Mit PDF
Mit PDF
Mit PDF
Doctor of Philosophy
at the MASSACHUSETTS INSTIItUTIE,
OF TECHNOLOGY
Massachusetts Institute of Technology
June 2004
LIBRARIES
© 2004 Massachusetts Institute of Technology. All rights reserved.
Accepted by ......
Ain A. Sonin
Chairman, Department Committee on Graduate Students
BARKER
Active Combustion Control:
Modeling, Design and Implementation
by
Sungbae Park
Abstract
The first building block concerns the modeling of hydrodynamics and its interactions with
the other components that contribute to combustion dynamics. The second is the impact of active
control on emissions even while suppressing the pressure instability. The third is the evaluation
of model-based active controllers in realistic combustors with configurations that include swirl,
large convective delays and unknown changes in the operating conditions. The above three
building blocks are investigated in the thesis experimentally in three different configurations.
The first is a 2D backward facing step combustor, constructed at MIT, with the goal of
investigating the flame-vortex interactions and the impact of active control on emissions. The
second is a dump combustor, constructed at University of Maryland, so as to reproduce more
realistic ramjet conditions. The third is an industrial swirl-stabilized combustor, constructed at
University of Cambridge, to mimic realistic industrial gas combustor configurations which
typically include large convective time delays, swirl, and on-line changes in the operating
conditions. Results obtained from these three configurations show that through an understanding
of the underlying physics and reduced-order modeling, one can design an appropriate actuation,
2
sensing and control algorithm, all of which lead to model-based active control that reduces
pressure oscillations to background noise.
3
Acknowledgments
I would like to first thank Professor Anuradha Annaswamy and Professor Ahmed Ghoniem
for their guidance through the five years. Their efforts have been invaluable to open my eyes
and carry out my research. I would also like to thank Professor Douglas Hart and Professor
Kamal Youcef-Toumi for their support and valuable suggestions during our discussions together.
I owe a great deal to Professor Kenneth Yu and Bin Pang at University of Maryland who
gave me the opportunity to investigate their dump combustor in 2003 and 2004.
I am grateful to Professor Ann Dowling and Dr. Alex Riley for giving me a change to
implement a Posicast controller in their rig in the summer of 2002.
I would like to thank Adam Wachsman with whom I collaborated for two years, for building
the combustor and spending time discussing our research. Also, I owe a great deal to Daehyun
Wee for his valuable advices on my research.
I am grateful to Dr. Jean-Pierre Hathout, my predecessor, for his generous help. He was the
only person whom I could ask about anything in my first year at MIT.
I would like to thank my colleagues; Ziaieh Sobhani, Debashis Sahoo, Youssef Marzouk,
Shanmugan Murugappan, Tongxun Yi, Chengyu Cao, Jaejeen Choi, Murat Altay, Raymond
Speth, Daniel Macumber and Mac Schwager for their continued support.
I am very grateful to my parents, Hunkyo Park and Boonle Kim for their love, support and
guidance. Finally, a warm and special thanks to my wife, Euene Kwon, for her love and support
during the past years.
This work has been sponsored by the National Science Foundation, contract no. ECS
9713415, and in part by the Office of Naval Research, contract no. N00014-99-1-0448.
4
Table of Contents
1. INTRODUCTION .................................................................................................................... 11
5
4.1.1 Combustor ........................................................................................................ 84
4.1.2 Instrum entation and A ctuation ........................................................................ 87
4.2 COMBUSTOR D YNAM ICS....................................................................................................... 89
4.2.1 Self-excited oscillations.................................................................................. 89
6
List of Figures
Figure 2-1 Photograph of the MIT backward facing step combustor with various sensors and
actu ato rs ..................................................................................................................... 21
Figure 2-2 Schematic of the backward facing step combustor with various sensors and actuators 22
Figure 2-3 Linear Photodiode Array Sensor (S2 ) Schematic. The flame image is filtered for CH*
chemiluminescence and focused onto the linear photodiode array........................ 24
Figure 2-4 Equivalence Ratio Sensor (S4) Setup........................................................................ 25
Figure 2-5 Moog DDV Bode Plot (Bandwidth is 444Hz).......................................................... 29
Figure 2-6 Schematic diagram of the acoustics, hydrodynamics and heat release dynamics and
their interactions..................................................................................................... . 31
Figure 2-7 The impact of varying equivalence ratio and Reynolds number on the overall sound
pressure level in Cases I and II. ............................................................................. 33
Figure 2-8 The impact of convective time delay on the overall sound pressure level in Cases I
and II. ......................................................................................................................... 35
Figure 2-9 Pressure (Si) and equivalence ratio (S4) frequency spectra at (a) Re=5300 and # =0.85
and (b) Re=8500, # =0.65 in Cases I and II.......................................................... 37
Figure 2-10 High speed CCD images of the flame in the backward-facing step combustor at
Re=6300 and # =0.85, in Cases I and II at different moments in the cycle. (a) p'=O
and dropping, (b) minimum p', (c)p'=0 and rising, and (d) maximum p'............. 39
Figure 2-11 High speed CCD images, and the CH* data captured by the linear photodiode,
images were captured at 500 Hz, and represent slightly more that one cycle........ 41
Figure 2-12 Pressure, velocity and heat release rate, with the numbering of the points
corresponding to fram es in Figure 2-11................................................................ 42
Figure 2-13 Schlieren images w ith CH * data............................................................................... 45
Figure 2-14 Dependence of the Overall Sound Pressure Level on the momentum ratio of the
static air injected near the step, for primary air flow velocity of 4.4 m/s, and a fixed
fuel flow. The overall equivalence ratio ranged from 0.58 to 0.47 depending on the
flow rate of the air jet and H 2. ............................................................................... 48
7
Figure 2-15 NO level dependence on the momentum ratio of the static air jet near the step for
primary air flow velocity of 4.4 m/s, and a fixed fuel flow. The overall equivalence
ratio ranged from 0.58 to 0.47 depending on the flow rate of air jet and H2. ........... 50
Figure 2-16 Correlations between the Overall Sound Pressure Level and the NO level with
changes in the air jet mom entum ............................................................................ 50
Figure 2-17 Dependence of the Overall Sound Pressure Level on the momentum ratio of the
modulated air jet without H2 addition injected near the step and the phase angle, for
primary air flow rates of 4.4 m/s, and a fixed fuel flow. The overall equivalence
ratio ranged from 0.52 to 0.48 depending on the flow rate of the air jet. .............. 52
Figure 2-18 Dependence of the Overall Sound Pressure Level on the flow rate of the modulated
air jet with H2 addition injected near the step and the phase angle, for primary air
flow rates of 4.4 m/s, and a fixed fuel flow rate. The overall equivalence ratio
ranged from 0.52 to 0.49 depending on the flow rate of the air jet......................... 53
Figure 2-19 Dependence of the mean CH* distributions on the phase angle of the controller for ,6
= 2.89 with the overall equivalence ratio of 0.48 and without H2 addition. .......... 54
Figure 3-1 Schematic diagram of a dump combustor................................................................. 60
Figure 3-2 A photograph of the dump combustor ....................................................................... 60
Figure 3-3 The Linear photodiode array arrangement and the imaged area in the combustor.......61
Figure 3-4 Schlieren images at different phase angles (0 degree corresponds to the maximum
pressure). The domain of the Schlieren image is 20.2 cm x 7.5 cm. .................... 64
Figure 3-5 Measured CH* fluctuations using the photodiode array at different phase angles (0
phase angle corresponds to pressure maximum). The viewing window is 100 x 146
mm .............................................................................................................................. 65
Figure 3-6 Cumulative energy in the POD modes .................................................................... 72
Figure 3-7 The first four POD m ode shapes.............................................................................. 72
Figure 3-8 The power spectral densities of the modal amplitudes ............................................ 73
Figure 3-9 Input and output data for system identification: the input, cc, is the amplitude of the
first POD mode, and the output is fuel injector signal.......................................... 75
Figure 3-10 The pressure spectra with and without closed-loop control. .................................. 80
8
Figure 3-11 Time history of the overall sound pressure level (OASPL) with and without POD
mode adaptation. OASPL is calculated from the pressure signal using a moving
window containing 10 pressure cycles (128 samples).......................................... 80
Figure 3-12 Mean CH* emission with and without control........................................................ 81
Figure 3-13 The normalized CH* intensity fluctuations, Q', /Q , using a 1 s time window with
and without control. Q, is taken from Figure 3-12............................................... 81
Figure 4-1 Schematic of the rig downstream of the choke plate, showing the plenum/combustion
chambers and the swirler unit. All dimensions in mm, diagram not to scale......... 84
Figure 4-2 Detailed schematic showing a cross-section of the swirler unit and the orientation of
the fuel injection bars............................................................................................ 86
Figure 4-3 Detailed schematic of the fuel system, together with the DDV and the plenum
cham b er ...................................................................................................................... 86
Figure 4-4 SPL spectrum of the self-excited combustion oscillations; m=0.04 kg/s, =0.7..... 90
Figure 4-5 Time series showing the low frequency fluctuations present in the combustor when
the adaptive Posicast controller is activated; ma=0.04 kg/s, $=0.65, -,,,=9.6 ms,
9
Figure 4-10 Change of the controller parameter, k, in case of Figure 4-9; ma=0.04 kg/s, =0.70,
r, 0,=9.6 ms, Lp=1.46 m, zero initial conditions for control parameters................. 98
Figure 4-11 SPL spectra showing the reduction in noise when the adaptive Posicast is turned on
in the longer plenum case; ma=0.04 kg/s, J=0.65, to, =9.6 ms, Lp=1.97 m............ 100
Figure 4-12 Change of the controller parameter, k, in a longer plenum case with different
adaptation gain, y, ; ma=0.04 kg/s, 4=0.65, r,, =9.6 ms, Lp=1.97 m. ...................... 101
Figure 4-13 SPL spectra showing the reduction in noise when the adaptive Posicast was used
with correct( ro,=9.6ms) and incorrect time delay (rt,=8.5ms); ma=0.0 4 kg/s,
4=0.7, r,0,=9.6 ms, Lp=1.73 m, controller input filter=120-500 Hz. ....................... 102
Figure 4-14 Time series showing the effect of the adaptive Posicast controller with changes in
the mass flow rate of air; ma=0.04 kg/s, #=0.7, Lp=1.73 m, controller input
filter= 120-500 Hz..................................................................................................... 104
Figure 4-15 Time series showing the effect of the adaptive Posicast controller with incorrect
initial condition in the control parameter; ma=0.04 kg/s, f=0.65, r,0 , =9.6 ms,
10
1. Introduction
Continuous combustion processes are used in many applications ranging from power
generation and heating to propulsion. One of the characteristics of these processes is growing
pressure oscillations that transition to a sustained limit cycle. These oscillations occur due to the
coupling between acoustics and heat release dynamics. In acoustics, the heat release oscillation
supplies volume expansion and this expansion acts as a driving force of the pressure oscillation
inside the combustor. This pressure oscillation generates velocity, temperature, and equivalence
ratio perturbations, and these perturbations again induce unsteady heat release through heat
release dynamics. If this feedback is positive, the combustion system becomes unstable, and the
Combustion instability occurs especially in a lean bum condition where the efficiency is
high and emission is low [2] or in a rich bum condition where the thermal output is high [3]. The
underlying mechanism is complex and changes with the geometry and operation condition
making it difficult to develop a single model that describes the dynamics in the whole region.
Instead, one needs to divide the operating condition into several sub-categories and develop
models that can represent distinct characteristics in those operating conditions [4]. Several
high Damkohler and low Reynolds number condition, a wrinkled thin flame model is used to
represent heat release dynamics [5]. In [5], the heat release oscillation is due to the change of the
flame area, and the flame area changes by the velocity oscillations. The flame area change by the
velocity perturbation is represented using G equation. At other regimes where the Damkohler is
11
low and the Reynolds number is high, the chemical reaction controls the heat release rate. In this
case, a Well Stirred Reactor model has been used to represent the heat release dynamics [6]. In
other regions, where hydrodynamics and other mechanisms are important, physics based
To avoid this instability without major modification of the hardware or limitations on the
operating conditions, active control has been proposed [7-10]. In active control, pressure or heat
flux sensors are utilized to detect the onset of the instability, and actuators such as loudspeakers
and fuel injectors, are incorporated to modulate the heat release rate and/or pressure. Early
attempts often utilized a phase-delayed form of the pressure sensor signal as an input to a pulsed
fuel injector, where the requisite phase delay was determined empirically [7,8]. In the last five to
ten years, model-based control has been attempted where it has been shown that an order of
description of the coupling between combustion dynamics and acoustics. This makes the
has been demonstrated successfully over the past several years. Fleifil et al. [5] demonstrated
that an analytical combustion instability model based on flame kinematics was able to correctly
predict experimentally observed unstable modes. Hathout et al. [11] designed a linear quadratic
regulator based on this model by minimizing a cost function of unsteady pressure and control
input, showing that pressure oscillations could be stabilized over a range of frequencies without
energizing secondary peaks. Annaswamy et al. [12] validated this control design in a 1 kW
bench-top experimental set up to demonstrate that faster settling time and reduced control effort
12
could be achieved. Mehta et al. [13] extended the flame kinematics model by incorporating the
impact of exothermicity and fuel transport. In [6,14,15], it was shown that at high intensities, the
model was shown to capture drastic changes in the stability characteristics as the operating
conditions approached the lean blow-out limit [6]. The same model was shown in [16] to be
stabilizable using a self-tuning controller which coped with over a 100% change in the system
System identification has also been used to develop dynamic combustion models using
input-output data. Brunell [17] used system identification to develop a model and model-based
controller for a near full-scale combustion rig under turbulent flow conditions. Murugappan et al.
