Centrifugal Compressors

Modeling and Control of Surge and
Rotating Stall in Compressors

Dr.ing. thesis
Jan Tommy Gravdahl
Report 98-6-W
Department of Engineering Cybernetics
Norwegian University of Science and Technology
N-7034 Trondheim, Norway
1998
Abstract
This thesis contains new results in the eld of modeling and control of rotating
stall and surge in compressors.
A close coupled valve is included in the Moore-Greitzer compression system model
and controllers for both surge and rotating stall is derived using backstepping.
Disturbances, constant and time varying, are then taken into account, and non-
linear controllers are derived. Stability results are given. Then, passivity is used
to derive a simple surge control law for the close coupled valve. This proportional
control law is shown to stabilize the system even in the presence of time varying
disturbances in mass ow and pressure.
A novel model for an axial compression system with non-constant compressor
speed is derived by extending the Moore-Greitzer model. Rotating stall and surge
is studied in connection with acceleration of the compressor.
Finally, a model for a centrifugal compression system with time varying compres-
sor speed is derived. The variable speed compressor characteristic is derived based
on energy losses in the compressor components. Active control of surge in con-
nection with varying speed is studied. Semi-global exponential stability of the
compression system with both surge and speed control is proven.
The main results of this thesis have been presented at international conferences.
Parts of the thesis has also been submitted for publication in international journals.
ii Abstract
Preface and
Acknowledgments
This thesis is submitted in partial fulllment of the requirements for the degree of
doktor ingenir at the Norwegian University of Science and Technology (NTNU).
The thesis is based on research done at the Department of Engineering Cybernet-
ics during the period January 1995 through January 1998. It has been nanced
by a scholarship from the Research Council of Norway, under grant 107467/410.
I am grateful to my supervisor, Professor Dr.ing. Olav Egeland, for introducing
me to the eld of compressor control and control of mechanical systems in general.
He has been a source of valuable advice, comments and inspiration throughout the
work on this thesis and the papers on which it is based.
The sta at the department and my colleagues are acknowledged for creating a
very pleasant environment in which to do research.
I would like to thank Dr.ing. Trygve Lauvdal with whom I shared an oce for the
past three years. Our oce has been a very pleasant place to work, and we have
had numerous useful and not-so-useful discussions. Trygve is also acknowledged
for his proofreading of this thesis.
The help of Dr.ing. Erling Aarsand Johannesen, in the form of many constructive
comments and proofreading, is greatfully acknowledged.
Finally, I would like to express my love and deepest gratitude to the three girls
in my life: To my wife Gril for her love, continued encouragement and support
during the work on this thesis, and to my daughters Irja and Mina for their un-
conditional love.
Jan Tommy Gravdahl

March 1998, Trondheim, Norway
iv Preface and Acknowledgments
Contents
Abstract i
Preface and Acknowledgments iii
List of Figures x
1 Introduction 1
1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.3 Stability of Compression Systems . . . . . . . . . . . . . . . . . . . 2
1.3.1 Surge . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.3.2 Rotating Stall . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.3.3 Stability Analysis . . . . . . . . . . . . . . . . . . . . . . . . 5
1.4 Modeling of Compression Systems . . . . . . . . . . . . . . . . . . 6
1.5 Control of Surge and Rotating Stall . . . . . . . . . . . . . . . . . 7
1.5.1 Surge/Stall Avoidance . . . . . . . . . . . . . . . . . . . . . 7
1.5.2 Active Surge/Stall Control . . . . . . . . . . . . . . . . . . 7
1.6 Contributions of this Thesis . . . . . . . . . . . . . . . . . . . . . . 8
1.7 Outline of the Thesis . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2 Close Coupled Valve Control of Surge and Rotating Stall for the
Moore-Greitzer Model 11
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.2 Preliminaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.2.1 The Model of Moore and Greitzer . . . . . . . . . . . . . . 13
2.2.2 Close Coupled Valve . . . . . . . . . . . . . . . . . . . . . . 16
2.2.3 Equilibria . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.2.4 Change of Variables . . . . . . . . . . . . . . . . . . . . . . 19
2.2.5 Disturbances . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.3 Surge Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.3.1 Undisturbed Case . . . . . . . . . . . . . . . . . . . . . . . 24
2.3.2 Determination of 0 . . . . . . . . . . . . . . . . . . . . . . 26
2.3.3 Time Varying Disturbances . . . . . . . . . . . . . . . . . . 29
2.3.4 Adaption of Constant Disturbances . . . . . . . . . . . . . . 34
2.4 Control of Rotating Stall . . . . . . . . . . . . . . . . . . . . . . . . 38
2.4.1 Undisturbed Case . . . . . . . . . . . . . . . . . . . . . . . 39
vi CONTENTS
2.4.2 Disturbed Case . . . . . . . . . . . . . . . . . . . . . . . . . 44
2.5 Simulations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
2.5.1 Surge Control . . . . . . . . . . . . . . . . . . . . . . . . . . 46
2.5.2 Rotating Stall Control . . . . . . . . . . . . . . . . . . . . . 48
2.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
3 Passivity Based Surge Control 51
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
3.1.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
3.1.2 Notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
3.2 Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
3.3 Passivity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
3.3.1 Passivity of Flow Dynamics . . . . . . . . . . . . . . . . . . 53
3.3.2 Passivity of Pressure Dynamics . . . . . . . . . . . . . . . . 53
3.3.3 Control Law . . . . . . . . . . . . . . . . . . . . . . . . . . 54
3.4 Disturbances . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
3.5 Simulations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
3.6 Concluding Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . 58
4 A Moore-Greitzer Type Model for Axial Compressors with Non-
constant Speed 61
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
4.2 Preliminaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
4.3 Modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
4.3.1 Spool Dynamics . . . . . . . . . . . . . . . . . . . . . . . . 64
4.3.2 Compressor . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
4.3.3 Entrance Duct and Guide Vanes . . . . . . . . . . . . . . . 67
4.3.4 Exit Duct and Guide Vanes . . . . . . . . . . . . . . . . . . 67
4.3.5 Overall Pressure Balance . . . . . . . . . . . . . . . . . . . 69
4.3.6 Plenum Mass Balance . . . . . . . . . . . . . . . . . . . . . 70
4.3.7 Galerkin Procedure . . . . . . . . . . . . . . . . . . . . . . . 71
4.3.8 Final Model . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
4.4 Simulations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
4.4.1 Unstable Equilibrium, = 0:5 . . . . . . . . . . . . . . . . . 74
4.4.2 Stable Equilibrium = 0:65 . . . . . . . . . . . . . . . . . . 78
4.5 Concluding Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . 79
5 Modelling and Control of Surge for a Centrifugal Compressor
with Non-constant Speed 81
5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
5.2 Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
5.2.1 Impeller . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
5.2.2 Diuser . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
5.3 Energy Transfer, Compressor Torque and Eciency . . . . . . . . 86
5.3.1 Ideal Energy Transfer . . . . . . . . . . . . . . . . . . . . . 86
5.3.2 Compressor Torque . . . . . . . . . . . . . . . . . . . . . . . 87
5.3.3 Incidence Losses . . . . . . . . . . . . . . . . . . . . . . . . 88
5.3.4 Frictional Losses . . . . . . . . . . . . . . . . . . . . . . . . 91
CONTENTS vii
5.3.5 Eciency . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
5.4 Energy Transfer and Pressure Rise . . . . . . . . . . . . . . . . . . 94
5.5 Choking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
5.6 Dynamic Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
5.7 Surge Control Idea . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
5.8 Controller Design and Stability Analysis . . . . . . . . . . . . . . . 99
5.9 Simulations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
5.10 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
6 Conclusions 109
Bibliography 118
A Nomenclature 119
A.1 Acronyms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
A.2 Subscripts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
A.3 Superscripts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
A.4 Mathematical Symbols and Denitions . . . . . . . . . . . . . . . . 120
A.5 Symbols . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
B Some Thermodynamic and Fluid Mechanical Relations 127
B.1 Flow and Pressure Coecients . . . . . . . . . . . . . . . . . . . . 127
B.2 Isentropic Processes . . . . . . . . . . . . . . . . . . . . . . . . . . 128
B.3 Mass Balance of the Plenum . . . . . . . . . . . . . . . . . . . . . . 129
B.4 Flow through a Nozzle . . . . . . . . . . . . . . . . . . . . . . . . . 130
B.5 Compressor Pressure Rise . . . . . . . . . . . . . . . . . . . . . . . 131
C Including a Close Coupled Valve in the Moore-Greitzer Model 133
D Numerical Values Used in Simulations 137
E Bounds on the Controller Parameters of Theorem 2.5 139
viii CONTENTS
List of Figures
1.1 Compressor characteristic with deep surge cycle, de Jager (1995). . 3
1.2 Physical mechanism for inception of rotating stall, Emmons et.al.
(1955). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.3 Schematic drawing of hysteresis caused by rotating stall. . . . . . . 5
2.1 Compression system. Figure taken from Moore and Greitzer (1986) 13
2.2 Cubic compressor characteristic of Moore and Greitzer (1986). The
constants W and H are known as the semi width and semi height,
respectively. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.3 Compression system with CCV . . . . . . . . . . . . . . . . . . . . 17
2.4 Compressor and throttle characteristics. . . . . . . . . . . . . . . . 19
2.5 Change of variables, principal drawing . . . . . . . . . . . . . . . . 20
2.6 Negative feedback interconnection of strictly passive system with pas-
sive system . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
2.7 Upper plot illustrates that V_2 + V_3 < 0 for > ; m . Lower plot
illustrates that det P > 0 for < choke . . . . . . . . . . . . . . . . 43
2.8 The throttle gain is set to = 0:61, and the compressor is surging.
The controllers are switched on at = 200. . . . . . . . . . . . . . 47
2.10 Disturbance induced surge stabilized by the adaptive controller (2.122). 47
2.9 Same situation as in Figure 2.8. However, here disturbances are
taken into account. The pressure and the mass ow disturbances
are both white noise varying between 0:05. . . . . . . . . . . . . . 48
2.11 Stabilization of rotating stall . . . . . . . . . . . . . . . . . . . . . . 49
2.12 Same simulation as in Figure 2.11, superimposed on the compressor
characteristic and in-stall characteristic . . . . . . . . . . . . . . . 49
2.13 Stabilization of rotating stall with pressure disturbances . . . . . . . 50
3.1 The closed loop system G1 G2 . . . . . . . . . . . . . . . . . . . . . 56
3.2 The closed loop system G1 G2 with disturbances . . . . . . . . . . . 56
3.3 Comparison of closed loop response with passivity based (solid lines)
and backstepping based (dashed lines) controllers. The controllers
were switched on at = 400. . . . . . . . . . . . . . . . . . . . . . . 58
4.1 Principal drawing of the compression system of (Moore and Greitzer
1986). The station numbers are used as subscripts in the following. 63
x LIST OF FIGURES
4.2 Simulation of the system (4.68)-(4.71). Low B leads to rotating
stall, and high B leads to surge. . . . . . . . . . . . . . . . . . . . . 76
4.3 Simulations result superimposed on the compression system char-
acteristics. The compressor characteristic, the in-stall characteris-
tic and the throttle characteristic are drawn with solid, dashed and
dash-dot lines respectively. . . . . . . . . . . . . . . . . . . . . . . . 76
4.4 The three rst harmonics of rotating stall. The rst harmonic J1
has the highest maximum value. This maximum value decreases with
mode number du to the eect of viscosity. The lower plot shows that
J2 dominates J1 during stall inception. . . . . . . . . . . . . . . . . 77
4.5 System response with low Ud, resulting in the compressor being stuck
in rotating stall. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
4.6 Stable equilibrium with and without B-dynamics . . . . . . . . . . . 78
5.1 Compression system with CCV. . . . . . . . . . . . . . . . . . . . . 82
5.2 Diagrammatic sketch of a radially vaned centrifugal compressor.
Shown here with a vaned diuser. . . . . . . . . . . . . . . . . . . . 83
5.3 Diagrammatic sketch of centrifugal compressor tted with a volute. 84
5.4 Velocity triangle at inducer. Section through inducer at radius r1 =
D1 =2. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
5.5 Velocity triangle at impeller tip. . . . . . . . . . . . . . . . . . . . . 85
5.6 Incidence angles at inducer. . . . . . . . . . . . . . . . . . . . . . . 89
5.7 Incidence angles at diuser. . . . . . . . . . . . . . . . . . . . . . . 90
5.8 Eciencies for compressor with vaned (upper plot) and annular
(lower plot) diuser. . . . . . . . . . . . . . . . . . . . . . . . . . . 93
5.9 Energy transfer for N = 35 000 rpm. Compressor with annular
diuser. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
5.10 Centrifugal compressor characteristic. The left plot is for an annu-
lar diuser, and the right for a vaned diuser. The dashed lines to
the right are choke lines, also known as stone walls. . . . . . . . . 96
5.11 Transient response of centrifugal compression system with annular
diuser. Without surge control, the compressor goes into surge,
shown with solid lines. The system response with the surge con-
troller is shown with dashed lines. . . . . . . . . . . . . . . . . . . . 105
5.12 (m(t) p(t)=p01 )-trajectories plotted together with the compressor char-
acteristic. N is the compressor speed in rpm. In the case of no surge
control, the surge cycle is clearly visible, but with surge control the
state converges to the intersection of the throttle and the compressor
characteristic. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
5.13 Transient response of centrifugal compression system with vaned dif-
fuser. Without surge control, the compressor goes into surge. This
is plotted with a solid line. The dashed lines is the system response
when the CCV surge controller is in use. . . . . . . . . . . . . . . . 107
Chapter 1
Introduction
1.1 Introduction
This thesis presents results from an investigation on nonlinear compressor control
where feedback is used to stabilize the compressor to the left of the surge line. The
work is motivated by the fundamental instability problem of surge and rotating
stall which limits the range of operation for compressors at low mass ows.
This instability problem has been extensively studied and industrial solutions
based on surge avoidance are well established. These solutions are based on keep-
ing the operating point to the right of the compressor surge line using a surge
margin. There is a potential for 1) increasing the eciency of compressors by
allowing for operation closer to the surge line than what is the case in current
systems, and 2) increasing the range of mass ows over which the compressor can
operate stably. The increase in eciency and mass ow range is in particular
possible with compressor designs where the design is done with such controllers
in mind. This, however, raises the need for control techniques, which stabilize the
compressor also to the left of the surge line, as disturbances or set point changes
may cause crossing of the surge line. This approach is known as active surge con-
trol. Active surge control is presently an area of very intense research activity, and
is also the topic of this thesis.
1.2 Background
Compressors are used in a wide variety of applications. These includes turbo-
jet engines used in aerospace propulsion, power generation using industrial gas
turbines, turbocharging of internal combustion engines, pressurization of gas and
uids in the process industry, transport of uids in pipelines and so on.
There are four general types of compressors: reciprocating, rotary, centrifugal and
2 Introduction
axial. Some authors use the term radial compressor when refering to a centrifu-
gal compressor. Reciprocating and rotary compressors work by the principle of
reducing the volume of the gas, and will not be considered further in this thesis.
Centrifugal and axial compressors, also known as turbocompressors or continuous
ow compressors, work by the principle of accelerating the uid to a high veloc-
ity and then converting this kinetic energy into potential energy, manifested by
an increase in pressure, by decelerating the gas in diverging channels. In axial
compressors the deceleration takes place in the stator blade passages, and in cen-
trifugal compressor it takes place in the diuser. One obvious dierence between
these two types of compressors is, in axial compressors, the ow leaves the com-
pressor in the axial direction, whereas, in centrifugal compressors, the ows leaves
the compressor in a direction perpendicular to the axis of the rotating shaft. In
this thesis both types of continuous ow compressor will be studied. The litera-
ture on compressors in general is vast, and a basic introduction is given by e.g.
Ferguson (1963) or Cohen et al. (1996), and more advanced topics are covered by
e.g. Cumpsty (1989).
The useful range of operation of turbocompressors is limited, by choking at high
mass ows when sonic velocity is reached in some component, and at low mass ows
by the onset of two instabilities known as surge and rotating stall. Traditionally,
these instabilities have been avoided by using control systems that prevent the
operating point of the compressions system to enter the unstable regime to the
left of the surge line, that is the stability boundary. A fundamentally dierent
approach, known as active surge/stall control, is to use feedback to stabilize this
unstable regime. This approach will be investigated in this thesis, and it will allow
for both operation in the peak eciency and pressure rise regions located in the
neighborhood of the surge line, as well as an extension of the operating range of
the compressor.
1.3 Stability of Compression Systems

Compression systems such as gas turbines can exhibit several types of instabil-
ities: combustion instabilities, aeroelastic instabilities such as utter and nally
aerodynamic ow instabilities, which this study is restricted to.
Two types of aerodynamic ow instabilities can be encountered in compressors.
These are known as surge and rotating stall. The instabilities limits the ow
range in which the compressor can operate. Surge and rotating stall also restrict
the performance (pressure rise) and eciency of the compressor. According to
de Jager (1995) this may lead to heating of the blades and to an increase in the
exit temperature of the compressor.
1.3.1 Surge
Surge is an axisymmtrical oscillation of the ow through the compressor, and is
characterized by a limit cycle in the compressor characteristic. An example of
1.3 Stability of Compression Systems 3
such a characteristic is shown as the S-shaped curve in Figure 1.1. The dotted
segment of the curve indicates that this section usually is an approximation of
the physical system, as it is dicult to measure experimentally. Surge oscillations
are in most applications unwanted, and can in extreme cases even damage the
compressor. As discussed by Erskine and Hensman (1975) and Greitzer (1981),
surge can also induce vibrations in other components of the compression system,
such as e.g. connected piping. It is common to distinguish between at least two
dierent types of surge: 1) Mild/Classic surge and 2) Deep surge. A combination
of surge and rotating stall is known as modied surge. For more information on
dierent types of surge, consult Greitzer (1981) or de Jager (1995).
The rst of these types is a phenomenon with oscillations in both pressure and
ow in the compressor system, while in the second type, the oscillations in mass
ow have such a large amplitude, that ow reversal occurs in the compression
system. A drawing of a typical deep surge cycle is shown in Figure 1.1. The cycle
starts at (1) where the ow becomes unstable. It then jumps to the reversed ow
characteristic (2) and follows this branch of the characteristic until approximately
zero ow (3), and then jumps to (4) where it follows the characteristic to (1), and
the cycle repeats. Surge can occur in both axial and centrifugal compressors.
pressure
(1)
(2)
(4)
comp. char.
(3)
mass ow
Figure 1.1: Compressor characteristic with deep surge cycle, de Jager (1995).
1.3.2 Rotating Stall

Rotating stall can occur in both axial and centrifugal compressors. Although
rotating stall is known to occur in centrifugal compressors, see e.g. Emmons et al.
(1955), there exists little theory on the subject, and according to de Jager (1995)
its importance is still a matter of debate. In this thesis, only rotating stall in axial
4 Introduction
compressors will be considered, and when it is referred to rotating stall it is to be
understood that a axial compressor is considered.
Rotating stall is an instability where the circumferential ow pattern is disturbed.
This is manifested through one or more stall cells of reduced, or stalled, ow prop-
agate around the compressor annulus at a fraction, 20-70% according to Greitzer
(1980), of the rotor speed. This leads to a reduction of the pressure rise of the com-
pressor, and in the compressor map this corresponds to the compressor operating
on the so called in-stall characteristic, see Figure 1.3.
Direction of
C stall propagation
B
A
Compressor
blade row
Figure 1.2: Physical mechanism for inception of rotating stall, Emmons et.al.
(1955).
The basic explanation of the rotating stall mechanism was given by Emmons et
al. (1955) and can be summarized as follows. Consider a row of axial compressor
blades operating at a high angle of attack, as shown in Figure 1.2. Suppose that
there is a non-uniformity in the inlet ow such that a locally higher angle of attack
is produced on blade B which is enough to stall it. The ow now separates from
the suction surface of the blade, producing a ow blockage between B and C.
This blockage causes a diversion of the inlet ow away from B towards A and C,
resulting in a increased angle of attack on C, causing it to stall. Thus the stall cell
propagate along the blade row.
It is common to distinguish between at least two types of rotating stall, full-span
and part-span. In full-span stall, the complete height of the annulus is stalled,
while in part-span rotating stall a restricted region of the blade passage is stalled.
Full-span stall is most likely to occur in high hub/tip ratio axial compressors. In
addition we can have various degrees of rotating stall depending of the size of
the area of the compressor annulus being blocked. In addition to the problem
1.3 Stability of Compression Systems 5
related to reduced pressure rise due to rotating stall , there is also the problem
of vibrations in the blades as stall cells rotate at a fraction of rotor speed. Thus,
the blades passes in and out of regions of stalled ow which can, according to
Horlock (1958), Greitzer (1981) and Pinsley et al. (1991), induce vibrations in
the blades. Moreover, if a natural frequency of vibration of the blades coincides
with the frequency at which the stall cell pass a blade, the result is resonance and
possible mechanical failure due to fatigue.
Another consequence of rotating stall is the hysteresis occurring when trying to
clear the stall by using the throttle. This phenomenon is depicted in Figure 1.3, and
might be described in the following manner: Initially the compressor is operating
stably (1), then a disturbance drives the equilibrium over the surge line resulting
in rotating stall, and a operating point on the low pressure in-stall characteristic
(2). By opening the throttle to clear the stall, result in a higher throttle opening
than initially (3), before the operating point is back on the stable compressor
characteristic (4). There are several degrees to how severe this hysteresis may
be. This depends on the so called skewness of the compressor characteristic. This
is treated in detail in recent papers by Wang and Krstic (1997a), Sepulchre and
Kokotovic (1996) and Protz and Paduano (1997).
pressure
(1) (4)
In-stall char. comp. char.

