ACTIVE STABILIZATION OF ROTATING STALL AND SURGE IN A

TRANSONIC SINGLE STAGE AXIAL COMPRESSOR

H. J. Weigl, J. D. Paduano, L. G. Frchette, A. H. Epstein, E. M. Greitzer

Gas Turbine Laboratory, Department of Aeronautics and Astronautics
Massachusetts Institute of Technology
Cambridge, MA 02139

M. M. Bright, A. J. Strazisar
NASA Lewis Research Center
Cleveland, Ohio 44135

ABSTRACT 1 INTRODUCTION
Rotating stall and surge have been stabilized in a transonic The concept of using rotating stall control to increase the
single-stage axial compressor using active feedback control. The stable compressor operating range has been demonstrated on sev-
control strategy is to sense upstream wall static pressure patterns eral research compressors (Day (1993), Paduano et al. (1993),
and feed back the signal to an annular array of twelve separately Haynes et al.(1993), and Behnken et al. (1995)) and is now fac-
modulated air injectors. At tip relative Mach numbers of 1.0 and ing issues which will determine feasibility in modern full-scale
1.5 the control achieved a 11% and 3.5% reduction in stalling devices. Whereas most experiments have been conducted on low-
mass flow respectively, with injection adding 3.6% of the design speed compressors, the three main issues addressed in this paper
compressor mass flow. The aerodynamic effects of the injection concern high-speed machines: (i) high bandwidth, high authority
have also been examined. actuation, (ii) modeling compressible rotating stall dynamics, and
At a tip Mach number, Mtip, of 1.0, the stall inception dy- (iii) demonstration of stabilization in a transonic compressor.
namics and effective active control strategies are similar to re- The issue of actuation is simply this: In a high-speed com-
sults for low-speed axial compressors. The range extension was pressor, are there feasible approaches to forcing pre-stall waves
achieved by individually damping the first and second spatial with sufficient bandwidth and authority to stabilize these waves?
harmonics of the pre-stall perturbations using constant gain feed- Most previous studies, conducted at much lower Mach numbers,
back. have not had as stringent actuator power and bandwidth require-
At a Mtip of 1.5 (design rotor speed), the pre-stall dynamics ments. This paper describes the first use of distributed inlet in-
are different than at the lower speed. Both one-dimensional (surge) jection actuation to excite, identify, and stabilize pre-stall waves
and two-dimensional (rotating stall) perturbations needed to be in a transonic compressor.
stabilized to increase the compressor operating range. At design For low tip Mach number devices, a two-dimensional, lin-
speed, the instability was initiated by approximately 10 rotor revo- earized hydrodynamic stability model of rotating wave dynamics
lutions of rotating stall followed by classic surge cycles. In ac- has proven useful for conceptualizing and implementing control
cord with the results from a compressible stall inception analysis, systems (see Epstein et al. (1989), Hendricks and Gysling (1994),
the zeroth, first, and second spatial harmonics each include more and Gysling and Greitzer (1994)). These studies have shown that
than one lightly damped mode which can grow into the large am- stabilization of pre-stall waves, which grow into rotating stall,
plitude instability. Forced response testing identified several can be used to increase the compressor operating range by allow-
modes traveling up to 150% of rotor speed for the first three spa- ing operation at normally unstable conditions.
tial harmonics; simple constant gain control cannot damp all of An important question is whether these modeling and con-
these modes and thus cannot stabilize the compressor at this speed. trol ideas carry over to modern high-speed devices, where com-
A dynamic, model-based robust controller was therefore used to pressibility is an important factor. In particular, we wish to know
stabilize the multiple modes which comprise the first three har- the character of the waves which lead to rotating stall when un-
monic perturbations in this transonic region of operation. stable, and the feasibility of actuating and stabilizing these waves

1
to increase the stable operating range. In short, does this concep- Compressible modes would not interest us if they remained
tual approach work in high-speed machines? These modeling stable and their frequency remained outside the bandwidth of con-
and control questions have recently been answered theoretically; trol. However the theory predicts these modes to be critical in high-
experimental demonstration of the theoretical predictions is the speed compressors, both in frequency and stability. The frequency
focus of this paper. is set by axial length rather than rotor frequency, because axial struc-
The actuators used in the present work are discrete jet injec- ture is primarily set by acoustic considerations. The incompressible
tors mounted on the compressor casing. The jet injectors are run (e.g. [1, 0]) mode frequency, in contrast, is set by rotor frequency
with a steady mean blowing rate so that positive and negative (typically [1, 0] rotates at between 0.2E and 0.6E). Thus, for in-
unsteady actuation can be commanded. As will be shown, steady stance, the [1, 0] and [1, 1] modes become closer as rotor frequency
mean injection has a strong positive impact on the compressor increases. At design speed in a modern high-speed compressor, the
stable operating range. This behavior has also been reported by [1, 1] mode generally rotates at about 1E.
Koch and Smith (1968) and Koch (1970), who investigated the This mode has been identified as a likely cause of instability
impact of upstream blowing and downstream bleed on the stable in various high-speed compressors by Tryfonidis et al. (1995).
operating range of a transonic compressor. Another concern is that excitations, such as 1E excitation or feed-
In the next section, we give a brief account of the theoretical back from active control systems, can cause large resonances that
developments which determine the requirements for successful con- can drive the system into rotating stall.
trol. Section 3 then describes the experimental setup used for the The set of first harmonic modes (... [1, -1], [1, 0], [1, 1], ...)
demonstrations, including the implementation and characterization are decoupled from the set of second harmonic modes (... [2, -1],
of injection actuation. Section 4 compares the dynamic behavior of [2, 0], [2, 1], ...) because they are spatially orthogonal. Thus we
the high-speed compressor model to forced response measurements do not expect these sets of modes to interact if we actuate the
carried out in the transonic stage. Section 5 gives stabilization re- harmonics independently. However, all of the lightly damped
sults using a constant-gain feedback control law at 70% speed. This modes with the same circumferential wave number affect the pre-
section also includes a brief overview of the robust H control law stall dynamics. When actuating the first harmonic, for instance,
design used at 100% speed followed by experimental results. Dis- the control law must properly account for the actuation effects on
cussion and conclusions are presented in Section 6. both the [1, 0] and the [1, 1] mode.
As shown in Section 4, predictions from the compressible
2 PRE-STALL WAVE DYNAMICS IN HIGH-SPEED COM- rotating stall theory are not yet accurate enough for a priori con-
PRESSORS trol law design. The models qualitative behavior does however
The two-dimensional, linearized stability model of Moore and drive the control system implementation and design. Because
Greitzer (1986) has been extended by Bonnaure (1991) and the [1, 1] mode appears to be important to compressor stability,
Hendricks et al. (1993) to the compressible flow regime. The analy- actuator bandwidth must be sufficient to stabilize this and other
sis has been further modified to include actuation by Feulner et al. lightly damped high frequency compressible modes. In addition,
(1994), who also converted the model to a form compatible with identification of the compressor behavior is needed to character-
control theory. The analysis describes pre-stall compressor ize the incompressible and compressible modes. Frequency shap-
eigenmodes: their circumferential () and axial shape, rotation fre- ing in the control law is then needed to simultaneously stabilize
quency, and growth rate (stability). Here we discuss the nature of (for instance) the [1, 0] mode, while avoiding instability in the
the compressible modes, and the implications for active control. [1, 1] and higher modes.
See Frchette (1997) for a physical explanation of these modes.
A qualitative difference in the pre-stall dynamics which is 3 EXPERIMENTAL SETUP
due to compressibility is the introduction of new modes with axial The experiments described here were conducted in the NASA
structure. In incompressible stability theory the modal perturba- Lewis Research Centers Single-Stage Axial Compressor Test Fa-
tions vary in (sinusoidally), but at a given location the axial cility, an open circuit facility for testing advanced compressor
velocity perturbations are uniform through the compressor. In stages. A new casing for the compressor was designed and built
compressible stability theory, on the other hand, the modes have to accommodate the actuators and sensors required for rotating
not only a sinusoidal circumferential shape but also axial struc- stall control as illustrated in Figure 1. Some of the basic consid-
ture so the axial velocity and mass flow vary with and with erations in the experimental setup are discussed here; for a com-
axial position in the compressor. We thus refer to the modes in plete description of the injection actuation development, see
terms of two wave numbers, a circumferential and an axial one. Berndt (1995) and Berndt et al. (1995).
The modes denoted [0, 0], [1, 0], [2, 0], ...[n, 0] are like those The test compressor, designated as NASA Stage 35, was origi-
originally described by Moore and Greitzer for incompressible nally designed as an inlet stage of an eight-stage 20:1 pressure
flow, and those denoted [n, -1], [n, 1], [n, 2], etc. have an nth ratio core compressor (Reid and Moore, 1978a). The design con-
harmonic shape in and various shapes axially. We refer to the ditions are a total pressure ratio of 1.82, a mass flow of 20.2 kg/s,
latter as compressible modes. a rotor tip speed of 455 m/s, and a rotation frequency of 286 Hz.