[18] developed a system identification model and a LQG-LTR model-based controller for a 30
kW swirl stabilized spray combustor, and succeeded in reducing the overall sound pressure level
developed an observer to estimate the frequencies and amplitudes of the resonant modes which
were used to generate control signals. Banaszuk et al. [20] used an extremum-seeking algorithm
Model-based control has also been demonstrated in combustion systems with a large
convective time-delay. Hathout et al. [21] applied a Posicast (positive forecasting) controller in
combustion control for the first time. Evesque et al. [22] developed a reduced order adaptive
positive cast controller based one a wave-based model formulation and a general heat-release
dynamics model
13
Through these several lab-scale studies, it has been shown conclusively that active control
is highly successful in suppressing the pressure oscillations. This success has set the stage for
transition of the technology from laboratories to large-scale applications in propulsion and power
generation. This thesis provides some of the building blocks for enabling this transition.
The first building block concerns the understanding of the interactions between
hydrodynamics and heat release dynamics and their impact on the underlying acoustics in the
combustion system. In the cases discussed in the above references, either the flow rates were
significantly low [5] or significantly high [6], causing the hydrodynamics to not be significant.
Often, however, it has been observed in experiments [23,24] that the hydrodynamic instability
does play a dominant role in the acoustic oscillations. A modeling of the interactions between
The second block is to address the optimization of emissions and other burning
indexes such as emission, efficiency and power density. However, to transform active control
strategy from laboratories into commercial applications, it is necessary that this strategy satisfies
more than a single performance index. This requires a sensing technology that can detect
conditions and external actuations since conventional pressure signal cannot be used to extract
other burning characteristics, e.g., efficiency, signature, emissions and etc. Also, new actuation
14
methods should be investigated since conventional fuel modulations have the possible emission
The third is the evaluation of model-based active controllers in realistic combustors with
configurations that include swirl, large convective time delays and on-line changes in the
operating conditions. In this thesis, I address these building blocks at three different combustor
configurations, each of which had its unique characteristics as shown in Table 1-1.
The first is a 2D backward facing step combustor (Configuration I), constructed at MIT, with
the goal of investigating the flame-vortex interactions and the impact of active control on
without requiring any image post processing. The impact of the vortices is bound to make the
heat-release quite distributed requiring a sensing technology that can detect spatio-temporal heat
release characteristics. For this purpose, a distributed heat release sensor using a ID photodiode
array is used.
For the purpose of simultaneous emission and stability control, air and hydrogen injections
limitations on this technology include the availability of high bandwidth fuel modulators and the
possible emission penalty associated with fuel injection into the combustion zone, where
maximum authority is expected. Although control algorithms that accommodate long time
delays and hence enable the use of fuel injection upstream of the combustion zone have been
formulated [21], there is no guarantee that hot spots would not arise. Instead, air forcing which
15
occurs very close to the dump plane can deliver high authority for actuation energy. Also, adding
additional air at the step tends to decrease the temperature, which is favorable for NOx reduction.
I examine the use of a steady and modulated air jet near the combustion zone and demonstrate its
effectiveness as a means for simultaneously suppressing oscillations and reducing NOx, in a case
in which the combustion instability is caused by strong flame-vortex interaction. In this regard,
an air jet is used to modify the flow structure without implementing major geometric redesign of
the flame-anchoring zone [25]. Under some conditions, using an air jet near the lean
flammability makes the flame susceptible to blow out. To remedy this situation, I experiment
with the addition of small amounts of hydrogen in the primary fuel. It has been shown that
adding a small amount of hydrogen to other hydrocarbons can extend the flammability limit and
increase resistance to flame extinction [26-31], with potential improvements in the emission
trend.
so as to reproduce more realistic ramjet conditions including axisymmetric flow, higher flow
velocity and an exit nozzle. In addition, the above-mentioned ID photodiode array is use to
model the flame-vortex interactions and capture the impact of control on spatial combustion
characteristics.
which typically include large convective time delays, swirl and on-line changes in the operating
conditions. Configuration III is a model combustor of a Rolls-Royce RB2 11 -DLE industrial gas
16
turbine [32]. Control is achieved by modulating the main fuel flow rate in response to a
measured pressure signal in the upstream of the combustor. The feedback control is an adaptive
controller which can accommodate changes in the resonant frequency and a time lag due to the
transport of the fuel [22]. I show from robustness studies that this controller retains control for a
20% change in frequency and a 23% change in air mass flow rate.
will be discussed in Configuration II. In Chapter 4, validation of the active control strategy in
17
Table 1-1 Configurations of the three combustors considered
FuelType Propane (main fuel) Ethylene (main fuel) Ethylene (Main fuel
Hydrogen(modulation) Ethanol(modulation) modulation)
18
2. Simultaneous Stability and Emissions Control
In this Chapter, I experiment air forcing as an actuator for the purpose of simultaneous
stability and emissions control. While fuel modulation is a common stabilization technique,
limitations on this technology include the availability of high bandwidth fuel modulators and the
possible emission penalty associated with fuel injection into the combustion zone, where
maximum authority is expected. Air forcing which occurs very close to the dump plane can
deliver high authority for actuation energy. Also, adding additional air at the step tends to
decrease the temperature, which is favorable for NOx reduction. Under some conditions, using an
air jet near the lean flammability makes the flame susceptible to blow out. To remedy this
situation, I experiment with the addition of small amounts of hydrogen in the primary fuel. I
show that, with air injection and a small amount of hydrogen addition, one can simultaneously
reduce pressure oscillations up to 15 dB and reduce NOx emission to sub-ppm levels. Section
2.1 describes the combustor setup. Section 2.2 discusses the combustion dynamics. Finally,
19
2.1 Experimental Setup
expansion with recirculation zone that anchors the flame, as shown in Figure 2-1 and Figure 2-2.
It consists of a rectangular stainless steel duct with a cross sectional area 40 mm high and 160
mm wide. Halfway along the length of the combustor, a ramp contracts the channel height from
The step height is 20 mm. The overall length of the combustor is 1.9 m, with the step located in
the middle. The air inlet to the combustor is choked. The exhaust gases are expanded at the end
of the chamber. The combustor is equipped with quartz viewing windows. An air compressor
supplies air up to 110 g/s at 883 kPa. Propane is used as the primary fuel. Fuel is supplied at
several spanwise holes from a fuel manifold located upstream of the step, at 17.5 cm or 35 cm.
Fuel is injected in the opposite direction to the flow to improve mixing. The sensors and
actuators are positioned as indicated in the figure. In the figure, p' is pressure fluctuations, U is
the mean air velocity in the upstream, u,' and u2 ' are velocity fluctuations at the point of fuel
injection and the step, respectively, Q' is the fluctuation in the heat release rate in the burning
zone, co' is vorticity fluctuations in the downstream from the step and Affiel and Aair are fuel and
20
Figure 2-1 Photograph of the MIT backward facing step combustor with various sensors and
actuators
Here, I describe the sensors used in the backward facing step combustor. Pressure
fluctuation is measured with the Kistler pressure sensors (SI). CH* chemiluminescence, which
can be related to heat release fluctuation is measured with the linear photodiode array (S2). 2D
flame structure is measured with a high-speed camera (S3 ). Equivalence ratio fluctuation is
measured with the laser and photodetector arrangement (S4 ). The pressure and equivalence ratio
measurements are logged at 1000 Hz by a Pentium III with a dSPACE data acquisition board
installed. The linear photodiode array voltages are captured by a Pentium IV with National
Instruments data acquisition hardware installed. The overall sample rate is 64 kHz to read all the
photodiodes, and each snapshot is captured at 500 Hz. In the following, detailed descriptions of
21
S3
I5 ( 6) S 1 (p')
S4 (
')
i
7
> II AH2 -> - U2
U Afuej
- -. P. OR
"'airL
S6 (NOx)
Sensors Variables
Figure 2-2 Schematic of the backward facing step combustor with various sensors and actuators
22
2.1.1 Sensors
Kistler pressure sensors are used to measure the dynamic pressure response from the
interior of the combustor. The 6061B ThermoCOMP Quartz Pressure Sensor can measure 0-2.5
bar up to 0-250 bar. It is water-cooled and designed especially for small combustion engines and
CH* chemiluminescence is measured spatially and temporally using a new sensor design
involving a linear photodiode array to capture distributed heat release characteristics [33,34]. An
NMOS linear image sensor (S3901-128Q) is available from Hamamatsu Photonics that provides
128 individual photodiodes in a linear array. Each pixel is 2.5 mm high and 45 um wide. A
flame image can be projected onto this array using the appropriate optics, and a "linear snapshot"
can be taken. This has an advantage over a single photodiode, because it provides spatial
information. It also has an advantage over a CCD camera, because the data can more easily be
streamed to a computer for analysis, and the amount of data can be handled in real-time for
control purposes. Each pixel integrates the light intensity over time, and resets when it is read.
In the experiment, the flame image passes through an optical bandpass filter centered at
430 nm, the wavelength of CH* chemiluminescence (Figure 2-3). Unlike most CCD arrays
which have peak sensitivity in the infrared region, the linear photodiode array has a high UV
23
sensitivity, making it suitable for this application. A bi-convex UV fused silica lens is used to
focus the image of the flame onto the chip. The photodiode array has high spatial resolution in
the streamwise direction, and integrates the light intensity in the vertical direction.
To capture 2D flame images at a high resolution and high speed, a MotionPro CMOS
camera from Redlake is used. The MotionPro system is designed to capture high-speed digital
images and deliver them directly into a PC for analysis and documentation with maximum frame
resolution of 1280 x 1024 pixels with record rates up to 1000 frames per second. Flexible
recording options permit using the PCI-board memory as a circular buffer into which specified
pre- and post-trigger frames are recorded, or dividing it into a segmented buffer for multiple
session operation.
430 nm
Optical Filter
Figure 2-3 Linear Photodiode Array Sensor (S2) Schematic. The flame image is filtered for CH*
chemiluminescence and focused onto the linear photodiode array.
24
Equivalence Ratio Sensor (S4)
The equivalence ratio sensor uses a laser and a photodetector. The laser emits a beam of
light of the wavelength (3.39 pm) that is absorbed by hydrocarbons like methane and propane
[35]. On the other side of the combustor, a detector is installed that is sensitive to that
wavelength of light. When fuel passes through the laser beam, it absorbs some of the laser light
and the detector signal is reduced. The intensity of the light can be related to fuel concentration
If= dx
I/Jo =10 0
where I is the intensity of incident monochromatic light, I is the intensity of transmitted light
through the absorbing species, e is the decadic molar absorption coefficient (cm 2/mol), 1 is the
absorption path length, and c is the concentration of absorbing species (mol/cm3). Figure 2-4
InAs Detector
I I
25
Air Flow Sensors (S)
Trak. The maximum flow rate is 173 g/s. The maximum pressure is 827 kPa. The unit is
powered by an 18-30 V DC power supply, and it outputs a signal from 0-10 V DC which is
performance characteristics such as NOx concentration and burning efficiency. Fuel modulation
is a common stabilization technique. The impact of fuel fluctuations on emissions and efficiency
has been measured, but the results have not been used in the feedback loop in a way that
Additionally, a study of emissions will provide insight into the possibility that air forcing
produces cleaner emissions and more complete burning than fuel modulation. For example, air
injection at the step may serve to cool the flame, reducing NOx.
Other uses for this equipment will be to correlate the linear photodiode array with
emissions characteristics. For example, it appears that the flame becomes more compact when
controlled with air injection. Compact flames are associated with low emissions because of
decreased residence time in which to form NOx. Preliminary analysis of linear sensor images
appears to show this compact flame shape after control is applied. If a correlation can be made
between emissions and linear sensor image, it is possible that the linear sensor could serve as an
26
An NOx emissions sensor has been installed in the combustor rig. The emissions probe is
located 62 cm downstream from the step in the exhaust section. The probe extends 20 mm (half
the combustion chamber height) into the chamber, through a 'A-NPT threaded boss. The probe is
attached to a Universal Analyzers Model 270S Stainless Steel Heated Stack Filter. The sample
line is connected to a Universal Analyzers Model 520 Single Channel Sample Cooler. The cooler
brings the sample down to 4 *C. A peristaltic pump removes to the exhaust trench the water that
is condensed by this operation. The cooled sample is sent to the Thermo Environmental
2.1.2 Actuators
In the following, I describe the air and fuel actuators. As indicated in Figure 2-2, the fuel
injection occurs several step heights upstream pointing upstream for uniform spanwise mixing.
Fuel can be modulated at the fundamental unstable frequency, but since the fuel has time to mix
in the streamwise direction before it reaches the flame, the authority of fuel modulation is
reduced. Also, this type of actuation introduces a delay between the time the fuel is modulated
and the time the fuel encounters the flame, which is not only moving, but also spatially
distributed. Air forcing which occurs very close to the dump plane can deliver high authority for
actuation energy. Also, adding additional air at the step tends to decrease the temperature, which
is favorable for NOx reduction. Detailed description of the air and fuel actuation is in the
following and closed-loop control results with air actuation will be shown in Section 2.3.