(2)
(3)
mass ow
Figure 1.3: Schematic drawing of hysteresis caused by rotating stall.
1.3.3 Stability Analysis

As a rule, stall and surge occur at the local maximum of the compressor char-
acteristic or at a point of the compressor characteristic with a certain positive
slope. When linearizing the nonlinear surge model of Greitzer (1976a), one gets
6 Introduction
a model similar to that of a damped linear oscillator. This surge model was also
studied by Stenning (1980). Calculating the eigenvalues of the oscillator system,
as done by Willems (1996) , reveals that the stability boundary is set by a relation
between the slope of the compressor characteristic, the slope of the load line, and
the Greitzer B-parameter. The surge point will be located some small distance to
the left of the peak. According to Cumpsty (1989), the peak of the compressor
characteristic provides a convenient working approximation for the surge point.
The same conclusion is drawn by Stenning (1980), where it is also pointed out
that rotating stall occurs at the peak of the compressor characteristic.
Another approach frequently used to investigate stability of compression systems
is bifurcation analysis. For studies on this topic, consult McCaughan (1989), Liaw
and Abed (1992) or Abed et al. (1993).
1.4 Modeling of Compression Systems

Low order models for surge in compression systems, both axial and centrifugal,
have been proposed by many authors. A classical reference is Emmons et al.
(1955). In the following, a compression system will refer to a system consisting of
a compressor pressurizing incompressible uid, which is discharged into a plenum
volume which discharges via a throttle valve.
The compression system model of Greitzer (1976a) has been widely used for surge
control design. It was derived for axial compression systems, but Hansen et al.
(1981) showed that it is also applicable to centrifugal compressors. The model has
two states, normalized mass ow and normalized pressure, and the compressor is
treated as an actuator disc, with a third order polynomial ow/pressure rise char-
acteristic. Other assumptions of this model are: one dimensional incompressible
ow in the ducts, isentropic compression process in the plenum and no spatial
variations of pressure in the plenum. Extensive experiments by Greitzer (1976b)
and Hansen et al. (1981) conrm that although it is of low order compared to the
complex phenomenon it models, the model is capable of describing both the qual-
itative and quantitative aspects of surge. The model was extended for centrifugal
compressors with varying rotor speed by Fink et al. (1992). Many other models of
compression systems have been presented in the literature, for an overview consult
Willems (1996).
Based on the work of Moore (1984b), the Moore-Greitzer model was derived in
Moore and Greitzer (1986). In developing this model, some of the assumptions
made were: Large hub/tip ratio, irrotational and inviscid ow in the inlet duct,
incompressible compressor mass ow, short throttle duct, small pressure rise com-
pared to ambient conditions and constant rotor speed. A Galerkin procedure was
used to approximate the PDEs describing the system dynamics by three ODEs.
This three state model is capable of describing both surge and rotating stall, the
third state being the stall amplitude. Many authors have extended and modi-
ed the model. Inclusion of higher order harmonics was studied by Mansoux et
al. (1994), and other types of compressor characteristics was used by Wang and
1.5 Control of Surge and Rotating Stall 7
Krstic (1997a).
An alternative model of rotating stall was presented by Paduano et al. (1995),
where rotating stall was described as a traveling wave, and spatial Fourier analysis
was used. Other models that are capable of demonstrating rotating stall are
those of Eveker and Nett (1991) and Badmus et al. (1995), but these models do
not include the stall amplitude as a state, but the presence of rotating stall is
manifested as a pressure drop.
1.5 Control of Surge and Rotating Stall

1.5.1 Surge/Stall Avoidance
As mentioned above, the state of art in control of compressors was until about a
decade ago, the method of surge avoidance, which is usually an open loop strategy
according to de Jager (1995). The compressor is prevented from operating in a
region near and beyond the surge line. This is achieved by e.g. recirculation of the
ow or blowing o ow through a bleed valve. As the compressor characteristic
and thus the surge line may be poorly known, it will often be necessary to have
a fairly conservative surge margin between the surge line and the surge avoidance
line. The compressor is then not allowed to operate between these two lines in the
compressor map. Accounting for possible disturbances also aects the size of the
surge margin.
The drawbacks of surge avoidance schemes are several: (1) Recycling and bleeding
of compressed ow lower the eciency of the system, (2) maximum eciency
and pressure rise may not be achievable at all, as they usually are achieved for
mass ows close to the surge line and (3) the surge margin limits the transient
performance of the compressor as acceleration of the machine tends to drive the
state of the system towards the surge line.
An alternative to surge avoidance is surge detection and avoidance. Using this
strategy, the drawbacks of the surge margin can be avoided by the activation of
the controller if the onset of instabilities is detected. De Jager (1995) concludes
that the main disadvantages of this strategy are problems associated with the
detection of the instability onset and the necessity of large control forced and
fast-acting control systems.
1.5.2 Active Surge/Stall Control

The approach of active surge/stall control aims at overcoming the drawbacks of
surge avoidance, by stabilizing some part of the unstable area in the compressor
map using feedback. This approach was introduced in the control literature by
Epstein et al. (1989). In the last decade, the literature on feedback stabilization of
compression systems has become extensive. This is partly due to the introduction,
and success, of the Moore-Greitzer model.
8 Introduction
In the literature on developments in the eld of jet engines for aircraft propulsion,
active stall/surge control is often said to become an important part of future
engines. For details see e.g. Covert (1995), Rues (1996) or DeLaat et al. (1996).
The term \smart engines" is often used when refering to these engines. Brown
et al. (1997) reports of successfull in-ight experiments on a F-15 aircraft with
active stall control on one engine. Experimental results on laboratory engines are
reported by Paduano et al. (1993), Ffowcs Williams et al. (1993), Hynes et al.
(1994) and Behnken and Murray (1997).
Among several possible actuators for stabilizing compression systems, the throttle
valve as studied by Krstic et al. (1995b) and Badmus et al. (1996), or bleed valves as
studied by Eveker and Nett (1991) or Murray (1997) has been the most commonly
used, at least in the control literature. Other possibilities are variable inlet guide
vanes as in Paduano et al. (1993), loudspeaker as in Ffowcs Williams and Huang
(1989), tailored structures as in Gysling et al. (1991), recirculation as in Balchen
and Mumme (1988), movable wall as in Epstein et al. (1989) or air injection as in
Day (1993) and Behnken and Murray (1997). However, the use of a close-coupled
valve (hereafter named CCV) was claimed by Simon et al. (1993) and van de Wal et
al. (1997) to be among the most promising actuators for compressor surge control.
The methods used for designing surge and stall controllers vary. Linearization and
complex-valued proportional control was used by Epstein et al. (1989), bifurcation
theory by Liaw and Abed (1996), feedback linearization by Badmus et al. (1996),
Lyapunov methods by Simon and Valavani (1991), backstepping by Krstic et al.
(1995b) and H1 by van de Wal et al. (1997) and Weigl and Paduano (1997). In
this thesis, backstepping, which is Lyapunov based, and passivity will be used.
Another important aspect of active surge and stall control is the disturbance rejec-
tion capabilities and robustness of the proposed controllers. Greitzer and Moore
(1986) recognized that research is needed on modeling of disturbances in compres-
sion systems. The reason for this being that disturbances may initiate surge or
rotating stall. Surge controllers with disturbance rejection was presented by Si-
mon and Valavani (1991) and stall/surge controllers by Haddad et al. (1997), and
stall/surge controllers for compressors with uncertain compressor characteristic
was studied by Leonessa et al. (1997b).
1.6 Contributions of this Thesis

The contributions of this thesis can be summarized as follows:
Close coupled valve control of the Moore-Greitzer model: Con-
trollers for a close coupled valve in a Moore-Greitzer type compression sys-
tem have been derived. The controllers allow for stabilization of rotating stall
and surge to the left of the compressors surge line. The eect of constant and
time varying disturbances has also been studied, and controllers have been
derived that ensures avoidance of stall and surge when these disturbances are
present in the compression system. The theory used is backstepping. This
1.7 Outline of the Thesis 9
work has been published in Gravdahl and Egeland (1997c) and Gravdahl
and Egeland (1997d), and is also reported in Gravdahl and Egeland (1998a).
Passivity based surge control: A surge control law has been derived for
the Greitzer model using passivity and a input-output model. The resulting
controller is very simple and is shown to render the system L2 -stable also
when disturbances are taken into account. Parts of this work has been
published in Gravdahl and Egeland (1997f) and submitted in Gravdahl and
Egeland (1998c).
Post stall modeling of axial compression systems with non-constant
speed: An extension of the Moore-Greitzer model has been derived to take
non-constant compressor speed into account. This new model also includes
higher harmonics of rotating stall. The work is published in Gravdahl and
Egeland (1997a) and is also reported in Gravdahl and Egeland (1997e)
Centrifugal compressor modeling and control: A speed dependent
compressor characteristic for a centrifugal compressor has been derived by
the use of energy transfer and losses in the compressor components. This
characteristic has been used in a dynamic model of a compression system
with time varying compressor speed, to derive controllers that simultaneously
suppress surge and ensures that desired speed is reached. This work has
been published in Gravdahl and Egeland (1997g) and Gravdahl and Egeland
(1998b), and is also under consideration for publication in Gravdahl and
Egeland (1997b).
The results reported of in this Thesis is also accepted for publication in Gravdahl
(1998).
1.7 Outline of the Thesis

The outline of this thesis is as follows
Chapter 2: The Moore-Greitzer model and the actuator used in this theses
for surge/stall control, the close coupled valve, are presented. Using back-
stepping, various controllers are derived. These includes controllers for surge
with time varying and constant disturbances (in the later case an adaptive
controller is derived) and controllers for rotating stall with time varying dis-
turbances. Stability results are given for each controller.
Chapter 3: Passivity is used to derive a surge control law. The controller
is shown to maintain certain stability properties even if disturbances are
considered.
Chapter 4: An extension of the familiar Moore-Greitzer model is presented.
This model includes Greitzer's B-parameters as a state, which is a conse-
quence of relaxing the assumption of constant compressor speed.
10 Introduction
Chapter 5: Here a novel model for a compression system with a variable
speed centrifugal compressor is derived. Using Lyapunov theory, controllers
are derived that ensures suppression of surge and regulation of speed. The
closed loop system is shown to be semi globally exponentially stable.
Chapter 6: Conclusions
Finally, references and ve appendices are given at the end of the thesis.
A nomenclature is given in Appendix A.
Chapter 2
Close Coupled Valve Control
of Surge and Rotating Stall
for the Moore-Greitzer
Model
2.1 Introduction
In this chapter controllers for surge and rotating stall using a CCV is derived.
Several possible actuators exists for stabilizing compression systems. Krstic et al.
(1995b), Badmus et al. (1996) and others suggest using a variable throttle valve,
Eveker and Nett (1991), Murray (1997) and others use bleed valves. These two
actuators have been the most commonly used, at least in the control literature.
However, there are many other possibilities: Paduano et al. (1993) use variable
inlet guide vanes, a loudspeaker is used by Ffowcs Williams and Huang (1989),
tailored structures in Gysling et al. (1991), recirculation is studied by Balchen and
Mumme (1988), a movable wall in Epstein et al. (1989) and nally Day (1993) and
Behnken and Murray (1997) employ air injection. However Simon et al. (1993)
claimed that the use of a close-coupled valve (hereafter named CCV) is among the
most promising actuators for active surge control.
The use of a CCV for control of compressor surge was studied by Dussourd et al.
(1976), Greitzer (1981), Pinsley et al. (1991), Simon and Valavani (1991), Simon et
al. (1993) and Jungowski et al. (1996). Experimental results of compressor surge
control using a CCV was reported by Erskine and Hensman (1975) and Dussourd
et al. (1977). Simon (1993) and Simon et al. (1993) compared, using linear theory,
this strategy to a number of other possible methods of actuation and sensing. The
conclusion was that the most promising method of surge control is to actuate the
system with feedback from the mass ow measurement to a CCV or an injector. In
Close Coupled Valve Control of Surge and Rotating Stall for the
12 Moore-Greitzer Model
line with this conclusion are the recent results of van de Wal and Willems (1996)
and van de Wal et al. (1997), where nonlinear controllers were derived based on
H1 performance specications.
Dussourd et al. (1976), Greitzer and Griswold (1976), Dussourd et al. (1977),
Greitzer (1977) and Greitzer (1981) conclude that downstream components in
compression systems have an impact on the onset of rotating stall. Dussourd et
al. (1977) used a CCV to achieve a signicant extension of ow range, and it was
concluded that the CCV also aected the onset of rotating stall as well as surge.
Osborn and Wagner (1970) report of experiments where rotating stall in an axial-
ow fan rotor was suppressed, at the cost of a drop in eciency, by a movable
\door" close coupled downstream of the rotor. The door had a similar geometry
to that of a axisymmetric nozzle, and the experimental setup simulated a turbofan
engine. Greitzer (1977) found that a downstream nozzle shifted the point of onset
of rotating stall to lower mass ows, however the eect was strongest for single
cell, full span stall. The Moore-Greitzer model, which is going to be used in this
chapter, assumes a high hub to tip ratio compressor, which is likely to exhibit full
span stall. Moreover, Greitzer (1977) found that the stabilizing eect of the nozzle
falls rapidly with increasing distance between compressor and nozzle, and thereby
emphasizing the importance of close coupling between compressor and actuator,
a point also made by Hendricks and Gysling (1994). Based on this, the CCV will
in this thesis not only be used as an actuator to stabilize surge, but also rotating
stall.
Simon and Valavani (1991) studied the stability of a compressor with CCV con-
trol by using a Lyapunov function termed the incremental energy. The control
law developed by Simon and Valavani (1991) requires knowledge of the compres-
sor characteristic, and additional adjustments to the controller dictated by the
Lyapunov analysis is performed in order to avoid a discontinuous controller.
Here we will use the backstepping methodology of Krstic et al. (1995a) to derive
a control law for a CCV which gives a GAS equilibrium beyond the original surge
line. Simon and Valavani (1991) studied the eect on stability of disturbances in
pressure rise. This will be considered also here, and in addition we will also consider
disturbances in the plenum outow. In the case of only pressure disturbances, we
will derive a controller that only requires knowledge of an upper bound on the slope
of the compressor characteristic in order to guarantee stability. Discontinuity is
not a problem with this controller. Under mild assumptions on the disturbances,
global uniform boundedness and convergence will be proven in the presence of
both pressure and mass ow disturbances. Constant disturbances or osets will
also be considered.
Krstic and Kokotovic (1995) and Krstic et al. (1995b) used backstepping to design
anti surge and anti stall controllers. The actuator used was a variable throttle.
Here, we use the pressure drop across the CCV as the control variable. Pinsley
et al. (1991) pointed out that in compression systems with large B-parameter (to
be dened), the use of a CCV is expected to be more eective than the use of the
throttle in control of surge. In this case the ow through the throttle is less coupled
with the ow through the compressor than is the case when the B-parameter is
2.2 Preliminaries 13
small. Pinsley et al. (1991), Krstic and Kokotovic (1995) and Krstic et al. (1995b)
use feedback from mass ow and pressure. As will be shown, the application of the
backstepping procedure to CCV surge and stall control, in the case of no mass ow
disturbances, results in a control law which uses feedback from mass ow only.
As opposed to throttle control, CCV control modies the compressor character-
istic. This allows for, at the cost of a pressure loss over the valve, recovery from
rotating stall beyond the surge line. Although the pressure rise achieved in the
compression system with a steady pressure drop across the CCV is comparable
with the pressure rise achieved when the machine is in rotating stall, the CCV
approach is to prefer as the possibility for blade vibration and high temperatures
is avoided.
2.2 Preliminaries
2.2.1 The Model of Moore and Greitzer
Several dynamic models for the unstable operation of compression systems have
been proposed in the last decade, but the model of Moore and Greitzer (1986)
stands out in the sense that rotating stall amplitude is included as a state, and
not manifested as a pressure drop which is the case in the other models. The low
01 E Station numbers
LI LE
Lc
HH
HH
H

ps
pT
Cx A
A
pT
A
Vp

Inlet duct
Exit duct
IGV Compressor Plenum Throttle
Figure 2.1: Compression system. Figure taken from Moore and Greitzer (1986)
order1 model of Moore and Greitzer (1986) captures the post stall transients of
a low speed axial compressor-plenum-throttle (see Figure 2.1) system. The main
assumptions made by Moore and Greitzer (1986) in deriving the model are: large
hub-to-tip ratio so that a two-dimensional description seems reasonable, incom-
pressible compressor mass ow, compressible ow in the plenum, spatially uniform
plenum pressure and short throttle duct. The three dierential equations of the
model arises from a Galerkin approximation of the local momentum balance, the
annulus-averaged momentum balance and the mass balance of the plenum. A
cubic compressor characteristic is assumed. The model is given by:

#_ = W=H ; 1 (#) H
4B 2 W W T lc
3 !
_ = H ; # ; c0 ; 1 ; 1 + 1 + 3 ; 1 1 ; J (2.1)
lc H 2 W 2 W 2
2 !
J_ = J 1 ; W ;1 ; J %
4
where
is the annulus averaged mass ow coecient (axial velocity divided
R by com-
pressor speed), where the annulus average is dened as 1 2 ( )d = 4
2 0
( ), and ( ) is the local mass ow coecient,
# is the non dimensional plenum pressure or pressure coecient (pressure
divided by density and the square of compressor speed),
J is the squared amplitude of rotating stall amplitude
T (#) is the throttle mass ow coecient and
lc is the eective ow-passage nondimensional length of the compressor and
ducts dened as
c
4 l + 1 +l
l = I E (2.2)
a
where the positive constant a is the reciprocal time-lag parameter of the
blade passage,
For a discussion of the employed nondimensionalization, consult Appendix B. The
constant B > 0 is Greitzer's B-parameter dened by Greitzer (1976a) as
r V
4 U
B= p
2as Ac Lc (2.3)
where U is the constant compressor tangential speed (in m/s) at mean diameter,
as is the speed of sound, Vp is the plenum volume, Ac is the ow area and Lc is
1 \Low order" refers to the simplicity of the model, three states, compared to the complex
uid dynamic system it models
the length of ducts and compressor. The constant % > 0 is dened as
% = (1 +3aH
ma)W (2.4)
where m is the compressor-duct ow parameter, H is the semi-height of the com-
pressor characteristic and W is the semi-width of the compressor characteristic.
The time variable used throughout this chapter is also nondimensional, and is
dened as 4 Ut=R
= (2.5)
where t is the actual time and R is the mean compressor radius. The notation _
is to be understood as the derivative of with respect to , that is _ = dd .
Relaxing the constant speed assumption is important for studying eects of set
point changes, acceleration, deceleration, etc. A model taking variable speed into
account will be developed in Chapter 4, but will not be considered further here.
In the case of pure surge, that is when J 0, the model reduces to that of Greitzer
(1976a):
_ = 1 (# () ; #)
lc c (2.6)
#_ = 4B12 l ( ; T (#)):
c
In Greitzer (1976a), the model was written
_ = B (#c() ; #)
#_ = 1 ( ; T (#)):
B
The discrepancy in the constants is due to Greitzer (1976a) dening nondimen-
sional time as = t!H where !H is the Helmholtz frequency. Here, nondimensional
time is dened according to (2.5), as was also done by Moore and Greitzer (1986).
The model (2.6) was derived for axial compression systems, but it was demon-
strated by Hansen et al. (1981) that the model also is applicable to centrifugal
systems.
The pressure rise of the compressor is a nonlinear function of the mass ow. This
function, #c ( ), is known as the compressor characteristic. Dierent expressions
for this characteristic have been used, but one that has found widespread accep-
tance in the control literature is the cubic characteristic of Moore and Greitzer
(1986):
3
1 3!
#c( ) = c0 + H 1 + 2 W ; 1 ; 2 W ; 1 (2.7)
where the constant c0 > 0 is the shut-o value of the compressor character-
istic. The cubic characteristic with the parameters c0 , W and H is shown in
Figure 2.2. Mansoux et al. (1994), Sepulchre and Kokotovic (1996) and Wang and
Krstic (1997a) suggest other compressor characteristics for axial compressors, and
Hansen et al. (1981) presents an alternative polynomial characteristic for centrifu-
gal compressors. However, the cubic seems to capture the general shape of the
compressor characteristic of a large class of compressors. The nondimensionaliza-
tion employed, transforms the usual family of curves in the compressor map, one
for each compressor speed, to one single characteristic given by (2.7). Nisenfeld
(1982) and Badmus et al. (1996) concludes that this is in fact a statement of the
Fan law relation. The surge line, which passes through the local maxima of the
family of curves is transformed to the local maximum of (2.7).