2

SHEET 3-HOLE
INJECTOR INJECTOR

R Side View Side View 30.0

Ps Ps Ps

Stator
Rotor
z Ps

FLOW

Ps High Response Wall Static Pressure Sensors 22.5

Actuator/Injector
View from
Area Traverse for Steady/Unsteady Measurements Downstream Bottom View

Figure 1: Stage 35 actuation and instrumentation schematic.

The blade tip diameter is approximately 50 cm with a hub-to-tip pressor mass flow when supplied with 100 psi air). The actua-
radius ratio of 0.7. The mean-line rotor and stator chord lengths tors must also have sufficient bandwidth to stabilize the lightly
are 56 mm and 40 mm. A detailed description of performance damped modes of interest. These actuators, developed by Moog
measurements from Stage 35 is given by Reid and Moore (1978b). Inc. and MIT and described by Berndt, have a large signal band-
Data acquisition and control are performed by a 90 MHz width of 400 Hz. It should be noted that this is only 1.4E, and
Pentium PC computer system, with 64 input and 16 output chan- that the theory predicts there are lightly damped modes traveling
nels. The sample rate used is 3 kHz, or roughly 10E, to insure significantly faster. The actuators have performed reliably over
that controller delays will not adversely affect performance. The three months of extensive testing which was a primary concern
optimized control loop can implement linear control laws with in such a harsh test environment.
up to 60 states at 3 kHz, while acquiring data in a circular buffer Two different injector geometries were used in this study (see
of up to 45 seconds in length (limited only by RAM). Inputs to Berndt (1995)). The injector is the portion of the actuator which
the control computer include high-response pressure transducer directs flow, modulated by a servo motor driven valve, from the
measurements, position sense signals from the actuators, and low actuator to the compressor face. Its shape is critical to the injection
frequency compressor performance measurements (mass flow, characteristics. After extensive wind tunnel testing, two types of
pressure rise, and throttle position). High-response sensors are injectors were designed for testing in this program. The first type
distributed circumferentially (8-12 wall static pressure ports at attempts to provide a radially well-mixed axial momentum flux at
each axial station) to allow decomposition of perturbations into the compressor face. There are three holes in the injector as shown
harmonics. The array of sensors immediately upstream of the in the inset in Figure 1. Each hole is at an angle of 30 relative to the
rotor provided the cleanest traveling wave measurements and were casing wall in the R-z plane. The objective is to inject over a signifi-
therefore used for forced response and control testing. cant portion of blade span because previous stall control experi-
The outputs of the control computer are 12 independently ments have been conducted with variable-pitch IGVs which affect
commanded mass flow injection rates. Figure 1 shows the loca- the entire blade span. The resulting injection profile covers the up-
tion of a circumferential array of 12 jet actuators placed 1.1 rotor per 40% of the annulus height. This injector is flush with the casing
chord lengths (63 mm) upstream of the compressor face. This wall - that is, it does not protrude into the flow.
axial distance is chosen to be as small as mechanically feasible, The second injector geometry, shown in the inset in Figure
to reduce the convective time delay between actuator fluid injec- 1, was designed to inject a sheet of high-momentum air along the
tion and compressor response. Measurement of the dynamics of casing wall into the rotor tip region. The objective of the sheet
fluid forcing indicate that this delay is one of the major dynamic injector is to provide maximum control authority taking advan-
effects which must be minimized. Berndt (1995) gives a com- tage of the fact that Rotor 35 is tip critical. Berndt found that a
plete description of the actuator design requirements; other im- large radius of turning was required to accomplish this, so the
portant issues are briefly discussed below. injector protrudes into the flow by 7% of the rotor inlet span. As
The actuators were designed to deliver a relatively high mass with the three-hole injector, the exit area must be sufficiently large
flow (they are capable of delivering 5.8% of the design com- to prevent mass storage, which results in delay, in the valve body.

3
 2 1.3 1.6
Solid Casing A
1.9 50% Injection B E
1.2 1.5
C
D
1.8

Total-to-Static Pressure Ratio

Total-to-Static Pressure Ratio

Total-to-Total Pressure Ratio

REF70
1.1 1.4
1.7

1.0 1.3
1.6
100% Speed

1.5 0.9 1.2

REF100
1.4
0.8 1.1
1.3 Solid Casing Solid Casing
0.7 0% Injection 1.0 0% Injection
1.2 50% Injection 50% Injection
70% Speed 100% Injection 100% Injection

1.1 0.6 0.9

8 10 12 14 16 18 20 22 8 10 12 14 16 16 17 18 19 20 21 22
Total Mass Flow, kg/s Total Mass Flow, kg/s Total Mass Flow, kg/s
(a) Pressure ratio at 70% and 100% speed. (b) Pressure ratio at 70% speed. (c) Pressure ratio at 100% speed.

Figure 2: Compressor characteristics for steady blowing with sheet injectors at -15 yaw angle.