27
Forcing Air (Aair)
Control actuation is accomplished using Dynamco D1B2204 Dash 1 direct solenoid poppet
air valves. This valve can supply 1.8 g/s of air when supplied with 689 kPa. The valve is
Another valve used for actuation is the Moog D633-7315 AIC Direct Drive Valve (DDV).
It has its own built in feedback loop to ensure the spool position using an LVDT. This feedback
loop is controlled by the Moog D143-098-013 Single Axis Electronic Controller. This unit is
powered by a 48V Condor Power Supply, which also powers the valve. The controller accepts
A transfer function for this valve was determined using system identification. White noise
with a bandwidth of 1000 Hz was the input. The spool position, measured with the LVDT was
the output. The spool position is related to mass flow rate by a calibration. The transfer function
is
where rh is mass flow rate (kg/s) and V is the input to the valve (Volt). The Bode plot in Figure
28
Bode Diagram
-90-
CA -160
Q
20
, 360
-450
1010 10
Frequency (rad/sec)
To extend the flammability limit and increase the resistance of the flame to extinction,
hydrogen is delivered at 0-9 mg/s through a Sierra Instrument mass flow controller. Hydrogen is
either mixed with propane and introduced through the fuel bar, AH2f, or mixed with the air jet and
29
2.2 Modeling of Combustion Dynamics
Figure 2-6 shows schematically how various flow, combustion and acoustics mechanisms
interact, and their contributions to the combustion dynamics and emissions. This model features
the following:
- The feedback mechanism that contributes to combustion dynamics, which couples heat-
" The heat-release perturbations that arise due to fluctuations in the equivalence ratio, #',
and hydrodynamic perturbations, co'.
Optimal actuation for control depends on the mechanism leading to combustion instability. For
example, if equivalence ratio fluctuations play a significant role, upstream fuel modulation may
work best. On the other hand, if flame-vortex interaction is the governing mechanism,
hydrodynamic-based actuation near the step, such as spanwise air injection, would be more
effective. A parametric study of the impact of the inlet conditions and the hydrodynamics on the
30
Hydrodynamics H eat 2
Emissions
Afuel ~ Inlet -
-yDyn ami cs NOX
S4
Ulf ~Acoustics -
Figure 2-6 Schematic diagram of the acoustics, hydrodynamics and heat release dynamics and
their interactions.
(Mixture Inhomogeneity)
I first examine the role of fuel/air mixture inhomogeneity, #' , in the combustion
dynamics. Equivalence ratio fluctuation occurs due to velocity perturbations, u', at the point of
fuel injection according to 0' = - , (if u' /U <1). #' appears at the burning zone
1+ -_
following a convective transport time delay, z,,, determined by the mean mixture velocity, U,
L
and the distance from the point of fuel injection to the burning zone, L , r,, = = [21,38].
U
31
The convective delay from the point of fuel injection to the burning zone determines the
phase between # ' and u' (and hence p'). The time delay can be modified by changing the fuel
injection location. To determine the impact of the mixture inhomogeneity, or # ' on stability,
the location of the fuel bar was varied and the pressure was measured at various air and fuel flow
rates. Moving the fuel injection further upstream has the following effects: 1) the convective
time delay is increased, and 2) the amplitude of mixture inhomogeneity is decreased due to
diffusion and reduction in u' (the fundamental acoustic mode in the combustor is a quarter-wave
mode with a closed end upstream and an open end downstream). Note that the hydrodynamic
I now consider two cases: I and II, where the fuel bar was located 17.5 cm and 35 cm
upstream from the step, respectively. In both cases, the uncontrolled combustor was operated at
different air-flow rates. Figure 2-7 shows the effect of varying the equivalence ratio and
Reynolds number on the overall sound pressure level (OASPL). In all these experiments, the
excited acoustic mode frequency, fa, was 38 Hz. In Figure 2-7(b), the pressure level almost
monotonically increases with the Reynolds number while it is insensitive to the changes in the
mean equivalence ratio. In contrast, in Figure 2-7(a), the impact of Reynolds number is non-
monotonic showing reduced pressure level at Re = 8500 and increasing suddenly at Re = 9500.
Also at Re = 9500, the pressure level is sensitive to the changes of the mean equivalence ratio.
Similar observations (pressure level increases with the increase in the Reynolds number and is
sensitive to the mean equivalence ratio change at high Reynolds number) have been made in Ref.
32
Re = 9500
162 -
160 -
Re = 6400 Re 7400
158 -
0 158
- -
152 Re = 8500
150
0.5 0.55 0.6 0.65 0.7 0.75 0.8 0.85 0.9 0.95 1
162 - Re 8500
160 -
- -Re 7400
158 -
Re = 6400
156 -
0
154 -
Re= 5300
152 -
150 -
0.5 0.55 0.6 0.65 0.7 0.75 0.8 0.85 0.9 0.95 1
Figure 2-7 The impact of varying equivalence ratio and Reynolds number on the overall
sound pressure level in Cases I and II.
To understand this discrepancy in Cases I and II, this stability map is redrawn with the
convective time delay, z,,, as the x axis as shown in Figure 2-8. The convective time delay from
33
the fuel injection point to the step is shown in the figure as multiples of r , where r is the
acoustic time scale, 1/fa = 26.3 ms. If the instability was controlled by the mixture
inhomogeneity only, a stability band should appear in these maps as the time delay is varied as
follows: stable zones when the transport delay is between nir and (n+0.5) r and unstable zones
when the delay is between (n+0.5) r and (n+1) z- [21,38]. No changes in the pressure
fluctuations were shown in Figure 2-8(b) during these transitions indicating that mixture
inhomogeneity does not play a role in Case II. However, in Case I, a pressure drop is observed
when the time delay is between r and (n+0.5) r indicating that the equivalence ratio oscillation
has a secondary role. This secondary role of the equivalence ratio oscillations in Case I explains
the non-monotonic relation between the pressure level and the Reynolds number observed in Fig.
Figure 2-7(a).
combustion instability. Equivalence ratio oscillations (S4 ) and the corresponding pressure signal
(Si) are compared in Cases I and II at two different operating conditions. First, at Re = 5300 and
# = 0.85, as shown in Figure 2-9(a), negligible equivalence ratio fluctuations are observed in
Case II due to streamwise diffusion, while significant equivalence ratio fluctuations are observed
in Case I. The pressure amplitude is higher in Case I indicating that the equivalence ratio
fluctuation increases the pressure oscillations at this operating condition. Note that Case I is in
the band between (n+0.5) z and (n+l) r . On the other hand, even without equivalence ratio
34
165 7
U 0.9
a 0.8
160 - 0.7
0.6
U M
U S0.5
0 U M
U U
* U
U
U
150L
0. 3 1 1.2 1.4 1.6 1.8
165
U U
U
160 U
U
U
155 0
150
3 3.5 4 4.5 5 5.5
' tot' /
Figure 2-8 The impact of convective time delay on the overall sound pressure level in
Cases I and II.
The power spectra obtained at a second operating condition, at Re = 8500 and # = 0.65,
are shown in Figure 2-9(b). This condition corresponds to a band between nr and (n+O.5) r in
35
both cases, where equivalence ratio oscillations may damp out pressure oscillations. While
significant equivalence ratio fluctuations are present, Case II has lower values since the fuel bar
is positioned further upstream, similar to the Re=5300 experiment. Since the equivalence ratio
fluctuations, under these conditions, provide negative feedback, Case I exhibits a lower pressure
amplitude. However, both cases show significant pressure oscillations, indicating that the
36
4
i'- Case I
L3 Case II
2-
C,
Q - - - - - - - - -- -- - -
0
0 10 20 30 40 50
Freq(Hz)
- Case I
0.4 - ---- Case 11
-\
0.2
Ji 000
0 10 20 30 40 50
Freq(Hz)
4
5Case
3 - Case 11
a>
2 1-
CL
0-
0 10 20 30 40 50
Freq(Hz)
- I- Case I
-- -- Case
0.4 - II
wU 0~
0 10 20 30 40 50
Freq(Hz)
Figure 2-9 Pressure (Si) and equivalence ratio (S4) frequency spectra at (a) Re=5300 and #=0.85
and (b) Re=8500, q#=0.65 in Cases I and II.
37
2.2.2 The Role of Hydrodynamics and Flame-Vortex
Interactions
Velocity perturbations can cause repeated shedding of vortical structures at the step. This,
in turn, causes the flame shape to change, thereby generating large amplitude of heat release
fluctuation. This vortex shedding has a preferred frequency quantified by a Strouhal number St =
fHH / U = 0(0.1) where H is the step height and fH is the frequency in Hz [40]. If the preferred
vortex shedding frequency is near the acoustic mode frequency, fa, the system is more prone to
combustion instability.
The strong dependency of combustion instability on the Reynolds number or the mean
velocity described above suggests that the primary mechanism causing combustion instability is
the hydrodynamics and its interaction with flame. As mentioned before, hydrodynamics has a
preferred frequency around fH = 0(0. lU / H). Interestingly, strong instability occurs above
Re=9500 in Case I, and Re=7400 in Case II, as shown in Figure 2-7, in which the preferred
frequency is fH =33Hz and 25Hz, respectively, both of which are close to the acoustic
frequency, fa,=38Hz. High speed CCD images in Figure 2-10, captured at 500 frames/s, show
strong flame-vortex interactions with no noticeable difference in the flame structure between the
two cases. These images confirm the role of flame-vortex interactions in the combustion
instability mechanism.
38
tream) Case II (fuel bar at 35cm upstream)
(a)
(b)
(d)
Figure 2-10 High speed CCD images of the flame in the backward-facing step combustor at
Re=6300 and # =0.85, in Cases I and II at different moments in the cycle. (a) p'=O and dropping,
(b) minimum p', (c) p'=O and rising, and (d) maximum p'.
Now, I now show the cyclical flame-vortex interactions using high-speed imaging. We start
the analysis of the mechanism from the moment of maximum velocity, according to Figure 2-12
(numbers between brackets correspond to the image number in the figure). Figure 2-11 shows 8
snapshots of the flame from the video record, successive images are separated by 4 ms, that is,
the cycles ends almost halfway between (7) and (8) since the cycle time is close to 25 ms.
Around the moment of maximum velocity, the flame appears as a vertical front moving
downstream of the step (1). Note that due to a combination of local small-scale turbulence and
relatively weak variation in the spanwise direction, the flames appear as thick zones. As the
velocity decreases gradually, the flame advances into the channel, forming two fronts, a leading
39
vertical front that accelerates forward on the top half of the channel, and a horizontal front that
extends back to the step (2). It is interesting to observe near-extinction at the step during this
part of the cycle, apparently due to the high flow velocity/strain rates there (2-3). As the velocity
becomes negative, the vertical flame curves slightly forward as it moves faster along the upper
wall than the lower wall, and the horizontal flame curves upwards, away from the step, and
forms a forward fold closer to the lower walls (4). As the flame departs from the step, it bums
vigorously into the inlet channel (4-5). Meanwhile, reactants are trapped in the fold, and in
between the fold and the leading curved flame (5). The inception of the fold can be seen in (3-4)
clearly, as well as the formation of reactants traps between the fold and the leading vertical flame
(4-5). During this part of the cycle, the curved horizontal flame propagates almost vertically
towards the top wall, pushing its leading and trailing edges downstream and upstream,
respectively, with the latter migrating further into the inlet channel (4-6). As the velocity
emerges from the negative region, the flames surrounding the fold start to converge inward,
burning the trapped reactants and closing the gaps (5-6). According to the images, this is the
moment of maximum total burning rate. Meanwhile the migration of the trailing edge of the
flame into the inlet channel, which started when the velocity delved into the negative domain, is
halted as the velocity recovers (5-7). Signs that the flame was pushed back towards the step
while the velocity was rising towards its maximum are seen in (7). The cycle is now over, and
40
1 5
2 6
3 7
4 8
Figure 2-11 High speed CCD images, and the CH* data captured by the linear photodiode,
images were captured at 500 Hz, and represent slightly more that one cycle.
41
41
2- -0
a-
00
4L e + 4-
44
-2 Il 0
Js- Pressure
3 --a- Velocity
--- Heat Release
4 ' -5
0 5 10 15 20 25 30
Time (ms)
Figure 2-12 Pressure, velocity and heat release rate, with the numbering of the points
corresponding to frames in Figure 2-11.
Now, we look closely at the CH* data to correlate the flame geometry, the heat release rate
and the pressure. At the beginning of the cycle, the CH* distribution peaks near the step, where
the flame front is located at that moment (1). As the flame propagates downstream, the CH*
distribution advances with it (2), stretching downstream with a flatter peak as the two parts of the
flame become distinguishable (3). Consistent with the video images that shows partial extinction
at the step, CH* signal there is near zero (2-3), until the flame is pushed away from the step and
into the inlet channel (4). We note the strong correlation between the brightness of the video
image and the local CH* signal, confirming the intended use of the former as a surrogate for
burning intensity. As the horizontal flame curves upward and away from the high strain zone,
42
near the step, a weak CH* signal starts to appear there, growing slowly in a nearly flat
distribution behind the leading peak (4-6), the latter corresponding to a more energetically
burning and folding flame that continues its forward propagation (5). One can clearly see the
sign of a flame advancing upstream in the CH* image in (5). As the flames converge towards
the centers of the folds, the leading peak in CH* rises significantly, heralding the peak heat
release rate (6). This is followed by a rapid drop in CH* throughout the channel, except in the
inlet channel where a flame is convected back downstream, following its brief early migration
there (7). The flame moving downstream the step is again captured in the CH* in (7-8).