2W
2H c()
c0
Figure 2.2: Cubic compressor characteristic of Moore and Greitzer (1986). The
constants W and H are known as the semi width and semi height, respectively.
The throttle mass ow T (#) is given by the throttle characteristic

p
T (#) = T # (2.8)
where T is the throttle gain. The inverse throttle characteristic
#T () = ;T 1 () = 12 2 (2.9)
T
is shown in Figure 2.5.
2.2.2 Close Coupled Valve

A compressor in series with a CCV will be studied in the following. According to
Simon and Valavani (1991), with \close-coupled" it is to be understood that the
distance between the compressor outlet and the valve is so small that no signicant
mass storage can take place, see Figure 2.3.
()
PP c
PP
PP
v ()
@; @;
;@ ;@ T ()

CCV Throttle

Compressor Plenum
Figure 2.3: Compression system with CCV
The assumption of no mass storage between the compressor and the valve allows
for the denition of an equivalent compressor. This term was introduced by Simon
and Valavani (1991). The pressure rise over this equivalent compressor is the sum
of the pressure rise over the compressor and the pressure drop over the valve.
The pressure drop over the valve will be used as the control. This will allow for
manipulation of the equivalent compressor characteristic, given by
#e () = #c () ; #v () (2.10)
where #c() and #v () are the compressor pressure rise and valve pressure drop
respectively and is the axial mass ow coecient. The motivation behind this, is
that the slope of the compressor characteristic determines the stability properties
of the equilibrium of the system, and this slope can be varied by varying the
pressure drop over the CCV. The use of the CCV as an actuator for surge control
is also elaborated upon in section 5.7, where the stabilizing eect of such a valve
is discussed in connection with the destabilizing eect of incidence losses.
The CCV has a characteristic given by
# () = 1 2
v 2 (2.11)
where > 0 is proportional to the valve opening. We now set out to repeat
the modeling and Galerkin approximation of Moore and Greitzer (1986) with the
equivalent characteristic #e replacing #c . Equation (5) in Moore and Greitzer
(1986), which gives the pressure rise across the compressor, is modied according
to
pE ; p1 = N F ( ) ; 1 2 @ + @ ;# ( )

s v (2.12)
| U {z 2a @ @ }
2
Equation (5) in Moore and Greitzer (1986)
where p1 and pE is the static pressure at the entrance and exit of the equivalent
compressor, is the constant inlet density, U is the compressor speed at mean di-
ameter, Ns is the number of compressor stages, F ( ) is the pressure rise coecient
in the blade passage, and is the angular coordinate around the wheel. Equation
(2.12) now gives the pressure rise over the equivalent compressor.
Using (2.12) as a starting point and following the derivation of Moore and Greitzer
(1986), the following model is found2 :

_# = W=H2 ; 1 T (#) H
4B W W lc

_ = H ; # ; c0 ; 1 ; 1 3 + 1
lc H 2 W
3
J
1 W 2J 2 !
+ 2 W ; 1 1 ; 2 ; 2 2H + H (2.13)
2 J 1 4W !

J_ = J 1 ; W ; 1 ; 4 ; 2 3H %
which will be used in design of stall/surge controllers in this chapter. In the simpler
case of pure surge, J is set to zero, and we are left with the model
#_ = 4B12 l ( ; T ( )) (2.14)
c
_ = l1 (#c() ; #v () ; #)
c
which will used in the study of surge control.
2.2.3 Equilibria
The compressor is in equilibrium when _ = #_ = J_ = 0. If J (0) = 0 then J 0
and the equilibrium values 0 and 0 are given by the intersection of #e () and
the throttle characteristic. If J (0) > 0, and the throttle characteristic crosses
#e to the left of the local maximum, the compressor may3 enter rotating stall
and the equilibrium values 0 and 0 are given by the intersection of the throttle
characteristic and the stall characteristic #es () which is found by analyzing the
J_-equation of (2.13). It is seen that J_ = 0 is satised for J = 0 or
2 !
;1
J = Je = 4 1 ; W ; 12 43WH : (2.15)
_
Inserting (2.15) in the -equation of (2.13) and setting _ = 0 gives the expression
for #es ( ) :
5 8 W W 2
#es () = #s () + H #v () ; H 2 1 ; 3H 2 2 (2.16)
2 The complete derivation is shown in Appendix C.
3 This depends on the numerical value of B . Greitzer and Moore (1986) showed that small B
gives rotating stall, and large B gives surge.
where !
#s () = c0 + H 1 ; 32 W ;1 + 5 ;1 3 (2.17)
2 W
is the stall characteristic found when the CCV is not present. In Figure 2.4 the
various characteristics are shown. As can be seen, the throttle line intersects #c in
a point of positive slope, that is in the unstable area of the compressor map, and
the compressor would go into rotating stall or surge. By introducing the CCV, the
throttle line crosses the equivalent characteristic #e in an area of negative slope.
This new equilibrium is thus stable.
0.8
0.7
0.6
#s () #c ()
0.5
#es ()
#
0.4
#v ()
0.3
0.2
;1 () #e ()
0.1
0
−0.2 −0.1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8

Figure 2.4: Compressor and throttle characteristics.
2.2.4 Change of Variables

The equilibrium of the compression system without the presence of the valve, is
at the intersection of the compressor characteristic #c() and the throttle char-
acteristic T (#). When introducing the CCV into the system, the equilibrium is
shifted to the intersection of the equivalent compressor characteristic #e () and
the throttle characteristic T (#). The equilibrium values are then related through
0 = ;T 1 ( 0 ): (2.18)

^
T 1 ()
;
^co
0 ^
e()
c()
0
^co v ()

0
Figure 2.5: Change of variables, principal drawing
To prepare for the analysis of the system, it is desirable to perform a change of

coordinates on the system equations so that the origin becomes the equilibrium
under study. The new coordinates are dened as
^ = # ; 0 (2.19)
^ = ; 0:
The system characteristics in the new coordinates are dened as

#^ e ( ^) = #e ( ^ + 0 ) ; #e ( 0 )
= #e ( ^ + 0 ) ; 0 (2.20)
^#0c ( ^) = #c ( ^ + 0 ) ; 0 (2.21)
#^ v ( ^) = #v ( ^ + 0 ) ; 0 (2.22)
^ T ( ^) = T ( ^ + 0 ) ; 0 : (2.23)
Using (2.7), the transformed compressor characteristic (2.21) can be calculated as
#^ 0c( ^) = ^co ; k3 ^3 ; k2 ^2 ; k1 ^ (2.24)
where
2H
^co = co ; 0 ; 2W 0 0
2 W ;3 (2.25)

k1 = 32HW 20 W0 ; 2 (2.26)

k2 = 23WH2 W0 ; 1 (2.27)
k3 = 2W H : (2.28)
3
It can be recognized that k3 > 0, while k1 0 if the equilibrium is in the unstable
region of the compressor map and k1 > 0 otherwise. The sign of k2 may vary. If
0 > W then k2 > 0, and if 0 < W then k2 < 0.
Using the denition of the equivalent compressor (2.10) and the denition of 0
in (2.18), it can be shown that
2H
0 = co ; 0
W ; 3 ; #v ( 0 ):
0 (2.29)
2W 2
Combining (2.29) with (2.25) we get the simple result
^co = #v ( 0 ) (2.30)
which could have been seen directly from Figure 2.5.
The Moore-Greitzer model (2.13) can now be written in the new coordinates as

_^ = 4B12l ^ ; ^ T ( ^) (2.31)
c 2
^_ = 1 #^ 0c( ^) ; ^ ; #^ v ( ^) ; 0 ; 3H J ; 1 ; W 2J
l c " 4 W 2
;1
2 J # 4W%
J_ = %J 1 ; W ; 4 ; 3H 2 J
By dening
#^ 0c( ^) = #v ( 0 ) + #^ c( ^) (2.32)
where 4 ;k3 ^3 ; k2 ^2 ; k1 ^
#^ c( ^) = (2.33)
the model (2.31) can be written

_^ = 4B12 l ^ ; ^ T ( ^) (2.34)
c
_^ = 1 #^ c( ^) ; ^ ; u ; 3H J ; 1 ; W 2 J
lc " 4
# W 2 2

; 1 2 ; J ; 4W% J
J_ = %J 1 ; W 4 3H 2
where the control u has been selected as
u = #^ v ( ^) + 0 ; #v ( 0 ) = #v () ; #v ( 0 ): (2.35)
In the case of pure surge, the model (2.34) reduces to
_^ = 4B12l ( ^ ; ^ T ( ^)) (2.36)
c
_^ = 1 (#^ c ( ^) ; u ; ^)
l c
Simon and Valavani (1991) suggested using the pressure drop across the valve as
the control variable u. This approach will also be taken here. Our aim will be to
design a control law u for the valve such that the compressor can be operated also
on the left side of the original surge line without going into surge or rotating stall.
That is, we are going to use feedback to move the surge line towards lower values
of , and thus expand the useful range of mass ows over which the compressor
can be safely operated.
As the pressure dierence across the valve always will be a pressure drop, the
valve must be partially closed during stable operation in order for the control u to
attain both positive and negative values. It is also evident that there must exist
a pressure drop over the valve when the compressor is operated in a previously
unstable area. The price paid for a larger operating range, is some pressure loss
at low mass ows. A further discussion of the steady state pressure loss can be
found in Simon and Valavani (1991).
2.2.5 Disturbances
As in all types of physical systems, disturbances will occur in the compression
system. Greitzer and Moore (1986) stated that this is a topic that need more
study, at least in the case of the disturbances initiating stall and surge. Some
research have been done in this area. Hynes and Greitzer (1987), DeLaat et al.
(1996) and others have studied the eect of circumferential inlet distortion on the
stability properties, and Simon and Valavani (1991), Haddad et al. (1997) and
others studied mass ow and pressure disturbances. From a control theory point
of view it is also important to investigate what performance the closed loop system
will have when disturbances is taken into account.
As in Simon and Valavani (1991), the eect of a pressure disturbance #^ d( ), and
a ow disturbance ^ d( ) will be considered here. The pressure disturbance, which
may arise from combustion induced uctuations when considering the model of a
gas turbine, will accelerate the ow. As pointed out by van de Wal and Willems
(1996), the ow disturbances may arise from processes upstream of the compressor,
other compressors in series or an air cleaner in the compressor duct. In the case
of an aircraft jet engine, large angle of attack or altitude variations may cause
mass ow disturbances according to van de Wal and Willems (1996) and DeLaat
et al. (1996). Also, DeLaat et al. (1996) reports of a number of aircraft maneuvers
(full-rudder sideslips, wind-up turns, etc.) causing inlet airow disturbances in
the jet engine of a F-15 ghter.
In the analysis of Simon and Valavani (1991) ^ d( ) is set to zero. Disturbances
in stall/surge control is also studied by Haddad et al. (1997), with disturbances
assumed to converge to zero. Here, both types of disturbances, mass ow and
pressure, will be considered. The disturbances are time varying, and the only
assumption made at this point is boundedness, that is k^ dk1 and k#^ dk1 exists.
In addition to time varying disturbances, constant, or slow varying, osets will be
introduced into the model. This is of particular interest when e.g. a constant neg-
ative mass ow disturbance pushes the equilibrium over the surge line, initiating
surge or rotating stall. The osets in mass ow and pressure rise is termed d and
d , respectively. The constant bias d in pressure can also be thought of as re-
ecting some uncertainty in the compressor characteristic #^ c( ^), and likewise and
the mass ow bias d can be thought of as reecting a uncertainty in the throttle
characteristic ^ ( ^). A study of surge/stall control for compressors with uncertain
compressor characteristic was also done by Leonessa et al. (1997b). With these
disturbances the model becomes:

_^ = 4B12 l ^ ; ^ T ( ^) ; ^ d ( ) ; d (2.37)
c
_^ = 1 ; ^ + #^ c( ^) ; #^ v ( ^) + #^ d ( ) + d
lc
; 3HJ ; 1 ; W 2J
4 W 2 2 !
2
J_ = %J 1 ; W ; 1 ; J ; 4W% J
4 3H 2
and in the case of pure surge

_^ = 4B12 l ^ ; ^ T ( ^) ; ^ d( ) ; d (2.38)
c
_^ = 1 ; ^ + #^ c( ^) ; #^ v ( ^) + #^ d( ) + d : (2.39)
l c
2.3 Surge Control
In this section controllers will be designed for the pure surge case. First the
undisturbed case is studied, then disturbances are added, and nally adaption will
be used to stabilize the system in the presence of constant disturbances.
2.3.1 Undisturbed Case

Theorem 2.1 The controller
u = c2 ( ; 0 ) (2.40)
where c2 > am and am is the maximum positive slope of the compressor charac-
teristic #^ c ( ^), renders the equilibrium ( 0 0 ) of (2.14) GAS.
Proof: The backstepping methodology of Krstic et al. (1995a) will be employed

in deriving the control law.
Step 1. Two error variables are dened as z1 = ^ and z2 = ^ ; . The control

Lyapunov function (clf) for this step is chosen as
V1 = 2B 2 lcz12 (2.41)
with time derivative along the solution trajectories of (2.36) given by

V_1 = z1 ;^ T (z1 ) + z2 + : (2.42)
The throttle is assumed passive, that is ^^ T ( ^) 0 8 ^. We have
^^ T ( ^) 0 ) ;z1 ^ T (z1 ) 0 (2.43)
As it is desirable to avoid cancelation of useful nonlinearities in (2.42), the stabi-
lizing function is not needed and accordingly = 0, which gives
V_1 = ;^ T (z1 )z1 + z1 z2 : (2.44)
Although the virtual control is not needed here, in the interest of consistency
with the following sections, this notation is kept.
Step 2.The derivative of z2 is

z_2 = l1 #^ c (z2 ) ; z1 ; u : (2.45)
c
The clf for this step is
V2 = V1 + l2c z22 (2.46)
2.3 Surge Control 25
with time derivative

V_2 = ;z1^ T (z1 ) + z2 #^ c(z2 ) ; u : (2.47)
Notice that V2 as dened by (2.46) is similar to the incremental energy of Simon
and Valavani (1991).
Control law. The control variable u will be chosen so that (2.47) is made negative
denite. To this end we dene the linear control law
u = c2 z2 (2.48)
where the controller gain c2 > 0 is chosen so that
z2 #c(z2 ) ; c2 z22 < 0: (2.49)
Using (2.33) this implies that c2 must satisfy

;k3 z22 z22 + kk2 z2 + k1 k+ c2 < 0: (2.50)
3 3
Finding the roots of the above bracketed expression, it is seen that (2.50) is satised
if c2 is chosen according to
2
c2 > 4kk2 ; k1 : (2.51)
3
Although (2.51) implies that the compressor characteristic must be known in order
to determine c2 , it can be shown that the knowledge of a bound on the positive
slope of the characteristic is sucient. Dierentiating (2.33) twice with respect to
^, reveals that the maximum positive slope occurs for
^ = ^m = ; k2 (2.52)
3k3
and is given by
^ ^

a = d#c^( )
2
= 3kk2 ; k1 = 23W
H: (2.53)
d ^=^m 3
Assuming that only an upper bound am on the positive slope of #^ c ( ^) is known,

a conservative condition for c2 is
2
c2 > am a > 4kk2 ; k1 : (2.54)
3
Thus the price paid for not knowing the exact coecients of the compressor char-
acteristic is a somewhat conservative condition for the controller gain c2 . Notice
also that no knowledge of Greitzer's B -parameter or its upper bound is required
in formulating the controller. The nal expression for V_2 is then
V_2 = ;z1 ^ T (z1 ) + #^ c(z2 )z2 ; c2 z22 = ;W (z1 z2 ) 0: (2.55)
The closed loop system can be written as
z_1 = 4B12 l (;^ T (z1 ) + z2) (2.56)
c
1
z_2 = l (;z1 + #^ (z2 ) ; c2 z2): (2.57)
c
It follows that the equilibrium point z1 = z2 = 0 is GAS, and the same result
holds for the equilibrium ( 0 0 ).
Remark 2.1 By combining (2.82) and (2.48), the following control law for the
CCV gain is found:
c2 ( ; 0 ) = #v () ; #v ( 0 ) (2.58)
+
r +
= 0
c2 : (2.59)
Notice that this control law requires measurement of mass ow only.
Remark 2.2 Although not showing stability, Bendixon's criterion can be used to
show that the controller (2.48) guarantees that no limit cycles (surge oscillations)
exists. Bendixon's criterion states (somewhat simplied) that no limit cycles ex-
ists in a dynamical system dened on a simply connected region D
IIR2 if the
divergence of the system is not identically zero and does not change sign in D. For
an exact statement of Bendixon's criterion and the proof, see any textbook on dy-
namical systems, e.g. Perko (1991). The divergence r f of the system x_ = f (x)
dened by (2.14) is
^ ^
^ ^
!
r f = ; 4B12l @ T ^( ) + l1 @ c^( ) ; @u^ : (2.60)
c @ c @ @
The slope of the throttle is always positive, so the rst term in (2.60) is always
negative. To make the second term also negative, it is sucient that @u @ ^ dominates
@ ^c (^) , which is exactly what is ensured by the controller in Theorem 2.1. Thus,
@ ^
according to Bendixon's criterion, no surge oscillations can exist for the closed
loop system.
2.3.2 Determination of 0
As a consequence of the controller (2.40) being designed after the system equations
are transformed to the new coordinates, its implementation depends on knowledge
of the equilibrium value 0 . The equilibrium is located at the intersection of the
equivalent compressor characteristic #e () and the throttle characteristic ;T 1 .
By combining (2.82), (2.40) and (2.59), it is seen that
#v ( 0 ) = (u + #v ( 0 ))j=0