About 50% of the circumference over the outboard 15% of the data acquired with 7-element pressure and temperature rakes lo-
span is affected by the injected sheets from the twelve actuators. cated downstream of the stage. The total-to-static pressure ratio is
plotted since its slope is analogous to the slope of total-to-static pres-
3.1 Steady State Injection Characteristics sure rise coefficient which is a relevant parameter for rotating stall
The actuators were designed so that the injection yaw angle in low-speed machines:
(incidence angle with respect to the rotor) can be varied in 15 Psexit Ptinlet
degree increments by rotating the injector on the valve body. In- Low-Speed:
U 2
jector type and injection yaw angle thus form a matrix of possible
injection configurations. The configuration providing the largest Psexit Ptinlet Psexit
High-Speed: = 1
increase in pressure rise and operating range was determined ex- Ptinlet Ptinlet
perimentally, using steady injection with a fixed supply pressure
of 100 psi and 0%, 50% and 100% valve opening. Unless other- The results of the injector configuration study are summarized
wise noted, all future references to "50% blowing" imply a 100 in Figures 2 and 3. For active control, we are primarily interested in
psi supply pressure with a 50% valve opening. modulating pressure rise at fixed mass flow, that is we would like
All of the compressor speedlines presented in this paper in- the speedlines to be separated vertically. The sheet injector directed
clude the combined effects of steady injection and of the compres- 15 against rotor rotation provided the largest and most consistent
sor itself. The inlet pressure is measured upstream of the actuators separation. Figure 2a shows the total-to-total pressure ratio maps
and the total corrected mass flow (kg/s) is the sum of the inlet flow for a solid casing and for this injector configuration with 50% of the
rate measured by an orifice far upstream of the compressor and the maximum blowing rate (3.6% of the design compressor mass flow).
injection mass flow rate measured by a separate venturi meter1. Total- Significant range extension is obtained with this amount of steady
to-total pressure ratio maps and inlet total-to-exit static pressure ra- blowing, which is the mean blowing level at which all of the control
tio maps were both constructed. The exit static pressure was deter- experiments are conducted. Mean blowing is introduced so that
mined by averaging inner and outer wall static pressure downstream positive and negative actuation can be commanded (similar to the
of the stage. Exit total pressure was determined by mass-averaging injection scheme used by Gysling and Greitzer (1994))2. The pres-
sure rise delivered by the compressor is increased by a substantial
1 The upstream orifice mass flow is corrected based on total pressure
and temperature measured far upstream of the compressor; the corrected 2 As discussed in Section 5.3, preliminary results indicate that the com-
injected mass flow is based on total pressure and temperature measure- pressor can be successfully stabilized using only 1-sided actuation (only
ments in the injector. adding injector mass flow).

4
 2.0 20
(a) Low-Speed Compressor, Mtip = 0.25
10
0

Gain dB
1.95 [1,0]
-10
-20
1.9 -30
-40
Total-to-Total Pressure Ratio

-2 -1.5 -1.0 -0.5 0 0.5 1.0 1.5 2.0

20
1.85 (b) Stage 35, 70% Speed, Mtip = 1.0
10
[1,1]
0 [1,-1] [1,0]

Gain dB
1.8 -10
-20
-30
1.75 -40
-2 -1.5 -1.0 -0.5 0 0.5 1.0 1.5 2.0
20
(c) Stage 35, 100% Speed, Mtip = 1.5
1.7 10
0 [1,0] [1,1]
[1,2]

Gain dB
-10 [1,-1]
Sheet 0 yaw, 100% injection
1.65 Sheet -15 yaw, 100% injection -20
Sheet -30 yaw, 100% injection -30
3-Hole 0 yaw, 100% injection
-40
1.6 -2 -1.5 -1.0 -0.5 0 0.5 1.0 1.5 2.0
16 17 18 19 20 21 22
Frequency Normalized by Rotor Speed
Total Mass Flow, kg/s

Figure 3: Pressure ratio characteristics for steady blowing Figure 4: Transfer functions for low and high-speed
with four injector configurations at 100% speed. compressors: measured, model fit.

fraction. Figures 2b and 2c detail the inlet total-to-exit static pres- ible rotating stall dynamics.
sure ratio for various levels of steady blowing with the same injec- It should be noted that all of the root loci and transfer func-
tor configuration. tions presented in the following sections are for complex coeffi-
Figure 3 shows a summary of the injector configuration study; cient models. The output is a complex spatial Fourier coefficient
in this figure the total-to-total pressure ratio is plotted to high- (SFC) of the nth harmonic pressure perturbation; the input is the
light the differences for the four injector configurations. All corresponding nth SFC of the injection. Since the n 0 models
speedlines are for 100% steady blowing. The sheet injector at - are complex, the root loci and transfer functions are not symmet-
15 yaw yields a higher total pressure ratio than both the sheet at ric for positive and negative frequencies which determine the di-
-30 yaw and the three-hole injector. At 0 yaw, the sheet injector rection of wave rotation.
provides higher total pressure ratio, but the mass flow could not
be increased above 20 kg/s corrected flow. This was thus not 4.1 Comparision of Predicted and Measured Dynamics
considered an acceptable configuration. Detailed investigation Figure 4 compares the forced response behavior of the low-
of the effect of steady injection on the compressor performance is speed compressor used by Haynes (1993) and Vo (1997) (with a
currently being conducted as part of this research program. relative tip Mach number of 0.25) to that of the current compres-
sor at 70% and 100% design speed ( Mtip of 1.0 and 1.5 respec-
4 OPEN LOOP DYNAMICS tively). The latter responses were obtained by exciting the sys-
In this section we compare qualitative and quantitative pre- tem with a rotating sinusoidal wave of actuation, superimposed
dictions of the compressible rotating stall model to forced response on the nominal 3.6% injection rate. The frequency of rotation
results obtained with unsteady injection. Our purpose is to dem- was swept from 10 Hz to 500 Hz in both directions of rotation.
onstrate that compressibility has a significant effect on the dy- Standard spectral analysis procedures yielded the transfer func-
namic behavior of compressors, and that the theory correctly ac- tion estimates shown. The measured input/output coherence was
counts for the differences between incompressible and compress- typically above 0.8, resulting in accurate transfer function esti-

5
mates. Figures 4b and 4c also include a fit using eigenvalues and Table 1: Measured and Theoretical First Harmonic
zeros (indicated by a dashed line). Note that this is not the model Rotational Frequencies
described in Section 2: this is an identified model based on the
Mode Measured Frequency/ Theoretical Frequency/
experimental measurements. In both the low and high-speed com-
Rotor Frequency Rotor Frequency
pressor tests, the actuation was inlet injection, although the de-
tails of the injection and measured response differ somewhat. [1, 0] 0.36 0.49
However, the goal here is to compare eigenmodes, which are in- [1, 1] 0.89 0.80
dependent of the forcing details. [1, 2] 1.13 1.10
The transfer functions exhibit peaks at the natural [1, -1] -0.72 -0.59
eigenfrequencies of lightly damped modes. As predicted by the [1, -2] -0.88 -0.83
incompressible theory, Figure 4a shows that the low-speed com-
pressor exhibits only one such peak in the first harmonic response.
The compressible theory, on the other hand, predicts that mul- out the compressor. The equivalent loss and deviation for each
tiple eigenfrequencies exist for each spatial harmonic. Further- blade row are determined from the mean-line calculation. Using
more, while the incompressible mode frequency scales with the a constrained optimization algorithm, the values of loss and de-
rotor frequency, the frequencies of the remaining (compressible) viation were changed iteratively within experimental error until
modes are set by the geometry of the compression system rather the computed flow field matched the temperature and pressure
than the rotor speed. As the rotor frequency (tip Mach number) measurements for each blade row. The optimization was aug-
increases, the frequencies of the compressible modes decrease mented to match the measured dynamics (eigenvalue locations)
with respect to the rotor frequency, as shown by the second peak as well as the steady temperature and pressure.
in Figures 4b and 4c. At a Mtip of 1.0, the [1, 1] peak appears at Table 1 shows that the rotation frequency of the theoretical and
1.38 times the rotor frequency, or 280 Hz in dimensional units. measured eigenvalues are in general agreement. As might be ex-
For Mtip=1.5, this peak appears at 0.89 times the rotor frequency, pected from the fact that geometry is the primary influence on fre-
or 260 Hz. The dimensional frequency is thus relatively con- quency, the theoretical frequencies do not depend strongly on the
stant. The frequency of the [1, 0] mode scales primarily with aerodynamic and performance inputs to the model. Growth rate
rotor frequency for all three cases (0.32 for the low-speed com- comparisons are much more sensitive to these parameters, and are
pressor, 0.32 for the low-speed Stage 35 test, and 0.28 at the less accurate. For example, the growth rate of the [1,0] mode is
design speed). Note also that the Stage 35 forced response mea- underestimated by 27% at this operating point. This error in stabil-
surements include other lightly damped modes traveling with and ity translates into an error in predicted stalling mass flow of 7%.
against the rotor direction (peaks at positive and negative fre- Figure 5 combines the frequency information in Table 1 with
quencies) at up to 2, which is consistent with the theory. growth rate information, for both the measured and calculated
Figure 4 shows that detailed dynamic information can be eigenvalues at operating point D in Figure 2c. The frequencies of
obtained through spatially distributed forced response testing; the measured eigenvalues are captured well. However, there is a
transfer functions similar to those shown in Figure 4 have been
measured for the 0th, 1st, and 2nd circumferential harmonics. Us-
1.5
ing this data, at least 20 dynamic modes of the system have been
Rotation Frequency / Rotor Frequency