The streamwise integral of CH* represents the instantaneous total heat release rate, shown
in Figure 2-12. The fast but more gradual rise in total heat release, followed by the faster drop
and slow recovery are clearly depicted in this figure. Because of the dynamics described in the
previous paragraphs, the total heat release rate peaks almost 100' ahead of the velocity at the
step, at a moment when the velocity near the step is still relatively low but rising. As can be seen
in the same figure, and because the dominant acoustic mode is a quarter wave mode, the pressure
also leads the velocity by 900, and the pressure and heat release rate are in phase, as expected in
an unstable combustor. The mechanism responsible for this coupling is clearly fluid dynamic in
nature, and is embedded in how the flow downstream the step causes the flame convolutions
shown in Figure 2-11. Next, we examine the Schlieren images to confirm the close relationship
between the bright zones in the direct video images and the burning zones, or "thick" flames in
Figure 2-13 shows the Schlieren images taken from another cycle but corresponding to the
same case shown in Figure 2-11, with the CH* concentration superimposed on the images as
43
before. Statistical variations are responsible for the small differences between the images shown
in Figure 2-11 and Figure 2-13, however the similarity is nevertheless very strong. In general,
the Schlieren images are better indicators of the flame front surface than the direct video since
they capture the density gradient. However they can be affected more by the spanwise averaging,
and by heating due to the proximity of walls, e.g., close to the step, which can introduce weak
local density gradients in the absence of flames. Despite that, the Schlieren images show strong
resemblance to the direct video images, and are also strongly correlated with the CH*
distributions.
44
1 I5
2 6
3 7
4 8
45
2.3 Simultaneous Control of Combustion Instability and
Emissions
In this section, I show open and closed-loop control results using the air jet and H2
addition, in order to simultaneously reduce pressure oscillations and NOx emissions. The fuel
bar is located 35 cm upstream of the step. In Section 2.3.1, I consider a static air jet (Aair) near
the step and H2 addition to the primary fuel (AH2f). In Section 2.3.2, I show the impact of
modulating the air jet (Aair) when it is mixed with hydrogen (AH2a). Primary fuel modulations
(Afuel) at the fuel bar were also attempted, but, as expected from the results of Section 2.2, its
The effect of a steady transverse air jet (Aair), introduced a short distance upstream of the
step, on the pressure fluctuations, and the impact of hydrogen addition to the primary fuel (AH2f)
in the presence of the transverse air jet as well are now examined. In these tests, the Reynolds
number was 6300, with a corresponding average velocity, U = 4.4 m/s. The primary fuel supply
was fixed. The transverse jet air velocity, u,, was varied from 0-8.4 m/s. The hydrogen addition
was at 1.6 and 3.2 % of the primary fuel, based on the lower heating value. The baseline
equivalence ratio, # =0.57. The overall equivalence ratio ranged from 0.58 to 0.47, depending on
46
Figure 2-14 shows OASPL as a function of momentum ratio of main air to cross jet flow,
#8=u /U 2
, at different H2 flow rates. For all cases, the pressure oscillations first increased for
#8< 0.5, where they reached a maximum. Above this value, the OASPL dropped rapidly.
Without hydrogen, the flame blows out for 8 < 1.8. However, with 8 > 1.8 the flame became
sustainable again, with a pressure reduction of up to 14 dB from the baseline case. With air
addition, the overall equivalence ratio was mostly below the nominal flammability limit of
propane/air mixture. Moreover, a relatively large momentum in the transverse jet was necessary
to reduce the pressure fluctuations and stabilize a flame near the step at yet lower equivalence
ratio. Thus, it appears that the impact of the transverse jet is aerodynamic. Studies of simple jet
in cross flow show that at small velocity ratios, the jet hardly penetrates into the stream, while as
the jet velocity increases, the jet penetration is pronounced and its impact is sustained several jet
widths downstream [41]. At high velocity ratios, the jet deflects the primary stream upwards,
keeping it away from the lower walls. By deflecting the stream away from the lower wall, less
heat is lost to the lower wall and combustion is more adiabatic. Since the transverse jet reduces
heat loss and lowering the heat loss promotes burning of lean to very lean mixtures [42], the air
Figure 2-14 shows that adding H2 to the primary air stream, in the presence of the
transverse jet, has a weak impact on the pressure level before the point of maximum pressure
oscillation is reached. Beyond this point, the effect of H2 addition is to support a more stable
flame, lowering pressure oscillation and preventing blowout. Moreover, adding hydrogen
enables burning over the entire range of the jet mass flow rate (we no longer see blowout around
/=1). This is consistent with previous observations that doping hydrocarbons with small amounts
47
of H2 extends their flammability limits and reduces their tendencies to extinguish under the
influence of strong strains [26-31]. It is interesting to observe that while 1.6 % H2, measured as
a percentage of the primary fuel based on the lower heating value, improves flame stability,
160 H2 (%)
O 1.6 3.2
158 o A
156
152
0)
150 H2 Increasing
H
CO148
146
144
I I I
0 0.5 1 1.5 2 2.5 3 3.5
Momentum Ratio (p)
Figure 2-14 Dependence of the Overall Sound Pressure Level on the momentum ratio of the
static air injected near the step, for primary air flow velocity of 4.4 m/s, and a fixed fuel flow.
The overall equivalence ratio ranged from 0.58 to 0.47 depending on the flow rate of the air jet
and H2.
Figure 2-15 shows that the introduction of an air jet generally reduces NO levels. The data
also show that the impact of the air jet may be non-monotonic when hydrogen is added to the
primary fuel; only beyond the point of maximum pressure oscillations does a further increase of
the air jet momentum guarantee substantial reduction in NOx. Before that, there is a complex
48
dependence of NOx on the transverse jet momentum, mostly because of the competing effects of
the dynamic flame structure, the equivalence ratio and the temperature fluctuations. Stable
combustion, associated with both leaner and more aerodynamically compact flame in the
presence of a transverse air jet, leads to a faster drop in NO. As in the case of pressure
fluctuations, it appears that high transverse jet momentum is necessary to achieve a measurable
reduction in NO formation. Note that hydrogen addition under the same conditions boosts NO
levels over the entire range, as the temperature rises in the combustion zone and in the products.
Thus, if H2 is to be used to lower the pressure level, it must be supplemented with additional air,
to be introduced strategically into the burning zone so as not to jeopardize the emissions.
Figure 2-16 shows the correlation between the NOx level and the OASPL with the change
in the momentum ratio of the air jet. It clearly shows that the air jet can simultaneously control
combustion instability and emissions. Without H2, the relationship between the NOx level and
the OASPL is almost linear. With H2 addition, we were able to increase the primary air flow by
extending the flammability limit. As a result, the NO level was reduced even further. In Ref.
[43], I showed that one can able to extend the flammability limit by using an air jet and H2
addition to stabilize the flame, all the way down to # = 0.41. Combustion remained stable, with
sub ppm NOx and 145 dB OASPL. This is ideal for applications requiring low NOx and stable
49
30 r
H 2 Increasing
-0 20
Blowout
0
CNJ
C10
CL10
H2 (%)
0
z 0 1.6 3.2
1 *I
n I I II I I
0 0.5 1 1.5 2 2.5 3 3.5
Momentum Ratio (p)
Figure 2-15 NO level dependence on the momentum ratio of the static air jet near the step for
primary air flow velocity of 4.4 m/s, and a fixed fuel flow. The overall equivalence ratio ranged
from 0.58 to 0.47 depending on the flow rate of air jet and H2.
30
E Primary Air:4.4m/s, without H2
o Primary Air:4.9m/s, 1.6% H2
25
0
~20
Ln 15
10
-r a increase
z
5
0
145 150 155 160
OASPL (dB)
Figure 2-16 Correlations between the Overall Sound Pressure Level and the NO level with
changes in the air jet momentum
50
2.3.2 Modulating the Air Jet (Aair) and H2 Injection (AH2a) at
the Step
Next, I investigate the impact of the transverse jet modulation on the pressure fluctuations.
I show that one advantage of using modulation is to minimize the mass flow rate of the jet by
controlling the phase using pressure feedback. I use a phase-shift controller to modulate the mass
flow rate of the transverse jet. In these tests, the Reynolds number was fixed at 6300 with a
nominal equivalence ratio of 0.57. The momentum ratios were 6= 0.64, 1.56 and 2.89. Pure air,
or an air-hydrogen mixture, in which H2 contributed 1.6 % of the total heat addition, were used
in the transverse jet. Hydrogen was premixed with air and injected through the slot. With air jet
modulation, adding H2 to the transverse jet at the slot was more effective than adding hydrogen
to the primary fuel. Paradoxically, without modulation, the opposite was true, i.e. adding H2 to
To examine the response of combustion dynamics to the transverse jet modulation, I used a
simple phase-shift controller with pressure feedback. A high pass filter with a cutoff-frequency at
5 Hz was incorporated to eliminate the drift in the pressure measurement and ensure that the duty
Figure 2-17 shows the OASPL as a function of fland the phase, with pure air injected at the
slot. For 8 > 1.56 substantial reduction in OASPL could be achieved, but the performance is
strongly dependent on the phase. Optimal phase angle in the phase-shift controller was found to
be around 200 degrees. In the case with ,= 1.56 the flame blew out at 60 and 120 degrees phase,
51
as indicated by the dotted line in the figure. This is similar to the results observed previously
where a small amount of cross jet was detrimental to stability and caused flame blowout. With /8
= 2.89 a pressure reduction up to 14 dB is realized at the same phase angle of 200. Increasing the
air jet velocity to 6 = 2.89 reduced pressure fluctuations throughout the entire range of phase
angles, widened the stability band around 250 degrees and eliminated the flame blowout
observed at lower velocities. It is interesting to observe that the maximum pressure reduction is
comparable to that obtained with a static transverse air jet, showing some saturation effect. This
was also observed in Figure 2-14 where at high velocity ratios, the pressure level reached a
plateau. Thus, regardless of the flow rate of the transverse jet, whether it is static or modulated,
160
152 -
-o Higher P
a 150 -
CL
U)148 -
146 - 3 = 0.64
o A r3=1.56
144 - = 2.89
0 50 100 150 200 250 300 350
Phase (degrees)
Figure 2-17 Dependence of the Overall Sound Pressure Level on the momentum ratio of the
modulated air jet without H2 addition injected near the step and the phase angle, for primary air
flow rates of 4.4 m/s, and a fixed fuel flow. The overall equivalence ratio ranged from 0.52 to
0.48 depending on the flow rate of the air jet.
52
Adding H2 in the transverse jet significantly increases the effectiveness of control, as
shown in Figure 2-18. With H2 addition, the transverse jet momentum required for flame
stabilization dropped by more than one half. By adding H2 at 1.6 % of the total heat-release, the
flame was stabilized at a momentum ratio of only 0.64, with a pressure drop up to 13 dB from
the unstable case. Note that the optimal phase angle is now 300 degree. While a pure air jet
reduces the burning velocity, the jet with H2 increases it, and we observed that H2 addition
16
2-
CL Higher p
50
C146 - p=O.64withH 2
>p= 1.56 with H2
144 . =2.89 with H2
0 50 100 150 200 250 300 350
Phase (degrees)
Figure 2-18 Dependence of the Overall Sound Pressure Level on the flow rate of the modulated
air jet with H2 addition injected near the step and the phase angle, for primary air flow rates of
4.4 m/s, and a fixed fuel flow rate. The overall equivalence ratio ranged from 0.52 to 0.49
depending on the flow rate of the air jet.
53
0.04
-4- Odeg
0.035 - 60deg
-+- 120deg
-ES- 180deg
0.03 .-- 240deg
300deg
0.025 --
0.02 -
0
0.015 -
0.01-
0.005
0
-2 0 2 4 6 8 10 12 14 16 18 0
Distance from the step(cm)
Figure 2-19 Dependence of the mean CH* distributions on the phase angle of the controller for p
= 2.89 with the overall equivalence ratio of 0.48 and without H2 addition.
Figure 2-19 shows the mean CH* distribution of the flame for the case of /p=2.89, over a
range of the controller phase angles. Clearly, the burning characteristics change significantly
with the phase-shift controller. At the optimal phase angles of 180 and 240 degrees, the burning
zone is closest to the step and the burning rate falls rapidly further downstream. In the most
unstable case at 60 degree lag, burning extends much further downstream, the peak drops by one
54
2.4 Summary
premixed backward-facing step combustor using a transverse air jet near the burning zone, and
H2 addition to the air jet. I first develop an insight into the interactions between hydrodynamics,
fuel/air mixture inhomogeneity and heat-release dynamics, and their impact on the instability of
the combustion system. A parametric study was conducted to investigate the impact of flame-
vortex interactions and fuel/air mixture inhomogeneity on the combustion instability. The
location of the fuel injection was varied, which changed the total convective time delay and the
level of fuel/air mixture inhomogeneity. Results show that when the primary fuel injection is
moved further upstream, the impact of fuel/air mixture inhomogeneity is negligible and
amplitude of pressure oscillations increase monotonically with increasing the Reynolds number.