= c2 ( ; 0 ) + +c2 2
0
0 =0
= c22 0 : (2.61)
At the equilibrium, we have
#c( 0 ) ; #v ( 0 ) = 12 20 (2.62)
or by using (2.61), 0 is found by solving the following 3rd order equation
3! c2
co + H 1 + 32 W0 ; 1 ; 12 W0 ; 1 ; 0 2 2
2 = 2 0 (2.63)
with respect to 0 , and its value is to be used in the control law (2.40). Solving
(2.63) requires knowledge of the compressor characteristic. If this is not the case,
alternatives to nding 0 explicitly is, using an adaption scheme like the one
suggested by Bazanella et al. (1997) or it is possible to use throttle control in
addition to the CCV control to control using 0 as the reference. If none of
these alternatives are attractive, an approximation for 0 can be used. In this case,
asymptotic stability cannot be shown, but convergence to a set and avoidance of
surge is easily shown. Dening % as
% = 0 ; aprx (2.64)
where aprx is the approximation used for feedback, and 0 is the actual and
unknown value of the equilibrium. By using the same Lyapunov function as is
Theorem 2.1, and
u = c2 ( ; aprx ) = c2 ( ; 0 + % ) (2.65)
the time derivative of V2 is found and upper bounded by
V_2 ;z1^ (z1 ) + z2 (#^ c (z2 ) ; c2 z2 )z2 ; z2 c2 % : (2.66)
Application of Young's inequality4 to the last term in (2.66) gives
2
;z2 c2 % c22 z2 + (% )2 0 (2.67)
0
4 In its simplest form Young's inequality states that
a b : ab 12 ( ac + cb2 ) c > 0:
2
8 8
where 0 is a constant, and it follows that
V_ 2 ;z1^ T (z1 ) + z2 (#^ c (z2 ) ; c2 (1 ; 21 )z2 )z2 + 0 (% )2
0
= ;W (z1 z2 ) + 0 (% )2 (2.68)
where the denition of W is obvious. By choosing 0 and c2 such that
c2 (1 ; 21 ) > am (2.69)
0
is satised, it can be shown that W (z1 z2 ) is radially unbound and positive denite.
Thus, V_2 < 0 outside a set R . This set can be found in the following manner:
According to Krstic et al. (1995a), the fact that V2 (z1 z2) and W (z1 z2) is positive
denite and radially unbounded, and V2 (z1 z2) is smooth, implies that there exists
class-K1 functions 1 , 2 and 3 such that
1 (jz j) V2 (z ) 2 (jz j) (2.70)
3 (jz j) W (z) (2.71)
where z = (z1 z2 )T . Following the proof of Lemma 2.26 in Krstic et al. (1995a),
we have that the states of the model are uniformly ultimately bounded, and that
they converge to the residual set
( )
;
R = z : jz j 1;1 2 3;1 (% )2 : (2.72)
From (2.68) it follows that V_ 2 is negative whenever W (z ) > 0 (% )2 . Combining

this with (2.71) it can be concluded that
;
jz( )j > 3;1 (% )2 ) V_2 < 0: (2.73)
; ;
This means that if jz (0)j 3 (% )2 , then
1
;
V2 (z ( )) 2 3;1 (% )2
(2.74)
which in turn implies that
;
jz ( )j 1;1 2 3;1 (% )2 :
(2.75)
; ;
If, on the other hand jz (0)j > 3 (% ) , then V2 (z ( )) V2 (z (0)), which
1 2
implies
jz ( )j 1;1 2 (jz (0)j): (2.76)
Combining (2.75) and (2.76) leads to the global uniform boundedness of z ( ):
;
kzk1 max 1;1 2 3;1 (% )2 1;1 2 (jz (0)j)
(2.77)
while (2.73) and (2.70) prove the convergence of z( ) to the residual set dened
in (2.72).
It is trivial to establish the fact that no limit cycles, and hence surge oscillations,
can exist inside this set using Bendixon's criterion. It is seen from (2.72), that the
size of the set R is dependent on the square of the equilibrium estimate error %
and the parameter 0 . A more accurate estimate aprx , or a smaller value of 0 ,
both implies a smaller set. A smaller value of 0 will, by equation (2.69), require
a larger controller gain c2 , which is to be expected.
2.3.3 Time Varying Disturbances
First, time varying pressure disturbances will be considered. That is, ^ d ( ), d
and d are set to zero as in Simon and Valavani (1991).
Theorem 2.2 (Time varying pressure disturbances)
The controller
u = (c2 + d2 )( ; 0 ) (2.78)
where c2 is chosen as in Theorem 2.1, and d2 > 0 guarantees that the states of the
model (2.38) are globally uniformly bounded, and that they converge to a set.
Proof: The controller will be derived using backstepping.

Step 1. Identical to Step 1 in proof of Theorem 2.1.
Step 2. The derivative of z2 is

z_2 = l1 #^ c( ^) ; z1 + #^ d( ) ; u : (2.79)
c
V2 is chosen as
V2 = V1 + l2c z22 (2.80)
where V_2 can be bounded according to

V_2 = ;^ T (z1 )z1 + z2 #^ c ( ^) + #^ d( ) ; u : (2.81)
Control law. To counteract the eect of the disturbance, a damping factor d2 > 0
is included and u is chosen as
u = c2 z2 + d2 z2 : (2.82)
c2 is chosen so that (2.54) is satised. Inserting (2.82) in (2.81) gives
V_2 = ;z1 ^ T (z1 ) + #^ c(z2 )z2 ; c2 z22 + #^ d( )z2 ; d2 z22: (2.83)
Use of Young's inequality gives
^2 ^
z2 #^ d( ) d2 z22 + #4dd( ) d2 z22 + k#4ddk1
2
(2.84)
2 2
and V_2 can be bounded according to
^2
V_2 ;W (z1 z2 ) + #4dd( ) ;W (z1 z2) + 41d k#^ dk21 (2.85)
2 2
where
W (z1 z2 ) = z1 ^ (z1 ) ; (#^ c(z2 )z2 ; c2 z22 ) (2.86)
is radially unbounded and positive denite. This implies that V_ 2 < 0 outside a set
R1 in the z1 z2 plane.
By using (2.70) and (2.71) again, it follows from similar calculations as in (2.73)-
(2.77), that z ( ) is globally uniformly bounded and that z ( ) converges to the
residual set ( ^ 2 !)
R1 = z : jzj 1 2 3 k#4ddk1 :
; 1 ;1 (2.87)
2
Remark 2.3 Notice that the controller (2.82) is essentially the same as (2.48),
with the only dierence being that (2.82) requires a larger gain in order to suppress
the disturbance. Consequently, Remark 2.1 also applies here.
It is now shown that an additional assumption on the disturbance ensures that

the controller (2.82) not only makes the states globally uniformly bounded, but
also guarantees convergence to the origin.
Corollary 2.1 (Convergence to the origin)
If the disturbance term #^ d( ) is upper bounded by a monotonically decreasing
non-negative function #d ( ) such that
j#^ d( )j #d ( ) 8 0 (2.88)
and
lim # ( ) = 0
!1 d
(2.89)
the controller (2.82) ensures that the states of the model (2.38), with pressure dis-
turbances, converge to the origin.
Proof: Inspired by the calculations for a simple scalar system starting on page 75
in Krstic et al. (1995a) , we introduce the signal
s(z ) = V2 (z )ec (2.90)
where c > 0 is a constant, for use in the proof:
d s(z ) = d V (z )ec
dt 2
dt_
= V2 (z ) + cV2 (z) ec
^ 2d( )
!
#
;W (z) + 4d + cV2 (z ) ec
2
^2
(;3 (jz j) + c2 (jz j)) ec + #4dd( ) ec : (2.91)
2
By choosing c according to
c 2;1 3 (jz j) 2;1 3 (kz k1 ) (2.92)
where the existence of kzk1 follows from (2.87), (2.91) gives
d V (z )ec #^ 2d ( ) ec : (2.93)
dt 2 4d2
By integrating (2.93) and using an argument similar to the one in the proof of
lemma 2.24 in Krstic et al. (1995a), it can be shown that
1 # 2 (0)e; c2 + # 2 (=2) :
V2 (z ( )) V2 (z (0))e;c + 4cd (2.94)
d d
2
Since lim!1 #d2 (=2) = 0 it follows that

lim V (z ( )) = 0:
!1 2
(2.95)
As V2 is positive denite it follows that
lim z ( ) = 0:
!1
(2.96)
Thus we have shown that under the additional assumptions (2.88) and (2.89) on
the disturbance term, z( ) converges to the origin. This also implies that ^ and ^
converge to the origin and that ( ) and #( ) converge to the point of intersection
of the compressor and throttle characteristic.
Notice that the positive constant c introduced in (2.90) is used for analysis only,
and is not included in the implementation of the control law.
At this point we include the ow disturbance ^ d ( ) in the analysis.
Theorem 2.3 (Time varying pressure and ow disturbances)
The controller
;
u = c2 z2 ; k3 3 + 3z22 ; k2 ^2 ; k1
d ^ d 2
1 ^
+ 4B 2 ;T (z1 ) + + d2 z2 1 + 4B12 (2.97)
where c2 > jk1 j guarantees that the states of the model (2.37) with both mass ow
disturbances and pressure disturbances is globally uniformly bounded and that they
converge to a set.
Proof: The backstepping procedure is as follows:

Step 1. As before two error variables z1 and z2 are dened as z1 = ^ and z2 = ^;.
Again, V1 is chosen as
V1 = 2B 2 lc z12 (2.98)
with time derivative

V_1 = z1 ;^ T (z1 ) + z2 ; ^ d ( ) + (2.99)
where (2.37) is used. The virtual control is chosen as
= ;d1 z1 (2.100)
where ;d1 z1 is a damping term to be used to counteract the disturbance ^ d( ).
V_1 can now be written as
V_1 = ;d1 z12 + z1 z2 ; ^ d ( )z1 ; ^ T (z1 )z1 (2.101)
and upper bounded according to
^
V_1 ;^ T (z1 )z1 + z1 z2 + k4ddk1 :
2
(2.102)
1
To obtain the bound in (2.102), Young's inequality has been used to obtain
^2 ^ 2
;^ d( )z1 d1 z12 + 4dd( ) d1 z12 + k4ddk1 : (2.103)
1 1
z_2 = l1 #^ c ( ^) ; z1 + #^ d( ) ; @z @ 1 ;^ (z ) + ^
T 1
c 1 4B 2 lc

@ ^ ( ) ; u :
+ 4B12l @z d (2.104)
c 1
From (2.100) it is seen that
@ = ;d : (2.105)
@z1 1
V2 is chosen as
V2 = V1 + l2c z22 : (2.106)
Using (2.102) and (2.104), an upper bound on V_2 is
^ 2
V_2 ;^ T (z1 )z1 + k4ddk1 + z2 #^ c( ^) + #^ d( )
1

d
+ 4B12 ;^ T (z1 ) + ^ ; 4dB12 ^ d ( ) ; u : (2.107)
Control law. To counteract the eect of the disturbances, a damping factor d2
must be included and u is chosen as
;
u = c2 z2 ; k3 3 + 3z 22 ; k2 ^2 ; k1
d ^ d 2
1 ^
+ 4B 2 ;T (z1 ) + + d2 z2 1 + 4B12 : (2.108)
The parameter c2 is now chosen according to
c2 > jk1 j: (2.109)
Inserting (2.97) in (2.107) gives
^
V_2 ;(c2 + k1 )z22 ; k3 (z24 + 32 z22) ; ^ T (z1 )z1 + k4ddk1
2
1
d 1 d 2
;d2 z22 + jz2 jk#^ dk1 + 4B 2 jz2 jk^ dk1 ; 4B 2 d2 z22 :
1 (2.110)
Using Young's inequality twice gives
^
jz2 jk#^ dk1 d2 z22 + k#4ddk1
2
(2.111)
2 !
d1 jz jk^ k 1 d2 d z 2 + k^ dk21 :
4B 2 2 d 1 4B 2 1 2 2 4d2 (2.112)
The nal upper bound for V2 can now be written as

V_2 ;W (z1 z2) + 1 k^ d k21 + 1 k#^ dk21
1 2 (2.113)
where
1 = 1 + 1 1 = 1

1
4d1 16B d2 2 4d2
2 (2.114)
and
W (z1 z2 ) = (c2 + k1 )z22 + k3 (z24 + 32 z22 ) + ^ (z1 )z1 (2.115)
_
is radially unbounded and positive denite. This implies that V2 < 0 outside a set
R2 in the z1 z2 plane. As in section 6.1, the functions V2 (z ) and W (z ) exhibit the
properties in (2.70). Again, it can be shown that this implies that z( ) is globally
uniformly bounded and that z ( ) converges to the residual set
( ^ 2 ^ 2 !)
R2 = z : jz j 1 2 3 k#dk1 + kdk1 :
; 1 ;1 (2.116)
1 2
Remark 2.4 Notice that the control law (2.97), as opposed to (2.82), requires
knowledge of the coecients in the compressor characteristic, the throttle charac-
teristic and the B-parameter.
Remark 2.5 Once the bounds on the disturbances k^ dk1 and k#^ dk1 are known,
the size of the set R2 in the z1 z2 plane can be made arbitrary small by choosing
the damping factors d1 and d2 suciently large. The same comment applies to the
set R1 dened in (2.87).
Corollary 2.2 (Convergence to the origin)
If the assumptions on #^ d ( ) in equations (2.117) and (2.118) hold, and the fol-
lowing assumption on ^ d ( ) is made:
j^ d ( )j d ( ) 8 0 (2.117)
and
lim ( ) = 0
!1 d
(2.118)
where d( ) is a monotonically decreasing non-negative function. Then, the states
of the model (2.38) converge to the origin.
Proof: By using the same arguments as in the proof of Corollary 2.1, but with
two disturbance terms, it can be shown that

V2 (z ( )) V2 (z (0))e;c + c1 d2 (0)e; 2 + d 2 (=2)
c
1 2 ; c1 2
+ c2 # (0)e 2 + # (=2) :
d d (2.119)
Now lim!1 #d 2 (=2) = 0 and lim!1 d2 (=2) = 0 implies that

lim!1 V2 (z ( )) = 0 and by the positive deniteness of V2 it follows that
lim z ( ) = 0:
!1
(2.120)
Thus we have shown that under the assumptions (2.88), (2.89), (2.117) and (2.118)
on the disturbance terms, z( ) converge to the origin. This also implies that ^( )
and ^( ) converges to the origin and that ( ) and #( ) converge to the point of
intersection of the compressor and throttle characteristic.
A simulation of the response of the control law (2.97) with both mass ow and
pressure disturbances is presented in Chapter 3, where it is compared to a passivity
based controller.
2.3.4 Adaption of Constant Disturbances

A constant or slow varying disturbance in mass ow can cause the equilibrium
of the compression system to be moved into the unstable area of the compressor
map. Therefore, the problem of being able to stabilize the system in this case is
a very important one. In addition constant disturbances in pressure is considered
simultaneously. Consider the following model

_^ = 4B12 l ^ ; ^ T ( ^) ; d
c
^_ = 1 #^ c( ^) ; ^ + d ; u
l (2.121)
c
where d and d are constant and unknown disturbances in mass ow and pres-
sure, respectively. Godhavn (1997) used adaptive backstepping for adaption of
constant or slow varying sea current disturbances in a surface ship model. The
same approach will now be employed here to stabilize (2.121). Two adaption laws
will be designed in order to estimate the unknown disturbances and allow them
to be counteracted by the control.
Theorem 2.4 The controller

u = #lc z1 + c2 z2 ; k3 d3 + 3d z22 ; k2 ^2 ; k1 d + d (2.122)
1
where the estimates d and d are updated with
d_ = ; #1 z1 (2.123)
1
d_ = ; z 1 (2.124)
#2 2
where #11 and #12 are adaption gains, makes the equilibrium of (2.121) globally
asymptotically stable. The states of the model converge to their equilibrium values
and the parameter error dynamics are GAS. asymptotically stable.
Proof: Backstepping is used

Step 1. The error variables z1 and z2 are dened as z1 = ^ and z2 = ^ ; . The
rst clf, V1 , is chosen as
V1 = 2B 2 lcz12 + #21 d~2 (2.125)
where 4 d ; d
d~ = (2.126)
is the parameter error and d is an estimate of d . The time derivative of V1 now
is
V_1 = z1 ;^ T (z1 ) + z2 ; d + ; #1 d~ d_ (2.127)
where (2.121) is used. Let the virtual control be chosen as
= d (2.128)
and the estimate d be updated as
d_ = ; #1 z1 : (2.129)
1
Thus, the terms including d~ in (2.127) are cancelled out, and V_1 can now be
written as
V_1 = z1z2 ; ^ T (z1 )z1 : (2.130)
Step 2. The second clf is chosen as
V2 = V1 + l2c z22 + #22 d~2 = 12 z T Pz + 21 d~ T ;;1 d~ (2.131)
where 4 d ; d
d~ = (2.132)
is the parameter error, d is an estimate of d ,
2
~d = (d~ d~ ) P = 4B lc 0 and ;;1 = #1 0 : (2.133)
0 lc 0 #2
Using (2.121), V_2 is calculated as
V_ 2 = ;^ (z1 )z1 + z2 (#^ c ( ^) ; u + #lc z1 + d ) ; #2 d~ d_ : (2.134)
1
Let the estimate d be updated as
d_ = #1 z2 (2.135)
2
and the control be chosen as
;
u = #lc z1 + c2 z2 ; k3 3 + 3z 22 ; k2 ^2 ; k1 + d : (2.136)
1
Using the calculations from the case of time varying disturbances it can be shown
that V_2 can be upper bounded as
V_2 ;(c2 + k1 )z22 ; k3 (z24 + 32 z22) ; ^ T (z1 )z1 (2.137)
By inserting the update laws (2.129) and (2.135) and the control (2.136) in equa-
tions (2.121), it can be shown that the error dynamics get the following form

z_ = P ;1 Az (z d) + P z z + d~ (2.138)
d~_ = ;;z (2.139)
where

Az (z d) = ;c2z2 ;;k^3T(z(3z1+) 32z2) and P z = ;01 1
0 : (2.140)
2
The stability result follows from application of LaSalle's theorem. From (2.137) it
is seen that
V_2 0 ) z 0 ) z_ 0: (2.141)
Inserting (2.141) into (2.138) we nd that d~ 0. Thus the origin of (2.138) is
GAS.
This result could also have been stated by using the more general Theorem 4.12
in Krstic et al. (1995a) or the results in Krstic (1996).
Remark 2.6 The stability properties of the proposed scheme of two parameter
update laws and control law, can also be established through passivity analysis of
the error dynamics.
The model (2.138)-(2.139) has some useful properties that will be taken advantage
of in the stability proof: The vector ;Az consists of sector nonlinearities, the
matrices C and ; are both diagonal and positive denite, and the matrix P z is
skew symmetric. Consider the positive function
S (z) = 21 zT Pz (2.142)
and calculate the time derivative of (2.142) along solution trajectories of (2.138):
d
dt S (z) = z P z_
T

= zT Az (z d) + P z z + d~
= zTAz (z d) + zT d~: (2.143)
The last step follows from the fact that P z is skew symmetric. As both elements
in Az (z d) are sector nonlinearities, the following inequality will always hold:
;zT Az (z d) > 0: (2.144)
Integrating (2.143) from 0 to t and using (2.144) results in
Zt Zt
zT()d~ ()d = S (z()) ; S (z(0)) + ;zT Az (z d)d: (2.145)
0 0
From this dissipation inequality it is concluded that the mapping d~ 7! z is strictly
passive with S (z) = z T Pz as storage function and ;zT Az (z d) as dissipation
rate. Inspection of (2.139) reveals that the mapping z 7! ;d~ is a passive integrator,
and thus the system (2.138)-(2.139) is a negative feedback interconnection of a
strictly passive and a passive system. This implies that the equilibrium (z d~ ) =
(0 0) is globally uniformly stable and that z ( ) converges to the origin as t ! 1.
The structure of the system is shown in Figure 2.6.
Remark 2.7 The result of this section also holds if either of the two constant
disturbances are set to zero. In the case of d 0 and d =6 0, the controller
(2.122) with parameter update law (2.124) is equivalent to an ordinary linear PI-
control law in z2 : Zt
u = c2 z2 ; 1 z2 ()d:
#2 (2.146)
0
R
By the same arguments as in the above proof, z = 0 is GAS and d = 0t #12 z2()d
converges to the true value of d .
h
d~ z
z_ = C (Az (z d) + Pz + d~ )
;
,
,, ;
ll
l
Figure 2.6: Negative feedback interconnection of strictly passive system with passive
system
2.4 Control of Rotating Stall

In this section CCV controllers will be derived for the full Moore-Greitzer model
(2.34). As in the section on surge control, two cases will be studied here as well.
First, a stabilizing stall/surge controller is designed for the undisturbed model.
Then, pressure disturbances are included in the model, and a second controller
is derived. For use in the stability proofs of this section, the following lemma is
needed:
Lemma 2.1 The squared amplitude of rotating stall has an upper bound:
9 Jmax < 1 such that J ( ) < Jmax 8 > 0 (2.147)
Proof:
The proof is similar to that of section 2.4 in Krstic et al. (1995a):
2 J 1 4W !

