynamics associated with combustion. Without control, the simulation showed the unbounded growth of a thermoacoustic instability. With closed-loop control, the instability was suppressed and the system was stable. The control system is undergoing scale-up design and will be tested on a 200 kw gas turbine combustor rig. The experimental apparatus is described in detail above and its baseline operating characteristics have been evaluated. The combustor exhibits a strong thermo-acoustic combustion instability which manifests itself through large pressure and flame emission fluctuation amplitudes near a frequency of 140 Hz. The instability tends to grow as the fuel and air flowrates are increased from those corresponding to the lean stability limit. Open-loop control experiments using main fuel flow modulation indicate that the combustion process can be manipulated to produce strong pressure fluctuations at the open-loop driving frequency. Closed-loop control experiments using pulse-width modulation on the main fuel injector are currently underway and will be reported in a future paper. Acknowledgment This work is supported by the Army Research Office, Engineering and Environmental Sciences Division (contract DAAG55-97-C-0031), under the supervision of Dr. David M. Mann. References 1. Dziomba, B. and Fiedler, H.E., Effect of initial conditions on two-dimensional free shear layers, J. Fluid Mech., Vol. 152, pp. 419-442, 1985. 2. McManus, K.R., Poinsot, T., and Candel, S.M., A Review of Active Control of Combustion Instabilities, Prog. Energy Combust. Sci., Vol. 19, pp. 1-29, 1993. 3. Candel, S.M., Combustion Instabilities Coupled by Pressure Waves and their Active Control, Invited lecture, Twenty-fourth Symposium (International) on Combustion, The Combustion Institute, pp. 1277-1296, 1992. 4. Zinn, B.T. and Neumeier, Y., An Overview of Active Control of Combustion Instabilities, AIAA Paper No. 97-0461, presented at the 35th Aerospace Sciences Meeting and Exhibit, Reno, 1997. 5. McManus, K.R., Magill, J.C., Miller, M.F., and Allen, M.G., Closed-Loop System for Stability Control in Gas Turbine Combustors, AIAA Paper No. 97-0463, presented at the 35th Aerospace Sciences Meeting and Exhibit, Reno, 1997. 6. Fleifil, M. Annaswamy, A.M., Ghoniem, Z. and Ghoniem, A.F., Response of a laminar premixed flame to flow oscillations: A kinematic model and thermoacoustic instability result, Combust. Flame, Vol. 106, pp. 487-510, 1996. 8