High-speed CCD images of the flame confirm that a strong flame-vortex interaction is the
dominant mechanism for heat release dynamics in all cases. For this reason, I focus the control
effort on modifying the flame-vortex interactions in the combustion zone using a transverse jet of
I study experimentally the impact of a static and modulated transverse jet on stability and
emissions. I demonstrate that with sufficient momentum, a static air jet injected perpendicular to
the incoming flow near the step favorably modifies flame-vortex interactions and reduces
55
also investigate the impact of the transverse jet modulation using closed-loop control with
pressure feedback. I show that modulating the air jet can achieve the same level of pressure
reduction with 50% of the air flow rate required in the static air jet case, but the control
effectiveness is strongly dependent on the phase angle. In both cases, adding small amounts of
56
3. Modeling and Control of Spatio-temporal Combustion
Dynamics
over a wide range of conditions without running the risk of thermoacoustic instability. In active
instability control, the state of the system is determined using point measurements of the pressure
and/or the heat release rate, and the signals are used in a control algorithm to determine the
output necessary to stabilize the system. Improving the burning characteristics, in addition to the
suppression of pressure oscillations, requires a sensing technology that can detect some spatial
information, e.g., the spatiotemporal burning distribution. In this Chapter, I explore the
application of a new sensing technology described in Section 2.1.1 capable of capturing the
spatial properties of the combustion zone as a feedback sensor. The sensor is composed of 128
photodiodes arranged in a row that can capture light intensity up to 3000 frames/s. It was tested
Hz, which coincided with the quarter-wave acoustic mode of the inlet tube. The sensor was used
to capture high frequency spatial and temporal combustion fluctuations. To exploit the spatial
to reduce the number of states to a manageable size [44]. The burning characteristics are then
described using the amplitude of the first POD mode and a system identification model. Closed-
loop control results show that the controller can reduce pressure oscillation by 23 dB, while
57
making the burning zone more compact and bringing it closer to the sudden expansion section. In
Section 3.1, I describe the combustor setup. In Section 3.2, I show uncontrolled combustion
characteristics using Schlieren and linear photodiode array images. In Section 3.3, reduced-order
modeling based on POD and System Identification (SI) is described. In Section 3.4, the control
algorithm and its implementation are presented. Finally, closed-loop control results are described
in Section 3.5.
58
3.1 Experimental Setup
The combustor consists of a 4 cm circular pipe expanding into a square dump whose height
and length are 10 and 61 cm, respectively (see Figure 3-1 and Figure 3-2). Ethylene is premixed
with the air at flow rates of 0.9 g/s and 23 g/s, respectively. Two secondary fuel injectors are
located at the dump plane, 45 degrees with respect to the flow directions. However, only one fuel
injector is used in this experiment. Liquid ethanol is used in the secondary fuel stream at flow
rate of 0.15 g/s. A pressure sensor is located 5 cm downstream from the dump plane actuator.
Measurements are recorded using a Keithley MetraByte DAS-1801AO data acquisition and
The schematic diagram of the photodiode array is shown in Figure 3-3. It is composed of
128 photodiodes arranged in a row at spacing of 50 ,um with a height of 2.5 mm. This sensor has
high UV sensitivity with excellent output linearity. The image domain in this experiment is 10 by
14.6 cm as shown in Figure 3-3, and 68 of the 128 sensors are used at 500 frames/sec. Bi-convex
UV fused silica lens with diameter of 25.4 mm andf=35 (the lens focal length) is used to project
the flame images on the photodiode array. Changing the distance between the lens and the
combustion zone can vary the magnification factor. To capture CH* chemiluminescence, which
can be correlated with the heat release rate [33,34], a CH* filter centered at 430 nm with
bandwidth of 8 nm is used. To digitize and process the analog signal from the photodiode array,
59
main fuel
choked flowmeter
39.55
0
88 \ 24,00
secondary controfler fuel\
Zho ed f tOwmeter
60
14.6cm
430 nm
Optical Filter
Figure 3-3 The Linear photodiode array arrangement and the imaged area in the combustor.
61
3.2 Uncontrolled Combustion Characteristics
(Flame-Vortex Interactions)
The combustor was naturally unstable at equivalence ratio # =0.6 and air flow rate of 23
g/s. At this condition, the fundamental mode was 38-40 Hz along with several higher harmonics.
The frequency of the fundamental mode coincided with the quarter-wave acoustic mode of the
inlet tube. Due to the location of the main fuel injection, 1 m upstream of the step, the impact of
equivalence ratio oscillations on the heat release dynamics and its coupling with the acoustics
leading to the combustion instability is expected to be small. Instead, the Schlieren images in
Figure 3-4 show the flame wrapped around a large-scale structure, which is periodically shed
near the expansion section. Although the detailed mushroom-shaped flame front is not clearly
shown, due to the presence of small-scale structures and the wall effects, a large-scale flame
From measurements using the linear photodiode array at different phase angles, depicted in
Figure 3-5, one can correlate the heat release fluctuation with the large-scale flame structure; the
periodic vortex shedding, and its convection to downstream. Note that the photodiode array is
primarily a 1-D sensor; it integrates the signal in the cross stream direction. The convective
velocity of the large-scale structure identified in Figure 3-5 can be estimated from the slope of
the line connecting the maximum CH* points at different phase angles. For the case analyzed
here, this velocity is 10 m/s which corresponds to the mean velocity of the flow (note that the
3600 on the x-axis corresponds to 25 ms.). The existence of a large scale structure, which is
62
captured in the Schlieren images and the linear photodiode array data in Figure 3-4 and Figure
3-5, makes it possible to develop a reduced-order model representing the spatio-temporal heat
release characteristics, using POD analysis, as will be done in the next section.
63
e) Phase angle: 202 deg
Figure 3-4 Schlieren images at different phase angles (0 degree corresponds to the maximum
pressure). The domain of the Schlieren image is 20.2 cm x 7.5 cm.
64
140 0.025
-0.02
120
0.015
E100 0.01
E
00
-8.0
0 -0.005
E 40
20 -0.025
20
-0.021
Figure 3-5 Measured CH* fluctuations using the photodiode array at different phase angles (0
phase angle corresponds to pressure maximum). The viewing window is 100 x 146 mm.
65
3.3 Reduced Order Modeling Using POD and System
Identification
To develop a model that can capture the spatio-temporal variation of the heat release due to
the control input, Proper Orthogonal Decomposition (POD) is used to systematically extract the
most energetic modes. These modes are used as basis functions for a System Identification based
model in order to reduce the solution space to the smallest linear subspace that is sufficient to
Proper Orthogonal Decomposition (POD) is a systematic approach for extracting the most
energetic modes from a set of realizations obtained from the plant model. These modes can be
used as basis functions for Galerkin projections of the governing equations in order to reduce the
solution space to the smallest linear subspace that is sufficient to describe the system. The
decomposition is 'optimal' in that a N-ordered POD base corresponds to the smallest projection
error compared to any other N-ordered base in a mean-squared sense. Over the years, it has been
applied in several disciplines including fluid mechanics, stochastic processes, image processing,
signal analysis, data compression, process identification and control in chemical engineering, and
oceanography. Particularly in fluid mechanics, the POD technique has been applied in the
analysis of coherent structures in turbulent flows and in obtaining reduced order models to
66
Ravindran [46] developed a POD and Galerkin projection based reduced order model for
minimizing the vorticity level of a backward facing step flow using blowing on the step corner.
The set of realizations for deriving POD basis was obtained from a finite element analysis of the
flow by considering solutions to the controlled problem with appropriate control inputs. The
author found that POD provides a systematic way to improve accuracy of the reduced model
while maintaining well-conditioned system matrices, when compared to other reduced order
techniques. Graham et al. [47,48] investigated the application of low-order models (using POD
and Galerkin projection) to control of the unsteadiness in a cylinder wake by changing the
cylinder rotational speed. The authors demonstrated that a poor choice of the set of realizations
for computing the basis functions of the low-order model can negate the benefits of control and
cause unsteadiness to rise. The technique to choose the 'right' basis function was left as an open
question by the authors. The process of reducing the plant model order using POD and designing
a controller that takes the system from its original state towards its optimal state does not
guarantee that the reduced order process will converge to the optimal control of the original
(large) system. Arian et al. [49] established convergence of the POD-based reduced-order
technique in the presence of control input, by embedding it into the concept of Trust-Region
(TR) methods. Most of these researches are through simulations due to the difficulty of
processing large amount of data in real time. Here, I apply this POD technique to a combustion
control problem using the linear photodiode array and validate it in a 50kW dump combustor.
One problem that arises when using POD algorithms in closed-loop control is that the modal
shape can change due to the control input. To overcome this difficulty, I use the Recursive POD
67
(RePOD) to recursively updated the POD mode online [50]. I first summarize the standard POD
where a, (t) is the ith temporal amplitude, i,(x) is the ith spatial basis function, 1 is the number
of modes chosen, and t and x are the temporal and spatial variables, respectively.
The set of realizations from a plant required for the POD basis calculation is typically
obtained experimentally or computationally and as such is discrete in nature. So, what follows in
Let m be the number of temporal points in the data ensemble, and n be the number of
spatial points. The POD method consists of finding #, (x) such that the projection error
{(# (x)}= = #e 9I", == 1,..,l. The optimization problem is then given by:
Min J, (01,---, )= Qi I
j=1 k=1
68
subject to: # f# , =(5,,1 i~j <1 0,#= [#1,---#0
Also, (Q,Tk) in the cost function is the k h temporal amplitude ak at time t .. So the expression
2
within the first summation sign is Q - aq, which refers to the projection error for the
k=1
data
The POD modal set can be obtained using the 'method of snapshots' as follow [51,52].
Denote Q (xk) = Q(,xk ). Also, let the matrix Y be formed as follows: Y = [Q',-,Q]. Then,
a singular value decomposition of the matrix Y gives : Y = BEA, where A and B are unitary
matrices, and
CT 1 !C 2 .. cT071
U' 1
Then
A(j,i)Q(x
J=1 Cy-
The eigenvalues corresponding to the POD modes are the squares of the singular values
7o, , ,---,a }, and represent the 'energy content' of the modes. Next we consider Recursive
69
The RePOD Algorithm
Suppose we begin at time t, with 0(m), where #(') represent the POD modes at time t,,,.
Given a new measurement Q' 1 at time t,+,,1 we seek to minimize AJ, defined as
ki1 Eq.(3-1 )
The assumption is that with the addition of the new data set, the number of POD modes does not
change.
Step 1: Find #'"*)that minimizes AJ(1)= Q' -(Q'TI#) 2 over all #1 under condition that
Ste#i=. v
Step i: Using the values of #f'"*l),..., '), find #,(") that minimizes the overall # ,
The minimization of AJ(') in Step i can be carried out using gradient techniques:
a i 1,>0
70
(M+0) (m) + -(~~ 1 T M+1) (M+1) ±~T()pm)}r())E.3-2)
Equation (3-2) is the RePOD algorithm. While this does not guarantee orthogonality of #1 " at
each time tm, it leads to an orthonormal set of modes as t -* oo . This RePOD algorithm is used
with closed-loop control when the combustion dynamics is expected to change drastically and
Results
In the POD analysis, 500 consecutive photodiode array images (m = 500) were used. The
number of sensors used was 68 out of 128 (n = 68) and the sampling rate was 500 frame/sec. To
match the overall equivalence ratio with the controlled case, a white noise signal was sent to the
secondary fuel injector with a duty ratio of 50%. The results are depicted from Figure 3-6 to
Figure 3-8. Figure 3-6 shows that the first and second modes contain almost 60 % and 30 % of
the total energy, respectively, indicating that with one or two basis functions one can accurately
describe the spatial variations of heat release. The shapes of the basis functions are shown in
Figure 3-7, with the frequency spectrum of modal amplitudes in Figure 3-8. The first two modes
correspond mainly to the flame oscillation at around 40 Hz and the other two modes to 40 and 80
Hz.
71
100
90
80
70
60
a)
-o 50
E)
40
30
20
10
2 4 6 8 10 12 14 16 18 20
Number of Modes
0.25
-0. 1<
-0.1 5,
-0.2
20 40 60 80 100 120 140
Distance from the step(mm)
72
-10 -10
-20- ----------- ----------------- -20 -----------------------------
M -30 --- ----- -- ----- M -30 - - -- ---- -- -- -- -- - --- -- ---
8-40 8-40
-50 -- - - ---- - - - -
--- -50 --- - -------------- --
-60
0 50 100 0 50 100
Frequency(Hz) Frequency(Hz)
-10 -10
-20 --------------- --------------- -20 --------------- ---------------
M -30 ------------ ----------------- M -30 -------------- --------------
-- - -
8-40 ---------- 8-40 -------------- r------- ---
-----
-50 -50
lo--
-60 -60
0 50 100 0 50 100
Frequency(Hz) Frequency(Hz)
POD is an "empirical" method applied to determine the basis functions by examining the
highest existing correlations. The Navier-Stokes equation could then be projected onto these
basis function spaces to obtain reduced order models. However, studies showed that this
projection suffers from many problems, such as unknown boundary conditions, insufficient
damping when smaller scales are neglected, mathematical complexities in the reacting flow case,
and numerical errors generated by differentiation. Alternative methods to develop reduced order
models have been attempted by Jeong [53] and Gillies [54]. Jeong used POD to generate basis
functions in wind-induced pressure on a low-rise building, and applied auto Regressive (AR)
73
models to fit the response of first 4 modes. Gillies used POD and neural network to furnish an
empirical prediction of the modal response of the wake to external forcing, in a flow past a
circular cylinder. I use a similar approach as Jeong and Gillies's to develop a reduced order
To perform SI, a white noise signal was generated, filtered by a band pass filter with a
bandwidth of 10-100 Hz, and transformed into a binary signal. This signal was sent to the
secondary fuel injector, and also used to trigger the photodiode array. Using a,, the amplitude of
the l't POD mode, as the output and V, the fuel injector signal, as the input, as shown in Figure
3-9, an SI model was developed using a linear AutoRegressive with external input (ARX)
model-structure [55] and the time delay, z-,, , was found to be 2 ms. The resulting transfer
G,(s) = a
--1 = 0.00187s 3 +22.51s 2 +1.303.10 4s+6.116.106
exp(-0.002s)
V S4 + 78.68s' + 3.406. -10Is2 + 1.229 -I0's + 1.707 -10'0 Eq.( 3-3)
74
I
-6
0
2.02
a 0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
0 time(s)
0
0 0.4
0.04
a- .F2
CD0.02
00
2-0.02
CL
E-0.04
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
time(s)
Figure 3-9 Input and output data for system identification: the input, xi, is the amplitude of the
first POD mode, and the output is fuel injector signal.