J_ = J 1 ; W ; 1 ; 4 ; 2 3H %
2
2 ; 2 ; %J 4W
= ; J4% ; %J W 2 W 3H 2
2
; J4 % + W 2% J jj + %J 4W jj
3H 2
2
; J8% ; %J8 J ; W 16 jj ; 32W jj
3H 2 (2.148)

W 2 , J ( ) will decay faster than the solution w( )
When J ( ) > 16j( )j W1 + 32H
of the dierential equation
2
w_ = ; w8 % (2.149)
2.4 Control of Rotating Stall 39
so that an upper bound of J ( ) is given by

4 Jmax
2W
J ( ) < J (0) % + 16 sup j( )j W1 + 3H 2 = (2.150)
1 + J (0) 8 0< min
where min > 0 is a lower bound on the throttle gain. Equation (2.150) states that
J is bounded if is bounded. An upper bound on is given by the choking of the
mass ow:
choke (2.151)
so that a conservative value for Jmax is
J (0) 1 2W

Jmax = + 16 choke W + 3H 2 : (2.152)
1 + J (0) %8 min
The following assumption on the mass ow coecient is also needed:

Assumption 2.1 The lower bound on is given by the minimum value obtained
during a deep surge cycle. This lower bound is negative, and is given by
9 m > 0 such that ( ) > ; m 8 > 0: (2.153)
The assumption is not a conservative one. Compressors in general are not intended
or designed for operation with reversed mass ow, so it is reasonable to assume
that the lowest value of mass ow is reached during the extreme condition of deep
surge, more precisely at the negative peak of a deep surge cycle.
2.4.1 Undisturbed Case

Theorem 2.5 Consider the system (2.34) evolving in the set

A = # J j # 2 IIR 2 '; m choke ] J 2 IIR+
(2.154)
with the controller
u = (c2 + c3 )( ; 0 ) (2.155)
where c2 > am and am is the maximum positive slope of the compressor charac-
teristic, and c3 > 0 is chosen according to
cmin max
3 < c3 < c3 : (2.156)
The origin of the closed loop system is then asymptotically stable with a region of
attraction equal to A.
Proof:
The backstepping methodology of Krstic et al. (1995a) is now employed to design
a controller for (2.34)
Step 1. Let the error variables z1 and z2 be chosen as

z1 = ^ and z2 = ^ ; (2.157)
and the CLF for this step be
V1 = 2B 2 lc z12: (2.158)
The time derivative of V1 along solution trajectories is

V_1 = z1 ;^ T (z1 ) + z2 + : (2.159)
As before, the throttle is assumed passive, such that ^^ ( ^) 0 8 ^. It follows
that
^^ T ( ^) 0 ) ;z1^ T (z1 ) 0: (2.160)
It is recognized that there is no need to cancel out terms in (2.159). Thus, the
stabilizing function is not needed and can be chosen as = 0, which in turn
gives
V_1 = ;^ T (z1 )z1 + z1 z2 : (2.161)
Although = 0 here, the notation of z1 and z2 is kept in the interest of consistency
with section 2.4.2

; 1 ; W 2J ; u :

z_2 = l1 ;z1 + #^ c(z2 ) ; 34H J W 2 2 (2.162)
c
The clf for this step is
V2 = V1 + l2c z22 + %J1 J (2.163)
max
and V_2 is calculated as
2
V_2 = ;z1^ T (z1 ) + z2(#^ c (z2 ) ; u) + J J ;1
1; W
max
!
; J4 ; 12 43W ; 3H ; 1 z J ; W 2 J z : (2.164)
H 4 W 2 2 2 2
By choosing u according to
u = (c2 + c3 )z2 (2.165)
where c2 > 0 and c3 > 0 are constants, V_2 can be written
4
X
V_2 = _ V2 i : (2.166)
i=1
The four terms in (2.166) are
_
V2 = ;z1 ^ T (z1 ) (2.167)
_ 1
V2 = z2 (#^ c (z2 ) ; c2 z2 ) (2.168)
_
2 2 !
V2 = JJ 1; W
;1 ; 12 43W
H (2.169)
3 max
_
V2

= ; J z2
P () J (2.170)
4 z2
where " 3H ; #
1 ; 1 + 4W 22 :
3H ;
P( )= 4Jmax 8 W (2.171)
8 W ; 1 + 4W 22 c3

Due to the passivity of the throttle, V_2 < 0. As shown in section 2.3, if c2 is
1
chosen as 2
c2 > am a > 4kk2 ; k1 (2.172)
3
_ am is the maximum positive slope of the compressor characteristic, then
where
V2 < 0.
2

By choosing the controller gains suciently large, V_2 can be made negative
3
for > 0. As the proposed controller (2.165) is of the same form as (2.40), the
_
expression (2.59) for , with gain (c2 + c3 ), can be used. For V2 to be negative
3
the following must be satised
2
;1
1; W ; 43W (c2 + c3 )
H + 0 <0 (2.173)
+ >0
3 H + 0
2 !
c2 + c3 > 4W ; W ;1 +1 (2.174)
4 G1 ()
H ( + ) 2 ; =

c2 + c3 > 43W 0 W W2 (2.175)
Simple calculations shows that the maximum of G1 () is reached for = W ; 20 .
The maximum is given by
3H 2
G 0
max
>0 1 () = 4W 2 4W + W + 0 : (2.176)
By choosing
2
c2 + c3 > 43WH2 4W0 + W + 0 (2.177)
which implies
2 4 C (c2 )
c3 > 43WH2 4W0 + W + 0 ; c2 = (2.178)
1
_
it follows that V2 is made negative for > 0.
3 _ _
For < 0 the sum V2 + V2 can be made negative for > ; m , that is
3 2
!
; 1 2 ; 1 4W
z2 (#^ c (z2 ) ; c2 z2) < ; J J 1 ; W (2.179)
max 2 3H
J < 1, and rearranging (2.179) gives
Using (2.59) and Jmax
c3
+ 0 > G2 ( c2 ) (2.180)
where
!
4 4W z2 #^ c (z2 ) ; c2 z 2 + 1 ; ; 1 2 ; c2 :
G2 ( c2 ) = (2.181)
3H 2 W + 0
It can be shown that it is sucient that (2.180) is satised at the end points of
the interval:
G2 (; m c2 ) < m c (2.182)
; 3
m 0
G2 (0 c2 ) < 0: (2.183)
By inspection of (2.181) it is easily shown that (2.183) is always satised. It is
assumed that 0 > m , so that the condition in (2.182) can be rewritten as
c3 < m ; 0 G2 (; m c2 ) =
4 C 2 (c2 m ): (2.184)
m

Plots of V_2 , V_2 and their sum are shown in Figure 2.7.
2 3
Finally, c3 must be chosen so that P is positive denite for ; m < < choke .
The determinant of P is given by
; 1 + W2
2
det P = 4Jc3 ; 38H W 4 2 > 0: (2.185)
max
A plot of det P is shown in the lower part of Figure 2.7. Again, a sucient
condition for det P > 0 for ; m < < choke is that det P ( choke ) > 0 and
det P (; m ) > 0. As shown in Appendix E, this leads to the following conditions
on c3 :
n o n o
max C 3 (c2 choke ) C 4 (c2 m ) < c3 < min C 3 (c2 choke ) C 4 (c2 m )
(2.186)
; m V_ 2
V_ 3
0.5 V_2 + V_ 3
V_
0
−0.5
−0.2 −0.1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8

0.1
0.08
choke
0.06
det P
0.04
0.02
0
−0.2 −0.1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8

Figure 2.7: Upper plot illustrates that V_2 + V_3 < 0 for > ; m . Lower plot
illustrates that det P > 0 for < choke .
Choosing c3 according to (2.186) ensures that

_
V2 4 < 0 ; m < < choke : (2.187)
The requirements (2.172), (2.178), (2.184) and (2.186) are summarized as

2
c2 > am a > 4kk2 ; k1 (2.188)
3
and
n o 4 min
c3 > max C 1 (c2 ) C 3 (c2 choke ) C 4 (c2 m ) = c3 (2.189)
n o 4 max
c3 < min C 2 (c2 m ) C 3 (c2 choke ) C 4 (c2 m ) = c3 (2.190)
Provided c2 and c3 are chosen according to (2.188) to (2.190), V_2 is upper bounded
as
V_2 ;U (z1 z2 J ): (2.191)
As V2 : A ! IIR, V2 (0) = 0, V2 is positive denite and continuously dierentiable
on A, and V_ (z ) < 0 for z 2 A ; f0g, the origin of the system is asymptotically
stable with region of attraction equal to A.
By doing the same calculations as in Remark 2.1, the control law for the CCV
gain is found to be: s
= (c ++ c0 ) : (2.192)
2 3
Notice that this control law requires sensing of mass ow only. The analysis
leading to this control law was conservative, resulting in a large value of the pa-
rameter c3 , that is c3 has to be chosen as c3 > cmin
3 . As was the case in Krstic et
al. (1995b), the controller should be implemented with a lower gain than dictated
by the Lyapunov analysis. Consider the closed loop Jacobian of the model (2.34)
with control law (2.155):
0 ; 12 a 1 0
1
4B lc 4B 2 l
Acl = B
@ ; l1c ; 4l3cHW 1 a ; cc2 +c3
lc lc 2
0 CA (2.193)
0 0 % ; W02 + 2W0 ; 4W%6(cH2 +c3 )
where
^ T (z1 )

a = @ @z and a = @ #^ c(z2 ) : (2.194)
1 z =0 2 z =0
@z
1 2
It can be shown that if Acl is Hurwitz for c3 > c3 , Acl is also Hurwitz for c3 = 0.
min
The closed loop system is therefore locally stable with c3 = 0.
2.4.2 Disturbed Case

The backstepping procedure is now used to design a controller that ensures bound-
edness of the states in the presence of pressure disturbances.
Theorem 2.6 Consider the model (2.37) with time varying pressure disturbances
only, that is #^ d( ) =
6 0 and ^ d ( ) 0, evolving on the set A. Let the controller u
be dened as
u = (c2 + c3 )z2 + d2 z2 (2.195)
where d2 > 0, c2 satises
c2 > am (2.196)
where am is an upper bound on the positive slope of the compressor characteristic,
and c3 satises
cmin
3 < c3 < c3
max (2.197)
where cmin max
3 and c3 are dened in (2.189) and (2.190). Then (2.195) makes the
states of the model (2.37) uniformly ultimately bounded and ensures convergence
to a set.
2.5 Simulations 45
Proof:
This result follows by combining the proofs of Theorem 2.3 and Theorem 2.5. The
Lyapunov function
V2 = V1 + l2c z22 + J 1 % J (2.198)
max
will have an upper bound on its time derivative given by
^2
V_2 ;U (z1 z2 J ) + #4dd( ) (2.199)
2
where 4
X
U (z1 z2 J ) = V_2 : i
(2.200)
i=1

and the four terms V_2 are given by (2.167)-(2.170) provided the control is
i
chosen as (2.195).
Provided c2 is chosen according to (2.188) and c3 satises (2.189) and (2.190),
then V_2 < 0 outside a set R3 . According to Krstic et al. (1995a), the fact that
V2 (z1 z2 J ) is positive denite, radially unbounded and smooth, and U (z1 z2 J )
is positive denite for ; m < < choke implies that there exists class-K1
functions 1 , 2 and a class-K function 3 such that
1 (jz j) V2 (z ) 2 (jz j)

3 (jz j) U (z) (2.201)
where z = (z1 z2 J )T . This implies that z( ) is uniformly ultimately bounded

and that z( ) converges to the set
( ^ 2
!)
R3 = z : jzj 1;1 2 3;1 k#dk1 : (2.202)
1
Remark 2.5 is also applicable to this case.

Remark 2.8 The case of stabilizing rotating stall in the case of disturbances in
mass ow as well as pressure rise, can be studied by combining Theorem 2.3 and
Theorem 2.5. It is straightforward to extend the results on convergence to the
origin and adaption of constant mass ow disturbances and pressure disturbances
from Section 2.3 to the control design in this section.
2.5 Simulations
In this section, the proposed controllers of this chapter are simulated. Results
form both surge and stall control, with and without disturbances, will be shown.
The compressor characteristic
3!
#c( ) = c0 + H 1 + 32 W
;1 ; 1 ;1
2 W (2.203)
of Moore and Greitzer (1986) is used in all simulations. The throttle will be set so
that the intersection of the throttle line and the compressor characteristic is located
on the part of the characteristic that has positive slope, resulting in an unstable
equilibrium. After some time, the controllers will be switched on, demonstrating
that the system is stabilized.
2.5.1 Surge Control
Surge control will now be demonstrated. A Greitzer parameter of B = 1:8 is used

in the simulations. The throttle gain is set to = 0:61 and thus the equilibrium
is unstable. The initial conditions of the systems were chosen as
( 0 0 ) = (0:6 0:6): (2.204)
In Figure 2.8, it is shown how the controller (2.40) with c2 = 1 stabilizes the
system.
In Figure 2.9 noise has been added, and the system is stabilized by the controller
(2.97) with c2 = 1, d1 = 0:3 and d2 = 3.
A simulation of surge induced by a constant disturbance is showed in Figure 2.10.

The compression system is initially operating stably with a throttle setting of
= 0:65 yielding a stable equilibrium. At = 200 the constant disturbances
d = ;0:1 and d = 0:05 are introduced into the system, resulting in the state
of the system being pushed over the surge line. Consequently, surge oscillations
emerge. At = 420 the adaptive controller (2.122) with update laws (2.123), and
as can be seen, the surge oscillations are brought to rest. The parameters of the
controller were c2 = 1:1, #1 = 9 and #2 = 20. The disturbances are unknown to
the controller, but as guaranteed by Theorem 2.4 their estimates converges to the
true values.
2.5 Simulations 47
0.6
0.4

0.2
0
−0.2
0 50 100 150 200 250 300 350 400
0.8
0.6
#
0.4
0.2
0 50 100 150 200 250 300 350 400
0.4
0.3
#v
0.2
0.1
0
0 50 100 150 200 250 300 350 400

Figure 2.8: The throttle gain is set to = 0:61, and the compressor is surging.
The controllers are switched on at = 200.
0.8 0.8
0.7
0.6
0.6
0.4 0.5
#

0.2 0.4
0.3
0
0.2
−0.2 0.1
0 200 400 600 800 0 200 400 600 800

0.1 0.4
0.3
0
d d
#v
0.2
−0.1
0.1
−0.2 0
0 200 400 600 800 0 200 400 600 800

Figure 2.10: Disturbance induced surge stabilized by the adaptive controller (2.122).
0.6
0.4

0.2
0
−0.2
0 50 100 150 200 250 300 350 400
0.8

0.6
#
0.4
0.2
0 50 100 150 200 250 300 350 400

0.2
#v
0.1
0
0 50 100 150 200 250 300 350 400

Figure 2.9: Same situation as in Figure 2.8. However, here disturbances are taken
into account. The pressure and the mass ow disturbances are both white noise
varying between 0:05.
2.5.2 Rotating Stall Control

In this section some simulation results of the proposed controllers for rotating
stall are presented. In Figure 2.11 the response of system (2.34) with controller
(2.165) is shown. The throttle gain is set at T = 0:61, resulting in an unstable
equilibrium for the unactuated system. The initial conditions of the systems were
chosen as
( 0 ) 0 J0 ) = (0:6 0:6 0:05): (2.205)
The B-parameter is set to B = 0:5 for all the rotating stall simulations. Thus, the
compressor enter rotating stall, and J increases until Je is reached. At = 100
the controller is switched on, and J decreases until J = 0 is reached. As discussed
at the end of Section 2.4.1, the controller gains are chosen as c3 = 0 and c2 = 1.
The simulation results are also plotted together with the compressor characteristic
and the in-stall characteristic in Figure 2.12. As the compressor is stabilized it
is seen that the pressure loss associated with the CCV is slightly less than that
associated with operating in rotating stall, which would have been the case if e.g.
the throttle control scheme of Krstic et al. (1995b) had been used.
The eect of pressure disturbances is shown in Figure 2.13. The B-parameter is
B = 0:5 and the compressor stalls. The pressure disturbance is white noise of
amplitude 0:1. Throttle setting and initial conditions were left unchanged. The
control law (2.195) with parameters c2 = 1 and d2 = 0:1 is switched on at = 100,
bringing the compressor out of rotating stall, and damping the disturbances.
2.5 Simulations 49
0.8
0.55
0.7
0.5
0.6
0.45
#

0.4 0.5
0.35
0.4
0.3
0.3
0.25
0.2 0.2
0 100 200 300 400 0 100 200 300 400

3 0.25
2.5 0.2
2
0.15
#v
J
1.5
0.1
1
0.5 0.05
0 0
0 100 200 300 400 0 100 200 300 400

Figure 2.11: Stabilization of rotating stall
0.7
0.6
0.5
#s
0.4
#
0.3 #c
0.2
0.1
T;1 ()

0 0.1 0.2 0.3 0.4 0.5 0.6
Figure 2.12: Same simulation as in Figure 2.11, superimposed on the compressor

characteristic and in-stall characteristic
0.8
0.55
0.7
0.5
0.6
0.45
#

0.4 0.5
0.35
0.4
0.3
0.3
0.25
0.2 0.2
0 100 200 300 400 0 100 200 300 400

3 0.25
2.5 0.2
2
0.15
#v
J
1.5
0.1
1
0.5 0.05
0 0
0 100 200 300 400 0 100 200 300 400