identified. Table 1 gives the identified frequencies of the lowest

5 first harmonic rotating stall modes at 100% speed, and com- 1.0 [1,2]
pares them to the theory described in Section 2. The comparison [1,1]
is shown for the operating point D in Figure 2c, where the most 0.5 [1,0]
coherent near-stall transfer function estimate was obtained.
Feulners model has been extensively refined in order to
model Stage 35 with steady 3.6% mean injection rate; this is the 0
first time that the compressible model has been directly compared Calculated Eigenvalues
Measured Eigenvalues
to forced response data (see Weigl (1997) and Frchette (1997) -0.5
[1,-1]
for details). The inputs to the dynamic model are the mean-line
flow quantities at 70% and 100% speed. Since standard correla-
-1.0
tions for total pressure loss and deviation do not apply to this
compressor due to the tip injection from the actuators, the steady
flow field could not be obtained from standard correlations. These -1.5
-0.4 -0.3 -0.2 -0.1 0 0.1
steady flow fields have therefore been derived from a mean-line Growth Rate / Rotor Frequency
data fit procedure based on total temperature, total pressure, and
static pressure measurements at various axial locations through- Figure 5: Calculated and measured eigenvalues at 100% speed.

6
significant difference in the growth rates which is primarily due (a) 70% Speed
to the sensitivity in stability to the steady model inputs. Further- 2.0

Rotation Frequency / Rotor Frequency

more, comparison of the measured and theoretical transfer func-
1.5
tion indicates that the model does not capture the zeroes accu- [1,1]
rately. We believe that the source of error in modeling the zeros 1.0
is that the two-dimensional model ignores the spanwise
nonuniformity of the tip injection. 0.5
Near -1 and 1.2, two additional eigenvalues (marked by [1,0]
B A
arrows in Figure 5) are predicted which were not identified from 0
the forced response tests. It is unclear if these eigenmodes are
-0.5
indeed erroneous. Detailed examination of the measured transfer
functions indicate that there are closely spaced eigenvalues and -1.0
zeros above 1 (see for example Figure 4c near 1.5) which are [1,-1]
not included in the experimental fit3. These eigenvalues and ze- -1.5
ros are so well damped and closely spaced that they contribute
only a small change in magnitude and phase to the transfer func- -2.0
-0.4 -0.3 -0.2 -0.1 0 0.1
tion. The magnitude of some of these eigenvalue/zero peaks is Growth Rate / Rotor Frequency
indistinguishable from the noise in the transfer function estimate.
The results shown in Table 1 and Figure 5 indicate that the (b) 100% Speed
2.0
main features of the measured modal dynamics, such as number,
frequency and, to a lesser degree, stability of the dominant eigen- Rotation Frequency / Rotor Frequency 1.5
values, are captured by the compressible stall inception theory. [1,2]
Comparable results were also found for the measured dynamics 1.0 [1,1]
at 70% of design speed. The theory is not yet an a priori tool for
0.5
stall prediction and control system design since the model inputs [1,0]
were derived from experimental measurements and the error in E D C
0
modal growth rate is significant.
-0.5
4.2 Critical Mode for Instability [1,-1]
The theory predicts that both the [1, 0] mode and the com- -1.0
pressible modes can become lightly damped as the compressor
-1.5
mass flow is reduced, and this has important implications for ac-
tive stabilization. To show the experimental trends of the eigen- -2.0
values, their frequencies and growth rates were found from the -0.4 -0.3 -0.2 -0.1 0 0.1
identified models at various mass flows: points A and B in Fig- Growth Rate / Rotor Frequency
ures 2b and points C, D, and E in Figure 2c. Figure 6 shows the Figure 6: Migration of eigenvalues with decreasing mass
measured eigenvalue migrations as stall is approached. flow (* indicates eigenvalues at lowest measured mass flow).
As the throttle is closed at design speed, several modes mi-
grate toward instability. The large change in stability of the [1, 0]
mode with mass flow indicates that it is the critical mode for stall [1, 1] and other compressible modes are more stable, and do not
inception (this is indeed the case as discussed in Section 5.3). move much as the compressor mass flow is decreased. The [1, 0]
The [1, 1] mode is also of concern since it is lightly damped and mode determines the stability of the first harmonic at this speed.
has a frequency close to 1E, and can be driven unstable by an 1E For this particular compressor, the [1, 1] and other compress-
excitation (Tryfonidis et al., 1995). Figure 6b indicates that there ible modes do not become unstable at the mass flows we have
are several lightly damped modes up to 1.5E. Since this is the reached with and without control, unlike some of the cases pre-
bandwidth limit of the actuator, there is more uncertainty in the sented by Tryfonidis et al. (1995). However, at 100% speed, the
exact location of these high frequency modes. At 70% speed, the lightly damped compressible modes play an important role in com-
pressor stabilization as discussed in the following section.
3 The closely spaced poles and zeros do not play an important role for
5 ACTIVE CONTROL
control (since these modes are effectively uncontrollable). They have
therefore been omitted to simplify the experimental fit procedure by fit- Discussion of the active stabilization results will be presented
ting a lower order model. in several parts to illustrate how the alteration in dynamics as