75
I
3.4 Closed-loop Control
3.4.1 Algorithms
A model-based controller called adaptive Posicast was used for pressure stabilization. The
adaptive Posicast controller has been developed to stabilize dynamic systems with large time
delays [22], by predicting the future pressure response in order to eliminate the delay effects on
the closed-loop and using an adaptive algorithm to adjust its parameters. It attempts to predict
future outputs using a Smith Controller, and uses a phase lead compensator to drive the future
output (parameter fluctuation) to zero. The controller implemented is a discrete form, where the
where k(t) and d(t) are the controller parameters and data vectors, respectively and defined as
where vj(t)= v(t), k, is the high frequency gain of Gp(s), F is the diagonal matrix whose
S +ZC
diagonal values, Y1 I Y 2 , -- 7n+2, are the adaptation gains, A, ... A, are the Smith controller
76
components to eliminate the delay from the closed loop and d,(t) = I d(t). n is determined
s+ a
by the time delay, r,,, in the SI model G,(s), ndt = i-l,, where dt is the time step. kJ, k2 , and
zC are components of the phase lead compensator k, zC k .+zC is selected to stabilize the
unstable mode of the delay free system, while ki and k2 are determined by the adaptation law.
POD is needed to generate the amplitude of the first mode for use in the Posicast controller.
However, POD modes will generally change in response to control action, so it is necessary to
update them. The POD method involves batch processing past data, and is not feasible in real
time. Therefore, a recursive POD (RePOD) method described in 3.3.1 is used in closed control
cases.
3.4.2 Implementation
Flame images were first captured by the linear photodiode array at 500 frames/s using
LabVIEW in the Pentium IV computer. The amplitude of the first POD mode, xi, was computed
by projecting the captured images onto the first POD mode shown in Figure 3-7, and ci1 was
used as an input in the adaptive Posicast controller in Eq. (3-4) to generate a control signal for
the fuel valve. Finally, the first POD mode was updated using Eq. (3-2).
77
3.5 Results
The controller described in Section 3.4 was implemented and the results are shown from
Figure 3-10 to Figure 3-13. Figure 3-10 shows the pressure response without and with closed-
loop control. In the baseline case, the standard pressure level (SPL) reached 155 dB at 38 Hz,
with its harmonic at 76 Hz, whose amplitude is as strong as that of the primary. Open-loop
modulation in the secondary fuel injector, at 50 Hz, decreased the pressure amplitude at 38 Hz
slightly. Openloop forcing at different frequencies ranging from 10 to 40 Hz was also attempted
without significant improvement. Next, closed-loop control was used with and without POD
adaptation. It resulted in pressure reduction by 18-23 dB at the primary mode and complete
stabilization in its harmonic. Closed-loop control with POD adaptation led to 5 dB more
reduction in the primary unstable mode. The adaptation gain, 6, used in the POD adaptation was
0.2.
The impact of the gain, 6, was also examined. Figure 3-11 shows the pressure time
response with different POD adaptation gains. A moving window containing 10 previous
pressure cycles computed the OverAll Sound Pressure Level (OASPL) at a given time in that
figure. For all cases, the controller was turned on at t=1 s. Without POD adaptation (6=0),
controller was turned on. With 6 = 0.1, the pressure drop during 1 < t < 2 is relatively slow
compared to the no-adaptation case. However, pressure fluctuations after t = 2 are lower than in
78
the no-adaptation case. Finally, with 6=10, the pressure drop during 1 < t < 2 is fastest, with a
small overshoots during 2 < t < 6 due to the high adaptation gain.
The heat release distribution can change with control. This can be easily identified using
the photodiode array, as shown in Figure 3-12 and Figure 3-13. The maximum average CH*
emission level was doubled with closed-loop control within the photodiode array domain, with
its peak moving closer to the step, as shown in Figure 3-12. Note that the fuel and air flow rates
were the same in three cases. Figure 3-13 shows the distributions of the normalized CH*
emission fluctuation levels, Q', /Q, , where Q',,rs is the rms and Q, is the mean value of the ith
photodiode with a 1 s time window and i=1 ... 68, with and without control. Clearly, the
maximum fluctuation point is also shifted upstream significantly, more than 5 step height. The
point of the maximum fluctuation in the uncontrolled case was observed at the end of the
viewing window, 146 cm downstream from the step. However, in the controlled case, the
maximum fluctuation occurs very close to the expansion section. Also, there is a slight
79
170 170
-- No control 1- With RePOD
Forcing at 50Hz -.-- With Standard POD
160 ------------ ------------ ----------- ------------- ----------- 160 --------------------------------------- ------------ -------------
M -- - - --
Cn----------------140
120 120
0 20 40 60 80 100 0 20 40 60 80 100
frequency, Hz frequency, Hz
(a) (b)
Figure 3-10 The pressure spectra with and without closed-loop control.
155
- 6=0
6=0.1
156 8=10
154 -
152 -
150 -
148 -
146
1 2 3 4 5 6
Time (s)
Figure 3-11 Time history of the overall sound pressure level (OASPL) with and without POD
mode adaptation. OASPL is calculated from the pressure signal using a moving window
containing 10 pressure cycles (128 samples).
80
0.014
0.012
-+
0.01
0.008
0.006
0.004
0
20 40 60 80 100 120 140
Distance from the step(mm)
L45
a 40
.9
C)
-~35
03
20-
02
30
20
20 40 60 80 100 120 140
Distance from the step(mm)
Figure 3-13 The normalized CH* intensity fluctuations, Q,,,, /Q,, using a 1 s time window with
and without control. Q is taken from Figure 3-12.
81
3.6 Summary
properties of the combustion zone. A linear photodiode array which can capture high speed
spatio-temporal heat release fluctuations has been tested as a feedback sensor in a 50 kW dump
combustor. Pressure measurements identified strong oscillations at 38 Hz, which is the quarter-
wave acoustic mode of the inlet tube. Images taken using the linear photodiode array and a
Schlieren system showed large-scale flame oscillations due to strong flame-vortex interactions at
38 Hz. The vortical structures are convected downstream at a speed of 10 m/s. To develop a
systematically extract the most energetic modes with the first and second POD modes capturing
about 60 and 30% of the total fluctuations, respectively. The first mode is used as a basis
function for a system identification based model. A model-based controller called adaptive
Posicast was used for pressure stabilization, which resulted in pressure reduction of up to 23 dB.
Heat release characteristics, determined using CH* emission, showed that the maximum value is
doubled due to the action of control, and the burning zone moved closer to the step. It should be
82
4. Control of Industrial Swirl Stabilized Combustor
In this Chapter, I apply the active control strategy in more realistic gas turbine conditions.
The design of the rig is based on the primary stage of the Rolls-Royce RB2 11 -DLE industrial gas
turbine. To verify whether the strategy is effective across all regimes of plant operation,
robustness studies with respect to changes in the resonant frequency, total time delays, changes
in operating conditions and initial conditions of control parameters are made. Control is achieved
by modulating the fuel flow rate in response to a measured pressure signal and the controller
used is the adaptive Posicast described in Section 3.4.1. The controller achieves a reduction of up
over an empirical control strategy (simple phase-shift controller) specifically tuned to the same
operating point. Robustness studies have shown that the controller retains control for a 20%
change in frequency and a 23% change in air mass flow rate. Section 4.1 describes the
combustor setup. Section 4.2 discusses the uncontrolled combustion dynamics. Closed-loop
control results are in Section 4.3. Finally, robust studies are in Section 4.4.
83
4.1 Experimental Setup
4.1.1 Combustor
The facility is a generic combustor designed to model the fuel injection/premix ducts of a
Rolls-Royce RB2 11 -DLE industrial gas turbine. The swirler unit is a scale model, however the
geometry of the plenum and combustor has been reduced to simple cylindrical pipes to have
Figure 4-1.
Air
I 12 8
70
U7 tcombustor
choke plenum swirler
Figure 4-1 Schematic of the rig downstream of the choke plate, showing the plenum/combustion
chambers and the swirler unit. All dimensions in mm, diagram not to scale.
A metered and steady air flow is supplied to the plenum (128 mm ID) through a choked
plate. This effectively decouples the air supply from pressure fluctuations in the working section,
enabling air to be supplied at a constant mass flow rate (ma) and providing a well-defined
acoustic boundary condition. The maximum mass flow rate available is approximately 0.2 kg/s,
however the typical operating range is 0.03-0.08 kg/s. The downstream end of the plenum
84
chamber incorporates an annular section prior to the swirler unit. The swirler exit is connected to
a quartz tube (70 mm ID) along which combustion takes place. This provides ideal optical access
to the flame region. The exit of the tube is open to atmospheric pressure and is not choked.
The fuel (ethylene, C2 H4 ) and air are mixed using a Rolls-Royce DLE counter-rotating,
radial swirler unit, scaled to fit the existing facilities (approx. 54%). Once the airflow has passed
into the annular section, it is split into two streams that flow through concentric channels. In the
channels, blades are fixed to induce two counter-rotating flows. The fuel is injected upstream
into the annular channels through eight cylindrical bars each fitted with two exit holes of 1.0 mm
diameter (see Figure 4-2) A pressurized commercial cylinder is used to supply the fuel for the
swirler unit. The fuel line comprises a pressure regulator, control valve and turbine flow meter
(see Figure 4-3). Together with suitably located pressure transducers and thermocouples, the
turbine flow meter is used to provide a mass flow reading. Fuel mass flow rates (mf) are in the
range 1.6-3.0 g/s in order to obtain equivalence ratios from 0.5-1.0 for the air mass flow rates
85
fuel
airAH p
fuel
th.ueLnecifuas/air
outer channel
inner channel
Figure 4-2 Detailed schematic showing a cross-section of the swirler unit and the orientation of
the fuel injection bars.
constant high
pressure supply
plenum
ethyl ene
flow
4
cylin der Metering
valve now
meter DDV
Solenoid valve
Filter
swirler unit
Figure 4-3 Detailed schematic of the fuel system, together with the DDV and the plenum
chamber
86
4.1.2 Instrumentation and Actuation
The feedback signal for the control algorithms is provided by a Kistler piezo-electric
pressure transducer (type 601A) measuring combustor pressure 700 mm downstream of the
swirler unit. The transducer is located in a side arm, 30 mm from the combustor wall.
The two closed-loop control algorithms proceed in three steps: acquisition of the unsteady
pressure signal, processing of the control signal by the control algorithm in real-time and then
sending of the resulting signal to drive the DDV valve. The algorithms have been implemented
on a 32-bit M62 digital signal processing (DSP) board, developed by Innovative Integration,
Unsteady measurements of pressures, light and valve displacements are recorded on a PC-
board from National Instruments. The signals are conditioned using an isolation amplifier and an
eighth-order elliptic low-pass filter. The cut-off frequency is automatically set to one-third of the
sampling rate (5 kHz). The SPL data presented are acquired over periods of up to 20 seconds and
Actuation for control was achieved using the same high frequency servo valve described in
Section 2.1.2 (DDV valve from Moog) to modulate the fuel flow rate into the swirler and hence
produce variations in equivalence ratio in the premix ducts as shown in the fuel system
schematic (Figure 4-3). The DDV uses a linear force motor where the stroke is proportional to
87
the applied voltage. The large pressure drop across the regulator and the inertia of the flow meter
means that the mass flow rate is effectively constant upstream of the plenum, in spite of the high
frequency modulations downstream. The plenum provides mass storage, so that the fuel flow rate
through the DDV can change as its open area is modulated. The pipe length between the DDV
and the swirler unit was kept as short as possible in order to decrease the attenuation and time
delay. Note that there is a significant time delay in the actuation due to the transport of the fuel
from the location to the burning one since that needs to be accounted for in the control strategy.
88
4.2 Combustor Dynamics
The instability characteristics of the rig have been investigated for a range of flow
conditions and plenum and combustor lengths. Within the typical envelope of the operating
parameters (ma=0.03-0.08 kg/s, #0=.5-1.0) several intense combustion instabilities with multiple
frequencies have been observed. Typical pressure spectra contain dominant peaks at one or two
low frequencies (100 - 300 Hz) and an additional high frequency peak (550 - 700 Hz). These
In the current set of feedback control tests, the experimental data has been obtained in the
ranges ma=0.03-0.05 kg/s and mpl.6-2.5 g/s, resulting in #=0.5-0.75. For the geometric
configuration shown in Figure 4-1 (Lp=1.73 in., L,=1.0 in.) the rig exhibits a 207 Hz plenum
mode instability. In order to obtain different unstable frequencies the plenum length was varied
(Lp=1.46/1.73/1.97 m.). The pressure spectra for these cases all contain dominant low frequency
peaks (<300 Hz) with magnitudes of up to 165 dB that set an ideal challenge for active control
(Figure 4-4).