Figure 2.13: Stabilization of rotating stall with pressure disturbances
2.6 Conclusion
In this chapter, anti surge and stall controllers for a close-coupled valve in series
with a compressor have been developed. First, surge was studied, and by the
application of the backstepping methodology, a control law which uses feedback
from mass ow only was derived. Global asymptotic stability was proven. Only an
upper bound on the slope of the compressor characteristic was required to imple-
ment this controller. The controller was used both in the case of no disturbances
and in the presence of pressure disturbances.
A more complicated surge control law was derived for the case of both pressure dis-
turbances and mass ow disturbances. In order to implement this controller, the
compressor characteristic and the B -parameter must be known. Global uniform
boundedness and convergence to a set was proven. By assuming the disturbances
upper bounded by a monotonically decreasing non-negative function, convergence
to the origin was proven. In order to stabilize the compression system in the
presence of constant disturbances, or biases, in mass ow and pressure, an adap-
tive version of the surge controller was derived. This controller ensures global
asymptotic stability.
Then, controllers for rotating stall were considered. The close coupled valve was
incooperated into the Moore-Greitzer model, and controllers were derived that
enables stabilization of rotating stall beyond the surge line. Without disturbances,
an asymptotically stable equilibrium is ensured, and in the presence of pressure
disturbances uniform boundedness was proven.
Chapter 3
Passivity Based Surge
Control
3.1 Introduction
In this chapter passivity will be used to derive a surge controller for a compression
system when time varying disturbances are considered in both pressure as well as
mass ow.
Passivity and input/output methods have been used in many control applications
such as mechanical systems in general, electrical machines, marine vehicles and so
on. To the authors best knowledge this is the rst attempt to apply this method
to the compressor surge control problem.
3.1.1 Motivation
In the previous chapter, backstepping was employed to derive a stabilizing control
law when time varying disturbances were taken into account. The control law
resulting from this approach was
;
u = c2 z2 ; k3 3 + 3z22 ; k2 ^2 ; k1 (3.1)
d ^ d 2
1 ^
+ 4B 2 ;T (z1 ) + + d2 z2 1 + 4B12 :
Of particular interest here, will be the use of input/output methods to design

controllers with disturbance rejection capabilities. This is motivated by the fact
that the surge controller (3.1) seems unnecessary complicated.
52 Passivity Based Surge Control
3.1.2 Notation
A brief introduction to passivity and L2 is given here. For a comprehensive
treatment of these concepts, consult van der Schaft (1996), from which the no-
tation is taken. The signal space L2 consists, somewhat simplied, of all functions
f : IIR+ ! IIR that satisfy Z1
jf ( )j2 d < 1: (3.2)
0
The truncation of f to '0 T ] is dened as
f ( ) 0<T
fT ( ) = 0 T (3.3)
and the set L2e , the extension of L2 , consists of all functions f such that fT 2 L2 .
A mapping G : u 7! y with input u 2 L2e and output y 2 L2e is said to be passive
if there exists a constant so that
ZT
u( )y( )d (3.4)
0
for all u 2 L2e and all T 0. The inner product on L2e is
ZT
hu yiT = u( )y( )d (3.5)
0
and the truncated norm is
kuk2T = hu uiT : (3.6)
A concept that will be used in this chapter is that of strict output passivity. The
mapping G : u 7! y is strict output passive if 9 > 0 and 9 such that
hy ui = hG u ui kG uk2T + 8u 2 L2e 8T 0: (3.7)
3.2 Model
A compression system consisting of a compressor, axial or centrifugal, in series with
a close-coupled valve, a plenum volume and a throttle is studied, consult gure 2.3.
The system is presented in section 2.2.4, and repeated here for convenience:
_^ = 4B12 l ( ^ ; ^ T ( ^)) (3.8)
c
^_ = 1 (#^ c ( ^) ; u ; ^)
l c
where u = #^ v , and the equivalent compressor characteristic is #^ e = #^ c ; #^ v . The
notation x_ is to be understood as dierentiation with respect to nondimensional
time = UtR as in the previous chapter.
3.3 Passivity 53
Assumption 3.1 The throttle is assumed to be a passive component, moreover the
constant 2 > 0 can always be chosen suciently small so that the characteristic
satises the sector condition:
8 ^ 9 2 such that ^ T ( ^) ^ 2 ^2 (3.9)
Our aim will be to design a control law u = #^ v ( ^) for the valve such that the
compressor also can be operated stably on the left side of the original surge line
without going into surge.
3.3 Passivity
3.3.1 Passivity of Flow Dynamics
Consider the non-negative function
V1 ( ^) = l2c ^2 (3.10)
The time derivative of (3.10) along solution trajectories of (3.8) is
V_1 = ; ^ ^ + #e ( ^) ^: (3.11)
Then, it is evident that

D E D ^ ^ Ê
; ^ ^ = ;#e( ) + V1 ( ) ; V1 (0)
T D ^ ^ ÊT
;#e( ) T
; V1 (0) (3.12)
Hence, the ow dynamics
G1 : ; ^ 7! ^ (3.13)
where G1 : L2e ! L2e is an input-output mapping, can be given certain pas-
sivity properties if the equivalent compressor characteristic #^ e ( ^) can be shaped
appropriately by selecting the valve control law #^ v ( ^).
3.3.2 Passivity of Pressure Dynamics

Proposition 3.1 The pressure dynamics
G2 : ^ 7! ^ (3.14)
where G2 : L2e ! L2e is an input-output mapping, are strictly output passive.
Proof : Consider the nonnegative function
V2 ( ^) = 2B 2 lc ^2 : (3.15)
Dierentiating V2 along the solution trajectories of (3.8) gives
V_2 = ^ ^ ; ^( ^) :
^ (3.16)
In view of Assumption 3.1 and (3.16) it follows that
D ^ Ê D^ E ZT
T = ( ^) ^ + V_2 d
T 0
ZT ZT
= ^ T ( ^) d
^ + V_2 d
0 0
ZT
2 2 ( )d + V2 ( ) ; V2 (0)
0
2 k ^k2T ; V2 (0): (3.17)
Hence, D E
G2 ^ ^ T 2 kG2 ^k2T ; V2 (0) (3.18)
and G2 is strictly output passive according to Denition 2.2.1 in van der Schaft
(1996).
Remark 3.1 In Simon and Valavani (1991), the incremental energy

V = l2c ^2 + 2B 2lc ^2 (3.19)
with dierent coecients due to the choice of nondimensional time, was used as a
Lyapunov function candidate. The selection of the functions V1 and V2 used here
is obviously inspired by this function.
3.3.3 Control Law

The following simple control law is proposed:
Proposition 3.2 Let the control law be given by
#^ v = c ^ (3.20)
2
where c > 4kk23 ; k1 + 1 and 1 > 0 is a design parameter. Then the equivalent
compressor characteristic ;#^ e( ^) will satisfy the sector condition
D E ZT
;#^ e ( ^) ^ T 1 ^2 ( )d = 1 k ^k2T (3.21)
0
3.3 Passivity 55
Proof :
The compressor characteristic is dened in equation (2.24). The equivalent com-
pressor characteristic #^ e ( ^) is then given by
#^ e ( ^) = ;k3 ^3 ( ) ; k2 ^2 ( ) ; (k1 + c) ^( ) (3.22)
Consider the inner product
D E ZT
;#^ e ( ^) ^ T
= ^( ) k3 ^3 ( ) + k2 ^2 ( ) + (k1 + c) ^( ) d
Z0 T
= ^2 ( ) k3 ^2 ( ) + k2 ^( ) + (k1 + c) d: (3.23)
0
4 k3 ^2 + k2 ^ + (k1 + c) has a minimum value for ^ = ; k2 .
It is noted that K ( ^) = 2k3
This minimum is calculated to be
2
K ( ^) = k3 ^2 + k2 ^ + (k1 + c) ; 4kk2 + k1 : (3.24)
3
With the choice
2
c 4kk2 ; k1 + 1 (3.25)
3
it it follows from (3.22) that
k3 ^2 ( ) + k2 ^( ) + (k1 + c) 1 : (3.26)
By inserting (3.26) into (3.23) we get
D E ZT
;#^ e ( ^) ^ T 1 ^2 ( )d = 1 k ^k2T (3.27)
0
Provided u = #^ v is chosen as (3.20), it follows that also the ow dynamics are
made strictly output passive, that is
D E
G1 (; ^) ; ^ T 1 kG1 (; ^)k2T ; V1 (0) (3.28)
We now state a stability reslult for the closed loop system G1 G2 , shown in g-
ure 3.1.
Theorem 3.1 The closed loop system G1G2 consisting of the model (2.36) and
control law (3.1) is L2 -stable.
h ;^ G1 ^
;
^
G2
Figure 3.1: The closed loop system G1 G2
Proof :
The closed loop system G1 G2 , shown in gure 3.1, is a feedback interconnection
of the two systems G1 : ; ^ 7! ^ and G2 : ^ 7! ^. As stated in (3.18) and (3.28),
the two mappings satisfy
D E
; ^ G1 (; ^) 1 kG1 (; ^)k2T ; V1 (0) (3.29)
D ^ Ê
G2 2 kG2 ^k2T ; V2 (0) (3.30)
for all T 0 and all ^ ^ 2 L2e . According to the passivity theorem, Theorem
2.2.6 in van der Schaft (1996), G1 G2 is L2 -stable.
3.4 Disturbances
^ d h ^ d ; ^ G1 ^
;
^ + ^ d h ^ d
G2
Figure 3.2: The closed loop system G1 G2 with disturbances
Consider the case when the compression system is subject to disturbances ^ d( )
in mass ow and #^ d( ) in pressure rise as in Section 2.2.5. The model is repeated
3.5 Simulations 57
here for convenience:
_^ = 4B12 l ( ^ + ^ d ( ) ; ^ T ( ^)) (3.31)

c
_^ = 1 (#^ c( ^) ; u ; ^ ; #^ d ( ))
l c
It is assumed that #^ d( ) ^ d ( ) 2 L2e . A stability result for the closed loop system
shown in gure 3.2 is now stated:
Theorem 3.2 The system (3.31) under control (3.20) is L2-stable if the distur-
bances ^ d ( ) in mass ow and #^ d ( ) in pressure rise are taken into account.
Proof :
Redene G1 as G1 : ;( ^ ; #^ d ) !
7 ^ and G2 as G2 : ^ + ^ d 7! ^. The result follows
by repeating the analysis in the preceding sections and replacing ; ^ with ; ^ + #^ d
when establishing the strict output passivity of G1 , and replacing ^ with ^ + ^ d
when establishing the strict output passivity of G2 .
The structure of the closed loop system is shown in gure 3.2. Disturbance rejec-
tion of L2 -disturbances in the Moore Greitzer model is also studied by Haddad
et al. (1997), where throttle control of both surge and rotating stall is consid-
ered. While achieving global results and disturbance rejection for disturbances in
both mass ow and rotating stall amplitude, the controller found by Haddad et
al. (1997) is of high order, requires detailed knowledge of the compression system
parameters and also needs full state feedback.
3.5 Simulations
Now the system (3.31) with control law (3.20) is simulated. The result is given in
gure 3.3. The controller gain was set to c = 1:1, and the controller was switched
on at = 400. The disturbances were
^ d ( ) = 0:15e;0:015 cos(0:2 )
#^ d ( ) = 0:1e;0:005 sin(0:3 ) (3.32)
which has the same structure as the L2 -disturbances considered in Haddad et al.
(1997).
0.6
0.4

0.2
0
−0.2
0 100 200 300 400 500 600 700 800
0.8

0.6
#
0.4
0.2
0 100 200 300 400 500 600 700 800

0.4 0.4
0.3 0.3
#v
#v
0.2 0.2
0.1 0.1
0 0
400 410 420 430 440 0 200 400 600 800

Figure 3.3: Comparison of closed loop response with passivity based (solid lines)
and backstepping based (dashed lines) controllers. The controllers were switched
on at = 400.
Also plotted in Figure 3.3 is the response of the controller (3.1), simulated with
the same disturbances. The parameters for (3.1) were chosen as c2 = 1, d1 = 0:3,
d2 = 0:1. The two set of responses are almost indistinguishable, but in the lower
left plot there is a blown up plot of the CCV pressure drop, and as can be seen there
is a small dierence. Figure 3.3 also illustrates the result of Corollary 2.2, where
the controller (3.1) guarantees convergence to the equilibrium in the presence of
disturbances upper bounded by a monotonically decreasing non-negative function.
The small dierence in control action is due to the low damping (di ) chosen for
(3.1), but with the current disturbances that is all that was needed.
One advantage of the backstepping controller (3.1) compared to the passivity based
(3.20) is that (3.1) ensures convergence to a set when the disturbances are not in
L2 , whereas the approach in this chapter demands that they are in L2 .
3.6 Concluding Remarks

In this chapter the passivity properties of the Greitzer model was used to derive
a surge control law for a close coupled valve. The results in this chapter, when
not taking disturbances into account, are similar to the results in section 2.3.1.
However, using input/output theory and passivity it was possible to show that
the proportional controller u = c ^ yielded an L2 -stable system in the presence of
3.6 Concluding Remarks 59
both mass ow disturbances as well as pressure disturbances. Compared to the
more complicated controller (3.1) the advantages of the input/output approach
over the Lyapunov based approach of backstepping, in this particular case, should
be evident.
However, it is believed that the simple controller u = c ^ also could have been
derived using backstepping, but with considerable more involved algebraic manip-
ulations, and possibly a new choice of Lyapunov function candidate.
It is straightforward to show that the stability result is still valid if the control
law (3.20) is changed as long as the sector condition (3.21) hold. Thus, a more
sophisticated controller could be used in order to e.g. minimize the steady state
pressure drop over the CCV or improve transient performance, and still stability
could be shown.
When comparing with the results on rotating stall control in Haddad et al. (1997),
the passivity based method used here shows promise for developing a simple, low
order, partial state feedback controller when rotating stall is also taken into ac-
count, and still achieving disturbance rejection. This is an interesting open prob-
lem.
Chapter 4
A Moore-Greitzer Type
Model for Axial
Compressors with
Non-constant Speed
4.1 Introduction
In this chapter we propose an extension to the Moore-Greitzer model. Non-
constant rotational speed of the compressor is taken into account. The conse-
quence of this is that the new model includes the B-parameter as a state. Higher
harmonics of rotating stall will also be included in the model.
In the original work of Moore and Greitzer, the compressor speed is assumed
constant. Greitzer and Moore (1986) concluded that low values of Greitzer's B-
parameter B lead to rotating stall, while high value of B leads to surge. High
and low in this context is dependent on the design parameters of the compressor
at hand, as Day (1994) points out. There exists a Bcritical for which higher B -
values will lead to surge and lower B -values will lead to rotating stall. However this
Bcritical is dierent for each compressor. As B is proportional to the angular speed
of the machine, it is of major concern for stall/surge controller design to include
the spool dynamics in a stall/surge-model, in order to investigate the inuence of
time varying speed on the stall/surge transients.
A model for centrifugal compressors with non-constant speed was presented by
Fink et al. (1992). In Gravdahl and Egeland (1997g) a similar model was derived,
and surge and speed control was investigated. However, both the models of Fink
et al. (1992) and Gravdahl and Egeland (1997g) were developed for centrifugal
machines, and do not include rotating stall as a state.
A Moore-Greitzer Type Model for Axial Compressors with
62 Non-constant Speed
The model of Moore and Greitzer (1986) is based on a rst harmonics approxi-
mation of rotating stall, using a Galerkin procedure. A fundamental shortcoming
of the low order three state More Greitzer model is the one mode approxima-
tion. Mansoux et al. (1994) found that higher order modes interact with the rst
harmonic during stall inception. Relaxation of the one mode approximation has
been presented by Adomatis and Abed (1993), Mansoux et al. (1994), Gu et al.
(1996) and Leonessa et al. (1997a), and control designs for such models have been
reported by several authors, see Table 4.1 below. Also, Banaszuk et al. (1997) re-
ports of design of stall/surge controllers without using a Galerkin approximation,
that is for the full PDE model. Leonessa et al. (1997a) highlights the importance
of including higher order modes of rotating stall in the model, and demonstrates
that the control law proposed by Krstic et al. (1995a), which was based on a one
mode model, fails when applied to a higher order model.
In Table 4.1, the development in stall/surge modeling and control is outlined.

Hansen et al. (1981) demonstrated that the model of Greitzer (1976a) also applies
to centrifugal compressors. It seems that the modeling and control of an axial
compression system including both rotating stall and spool speed is an open prob-
lem. The problem of non-constant speed is also listed among topics for further
research in Greitzer and Moore (1986).
Reference(s) states A/C M/C/S

Greitzer (1976a) # A M
Hansen et al. (1981) # C M
several, see de Jager (1995), incl. # A/C C
Gravdahl and Egeland (1997c)
Fink et al. (1992) # B C M
Gravdahl and Egeland (1997g) # B C MCS
Moore and Greitzer (1986) # J A M
Eveker and Nett (1991) # B A MC
several, see de Jager (1995), incl. # J A C
Gravdahl and Egeland (1997d)
Mansoux et al. (1994) # Ji A M
Leonessa et al. (1997a) # Ji A MC
Adomatis and Abed (1993) # Ji A C
Hendrickson and Sparks (1997)
Humbert and Krener (1997)
Gravdahl and Egeland (1997a) # J B A MS
Gravdahl and Egeland (1997e) # Ji B A MS
Table 4.1: Outline of the development in compressor stall/surge modeling and
control. A=Axial, C=Centrifugal, M=Modeling, C=stall/surge control, S=Speed
control. The model of Eveker and Nett (1991) actually uses states with dimensions.
01 E Station numbers
LI LE
Lc
HH
HH
H

ps
pT
Cx A
A
pT
A
Vp

Inlet duct
Exit duct
IGV Compressor Plenum Throttle
Figure 4.1: Principal drawing of the compression system of (Moore and Greitzer
1986). The station numbers are used as subscripts in the following.
4.2 Preliminaries
The compression system is essentially the same as in Chapter 2, and consists of an
inlet duct, inlet guide vanes (IGV), axial compressor, exit duct, plenum volume
and a throttle. The dierences compared to Chapter 2 is that the compressor speed
is non-constant, and a CCV is not included. The system is shown in Figure 4.1.
Our aim is to develop a model for this system in the form
z_ = f (z) (4.1)
where z = ( # Ji B )T 2 IIRq and
is the circumferentially averaged ow coecient
# is the total-to-static pressure rise coecient
Ji is the squared amplitude mode i of angular variation (rotating stall)
B is Greitzer's B-parameter which is proportional to the speed of the com-
pressor.
i = 1 : : : N is the rotating stall mode number. N is the maximum number
of stall modes determined by the gas viscosity.
The dimension of the state space is q = N + 3:
The modeling of the compression system relies heavily on the modeling in Moore
and Greitzer (1986). However, the assumption of constant speed U , and thus
constant B , is relaxed and a momentum balance for the spool is included. The gas
viscosity as introduced by Adomatis and Abed (1993) is also taken into account,
and N modes are included in the model as opposed to the work of Moore and
Greitzer (1986) where a rst harmonic approximation was used.
Moore and Greitzer (1986) dene nondimensional time as
MG = Ut
R (4.2)
where U is the rotor tangential velocity at mean radius, R is the mean compressor
radius and t is actual time in seconds. As we here consider time varying U , this
normalization will not be used. Instead we propose to use
= URdt (4.3)
where Ud is the desired constant velocity of the wheel. Note that if Ud = U = const,
we have that = MG . All distances are nondimensionalized with respect to R,
that is the nondimensional duct lengths, see Figure 4.1, are dened as
lI = LRI and lE = LRE : (4.4)
The axial coordinate is denoted and the circumferential coordinate is the wheel
angle . Greitzer's B-parameter is dened as
r V
4 U
B= p
2as Ac Lc (4.5)
where as is the speed of sound, Vp is the plenum volume, Ac is the compressor
duct ow area and Lc is the total length of compressor and ducts. B and U are
related as
U = bB (4.6)
where s
4
b = 2as AVc Lc (4.7)
p
is a constant.
4.3 Modeling
4.3.1 Spool Dynamics
The momentum balance of the spool can be written
I d!
dt = t ; c (4.8)
4.3 Modeling 65
where ! is the angular speed, I is the spool moment of inertia, and t and c are
the drive (turbine) torque and compressor torque respectively. Using ! = U=R,
(4.3) and (4.5), the spool dynamics (4.8) can be written
s
2as IUd Ac Lc dB = ; : (4.9)
R2 V d t c
p
As in Fink et al. (1992), torques are nondimensionalized according to
t ; c
; = ;t ; ;c = A (4.10)
RU 2 c
where is the constant inlet density. Now, (4.9) can be written
dB = + B 2 (; ; ; ) (4.11)
d 1 t c
where the constant +1 is dened as
4 R Ac b:
+1 =
3
(4.12)
IU d
As the compressor torque equals the change of angular momentum of the uid,
see for instance Mattingly (1996), the compressor torque can be written
c = mc R(C2 ; C1 ) (4.13)
where C1 and C2 are the tangential uid velocity at the rotor entrance and exit
respectively. Assuming that Cx , the ow velocity in the axial direction, is the
same at entrance and exit,
c = mc RCx (tan 1 ; tan 2 ) (4.14)
where 1 and 2 uid angles at the rotor entrance and exit respectively. The uid
angles are time varying. However, Cohen et al. (1996) show that
tan 1b ; tan 2b = tan 1 ; tan 2 (4.15)
where 1b and b2 constant blade angles at the rotor entrance and exit respectively.
The compressor mass ow is given by
mc = Ac U : (4.16)
Combining (4.14), (4.15) and (4.16) gives
c = R Ac 2 U 2 (tan 1b ; tan 2b ) (4.17)
and by (4.10) the nondimensional compressor torque is
;c = 2 (tan 1b ; tan 2b ): (4.18)
In terms of the compressor speed U , the dynamics of the spool can be written
dU = +1 ;U 2 : (4.19)
d b
4.3.2 Compressor
Moore (1984a) gives the pressure rise over a single blade row as
pE ; p1 = F ( ) ; d (4.20)
1 U 2 dt
2
where
= CUx (4.21)
is the local axial ow coecient, F ( ) is the pressure rise coecient in the blade
passage, Cx is the velocity component along the x-axis, and is a coecient of
pressure rise lag. According to Moore and Greitzer (1986), d dt can be calculated
as
d = @

dt @t + @@t : (4.22)
rotor stator
Using (4.3) it is seen that
@
@ @ + @ @
@t rotor = @ @t @ @t
U
= Rd @@ + @@ UR(t) (4.23)
@
Ud @
@t = R @ (4.24)
stator
where the unsteadiness of the ow through the stator passage reects the acceler-
ations associated with transients eects. For the rotor there is also unsteadiness
due to the rotor blades moving with velocity U (t) through a circumferentially
nonuniform ow. Considering a compressor of Ns stages, we get
pE ; p1 = N F ( ) ; 1 2 @ + U @