7
rotor speed is increased affects the control problem. We first 5.1 Rotating Stall Stabilization at 70% Speed
present the results of constant-gain feedback at 70% speed with a Figure 7 shows the 70% speed total-to-static pressure ratio
tip Mach number of 1.0. We also present the results of reducing characteristic. The solid line corresponds to steady injection at
the supply pressure to the injectors to demonstrate that the con- 50% of full blowing, while the dashed line shows the baseline
trol power requirements for stabilization are not prohibitive. We characteristic without injectors (solid casing). Steady blowing
then describe results at 100% speed where a simple constant-gain without feedback extends the stable operating range of the com-
control law is not sufficient to stabilize the various modes that pressor by 17.4%. A simple constant gain feedback control, like
can lead to stall and surge. We thus introduce the design of a those used by Paduano (1993) and Haynes (1993), was imple-
model-based H control law which successfully stabilized the mented on the measured first harmonic. This feedback law has
compressor. Results are then given showing the range extension the following form:
achieved and details of the controlled behavior. Finally, we illus- u1 = Zp1 ,
trate the robustness of the controlled compressor by demonstrat-
ing recovery from a growing rotating stall disturbance and from where p1 is the first spatial Fourier coefficient (SFC) of the static
fully developed surge. Only the main results are presented here; pressure perturbations measured at the sensor location 44% of
a complete overview of the control experiments including con- chord upstream of the rotor as shown in Figure 1, and u1 is the
trol design trade-offs and limitations, and detailed pre-stall tran- corresponding first SFC of the injection wave, which is intro-
sient and spectral analysis can be found in Weigl (1997). duced as a first harmonic sinusoidal wave. Z, the proportionality
In Sections 5.1 and 5.3 we present the percentage decrease constant between the measured and actuated wave, is a complex
in stalling mass flow referenced to the stalling mass flows at 70% number. The amplitude of Z determines the feedback gain of the
and 100% speed with a solid casing (no injectors installed). Table control law, and the phase of Z determines the spatial phase shift
2 in Section 6 summarizes these results and also lists the improve- (in radians) between the measured and actuated wave. Z was
ment in stall margin for each of the following cases. experimentally optimized to the value Z = 2ej*2.36 for the results
in Figure 7. Constant-gain feedback control of the first mode
1.3 increased the stable operating range of the compressor an addi-
tional 8.3% above steady blowing, for a total range increase of
25.7%. The compressor characteristic was extended smoothly
without a drop in pressure rise.
}

1.2 The control activity required to obtain the results in Figure 7

Range
Extension with a 100 psi injector supply pressure was approximately 5%
With Control
of the available 50% valve motion. The small valve motion
motivated an experiment in which the supply pressure to the valves
1.1
was reduced to 30 psi and the mean blowing rate was 1.5% of the
Total-to-Static Pressure Ratio

design compressor mass flow. This pressure was the minimum

value that could be used effectively in the valve, and represented
1.0 roughly the supply pressure available from downstream of the
compressor at design speed. Figure 8 shows the constant gain
feedback stabilization results for this experiment. Reducing the
steady injection rate changes the compressor characteristic dra-
0.9
matically which in turn alters the stall inception dynamics. The
constant gain control law was therefore retuned experimentally
to provide the range extension shown.
0.8 The compressor characteristic with mean blowing is shifted
only slightly from the characteristic with no blowing. Steady
blowing extends the operating range by about 2.8%. Feeding
50% Steady Blowing, No Control back the first Fourier coefficient extends the stable operating range
0.7
1st Harmonic Control by an additional 7%, and first and second SFC feedback extends
Solid Casing, No Injectors the range 4% more, for a total of 13.8% range extension. In this
case, the feedback gains for the first and second harmonics are
0.6 Z1 = 4ej*2.36 and Z2 = 4ej*4.10.
8 9 10 11 12 13 14 15 16 17 These results show that constant gain control can achieve a
Total Mass Flow, kg/s range improvement for this high-speed compressor at 70% speed.
Figure 7: Stabilized compressor characteristic at 70% The results are similar to those obtained by Paduano et al. (1993)
speed with 3.6% injected mass flow rate. and Haynes et al. (1993) on much lower speed compressors in

8
 1.3
pressor at 100% speed. This section briefly describes some of
the challenges faced and highlights the use of a new design tech-
nique to address eigenvalue uncertainty. Details of the control
1.2 law are left for a future article (see also Weigl (1997)). The reader
can refer to Kwakernaak (1993) for a tutorial on H control law
}
design. The theoretical development of the new eigenvalue per-
Range
Extension turbation procedure is presented in Smith (1995); Smith (1994)
1.1 With Control describes the application of the technique to control of a flexible
Total-to-Static Pressure Ratio

structure. Although the zeroth, first, and second harmonics have

been stabilized, we will again focus our discussion on the first
1.0 harmonic. The design issues and procedure carry over directly to
the zeroth and second harmonics. A separate H control law was
implemented for each of these decoupled harmonics.
The reason that constant gain control cannot stabilize the first
0.9 harmonic is straightforward: the particular gain and phase that
damp the [1, 0] mode destabilize the higher frequency modes. To
damp all of the dominant modes simultaneously, a dynamic con-
0.8 troller tuned to provide different gain and phase at various fre-
quencies is required. Such a dynamic control law can only be
designed based on a model of the dynamics to be controlled (the
50% Steady Blowing, No Control two parameters of the constant gain control law were simply tuned
0.7 1st Harmonic Control experimentally). Unfortunately, we do not have an accurate model
1st & 2nd Harmonic Control of the unstable stall inception dynamics for two reasons. First,
Solid Casing, No Injectors
the theoretical model discussed in Section 2 is not yet accurate
0.6 enough for control design. Second, it is difficult to experimen-
8 9 10 11 12 13 14 15 16 17 tally measure the unstable dynamics even when the compressor
Total Mass Flow, kg/s has been effectively stabilized. We therefore estimated the un-
Figure 8: Stabilized compressor characteristic at 70% stable dynamics from the identified stable dynamics by extrapo-
speed with 1.5% injected mass flow rate. lating the stable eigenvalue (and zero) locations shown in Figure
6b to a lower mass flow for which the [1, 0] mode is unstable.
that stabilizing the incompressible-type modes increases the op- This method of estimating the unstable dynamics introduces
erating range. At this rotor speed the compressible modes, such uncertainty in the control design model. However, the main at-
as the [1, 1] and [2, 1] mode, are sufficiently stable that stabiliz- tribute of many modern control design algorithms, such as H
ing only the [1, 0] and [2, 0] mode with constant gain feedback control, is robustness to errors in the control design model: to
provides significant range extension. effectively meet the control specifications despite modeling er-
Study of the pre-stall waves at the limit of stabilization indi- rors. In this case, the goal is to lightly damp all of the dominant
cates that the compressible [2, 1] mode becomes lightly damped first harmonic modes even though their precise stability and rota-
with constant gain control, and that additional compensation for tion frequency are not known. To directly address uncertainty in
the [0, 0] and [1, 0] modes may further increase the stable range the eigenvalue locations, a new technique developed by Smith
at 70% rotor speed. Rather than refining this control design, how- (1994, 1995) was used within the H control design framework.
ever, our investigation turned to control at 100% design speed. This procedure designs to a range of eigenvalue locations as il-
At this speed, no range extension was achieved with constant gain lustrated in Figure 9. The estimated eigenvalue locations are shown
feedback on the zeroth and first spatial Fourier coefficients. The by the xs; note that only the [1, 0] mode is assumed to be un-
reason is that both the [1, 0] and the [1, 1] mode are lightly damped stable based on the identified modes in Figure 6b. The circles
at the stall point and a constant-gain controller cannot stabilize represent uncertainty in the eigenvalue locations; the control law
both modes simultaneously. A dynamic controller, which can in- will stabilize the compressor as long as the actual first harmonic
troduce the proper gain and phase at different frequencies, is thus eigenvalues lie within these uncertainty areas. [Note that only
required. circular uncertainty regions can be implemented though the scales
in Figure 9 makes them appear to be ellipses.] The large [1, 0]
5.2 H Control Law Design with Eigenvalue Perturbations mode disk ensures that the controller will not destabilize the sys-
Using the identified models presented in Section 4, a robust, tem at higher mass flows when the compressor is open-loop stable,
linear, dynamical control law was designed to stabilize the com- yet will stabilize the [1, 0] mode at lower mass flows where the

9
 2.0 1.6
H G
F
Rotation Frequency / Rotor Frequency

}
1.5 Range
1.0 [1,1] Extension
With Control
[1,0]