These dominant modes correspond to the second harmonic of the plenum (wavelength=Lp).
The fundamental and the higher harmonics are also present. In addition, for the Lp=1.97 m case
89
there is a peak at 244 Hz which corresponds to the quarter wavelength mode of the combustor
tube. The appearance of the combustor modes is not as predictable as the plenum modes.
170
- L = 1.73 m
p
L = 1.46 m
160 - .. , L= 1.97 m -
P9
Ip ;
150.....I.............................. ........... .
_J14 0 - - - --- - - -- --
CO)
130 -
120 -
110
0 100 200 300 400 500
frequency, Hz
Figure 4-4 SPL spectrum of the self-excited combustion oscillations; ma=0.04 kg/s, =0.7.
90
4.2.2 Determination of z, 0,
The total time delay, z,, is needed to determine the adaptation law in Eq.(3-4) and n,
respectively. This can be determined through open-loop forcing and investigating the transfer
function Gp(s) = p' [V at a range of frequencies through system-identification tests [18]. At the
desired operating condition, a simple phase shift controller was applied in order to minimize the
self-excited oscillation. In addition to this, a sine sweep of V(t), with changing frequency from
40 Hz to 430 Hz, was also supplied to the DDV valve. The resulting pressure data was post
processed with a bandpass filter [80 Hz - 230 Hz] to better capture dynamics around the unstable
frequency. Gp(s) was obtained using a linear ARX model-structure [55] and zO, was found to be
9.6 msec.
Initial tests with the Posicast were conducted using a system-identification reduced order
model to obtain optimal starting values for the control parameters. To obtain this another system-
identification was carried out using V(t-r,0 ,) and p'(t) to yield an ARX model, which resulted
in a Gp(s) given by
where sign(ko) = -1. This negative high frequency gain was an unexpected result and might be
the result of additional dynamics in the fuel supply system that were not addressed in the model.
91
For the current tests the sign of ko was changed in the adaptation scheme. In the current tests, the
user inputs into the adaptive Posicast were a sampling rate of 2 kHz, n=19, a=Z,=1000,
92
4.3 Closedloop Control Results
Initial tests of the controller produced reductions in noise that were 5-15 dB larger than
those gained with the appropriately tuned fixed phase shift controller. However, the intermittent
presence of a low frequency mode of the order of 1 Hz degraded the performance. Figure 4-5
shows a typical time series indicating the presence of the low frequency mode when the
controller is on.
On Off
1
0 ---- ...
------ - ----- .--. -- --.--..-.-
...-.
0
0
(,
CL,
-201
0 2 4 6 8 10 12 14 16 18 20
time, secs
100
80 -L -----
0 2 4 6 8 10 12 14 16 18 20
time, secs
Figure 4-5 Time series showing the low frequency fluctuations present in the combustor when
the adaptive Posicast controller is activated; ma=0.04 kg/s, =0.65, ro,=9.6 ms, Lp=1.73 m, zero
initial conditions for control parameters.
93
Analysis of the valve time series suggested that there was significant control action around
100 Hz, once the main instability was stabilized, prior to the controller divergence. A bandpass
filter (120-500 Hz) was introduced between detection and the controller input to remove the low
frequency dynamics. Figure 4-6 shows the resulting improvement in performance. The controller
is turned on at 4.4 seconds and stabilization is achieved in about 3.6 seconds. The control
parameters such as k, and k2 start at zero and converge to a specific value, as shown for k, in
Figure 4-6. The zero initial conditions simply imply that the controller does not know the
combustion dynamics initially and these values are automatically tuned by the adaptive
controller to give optimal performance. However, it should be mentioned that the settling time
for other test cases, under nominally the same operating conditions, can be large (up to 7
seconds). This was due to two reasons; 1) higher pressure oscillations (more than 10 kPa) made
the valve saturate (already present to some degree in Figure 4-6) and 2) zero initial conditions of
the control parameters requires time for the controller to search for the optimal control
parameters. When the pressure amplitude was smaller than 10 kPa and the initial conditions were
chosen based on Gp(s), the settling time could be reduced to less than 1 second as shown in
Figure 4-7. Figure 4-8 shows typical pressure spectrum for the control on/off cases where there is
a reduction of approximately 15 dB by the phase shift controller and 30 dB for the adaptive
Posicast controller at the 207 Hz instability'. One of the reasons for the limited performance of
the delay controller was due to a low frequency mode, which can be seen in Figure 4-8 and is
similar to that seen in Figure 4-5. In some cases, when this low frequency mode was not present,
the phase shift controller was able to reduce pressure oscillations up to 26 dB.
94
On Off
20
a_ 0
Co
-20L
0 2 4 6 8 10 12 14 16 18 20
'I,
10 2
50
0
CD 2 5 -- . -- ---...
-.. -. -.--- .. . .
.-.-
01
0 2 4 6 8 10 12 14 16 18 20
-401
0 2 4 6 8 10 12 14 16 18 20
time, secs
Figure 4-6 Time series showing the improvement on the pressure fluctuations in the combustor
gained by filtering the adaptive Posicast controller input; ma=0.04 kg/s, =0.65, r,=9.6 ms,
Lp=1.73 m, controller input filter=120-500 Hz.
10
10 -
20
3 -2 -1 0 1 2 3 4 6
-10
-10001
-10.002
-10.003
-10.004
-10.005 Controller On
- 10.006
3 -2 -1 0 1 2 3 4 5
timne(s)
Figure 4-7 Time series showing the improvement in the settling time of the adaptive Posicast
controller when appropriate initial conditions are chosen; ma=0.04 kg/s, 4=0.65, 1," =9.6 ms,
Lp=1.73 m, initial conditions of kr=-10, k2=10.
95
170 1 1 1
No control
Phase shift control
-- Adaptive Posicast
160 - -.. -- -- - - .. ..
- -. - --. . .. ..- - --...
. . .. . .. . ..-.-- - ....-- .-....
-.- ..-.
10
130 . . .. .
1. . .
2. 2. .4.4 .. .
120 A V
100
0 s0 100 150 200 250 300 t~ 40 450 50
frequency, Hz
Figure 4-8 SPL spectra showing the reduction in noise when the adaptive Posicast and the phase
shift controllers are turned on; ma=0.04 kg/s, 0=0.65, r, ,9.6 ms, L,=1.73 m, controller input
filter=120-500 Hz.
96
4.4 Robustness Studies
As mentioned previously, the adaptive Posicast controller can automatically tune its control
parameters when the combustion dynamics change. In this section, I carry out extensive
robustness studies with respect to changes in the resonant frequency, total time delays, changes
The resonant frequency varies as the temperature of a combustor changes. Banaszuk et al.
[56] reported a 20% change in resonant frequency over a 9 second period during warming up of
an experimental combustor. As a result, at a start up or during thrust change, the controller needs
to be able to cope with possible changes in frequency. As explained in Section 4.2.1, the resonant
frequency was changed by changing the plenum length, Lp. Initially, Lp was changed from 1.73
to 1.46 m and both the delay and the Posicast controllers were tested with parameters
predetermined in the nominal case (Lp=1.73 m). Figure 4-9 shows the pressure spectrum and as
expected, the resonant frequency moved to 240 Hz. The controller produced a 28 dB reduction at
the unstable frequency. It is worth noting that k, converges to positive values for this case as
shown in Figure 4-10, whereas it converges to a negative value in the nominal case in Figure 4-6.
The phase shift controller, using time delay/gain inputs determined in the nominal case, did not
have any effect and has not been included in Figure 4-10.
97
170
- - No ccontrol
-- Ada ptive Posicast
16 0
.------. --..---.-..--
.......... - -.--- -.- -...-- -
150
C,)
130
- - - - - - -- .. . .. . . . . .. . ....
120
110
10
50 100 150 200 250 300 350 400 450 500
frequency, Hz
Figure 4-9 SPL spectra showing the reduction in noise when the Posicast is turned on in the
shorter plenum case; ma=0.04 kg/s, 0=0.70, z-,,,=9.6 ms, Lp=1.46 m, controller input filter=120-
500 Hz.
25
20
15
10
0 2 4 6 8 10 12 14 16 18 20
time(s)
Figure 4-10 Change of the controller parameter, k, in case of Figure 4-9; ma=0.04 kg/s, =0.70,
Tt, ,= 9 6 4 6 m, zero initial conditions for control parameters.
0 . ms, Lp=1.
98
Next, Lp was increased to 1.97 m, and the phase shift and the adaptive Posicast controllers
were tested with parameters predetermined in the nominal case. Once again the phase shift
controller had no significant effect. Figure 4-11 shows the pressure spectrum without control and
with the adaptive Posicast cases and interestingly, two modes at 180 and 244 Hz were observed
in the no control case. The latter is the quarter wavelength mode for the combustor tube. The
optimal k, values are different for these two modes and hence k, is found to oscillate as it tries to
control first one and then the other mode (Figure 4-12). Finally, k, converges to a value that is in
between these two optimal values (Note: average k, negative in this case). Large adaptation gains
may generate overshoot during convergence, but this can be reduced using a smaller adaptation
gain, yr. When the adaptation gain was halved to yi =50000, k, showed less oscillatory behavior
(Figure 4-12) and the resulting pressure reduction was 15 dB (4 dB lower that that achieved with
yr =100000).
For both cases described above, the performance of the phase shift controller could be
improved by optimizing the time delay input. However, for Lp=1.46 m. only a 7 dB reduction
was achieved, compared with 28 dB for the adaptive Posicast. For L=1.97 m., a 22 dB reduction
was achieved, compared with only 15 dB for the adaptive Posicast. In this case, though, there
99
170
- No control
- Adaptive Posicast
160- - -- . - - -- -- -- -- -.-- - - -- --.- - - --
.-
150 - - -. -- - ---
.. - -- -
* fl:
140
CL
UI) 130 - - - - - -.- 1 -.-.- - - - .-1- . . . - - I ...-.
-. - .. -..
... -- ---
120 I - - o,
- L - - - -
.........
110
.J L. J
. . . ... J ..------
L ....... -----... .------ -------
100
0 50 100 150 200 250 300 350 400 450 500
frequency, Hz
Figure 4-11 SPL spectra showing the reduction in noise when the adaptive Posicast is turned on
in the longer plenum case; ma=0.04 kg/s, +=0.65, rtot =9.6 ms, Lp=1.97 m
100
--- =100000
.-- .7=50,000
-1
-2-
-3 -
-5
-6-
-7
0 5 10 15 20 25 30
time(s)
Figure 4-12 Change of the controller parameter, k, in a longer plenum case with different
adaptation gain, y, ; ma=O.04 kg/s, $=O.65, tot =9.6 ms, Lp=1.97 m.
The impact of a change in the total time delay on the performance of the adaptive Posicast
controller was also evaluated. Such a change can occur with a change in the flow rate or a
change of the burning zone location. Instead of changing the operating conditions, the total time
delay in the Posicast was changed by z-A .Experimental results show that the controller still
maintains control up to a maximum of r =1. 1msec less than the vr,, derived from the system
101
identification, as shown in Figure 4-13. However, control was lost if z-tt was increased beyond
9.6 ms.
170
N--No control
-- Control with x=9.6m s
Control with =8.5ms
160 --.
- --
- - -- - - -.
-- - - -- -- -.-- -- - - -- - - ....--..
- -- - -..... - -.--.. . -. . -- -
150
J.......... . . .
140
J,
CL
Mj
130
.... ..
.. ..
.. ......
120
- - - -- -
110
100
0 50 100 150 200 250 300 350 400 450 500
frequency, Hz
Figure 4-13 SPL spectra showing the reduction in noise when the adaptive Posicast was used
with correct( r,0 ,=9.6ms) and incorrect time delay (, 0 , =8.5ms); ma=0.0 4 kg/s, =0.7, Ir,-9.6
Ins, Lp=1.73 m, controller input filter=120-500 Hz.
The robustness of the adaptive Posicast controller was investigated with respect to changes
in the mass flow rate of the air. The controller was turned on at the nominal case with an air flow
rate of 0.04 kg/s, which was then decreased slowly by 23% (0.009 kg/s) over 100 seconds while
keeping the equivalence ratio constant. Figure 4-14 shows that the controller was able to
102
maintain control as the flow rate changes. After 100 seconds, once the controller was turned off,
the pressure amplitude returned to previous levels indicating that the final operating condition
was also unstable. The air mass flow rate was also increased from 0.04 kg/s to 0.055 kg/s.
The rig exhibits an unstable mode at 207 Hz when 4=0.6-0.7, and 4=0.6 is very close to the
flammability limit. The controller maintains control as 4 is varied within this range. At 4 larger
than 0.75, a combustor tube mode above 600 Hz was excited. As a result, the controller can only
be tested in the range =0.6-0.7 due to bandwidth limitation of the DDV valve and the
flammability limit.
103
15
10
5
ID
0
-1 0
-1
0 10 20 30 40 50 60 70 80 90
time(s)
42- I I I I I I I I I
40-
038-
636-
0
~34-
32-
30-
0 10 20 30 40 50 60 70 80 90
time(s)
Figure 4-14 Time series showing the effect of the adaptive Posicast controller with changes in
the mass flow rate of air; ma=0.04 kg/s, =0.7, Lp=1.73 m, controller input filter-120-500 Hz.