1 U 2 s 2a @ Ud @ (4.25)
2
where
a=4 R (4.26)
Ns Ud
is a constant. Note that if U ( ) = const and Ud U such that = MG , equation
(4.25) is reduced to equation (5) in Moore and Greitzer (1986). It is noted that
the ow coecient can depend on both and , even though the atmospheric
stagnation pressure pT is constant. The average of around the wheel is dened
1 Z 2 ( )d =
as
4 ( ): (4.27)
2 0
Further
= ( ) + g( ) and h = h( ) (4.28)
where h is a circumferential coecient. As no circulation occurs in the entrance
duct it is clear that the averages of g and h vanish:
Z 2 Z 2
g( )d = h( )d = 0: (4.29)
0 0
4.3 Modeling 67
4.3.3 Entrance Duct and Guide Vanes
The fact that the rotational speed of the wheel now is assumed time varying does
not change the conditions upstream of the compressor. Therefore the equations
stated in Moore and Greitzer (1986) are still valid, and will be presented here.
The pressure dierence over the IGVs, where the ow is axial, can be written
p1 ; p0 = 1 K h2 (4.30)
U 2 2 G
where 0 < KG 1 is the entrance recovery coecient. If the IGVs are lossless
KG = 1. Upstream of the IGV irrotational ow is assumed so that a (unsteady)
velocity potential ~ exists. The gradient of ~ gives axial and circumferential ve-
locity coecients everywhere in the entrance duct. At the IGV entrance point
(denoted by subscript '0') we have
( ~ )0 = ( ) + g( ) and ( ~ )0 = h( ) (4.31)
where partial dierentiation with respect to and is denoted by subscripts.
Bernoulli's equation for unsteady, frictionless and incompressible ow will be used
to calculate the pressure drop in the entrance duct. As in Moore and Greitzer
(1986) and White (1986), this can be written
pT ; p0 = 1 ( 2 + h2 ) + ( ~ ) (4.32)
U 2 2 0
where the term ( ~ )0 is due to unsteadiness in and g. A straight inlet duct of

nondimensional length lI is considered, and the velocity potential can be written
~ = ( + lI )( ) + ~0 ( ) (4.33)
where ~0 is a disturbance velocity potential such that

~0 = 0 ( ~0 )0 = g( ) and ( ~0 )0 = h( ): (4.34)
=;lI
Equation (4.32) can now be written
pT ; p0 = 1 ( 2 + h2 ) + l d + ( ~0 ) : (4.35)
U 2 2 I d 0
4.3.4 Exit Duct and Guide Vanes

Downstream of the compressor the ow is complicated and rotational. As in Moore
and Greitzer (1986), the pressure p in the exit duct is assumed to dier only slightly
from the static plenum pressure ps ( ), such that the pressure coecient P satises
Laplace's equation.
P= 4 ps ( ) ; p r2 P = 0: (4.36)
U 2
The axial Euler equation, see for instance White (1986), is used to nd the pressure
drop across the exit duct. The Euler equation is the inviscid form of the dierential
equation of linear momentum. If integrated along a streamline of the ow, the
Euler equation will yield the frictionless Bernoulli equation. The Euler equation
in the x-coordinate, neglecting gravity, can be written as
dp = dCx :
; dx (4.37)
dt
Employing the chosen nondimensionalization of time and distance,
d = Ud d and d = 1 d (4.38)
dt R d dx R d
we get
dpE = ; U 2 1 (P ) (4.39)
dx Rn E o
dCx = Ud d ( ~ ) U : (4.40)
dt R d 0
Inserting (4.39) and (4.40) in (4.37) we get the following expression for the axial
Euler equation, evaluated at E , where time varying U has been taken into account
n o
(P )E = Ud2 d ( ~ )0 U
U d
= UUd2 ( ~ )0 U + ( ~ )0 dU
d : (4.41)
From (4.31) we have
( ~ )0 = ( ) + ( ~0 )0 : (4.42)
Inserting (4.31) and (4.42) into (4.41) we get

(P )E = UUd dd + ( ~0 )0 + UUd2 dU d ( ) + ( ~0 )0 : (4.43)
At the duct exit, = lE , we want that p = ps , that is P = 0. Thus, when
integrating (4.43), the constant of integration is chosen such that

P = UU(d ) ( ; lE ) dd ; ~0 + UUd2 dUd ( ; l E )( ) ; ~0 : (4.44)
Finally, we get
ps ; pE = (P ) = Ud ;l d ; (m ; 1)( ~0 )

U 2 E U ( ) E d 0

+ UUd2 dU
d ;l E ( ) ; ( m ; 1)( ~0 )0 (4.45)
where, as in Moore (1984b) and Moore and Greitzer (1986), the compressor duct
ow parameter1 m has been included. It is noted that if U ( ) =const and Ud U
such that = MG , equation (4.45) is reduced to equation (20) in Moore and
Greitzer (1986).
1 m = 1 for a very short exit duct, and m = 2 otherwise.
4.3 Modeling 69
4.3.5 Overall Pressure Balance
Using the preceding calculations, we now want to calculate the net pressure rise
from the upstream reservoir total pressure pT to the plenum static pressure ps at
the discharge of the exit duct. This is done by combining equations (4.25), (4.30),
(4.35) and (4.45) according to
ps ; pT = (NF ( ) ; 1 2 ) ; (l + l Ud + 1 ) d
U 2 2 I E U a d

+ (1 ; m) UUd ; 1 ( ~0 )0 ; 21 (1 ; KG)h2

+ UUd2 dU
d ;lE ( ) ; (m ;1)( )0
~0
; 21a 2( ~0 )0 + UU ( ~0 )0 (4.46)

d
where
2 @@ + @@ = 2 @@ + 2 @g
@ + @
@g
= 2 @@ + 2( ~0 )0 + ( ~0 )0 (4.47)

has been used. By dening 2
; pT
#( ) = ps U 2
#c ( ) = NF ( ) ; 21 2
lc (U ) = lI + lE UUd + a1
mU (U ) = (1 ; m) UUd ; 1 (4.48)
and assuming KG 1, (4.46) can be written
#( ) = #c( ) ; lc(U ) dd + mU (U )( ~0 )0 (4.49)

+ UUd2 dU
d ;lE ( ) ; (m ; 1)( )0
~0
; 21a 2( ~0 ) + UU ( ~0 ) ; ( ~0 )

d 0
which, when assuming U = Ud and = 0, reduces to equation (26) in Moore and
Greitzer (1986). In (4.49) the -term is included in order to account for viscous
transportation of momentum in the compressor. Viscosity was rst introduced
2 Notice that l = L R. As in Fink et al. (1992) L , used in the denition of B , is a constant.
c 6 c c
See equation (4.5).
into the Moore-Greitzer model by Adomatis and Abed (1993), and included in
the analysis of Gu et al. (1996) and Hendrickson and Sparks (1997). Adomatis
and Abed (1993) concludes that the eect of viscous damping must be included
in a multi mode Moore-Greitzer model. This to ensure that the large velocity
gradients associated with higher modes will be damped out. Without viscosity, all
the modes would have the same amplitude in fully developed rotating stall.
Equation (4.49) requires knowledge of the disturbance velocity potential ~0 and its
derivatives. As ~0 satises Laplace's equation r2 ~0 = 0 it has a Fourier series. This
Fourier series has N terms, where the number N is dependent on . According to
Adomatis and Abed (1993) and Gu et al. (1996),
1 for = 0
N = nite for > 0 (4.50)
By using (4.6), equation (4.49) can be written in terms of B , and by integration
of (4.49) over one cycle with respect to , we get
d U d ;+ 1 1 Z 2
#( ) + lc (B ) d + lE b ( ) = 2 c ( )d (4.51)
0
which is the annulus averaged momentum balance.
4.3.6 Plenum Mass Balance

The mass balance in the plenum can be written
d
d ( p Vp ) = mc ; mt (4.52)
where p is the plenum density, mc is the mass ow entering the plenum from the
compressor, and mt is the mass ow leaving through the throttle. Assuming the
pressure variations in the plenum isentropic, we have
dpp = a2s (m ; m ) (4.53)
d Vp c t
which is shown in detail in Appendix B. By nondimensionalizing pressure with
U 2 , mass ow with UAc , transforming to nondimensional time , and taking
account for (4.5), (4.53) can be written
d# = +2 ( ; ) ; 2 dB # (4.54)
d B T B d
where the constant +2 is dened as
4 R b:
+2 = (4.55)
LU c d
Using (4.11) we get
d# = +2 ( ; ) ; 2+ ;B #: (4.56)
d B T 1
4.3 Modeling 71
The model of the compression system now consists of the torque balance for the
spool (4.11), the local momentum balance (4.49), the annulus averaged momentum
balance (4.51) and the mass balance of the plenum (4.56).
The compression systems characteristics are taken from Moore and Greitzer (1986).
The usual third order polynomial steady state compressor characteristic presented
in Chapter 2 is used, and repeated here for convenience:

#c( ) = c0 + H 1 + 23 ( W ; 1) ; 21 ( W ; 1)3 (4.57)
where the parameters c0 , H and W are dened in Chapter 2.
The throttle characteristics is again taken to be
p
T (#) = T #: (4.58)
4.3.7 Galerkin Procedure

In order to transform the PDE (4.49) into a set of ODEs, a Galerkin approximation
is employed. The disturbance velocity potential ~0 is represented by the function
( ~0 ) :
X
N
( ~0 ) = () W en An ( ) sin(n ; rn ( ))
n (4.59)
n=1
where An ( ) is the amplitude of mode number n of rotating stall and rn ( ) n =
1 : : : N () are unknown phase angles. The residue Rn , for use in the Galerkin
approximation, is dened as
4 ( ~0 ) ; ( ~0 )0
Rn = (4.60)
0
and using (4.49) and (4.59), Rn can be calculated as

Rn = Wn dA n drn
d sin n ; An d cos n (4.61)
; m 1(B ) #( ) + lc(B ) dd + lE Ud;+ b

1 ; # ( + WA ( ) sin )
c n n
B
+ U d ;+1 (m ; 1) W W dA n dr n
An ( ) sin n + 2a 2 d sin n ; An d cos n
bn
Un 2
+ U An ( ) cos n + n An ( ) sin n
d
where
4 n ; rn ( ):
n = (4.62)
The Galerkin approximation is calculated using the weight functions
h1 = 1 h2 = sin n h3 = cos n (4.63)
and the inner product
1 Z 2
hRn hi i = 2 Rn ( )hi ( )d: (4.64)
0
Calculating the moments
Mi = hRn hi i for i = 1 2 3 (4.65)
results in
M1 = 21 #( ) + lc(B ) dd + lE Ud;+ 1
b ( )
Z 2
+ #c ( + WAn ( ) sin n ) dn
0
(m ; 1) A ( ) + n2 A ( )
M2 = 21 Ud;+1bn n 2Wa n

+ dA n 1 ; mB (B )
d a n (4.66)
Z 2
+ #c ( + WAn ( ) sin n ) sin n dn
0
M3 = 21 ; dr n 1 mB (B ) ; bBn A( )
d a ; n 2aUd
Z 2
+ #c ( + WAn ( ) sin n ) cos n dn
0
It is recognized that, due to #c( ) being even in , the last term in M3 vanishes.
Demanding M3 = 0 and assuming An 6= 0 the phase angles rn must satisfy
drn = n2 Ubd B = n2 b
d 1 ; mBn(B)a 2Ud(n ; mB (B )a) B: (4.67)
Notice that time varying B implies that the phase angles rn are not constant, which
was the case in the constant speed, rst harmonic model of Moore and Greitzer
(1986) and the constant speed, higher order models of Adomatis and Abed (1993)
and Gu et al. (1996).
4.3.8 Final Model

By evaluating the integrals in (4.66) using (4.57), demanding M1 = M2 = 0, and
using (4.11)and (4.56) the following model for the compression system is found:
4.3 Modeling 73
d = H ; # ; c0 + 1 + 3 ( ; 1)(1 ; J )
d lc (B ) H 2 W
2
; 21 ( W ; 1)3 ; lE UbH
d ;+1 (4.68)
d# = +2 ( ; ) ; 2+ ;B # (4.69)
d B T 1
dJn = J 1 ; ( ; 1)2 ; Jn ; n2 W
d n W
2 U ;+ ( m ; 1) W
4 33aH aHn
; d 31bHn (n ; mB (B )a)W (4.70)
dB = + ;B 2 (4.71)
d 1
where n = 1 : : : N (), Jn is dened as the square of the stall amplitude An

4 A2 ( ):
Jn ( ) = (4.72)
n
and
4 1 X
N
J ( ) = N () ()Jn ( ): (4.73)
n=1
The model (4.68)-(4.71) is in the desired form of (4.1).
In the case of pure rotating stall, dJdn = 0 and ; = 0, the equilibrium values of Jn
is found from (4.70) to be
2 2W
!
;1
Jne = 4 1 ; W ; n3aH (4.74)
which corresponds to the result by Gu et al. (1996).

If the Galerkin approximation is carried out with only one term in the Fourier
expansion of ~0 and assuming = 0, as was done in Gravdahl and Egeland (1997a),
equation (4.70) is changed to
dJ1 = J 1 ; ( ; 1)2 ; J1 ; 2Ud;+1(m ; 1)W 3aH

d 1 W 4 3bH (1 ; mB (B )a)W : (4.75)
The rest of the model is left unchanged, and the model is similar to that of Moore
and Greitzer except for time varying B .
It should be noted that the model developed in this chapter also can be reduced
to other well known compression system models. This is summed up in table 4.2.
Changes made Redene Gives the model of
to (4.68)-(4.71) nondim.
time acc. to
U = Ud, ; = 0
J 0, dB
d = 0 := Ut
R Greitzer (1976a)
N = 1, = 0
;=0
dB = 0, U = Ud := Ut Moore and Greitzer (1986)
d R
N = 1, = 0
J 0 := t!H Fink et al. (1992)
N = 1, = 0 (Centrifugal)
N = 1, = 0 { Gravdahl and Egeland (1997a)
U = Ud, ; = 0
dB = 0 := Ut Gu et al. (1996)
d R
m = tanh(nlI )
U = Ud, ; = 0 := Ut
R Adomatis and Abed (1993)
dB = 0
d
Table 4.2: In this table it is shown how the model derived in this chapter can be
reduced to other known 12 form the literature. !H is the Helmholtz frequency
Acmodels
dened as !H = as LcVp .
4.4 Simulations
Here some simulations of the model developed in this chapter will be presented.
For speed control, a simple P-type controller of the form
;t = cspeed (Ud ; U ) (4.76)
will be used. The nondimensional drive torque ;t is used as the control, and
feedback from compressor speed U is assumed. In Gravdahl and Egeland (1997g)
compressor speed was controlled in a similar manner, and stability was proven
using Lyapunov's theorem. The desired speed was set to Ud = 215m/s in both the
following simulations. Numerical values for the parameters in the model are given
in Appendix D.
4.4.1 Unstable Equilibrium, = 0:5

In Figure 4.2 the response of the model (4.68)-(4.71) with speed control (4.76) and
cspeed = 1 is shown. Only the rst harmonic J1 of rotating stall is included in this
simulation. Initial values were chosen as
( # J1 B )0 = (0:55 0:65 0:1 0:1) (4.77)
4.4 Simulations 75
such that the ( #)-trajectory starts on the stable part of the compressor char-
acteristic as can be seen in Figure 4.3.
The throttle gain was set at = 0:5 so that the equilibrium is to the left of the local
maximum of the characteristic. As can be seen from Figure 4.2, the compressor
goes into rotating stall as B (and thus compressor speed U ) is low. Moreover,
when the applied torque from the speed controller cause B to increase, the stall
amplitude J falls o and the compressor goes into surge. This is what could be
expected according to Greitzer and Moore (1986). The surge oscillations have a
period of 180, which correspond to a surge frequency of about 10Hz. A desired
speed of Ud = 215m=s corresponds to a desired B-parameter of Bd = Ud =b = 2:23.
After 1500 this value is reached. As the compressor torque ;c varies with ,
see equation (4.18), we would expect oscillations in speed U as the compressor is
in surge. This is conrmed by the lower right plot in Figure 4.2. In the upper
plot of Figure 4.3, the trajectory starts on the stable part of the characteristic,
then rotating stall occurs and the trajectory approaches the intersection of the
throttle and in-stall characteristics. As B increases the resulting surge oscillations
are clearly visible.
Now, the model is simulated using N = 3, that is three harmonics. Initial values
were chosen as
( # B J1 J2 J3 )0 = (0:55 0:65 0:1 0:1 0:05 0:01): (4.78)
The upper plot in Figure 4.4 shows the response of the three rst harmonics of
the squared amplitude of rotating stall J1 , J2 and J3 . The response is otherwise
similar to that of Figure 4.2. The lower plot is a magnied version of the 200
rst time units of the upper plot, and shows that during stall inception the second
harmonic J2 dominates the rst harmonic J1 . This emphasizes the importance of
using higher order approximations of the Moore-Greitzer model. This phenomenon
was also observed by Mansoux et al. (1994) for constant speed compressors.
The previous simulations have both used a desired speed of Ud = 215m=s resulting
in a high B-parameter and driving the compressor into surge. Now, the desired
speed is changed to Ud = 75m=s which corresponds to a B-parameter of 0:78.
This is suciently low for the compressor studied here to become stuck in rotating
stall. The simulation is shown in Figure 4.5 using three harmonics of the squared
amplitude of rotating stall. The rst harmonic J1 has the highest equilibrium
value. This equilibrium value decreases with mode number n due to the eect of
viscosity, as stated in equation (4.74).
0.8 1
0.6 0.8
#

0.4 0.6
0.2 0.4
0 0.2
−0.2 0
0 500 1000 1500 0 500 1000 1500

4 2
B
1
3
0
0 500 1000 1500
J
2 216
U
215
0 214
0 500 1000 1500 800 1000 1200 1400

Figure 4.2: Simulation of the system (4.68)-(4.71). Low B leads to rotating stall,
and high B leads to surge.
0.8
#
0.6
#s
0.4
#c
0.2
0
−0.2 −0.1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8

Figure 4.3: Simulations result superimposed on the compression system character-
istics. The compressor characteristic, the in-stall characteristic and the throttle
characteristic are drawn with solid, dashed and dash-dot lines respectively.
4.4 Simulations 77
J1 J2 J3
2

0 500 1000 1500
3
J1 J2 J3

0 20 40 60 80 100 120 140 160 180 200
Figure 4.4: The three rst harmonics of rotating stall. The rst harmonic J1 has
the highest maximum value. This maximum value decreases with mode number du
to the eect of viscosity. The lower plot shows that J2 dominates J1 during stall
inception.
0.8 1
0.6 0.8
#

0.4 0.6
0.2 0.4
0 0.2
−0.2 0
0 500 1000 1500 0 500 1000 1500

4 1
0.8
3
J1 J2 J3
0.6
B
2
0.4
1
0.2
0 0
0 500 1000 1500 0 500 1000 1500

Figure 4.5: System response with low Ud , resulting in the compressor being stuck
in rotating stall.
4.4.2 Stable Equilibrium = 0:65
In Figure 4.6, the impact of the spool dynamics on a stable equilibrium of the
compression system is illustrated. Only J1 is used in this simulation, and the
initial values in (4.77) were used. Similar plots can be produced including higher
harmonics. The throttle gain was set at = 0:65, giving a stable equilibrium, and
the speed controller gain was again chosen as cspeed = 1. The solid trajectories
show the system response to a speed change from U = 0:05m=s to U = 215m=s. It
can be seen that the acceleration of U aects the other states of the model. This is
due to the couplings with speed U and torque ; in the model. Of special interest
here, is the stall amplitude. The initial value of J (0) = 0:05 grows to nearly fully
developed rotating stall as the machine is accelerating, but is quickly damped out
as desired speed is reached. Simulations show that this stalling can be avoided by
accelerating the compressor at a lower rate, that is by using a smaller cspeed . This
could possibly also be achieved with other, more advanced speed controllers, and
is a topic for further research.
In contrast, the dashed trajectories in Figure 4.6 show the response without the
spool dynamics. Now, the initial value J (0) = 0:05 is damped out very quickly,
and and # converges to their equilibrium values. The transient eects observed
are due to the initial conditions.
0.8 0.8
0.7 0.7
0.6 0.6
#