0 1.4

Total-to-Static Pressure Ratio

[1,-1]
-1.0 1.3

-2.0 1.2
-0.4 -0.3 -0.2 -0.1 0 0.1
Growth Rate / Rotor Frequency

Figure 9: Eigenvalue perturbations for first harmonic 1.1

control design.
50% Steady Blowing, No Control
F No Control & 1st Harmonic Control
dynamics have been estimated. The smaller circles around the G 0th Harmonic Control
lightly damped high frequency modes account for changes in these 1 0th & 1st Harmonic Control
eigenvalue locations as the compressor mass flow changes, al- 0th, 1st & 2nd Harmonic Control
Solid Casing, No Injectors
though these perturbations do not allow for unstable high fre-
quency modes. 0.9
This H control design has been used to stabilize the first 16 17 18 19 20 21 22
three harmonic perturbations at 100% speed. The following sec- Total Mass Flow, kg/s
tion presents the results for the nominal control configuration.
Figure 10: Stabilized compressor characteristic at 100%
We have found that the control design is also robust to moderate
speed with 3.6% injected mass flow rate.
changes in the steady state operating point of the compressor.
This was verified by testing the control law with less injected air
and at a 95% speed as discussed in Weigl (1997). The controlled sure traces at the different stall inception points shown in Figure
compressor has also proven to be robust to large amplitude dis- 10: a) F: no control, b) F: 1st harmonic control only, c) G: 0th
turbances as shown below. harmonic control only, and d) H: 0th and 1st harmonic control
combined. Note that the mass flow is different in each case, cor-
5.3 Rotating Stall and Surge Stabilization at 100% Speed responding to the mass flow at which the most unstable uncon-
The range extension achieved at 100% speed is shown in trolled harmonic goes unstable. The pressure perturbations have
Figure 10. As in Figure 7, the total-to-static pressure ratio map is been non-dimensionalized by a measure of the mean dynamic
presented for the nominal 3.6% injection mass flow rate along head in the inlet duct (Ptinlet - Psinlet). In all four cases, large scale
with the original speedline. This figure also shows the results of instability is initiated by approximately 10 rotor revolutions of
4 different control runs, which demonstrate the importance of rotating stall followed by classic surge cycles. During the large
controlling the 0th and 1st harmonic simultaneously. Controlling amplitude rotating stall, the peak-to-peak amplitude of the non-
either the 0th harmonic or 1st harmonic alone results in almost no dimensional pressure perturbations is approximately 4.
further range extension. With 50% steady injection, the stalling Figure 11a shows that both 0th and 1st harmonic pertubations
mass flow is reduced by 4.3%. With both 0th and 1st harmonic exist in uncontrolled pre-stall traces indicating lightly damped
control, about 3.0% range extension is achieved; an additional 0th and 1st harmonic modes. In Figure 11b, the 1st harmonic pre-
0.5% is gained by stabilizing the 2nd harmonic. stall perturbations have been effectively damped, leaving 0th har-
To illustrate that instability in any of these harmonics (0th, monic perturbations which eventually initiate rotating stall no
1 , or 2nd) can lead to stall, stall inception transients were re-
st operating range extension is achieved. In Figure 11c, a small
corded with various combinations of harmonics stabilized. The range extension is achieved by controlling only 0th harmonic per-
harmonic left uncontrolled is always the one responsible for stall turbations; here pre-stall perturbations contain only the 1st har-
inception. Figure 11 shows the evolution of three wall static pres- monic. As discussed by Hendricks et al. (1996), both the 0th and

10

a) No Control, m
stall = 17.92 b) 1st Control, m stall = 17.92

360 360

0 0
}

n =0 n =1 n =0

-20 0 20 -20 0 20
Rotor Revs Rotor Revs

c) 0th Control, m
stall = 17.85 d) 0th & 1st Control, m stall = 17.36

4
360 360

0 0
}
}

n =1 n=2

-20 0 20 -20 0 20
Rotor Revs Rotor Revs
Figure 11: Wall static pressure perturbations non-dimensionalized by (Ptinlet - Psinlet) at 100% speed.

1st harmonics need to be damped to extend the transonic com- Roughly the same additional range extension (4.1%) was achieved
pressor operating range. Figure 11d shows stall inception with with one-sided actuation as with two-sided, suggesting that it may
0th plus 1st harmonics control; a 2nd harmonic perturbation is now be possible to stabilize the compressor without adding a large
evident prior to stall. Stabilizing the 2nd harmonic provides a steady amount of injected mass flow. Such a test, allowing the
little more range extension. Spectral analysis of the stall incep- actuators to open from 0-50% unsteadily, has not yet been at-
tion for these 4 cases (as in Tryfonidis (1995)) indicates that the tempted on Stage 35.
low frequency incompressible, uncontrolled modes ([0, 0], [1, 0], We have discussed the robustness properties of the control
or [2, 0]) become unstable in this compressor. For the control law with respect to design model inaccuracies. Another form of
laws that have been tested, some resonating high frequency modes robustness that has received much attention is disturbance rejec-
are also evident but do not appear to be the cause of instability. tion (Eveker et al., 1995). Fundamentally, disturbances drive the
Control tests were also conducted with 1-sided actuation system away from its equilibrium, requiring the controller to ex-
where the valves were thresholded to only move from 50-100% ert extra effort to recover stable operation. To test robustness, the
open by clipping the negative portion of the sinusoidal command. mass flow was reduced below the open-loop neutral stability point
With this 1-sided unsteady valve motion, the time-averaged in- with the control activated. Control was then switched off for some
jected mass flow rate did increase from 3.6% to 4.0% so that the time. During the period with no control, the unstable perturba-
controlled speedline shifted to a slightly higher pressure rise. tions grew to large amplitude. When the controller was switched

11
 a) Rotating Disturbance b) Fully-Developed Surge
2 2
Pressure Perturbation

Pressure Perturbation
1 1
Non-Dimensional

Non-Dimensional
0 0

Control
-1 Off -1
Control
On
Open Open
100% 100%
Actuator Position

Actuator Position
0% 0%
Closed Closed
Control Control
Off On
0 50 100 0 50 100
Rotor Revs Rotor Revs

c) Rotating Disturbance d) Fully-Developed Surge

0.5 0.5

0.4 0.4 Time

0.3 0.3
~
p1 Control On ~
p1 Control
On
Time
0.2 0.2 Equilibrium
State

0.1 0.1

Equilibrium State
0 0
0 0.1 0.2 0.3 0.4 0.5 0 0.1 0.2 0.3 0.4 0.5
~
p0 ~
p0

Figure 12: Recovery from large amplitude disturbance (pressure perturbations are
non-dimensionalized by (Ptinlet - Psinlet)).

back on, it was required to re-stabilize with large initial condi- the measured valve position from one actuator. The accompany-
tions, thereby approximating the effect of a large amplitude dis- ing Lissajous figures, 12c and 12d, plot 0th against 1st harmonic
turbance. amplitude, to show the trajectory of the system state back toward
Figure 12 shows the results of the disturbance rejection tests. equilibrium. The units of pressure are (Ptinlet - Psinlet) so that the
In Figure 12a, rotating stall perturbations are recovered, and in perturbations can be seen to be a substantial fraction of the mean
Figure 12b, a fully developed surge is recovered. The top trace is inlet dynamic pressure. The first point on these plots is the ampli-
the wall static pressure from one sensor and the bottom trace is tude of the perturbations at the instant of control reinitiation; this