As mentioned earlier, carefully chosen initial conditions can reduce the settling time of the
adaptive Posicast controller significantly. We can also test the sensitivity of the controller to the
initial conditions of control parameters, by deliberately setting the control parameters far from
the optimum values. The control gain, k1 , was set to +10 initially, which is opposite in sign to the
optimal value. Figure 4-15 shows that the controller still converges to the optimal parameters and
that it is robust with respect to the initial conditions of the control parameters. However, the
104
settling time increased by an order of magnitude over that shown in Figure 4-15 where the initial
20
10
- 10
-20
0 2 4 6 8 10 12
time(s)
10
0--
-10 --
-20 -
-30-
-40
0 2 4 6 8- 10 12
time(s)
Figure 4-15 Time series showing the effect of the adaptive Posicast controller with incorrect
initial condition in the control parameter; m,=0.04 kg/s, 0-=0.65, ,,=9.6 ms, Lp=1.73 m,
controller input filter=120-500 Hz.
105
4.5 Summary
Forced unsteady combustion has been successfully achieved by using a high frequency
valve to modulate the rate of fuel injection into a Rolls Royce RB2 11 -DLE swirler unit. Tests on
the current combustion rig, which has a significant time delay (9.6 ms), have shown that it can
successfully reduce noise levels by an additional 5-15 dB compared to a tuned fixed phase shift
controller and by pulsing the same amount of fuel (2%). Better performance of the Posicast
controller in a nominal condition can be attributed as follows: 1) Since the closed loop is
essentially a delay free system due to the design of the adaptive Posicast, the transport delay does
not interact with the inherent dynamics of the combustor. Hence, it is not likely to introduce
secondary peaks. 2) Due to 1), larger gain is permitted which results in larger pressure reduction
Robustness studies have been started to test the adaptive Posicast controller over a wide
range of frequencies and operating conditions. The initial results have been encouraging. With
changes in the resonant frequency of 20%, the adaptive Posicast controller was able to adjust its
control parameters and produce a significant pressure reduction. Even with two unstable modes,
the controller adjusted its parameters to compromise and produce a pressure reduction in both
modes. The algorithm is not currently capable of adapting to a variable ro,, however 1.1msec
delay change was achievable. For air flow rate and equivalence ratio, 23% and 14% changes,
106
Due to limited authority of the actuator and no prior knowledge of combustion dynamics
(zero initial conditions of control parameters), the settling time could be large (up to 7 seconds).
Also, when two unstable modes are present, large adaptation gain produced overshoot in control
parameters. These suggest that adaptation gain and initial conditions of control parameters need
to be selected carefully to make the adaptive Posicast controller better perform in the 'transient
region'. After the control parameters were converged, the performance of the controller was not
changed with changes in the adaptation gain and the initial conditions.
107
5. Conclusions
This thesis addressed current challenges in the area of combustion control in three distinct
directions for transition of the active control strategy to industrial applications. First developing
an understanding of the interactions between hydrodynamics and heat release dynamics and their
impact on the underlying acoustics in the combustion system. Secondly addressing the
instabilities in a combustor. Thirdly verifying the active control strategy at more realistic
conditions including swirl, a large convective time delay in the control input and changes in the
resonant frequency and operating conditions. I addressed these challenges in three different
premixed backward-facing step combustor (Configuration I) using a transverse air jet near the
step. First, a parametric study was conducted to investigate the responsible mechanism for
negligible and the hydrodynamics is the dominant mechanism for combustion instability. For this
reason, I focused the control effort on modifying the flame-vortex interactions in the combustion
zone using a transverse air jet. I studied experimentally the impact of static and modulated
transverse jet on stability and emissions. With sufficient momentum, a static air jet injected
perpendicular to the incoming flow near the step favorably modified flame-vortex interactions
and reduced pressure oscillations by up to 14 dB, while reducing NOx emission by an order of
108
magnitude. Using closed-loop control with pressure feedback, the air jet achieved the same level
of pressure reduction with 50% of the air flow rate required in the static air jet case. However,
control effectiveness was strongly dependent on the phase angle. Adding small amounts of H2
either in the air jet or in the primary fuel led to further reduction in OASPL and extension of the
flammability limit.
of the combustion zone in the axisymmetric dump combustor (Configuration II). A linear
photodiode array which can capture high speed spatio-temporal heat release fluctuations was
oscillations at 38 Hz, the quarter-wave acoustic mode of the inlet tube. Images taken using the
linear photodiode array and a Schlieren system showed large-scale flame oscillations due to
strong flame-vortex interactions at 38 Hz. To develop a model of this combustion system, Proper
Orthogonal Decomposition (POD) was used to systematically extract the most energetic modes
with the first and second POD modes capturing about 60 and 30% of the total fluctuations,
respectively. The first mode was used as a basis function for a System Identification based
model. A model-based controller was used for pressure stabilization, which resulted in pressure
reduction of up to 23 dB. Heat release characteristics, determined using CH* emission, showed
that the maximum value of CH* is doubled due to the action of control, and the burning zone
Finally, to validate the model-based active control strategy in industrial gas turbine
combustion, a Rolls-Royce RB2 11 -DLE industrial gas turbine (Configuration III) was used in
Chapter 4. The combustor exhibited a significant time delay (9.6msec) in the control input due to
109
the location of the fuel injection. To accommodate this time delay, the adaptive Posicast
controller was used. Tests showed that it could successfully reduce noise levels 5-15 dB beyond
the level achieved by a phase-shift controller while pulsing the same amount of fuel. Robustness
studies showed that the adaptive Posicast controller could also accommodate changes in the
Results obtained from these three configurations show that through an understanding of the
underlying physics and reduced-order modeling, one can design an appropriate actuation,
sensing and control algorithm, all of which lead to model-based active control that reduces
pressure oscillations to background noise. In addition, active control improves other performance
matrices, such as reduction in NOx emissions, increase in the volumetric heat release and the
extension of the flammability limit, providing key building blocks for transition of the active
generation.
110
References
[1] J.W.S. Rayleigh, "The theory of Sound", Volume 2, Dover, New York, 1945.
[ 6 ] S. Park, A.M. Annaswamy and A.F. Ghoniem, "Heat release dynamics modeling of
kinetically controlled burning," Combustion andFlame, Vol.128, pp. 2 1 7-231, 2002.
[ 7 ] K. Yu, K.J. Wilson and K.C. Schadow, "Scale-up experiments on liquid-fueled active
combustion control," AIAA Paper 98-3211, 34th AIAA/ASME/SAE/ASEE Joint Propulsion
Conference & Exhibit, Cleveland, OH, 1998.
[ 8 ] E. Gutmark, T.P. Parr, K.J. Wilson, D.M. Hanson-Parr and K.C. Schadow, "Closed-loop
control in a flame and a dump combustor," IEEE ControlSystems, Vol. 13, pp.73-78, 1993.
[ 9 ] P. J. Langhorne, A.P. Dowling and N. Hooper, "Practical active control system for
combustion oscillations," Journal of Propulsionand Power vol.6, No. 3, pp324-333, 1990.
111
[ 11 ] J. P. Hathout, A. M. Annaswamy, M. Fleifil and A. F. Ghoniem, "A model-based active
control design for thermoacoustic instability," Combustion Science and Technology, Vol. 132,
pp. 99-138, 1998.
[ 12 ] A.M. Annaswamy, M. Fleifil, J.W. Rumsey, R. Prasanth, J.P. Hathout and A.F. Ghoniem,
"Thermoacoustic instability: model-based optimal control designs and experimental validation,"
IEEE Transactionson ControlSystems Technology, Vol. 8, No. 6, November 2000.
[ 15 ] T. Lieuwen, Y. Neumeier and B. T. Zinn, "The role of unmixedness and chemical kinetics
in driving combustion instabilities in lean premixed combustors," Combustion Science and
Technology, Vol. 135, pp.193-211, 1998.
[ 18 ] S. Murugappan, S. Park, A.M. Annaswamy, A.F. Ghoniem, S. Acharya and D.C. Allgood,
"Optimal control of a swirl stabilized spray combustor using system identification approach,"
Combustion Science and Technology , Vol.175, pp.55-81, 2003.
[ 19 ] Y. Neumeier, N. Markopoulos and B.T. Zinn, "A procedure for real-time mode
decomposition, observation and prediction for active control of combustion instabilities," In
Proceedings of the IEEE Conference on ControlApplications, paper No. 97318, Hartford, CT,
1997.
112
[ 21 ] J.P. Hathout, M. Fleifil, A.M. Annaswamy and A.F. Ghoniem, "Combustion instability and
active control using periodic fuel injection," AIAA Journalof Propulsionand Power, Vol. 18,
No. 2, pp.3 9 0 -399, March-April 2002.
[ 22 ] S. Evesque, A.P. Dowling and A.M. Annaswamy, "Self-tuning regulators for combustion
oscillations," Royal Society JournalProceedingsof the Mathematical,Physicaland Engineering
Sciences, Vol. 459, Issue 2035, pp 1709-1749, 2003.
[ 25 ] K.C. Schadow, E. Gutmark, K.J. Wilson and R.A.Smith, "Multistep dump combustor
design to reduce combustion instabilities," Journalof Propulsionand Power, Vol. 6, No. 4,
pp.407-411, 1990.
[ 26 ] J.-Y. Ren, W. Qin, F. N. Egolfopoulos and T.T. Tsotsis, "Strain-rate effects on hydrogen-
enhanced lean premixed combustion," Combustion and Flame, Vol.124, pp. 717-720, 2001.
[ 28 ] G.S. Jackson, R. Sai, J.M. Plaia, C.M. Boggs, K.T. Kiger, "Influence of H2 on the response
of lean premixed CH 4 flames to high strained flows," Combustion and Flame, Vol.132, pp.503-
511, 2003.
[ 30 ] J.R. Maughan, J.H. Bowen, D.H. Cooke and J.J. Tuzcon, "Reducing gas turbine emissions
through hydrogen-enhanced steam-injected combustion," Journal of EngineeringforGas
Turbines and Power, Vol.118, pp. 78-85, 1996.
[ 31 ] J.L. Gauducheau, B. Denet, G. Searby, "A numerical study of lean CH 4/H2/air premixed
flames at high pressure," Combustion Science and Technology, Vol.137, pp. 81-99, 1998.
113
[ 32 ] A.J. Riley, S. Park, A.P. Dowling, S. Evesque and A.M. Annaswamy, "Adaptive closed-
loop control on an atmospheric gaseous lean-premixed combustor," ASME IGTI Turbo Expo,
Atlanta, GA, June 2003.
[ 33 ] B. Higgins, M. Q. McQuay, F. Lacas and S. Candel, "An experimental study on the effect
of pressure and strain rate on CH chemiluminescence of premixed fuel-lean methane/air flames,"
Fuel 80, pp. 1583-1591, 2001.
[ 37 ] J. Brouwer, B. A. Ault, J. E. Bobrow, and G. S. Samuelsen, "Active control for gas turbine
combustors," 23rd Symposium (International)on Combustion/The Combustion Institute, pp.
1087-1092, 1990.
[ 38 ] T. Lieuwen, H. Torres, C. Johnson and B.T. Zinn, "A mechanism of combustion instability
in lean premixed gas turbine combustors," Journalof EngineeringforGas Turbine and Power,
Vol. 123, No. 1, pp. 182-189, 2001.
[ 39 ] K.K. Venkataraman, L.H. Preston, D.W. Simons, B.J. Lee, J.G. Lee, D.A. Santavicca,
"Mechanism of combustion instability in a lean premixed dump combustor," Journalof
Propulsionand Power, Vol.15, No.6, pp.909-918, 1999.
[ 41 ] S. H. Smith and M. G. Mungal, "Mixing, structure and scaling of the jet in crossflow,"
Journalof FluidMechanics, Vol. 357, pp. 83-122, 1998
114
[ 42 ] K. Hanamura, K. Bohda and Y. Miyairi, "A study of super-adiabatic combustion engine,"
Energy Conversion and Management, Vol. 38, No. 10-13, pp.1259-1266, 1997.
[ 43 ] A.F. Ghoniem, A.M. Annaswamy, S. Park, Z. C. Sobhani, "Stability and emissions control
using air injection and H2 addition in premixed combustion," Proceedings of the Combustion
Institute, Vol. 30, 2004.
[ 46 ] S. Ravindran, "A reduced order approach to optimal control of fluids using proper
orthogonal decomposition," InternationalJournalforNumerical Methods in Fluids, Vol. 34, pp.
425-448, 2000.
[ 47 ] W. R. Graham, J. P. Peraire and K. Y. Tang, "Optimal control of shedding using low order
models, Part I- Open-loop model development," InternationalJournalforNumerical Methods in
Engineering,Vol. 44, pp. 945-972, 1999.
[ 48 ] W. R. Graham, J. P. Peraire and K. Y. Tang, "Optimal control of shedding using low order
models, Part II- Model-based control," InternationalJournalforNumerical Methods in
Engineering,Vol.44, pp. 973-990, 1999.
[ 52 ] D. Tang, D. Kholodar, J.-N. Juang and E. H. Dowell, "System identification and proper
orthogonal decomposition method applied to unsteady aerodynamics," AIAA Journal,Vol.39,
2001.
115
[ 53 ] S. H. Jeong, B. Bienkiewicz. "Application of autoregressive modeling in proper
orthogonal decomposition of builing wind pressure," Journalof Wind Engineeringand
IndustrialAerodynamics, 69-71, pp.685-695, 1997.
[ 55 ] L. Ljung, System Identification, second edition, Prentice Hall, Upper Saddle River, NJ,
1999.
116