12
is the relevant measure of disturbance rejection. In the first case, stall margin as we are in comparing the changes in stall margin
a primarily 1st harmonic initial condition is stabilized; in the sec- due to steady and controlled injection.
ond the 0th harmonic is also large. Injection can influence the overall behavior of the compressor
in two ways. The first is simply the additional momentum added
6 DISCUSSION AND CONCLUSIONS by the jets which increases the total pressure at the rotor inlet.
This machine is the first high-speed compressor with The second effect is a change in compressor performance which
circumferentially distributed actuation of sufficient authority and increases the pressure rise across the stage. In order to separate
bandwidth to excite and control the modal system dynamics. We the two effects, we have also calculated the stall margin based on
have described for the first time the impact of unsteady blowing the total pressure ratio from the rotor inlet to the compressor exit.
on compressor performance. The flexibility to excite distributed To estimate the total pressure at the rotor inlet (with the added jet
modal dynamics enabled the first detailed look at rotating stall momentum), the inlet total pressure and injector total pressure
fluid mechanics in a high-speed environment. We have demon- were mass-averaged. This conservative estimate assumes that
strated active stabilization of rotating stall and surge in a tran- the injected fluid does not mix with the primary inlet flow. Re-
sonic compressor and achieved a significant extension of the op- calculating the stall margin in this manner indicates that the addi-
erating range. Finally, these results demonstrate the effective- tional momentum contributes approximately 2% of the stall mar-
ness of inlet injection for forcing and controlling high-speed com- gin improvements shown in Table 2 while the change in com-
pressor dynamics. pressor performance is only about 0.5%.
Table 2 summarizes the operating range improvement for the Perhaps the most promising aspect of these experiments is
cases we have presented in terms of the change in stalling mass that the pre-stall dynamics behaved qualitatively as expected from
flow referenced to the stall points without injectors. We also tabu- hydrodynamic stability theory, adapted for active control work.
late the increase in the NASA standard stall margin as defined by Although gaps still exist in ab initio predictive capability, the
Reid and Moore (1978b): theoretical framework for modeling and control captures the
physical processes, and provides guidance in selecting actuator
Total Pressure Ratio (stall) Mass Flow (ref) and sensor type, number and placement, bandwidth requirements,
SM = 1 100
Total Pressure Ratio (ref) Mass Flow (stall) and control law design procedures. Neither the predictive capa-
bility nor the accuracy of the model is currently sufficient for
For 70% and 100% speeds, the peak efficiency points with a direct control law design.
steady 3.6% injection rate (the nominal configuration for con- Using distributed inlet injection actuators allows us to vali-
trol) have been chosen as the reference points. These reference date the modeling concepts and proposed physical mechanisms
operating points are labeled as REF70 and REF100 in Figures 2b for stall initiation when compressibility is a factor. In particular,
and 2c. Although this choice of reference points is somewhat ar- we have directly verified that eigenmodes which rotate at a fre-
bitrary, we are not so much interested in the absolute value of quency near the rotor frequency exist and become lightly damped

Table 2: Summary of Steady Blowing and Control Experiments

Mass Flow % Decrease Pressure Ratio Stall

Configuration at Stall Mass Flow at Stall at Stall Margin %

70% Speed
Solid casing 11.86 1.3450 14.2
1.5% blowing, no control 11.53 2.8 1.3370 16.8
3.6% blowing, no control 9.80 17.4 1.3370 37.4
5.8% blowing, no control 8.83 25.5 1.3640 55.7
1.5% blowing, 1st, 2nd control* 10.22 13.8 (+11.0) 1.2940 27.5 (+10.7)
3.6% blowing, 1st harmonic control* 8.81 25.7 (+8.3) 1.3370 51.1 (+13.7)

100% Speed
Solid casing 18.73 1.9010 12.2
3.6% blowing, no control 17.92 4.3 1.8850 17.9
5.8% blowing, no control 16.86 10.0 1.8910 24.0
3.6% blowing, 0th, 1st, 2nd control* 17.27 7.8 (+3.5) 1.8650 19.4 (+1.5)
3.6% blowing, 1-sided actuation* 17.16 8.4 (+4.1) 1.8670 20.3 (+2.4)
* The values in parentheses indicate the portion of the total stall margin improvement due to active control.

13
near the peak of the pressure-rise map. The existence of these sor, increasing its pressure rise and range of operation as previ-
modes was strongly supported by the unforced data and calcula- ously reported by Koch and Smith (1968). Further work is needed
tions of Tryfonidis et al. (1995), but room for conjecture remained to investigate the trade-offs between steady injection and active
because these modes are close to the rotor frequency. With forced control. Steady blowing requires a large amount of injected air.
response experiments, we are able to make the following addi- Active stabilization requires a feedback control law with high-
tional statements about the linearized dynamic model of Bonnaure bandwidth sensing and actuation. One must also study the per-
(1991), Hendricks (1993), and Feulner (1994): formance impact of both stabilization schemes before consider-
The primary physical mechanisms associated with pre-stall ing any implementation in an engine. In addition, work directed
dynamics in high-speed compressors are captured by the com- towards developing robust (durable) actuators for engine envi-
pressible model. This is verified by eigenvalue analysis of ronments is a necessity if active control is to become a viable
data. technology.
Control design based on the understanding afforded by the We also note that forced response testing with high-author-
theory is effective for stabilization of rotating stall. The theo- ity, high-bandwidth, distributed actuation provides valuable in-
retical model is validated from a conceptual standpoint although formation that cannot be obtained in other ways. By applying
more work is needed for it to be usable as a stall prediction and system identification methods to input-output data, we are able to
control design tool. identify the modal content of the wave dynamics of a high-speed
A useful way to summarize the results is to compare and compressor. Prior to these studies, compressor wave dynamics
contrast stabilization of this compressor to stabilization of low- were thickly veiled by noise, turbulence, and secondary sources
speed compressors by similar means, reported in Paduano et al. of forcing (rotor-stator interaction noise, excitation at multiples
(1993), Haynes et al. (1993), and Gysling and Greitzer (1994). of the rotor frequency, wake noise, etc.). Regardless of the impli-
Control results are the same in the following ways: cations for control of rotating stall, the value of this type of ex-
Significant range extension can be obtained through stabiliza- perimental setup for the investigation of the compressor dynam-
tion of small amplitude, pre-stall waves. ics which lead to rotating stall is apparent.
The incompressible modes observed in low-speed compres-
sors are the first to become unstable in this high-speed com- 7 ACKNOWLEDGMENTS
pressor although this may not be the case in general as shown This project is a collaboration between NASA Lewis and MIT.
by Tryfonidis (1995). The authors would like to thank D. Williams, B. Piendl, M. Goin,
With a circumferentially uniform background flow, the dynam- D. Van Zante, and R. Bruckner for their help during compressor
ics of the circumferential harmonics are decoupled and can be testing, as well as V. Dubrowski, J. Letendre, and M. Hellwig for
treated as separate entities for the purpose of control. their assistance developing the experimental hardware. We would
As more eigenmodes of the system are stabilized, more range also like to thank H. D. Vo for providing low-speed compressor
extension is obtained. Instability typically results when an forced response data. Special thanks also to D. Park for prepar-
eigenmode which is not being stabilized goes unstable. ing this manuscript. This work was conducted under NASA grant
Active stabilization extends the compressor characteristic NAG3-1457.
smoothly, without a sharp drop in pressure rise.
The differences between high and low-speed applications are: REFERENCES
Although circumferential harmonics are decoupled, multiple Behnken, R. L., DAndrea, R., Murray, R. M., 1995, Con-
eigenmodes exist for each circumferential mode due to com- trol of rotating stall in a low-speed axial flow compressor using
pressibility effects. Understanding the compressible modes pulsed air injection: modeling, simulations, and experimental vali-
is critical to designing effective control laws at high speed. dation, 34th IEEE Conference on Decision and Control, New
Because of the multi-modal nature of the dynamics, simple Orleans, December 13-15.
proportional gain feedback is not always sufficient to stabilize Berndt, R. G., 1995, Actuation for Rotating Stall Control of
the dynamics associated with a given circumferential harmonic. High Speed Axial Compressors, M. S. Thesis, MIT Department
In this study, a dynamic control law using eigenvalue pertur- of Aeronautics and Astronautics.
bations achieved good results. Berndt, R.G., Weigl, H.J., Paduano, J.D., and Epstein, A.H.,
Finally, several conclusions were reached concerning the ac- 1995, Experimental Techniques for Actuation, Sensing, and
tuation mechanism itself. This experiment constitutes the first Measurement of Rotating Stall Dynamics in High-Speed Com-
demonstration that unsteady injection provides effective and reli- pressors, SPIE International Symposium on Aerospace/Defense
able actuation in a high-speed compressor (previous injection stud- Sensing and Dual-Use Photonics, April 17-18, Orlando, FL.
ies such as Day (1993), Behnken et al. (1995), and Vo (1997) Bonnaure, L.P., 1991, Modeling High Speed Multistage
have been done on low-speed compressors). Injecting high-mo- Compressor Stability, M.S. Thesis, MIT Department of Aero-
mentum flow near the tip region of the rotor was found to have a nautics and Astronautics.
significant effect on the steady-state performance of the compres- Day, I. J., 1993, Active Suppression of Rotating Stall and

14
Surge in Axial Compressors, Journal of Turbomachinery, Vol. Reid, L., and Moore, R. D., 1978a, Design and Overall Per-
115, pp. 40-47. formance of Four Highly Loaded, High Speed Inlet Stages for an
Eveker, K. M., Gysling, D. L., Nett, C. N., and Sharma, O. Advanced High Pressure Ratio Core Compressor, NASA TP-
P., 1995, Integrated Control of Rotating Stall and Surge in a Low- 1337.
Speed Axial Compression System, SPIE International Sympo- Reid, L., and Moore, R. D., 1978b, Performance of Single-
sium on Aerospace/Defense Sensing and Dual-Use Photonics, Stage Axial-Flow Transonic Compressor With Rotor and Stator
April 17-18, Orlando, FL. Aspect Ratios of 1.19 and 1.26, Respectively, and With Design
Epstein, A. H., Ffowcs-Williams, J. E., and Greitzer, E. M., Pressure Ratio of 1.82, NASA TP-1338.
1989, Active Suppression of Aerodynamic Instabilities in Tur- Setiawan, J. D., 1996, Distortion Tolerance of Stall/Surge
bomachines, Journal of Propulsion, Vol. 5, No. 2, pp. 204-211. Controllers in Axial Compression Systems, M. S. Thesis, MIT
Frchette, L. G., 1997, Implications of Stability Modeling Department of Aeronautics and Astronautics.
for High-Speed Axial Compressor Design, M. S. Thesis, MIT Smith, R. S., Chu, C., Fanson, J.L., 1994, The Design of
Department of Aeronautics and Astronautics. Controllers for an Experimental Non-collocated Flexible Struc-
Feulner, M. R., Hendricks, G. J., and Paduano, J. D., 1994, ture Problem, IEEE Trans. on Control Systems Technology, Vol.
Modeling for Control of Rotating Stall in High Speed Multi- 2, No. 2, pp. 101-109.
Stage Axial Compressors, ASME Paper No. 94-GT-200. Smith, R. S., 1995, Technical Notes and Correspondence:
Gysling, D. L., and Greitzer, E. M., 1994, Dynamic Con- Eigenvalue Perturbation Models for Robust Control, IEEE Trans.
trol of Rotating Stall in Axial Flow Compressors Using on Automatic Control, Vol. 40, No. 6, pp. 1063-1066.
Aeromechanical Feedback, ASME Paper No. 94-GT-292. Tryfonidis, M., Etchevers, O., Paduano, J.D., Epstein, A.H.,
Haynes, J. M., Hendricks, G. J., and Epstein, A. H., 1993, and Hendricks, G.J., 1995, Pre-Stall Behavior of Several High-
Active Stabilization of Rotating Stall in a Three-Stage Axial Speed Compressors, Journal of Turbomachinery, Vol. 117, pp.
Compressor, ASME Paper No. 93-GT-346. 62-80.
Hendricks, G. J., and Gysling, D. L., 1994, A Theoretical Vo, H.D., 1997, "Active Control of Rotating Stall in a Three-
Study of Sensor-Actuator Schemes for Rotating Stall Control, Stage Axial Compressor with Jet Actuators," M.S. Thesis, MIT
Journal of Propulsion and Power, Vol. 10, No. 1. Department of Aeronautics and Astronautics.
Hendricks, G. J., Bonnaure, L. P., Longley, J. P., Greitzer E. Weigl, H. J., 1997, Active Stabilization of Rotating Stall
M., Epstein A. H., 1993, Analysis of Rotating Stall Onset In and Surge in a Transonic Single Stage Axial Compressor, Ph. D.
High-Speed Axial Flow Compressors, AIAA Paper No. 93-2233. Thesis, MIT Department of Aeronautics and Astronautics.
Hendricks, G. J., Sabnis, J. S., and Feulner, M. R., 1996,
Analysis of Instability Inception in High-Speed Multi-Stage
Axial-Flow Compressors, ASME Paper No. 96-GT-360.
Koch, C.C., and Smith, L.H., 1968, "Experimental Evalua-
tion of Outer Case Blowing or Bleeding of Single Stage Axial
Flow Compressor. Part III - Performance of Blowing Insert Con-
figuration No. 1," NASA CR-54589, GE R68AEG318.
Koch, C.C., 1970, "Experimental Evaluation of Outer Case
Blowing or Bleeding of Single Stage Axial Flow Compressor.
Part IV - Final Report," NASA CR-54592, GE R69AEG256.
Kwakernaak, H., 1993, Robust Control and Hinf-Optimi-
zation - Tutorial Paper, Automatica, Vol. 29, No. 2, pp. 255-273.
Longley, J. P., 1993, A Review of Non-Steady Flow Mod-
els for Compressor Stability, ASME Paper No. 93-GT-17.
Moore, F. K. and Greitzer, E. M., 1986, A Theory of Post-
Stall Transients in Axial Compression Systems, Part I - Develop-
ment of Equations, and Part II - Application, ASME Journal of
Engineering for Gas Turbines and Power, Vol. 108, pp. 68-97.
Paduano, J. D., Valavani, L., Epstein, A. H., and Greitzer, E.
M., 1994, Modeling for Control of Rotating Stall, Automatica,
Vol. 30, No. 9, September.
Paduano, J. D., Epstein, A. H., Valavani, L., Longley, J. P.,
Greitzer, E. M., and Guenette, G. R., 1993, Active Control of
Rotating Stall in a Low-Speed Axial Compressor Journal of Tur-
bomachinery, Vol. 115, pp. 48-56.

15