0.5 0.5
0.4 0.4
0.3 0.3
0.2 0.2
0 500
1000 1500 0 500
1000 1500
3 2.5
2.5 2
2
1.5
B
J
1.5
1
1
0.5 0.5
0 0
0 500
1000 1500 0 500 1000 1500
Figure 4.6: Stable equilibrium with and without B-dynamics

4.5 Concluding Remarks 79
4.5 Concluding Remarks
In this chapter, a multi mode Moore-Greitzer axial compressor model with spool
dynamics was derived. This resulted in a model with time varying B -parameter.
Through simulations it was demonstrated that the model was capable of demon-
strating both rotating stall and surge, and that the type of instability depended
on the compressor speed. Compressor speed was controlled with a simple pro-
portional control law. In the original Moore Greitzer model only the rst mode
of rotating stall is included. The simulations in this chapter show that during
stall inception, higher order modes can dominate the rst mode. This is in accor-
dance with known results, and is shown here to be valid also for variable speed
compressors.
Further work on this topic includes 1) Stability analysis and stall/surge control
design for variable speed axial compressors and 2) The use of simultaneous speed
control and stall/surge control to achieve rapid acceleration without stalling the
compressor.
Chapter 5
Modelling and Control of
Surge for a Centrifugal
Compressor with
Non-constant Speed
5.1 Introduction
In this chapter the compressor characteristic for a variable speed centrifugal com-
pressor is investigated. This characteristic is incooperated in a model similar to
that of Fink et al. (1992), and surge phenomena is studied in connection with
varying compressor speed. Energy losses in the compressor components is used to
derive the characteristic, making a departure from the usual cubic characteristic
most commonly encountered in the surge control literature. Inspired by Ferguson
(1963) and Watson and Janota (1982), uid friction and incidence losses, as well as
other losses, in the compressor components are modeled, and a variable speed com-
pressor characteristic is developed based on this. Both annular and vaned diusers
are studied. Surge and speed controllers for variable speed centrifugal compressors
are presented and analyzed. The speed is controlled with a PI-control law.
Axial and centrifugal compressors mostly show similar ow instabilities, but in
centrifugal compressors, the matching between components, such as impeller and
diuser, inuences the stability properties. According to Emmons et al. (1955),
de Jager (1995) and others, rotating stall is believed to have little eect on cen-
trifugal compressor performance. The modeling and analysis in this chapter is
thus restricted to surge. Hansen et al. (1981) showed that the model of Greitzer
(1976a) is also applicable to centrifugal compressors, and this model will also be
used here.
Since compressors are variable speed machines, it is of interest to investigate the
Modelling and Control of Surge for a Centrifugal Compressor with
HH
HH pp
p01 p2
m c A A mt
T01 A A
CCV Vp

Lc
Figure 5.1: Compression system with CCV.
inuence of speed transients on the surge dynamics. Models describing this in-
teraction was developed in Eveker and Nett (1991) for axial compressors and in
Fink et al. (1992) for centrifugal compressors. As surge can occur during acceler-
ation of the compressor speed, it is of major concern to develop controllers that
simultaneously can control both surge and compressor speed.
For surge control, the CCV is again employed. Semi-global exponential stability
results for the proposed controllers are given using Lyapunovs method. The results
are conrmed through simulations.
5.2 Model
We are considering a compression system consisting of a centrifugal compressor,
close coupled valve, compressor duct, plenum volume and a throttle. The throttle
can be regarded as a simplied model of a turbine. The system is showed in gure
5.1. The model to be used for controller design is in the form
2
p_p = aV01 (m ; mt )
p
A
m_ = L1 (p2 ; pp ) (5.1)
c
!_ = 1I (t ; c )
where m is the compressor mass ow, pp is the plenum pressure, p2 is the pressure
downstream of the compressor, a01 is the inlet stagnation sonic velocity, Lc is
the length of compressor and duct, A1 is the area of the impeller eye (used as
reference area), I is the spool moment of inertia, t is the drive torque and c
is the compressor torque. The two rst equations of (5.1) are equivalent to the
model of Greitzer (1976a) which was also used in Chapter 2 and 3, whereas the
whole model (5.1) is similar to the model of Fink et al. (1992). In addition to
the assumptions used in the derivation of model of Greitzer (1976a), it is now
5.2 Model 83
assumed that the gas angular momentum in the compressor passages is negligible
compared to rotor angular momentum. Note that in the model used throughout
this chapter, the states are with dimension, as opposed to the previous chapters,
where nondimensional states were used.
The angular speed of the compressor ! is included as a state in addition to mass
ow and pressure rise which are the states in Greitzer's surge model. The equation
for p_p follows from the mass balance in the plenum, assuming the plenum process
isentropic, the derivation is shown in Appendix B. The equation for m_ follows
from the impulse balance in the duct. In the following, the model (5.1) will be
developed in detail. In particular, expressions must be found for the terms p2
and c . It will also be shown that an expression for the compressor characteristic
results from this derivation.
vaned diuser
!
r2
r1
t
r h1
impeller impeller
eye (inducer)
Figure 5.2: Diagrammatic sketch of a radially vaned centrifugal compressor. Shown

here with a vaned diuser.
The calculation of the compressor pressure rise will be based on energy transfer and
energy losses in the various parts of the compressor. In the following sections, the
dierent components of the centrifugal compressor will be studied. The centrifugal
compressor consists essentially of a rotating impeller which imparts a high velocity
to the gas, and a number of xed diverging passages in which the gas is decelerated
with a consequent rise in static pressure. A schematic drawing of a compressor is
shown in Figure 5.2. The innermost part of the impeller is known as the inducer,
or the impeller eye, where the gas is sucked into the compressor. The part of the
compressor containing the diverging passages is known as the diuser. The diuser
can be vaned, as in Figure 5.2, or vaneless. A vaneless diuser (also known as an
annular diuser) is a simple annular channel with increasing area. The choice
of diuser type depends on the application of the compressor. After leaving the
diuser, the gas may be collected in a volute (also known as a scroll), as shown in
Figure 5.3.
!
Impeller and
diuser
Figure 5.3: Diagrammatic sketch of centrifugal compressor tted with a volute.
The energy transfer to the gas takes place in the impeller. In the ideal case, this
energy is converted into a pressure rise. However, a number of losses occur in
the compressor, the main ones being friction losses and incidence losses in the
impeller and the diuser. These losses shape the compressor characteristic, and
the incidence losses are the cause of the positive slope of the characteristic, which
in turn determines the area of the compressor characteristic where surge occurs.
Therefore, these components will be studied in detail in the following. A number of
other losses, e.g. losses in the volute, are taken into account as drops in eciency.
5.2.1 Impeller
Incoming gas (air) enters the impeller eye (the inducer) of the compressor with
velocity C1 , see Figure 5.4. The mass ow m and C1 is given by
C = 1 m
1 01 A1 (5.2)
where 01 is the constant stagnation inlet density. The tangential velocity U1 , at
diameter D1 , of the inducer is calculated as
U1 = D21 ! = D1 N (5.3)
where ! is the angular velocity of the impeller and N is the number of revolutions
per second. The average diameter D1 is dened according to
D12 = 21 (Dt21 + Dh2 1 ) (5.4)
5.2 Model 85
where Dt1 and Dh1 are the diameters at inducer tip and hub casing respectively.
The circle with diameter D1 and area A1 divides the inducer in two annuli of equal
area.
C1
U1
U1
Q
Q
Q 1b
Q
Q
Q Ca1
Q C1
W1 QQ
Q 1
Q
Figure 5.4: Velocity triangle at inducer. Section through inducer at radius r1 =

D1 =2.
From Figure 5.4 it is seen that the gas velocity W1 relative to the impeller is
W12 = C12 + U12 ; 2U1C1 : (5.5)
The gas leaves the impeller at the impeller tip with velocity C2 as shown in Fig-
ure 5.5. The diameter at the impeller tip is D2 and the tangential tip velocity is
U2 .
C2
U2
b
b
b
b
W2 bb C a2
b C2
2 bb 2
b
! ;
;
;
D2
Figure 5.5: Velocity triangle at impeller tip.

5.2.2 Diuser
Centrifugal compressors are usually tted with either a vaneless (annular) or a
vaned diuser. The inuence of the diuser upon compressor performance cannot
be over emphasized, as a considerable proportion of the uid energy at the im-
peller tip is kinetic energy and its ecient transformation into static pressure is
important Ferguson (1963).
Annular Diuser
The annular diuser is a simple annular channel in which the uid loses velocity
and gains static pressure. One disadvantage of the annular diuser is its size, as
its outlet radius must be twice its inlet radius if the velocity is to be halved in it
Ferguson (1963). Its advantages are its price and wide range of operation.
Vaned Diuser
In a vaned diuser, vanes are used to guide the ow so that he overall rate of
diusion is higher than in an annular diuser. This leads to a smaller size, but
higher production costs. The vaned diuser has a higher eciency but less mass
ow range than the annular diuser. This is due to stalling of the diuser vanes
for low mass ows.
5.3 Energy Transfer, Compressor Torque and Ef-

ciency
5.3.1 Ideal Energy Transfer
For turbomachines, applied torque equals the change in angular momentum of the
uid:
c = m(r2 C2 ; r1 C1 ) (5.6)
where c is the compressor torque, r1 = D21 , r2 = D22 and C2 is the tangential
component of the gas velocity C2 . Power delivered to the uid is
W_ c = !c = !m(r2 C2 ; r1 C1 )
= m(U2 C2 ;1 C1 ) = m%h0cideal (5.7)
where %h0cideal is the specic enthalpy delivered to the uid without taking ac-
count for losses. Equation (5.7) is known as Euler's pump equation. For simplicity
the following two assumptions are made:
1. A radially vaned (no backsweep) impeller is considered with 2b = 90 , and
2. There is no pre-whirl, that is 1 = 90 ) C1 = 0.
5.3 Energy Transfer, Compressor Torque and Eciency 87
The slip factor is dened as
4 C 2 1 ; 2 :
= (5.8)
U 2 i
The approximation is known as Stanitz' formula where i is the number of compres-
sor blades. From (5.7), and using (5.8), we have that the ideal specic enthalpy
delivered to the uid is
_
%h0cideal = Wcideal
m = U2 :
2 (5.9)
Notice that %h0cideal is independent of mass ow m, and ideally we would have the
same energy transfer for all mass ows1. However, due to various losses, the energy
transfer is not constant, and we now include this in the analysis. According to
Watson and Janota (1982), Ferguson (1963), Nisenfeld (1982) and other authors,
the two major losses, expressed as specic enthalpies, are:
1. Incidence losses in impeller and diuser, %hii and %hid
2. Fluid friction losses in impeller and diuser, %hfi and %hfd
The incidence losses and uid friction losses play an important role in determining
the region of stable operation for the compressor. Other losses, such as back ow
losses, clearance losses and losses in the volute will be taken into account when
computing the eciency of the compressor. There also exist other losses such as
inlet casing losses, mixing losses and leakage losses, but these will be ignored in the
following. For a further treatment on this topic, some references are Balje (1952),
Johnston and Dean, Jr. (1966), Whiteld and Wallace (1975) and Cumpsty (1989).
5.3.2 Compressor Torque

Using the assumption of no pre whirl and equation (5.6), the compressor torque is
c+ = mr2 C2 = mr2 U2 m > 0: (5.10)
The torque calculated in (5.10) is for forward ow. However, the compressor may
enter deep surge, that is reversal of ow, and there is need for an expression for the
compressor torque at negative mass ow. According to Ko and Greitzer (1986),
an axial compressor in reversed ow can be viewed as a throttling device. Here
it is assumed that a centrifugal compressor in reversed ow can be approximated
with a turbine. This allows for the use of Euler's turbine equation:
c; = m(r1 C1 ; r2 C2 ) = ;mr2 U2 m < 0: (5.11)
Combination of (5.10) and (5.11) gives
c = jmjr2 U2 8m (5.12)
which is in accordance with the compressor torque used by Fink et al. (1992).
1 If backswept impeller blades, < 90 , were considered, h would decrease with
b2 0cideal
increasing m.
Disc Friction Torque
The torque set up by the rotation of the impeller in a uid is modeled as rotation
of a disc. When a disc is rotated in a uid, a resistive torque is set up given by
Z r2
df = 2 tan 2r2 dr (5.13)
r1
where r1 is the radius of the shaft, r2 is the radius of the disc and tan is the
tangential component of the shear stress between the disc and uid. According to
Ferguson (1963) the torque can be approximated by
5 2 5
df = Cm r2 2 ! = 2CDm2 r2 U12 (5.14)
1
where Cm is a torque coecient depending on the disc Reynolds number and the
space between the disc and the casing, is the density of the uid and ! is the
angular velocity of the disc. In (5.14) it has been assumed that r1 << r2 .
5.3.3 Incidence Losses

The losses due to incidence onto the rotor and vaned diuser play an important
role in shaping the compressor characteristic. There exists several methods of
modeling this loss, and a comparative study is given by Whiteld and Wallace
(1973). The two most widely used approaches are:
(1) The so called \NASA shock loss theory" reported in Watson and Janota
(1982) and Whiteld and Wallace (1973), which is based upon the tangential
component of kinetic energy being destroyed.
(2) A constant pressure incidence model reported by Whiteld and Wallace
(1973) where it is assumed that the ow just inside the blades has adapted
to the blades via a constant pressure process.
Watson and Janota (1982) concludes that for centrifugal compressors, the dier-
ences between the two models are small. According to Whiteld and Wallace
(1973), the main dierence lies in the prediction of the incidence angle at which
zero loss occurs. For model (1) zero loss is predicted when the ow angle at the
inlet equals the blade angle. This is not the case for model (2). Based on this,
and the simplicity of (1), the NASA shock loss theory is used here. As Ferguson
(1963) points out, the term shock loss is misleading as nothing akin to shock oc-
curs in practice, but the simple notion of shock is used to explain the shape of the
compressor characteristic.
Depending on whether the mass ow is lower or higher than the design ow,
positive or negative stall is said to occur. The use of model (2) leads to a loss
varying with the square of the mass ow, symmetrical about the design ow.
Ferguson (1963) states that the incidence loss in practice increase more rapidly
C1
U1
Q
e eeQ
b 1

e eQ1Q

e e Q

e e i QQ C a 1
C1
W1b e e W1 QQ
e e Q 1
e e W 1 Q
e e
e
e
e
e
e
Figure 5.6: Incidence angles at inducer.
with reduction of ow below design ow, than with increase of ow above the
design ow. This leads to a steeper compressor characteristic below the design
point than above. According to Sepulchre and Kokotovic (1996) and Wang and
Krstic (1997b), such a characteristic is said to be right skew.
Impeller
The velocity of the incoming gas relative to the inducer is denoted W1 . In o-
design operation there will be a mismatch between the xed blade angle 1b and
the direction of the gas stream 1 = 1 (U1 C1 ), as shown in Figure 5.6. The angle
of incidence is dened by
4 1b ; 1 :
i = (5.15)
As the gas hits the inducer, its velocity instantaneously changes its direction to
comply with the blade inlet angle 1b . The direction is changed from 1 to 1b , and
the kinetic energy associated with the tangential component W1 of the velocity
is lost. That is, the incidence loss can be expressed as
2
%hi = W21 : (5.16)
From Figure 5.6 it is easily seen that
; C1 and sin = Ca1 :
cos 1 = U1 W (5.17)
1 W
1 1
Furthermore,
1b ; 1 ) W = (cos ; cot sin ) W :
W1 = sin(sin (5.18)
1 1 1b 1 1
1b
e
e
e
e
e
C2i e
; e
i; e
C2b ; Ca2 e
; 2 e
; 2b e
; e
C2
U2
Figure 5.7: Incidence angles at diuser.
Inserting (5.17) in (5.18) gives

W1 = U1 ; C1 ; cot 1b Ca1 : (5.19)
and the incidence loss (5.16) can be written
2
1b m
%hi = 21 (U1 ; C1 ; cot 1b Ca1 )2 = 12 U1 ; cot
01 A1 (5.20)
where the second equality if found using (5.2). Similar results are presented in
Chapter 5 in Ferguson (1963).
Diuser
According to Watson and Janota (1982), the losses in the vaned diuser can be
modeled with friction/incidence losses in a similar manner as in the impeller.
Similar to the inducer incidence loss, it is assumed that the velocity of the uid
entering the diuser is instantaneously changed to comply with the xed diuser
inlet angle 2b . The direction is changed from 2 to 2b , and the kinetic energy
associated with the tangential component C2i of the velocity is lost, see Figure 5.7.
That is, the incidence loss can be expressed as
2
%hid = C22i : (5.21)
Using Figure 5.7 it is seen that
%hid = 12 (C2 ; cot 2b Ca2 )2
= 12 (U2 ; cot 2b Ca2 )2 : (5.22)
For simplicity the choice2 Ca1 = Ca2 is made. The diuser inlet angle 2b , is now
designed such that there is minimum incidence loss in both impeller and diuser
for the same mass ow m. For i = 0, we have that
U1 = Ca1 cot 1b ) Ca2 = U1 tan 1b : (5.23)
From Figure 5.7 and (5.23), it follows that
tan = Ca2 = U1 tan 1b
2b C2 U2 (5.24)
and
2b = atan D1D
tan 1b (5.25)
2
and consequently the diuser incidence loss (5.22) can be written

%hid = 12 DD2 U1 ; m cotA2b : (5.26)
1 01 1
5.3.4 Frictional Losses

Impeller
According to Ferguson (1963) loss due to friction can be calculated as
2
%hfi = Ch Dl W21b (5.27)
where Ch is the surface friction loss coecient3 , l is the mean channel length and
D is the mean hydraulic channel diameter. This friction loss is actually calculated
for constant area pipes of circular cross section. The friction loss coecient Ch is
dened as in Watson and Janota (1982):
Ch = 4f (5.28)
where the friction factor f depends on the Reynolds number. Many dierent
formulas for the friction factor have been published, see e.g. Ferguson (1963) or
White (1986). Here we will use Blasius' formula:
f = 0:3164(Re);0:25: (5.29)
According to White (1986), (5.29) was found empirically for turbulent ow in
smooth pipes with Reynolds number Re below 100:000. The mean hydraulic chan-
nel diameter D is dened as
D = 4A a (5.30)
2 This is a design choice, and other choices will lead to dierent expression for the diuser
angle 2b
3 The friction loss considered here is due to friction in the impeller. According to Watson and
Janota (1982) diusion losses in the impeller are small compared to impeller friction losses, but
they may be included in the analysis by choosing C to suit.
h
where the cross section area A and perimeter a are mean values for the passage.
The mean hydraulic diameter D corresponds to a circle with area A and perimeter
a. Although the passages between the blades in the compressor are neither circular
nor of constant area, Balje (1952) reports of good agreement between theory and
measurement using (5.29).
Using Figure 5.6, it is seen that
W1b W1
sin 1 = sin 1b (5.31)
and using sin 1 = CWa11 we get
W1b = sinC1 : (5.32)

1b
Inserting (5.2) and (5.32) in (5.27) gives
%hfi = Ch l m2 = kfi m2 : (5.33)
2D 1A21 sin2 1b
2
As can be seen the friction losses are quadratic in mass ow and independent of
wheel speed U . Equation (5.33) represents the loss due to friction of a mass ow
m through a pipe of hydraulic diameter D.
Diuser
The loss due to uid friction in the diuser can be modeled in a similar manner
as in the impeller:
%hfd = kfd m2 : (5.34)
In the vaned diuser a pipe friction loss is calculated for each diuser passage.
5.3.5 Eciency
The isentropic eciency of the compressor is dened as, see e.g. Cumpsty (1989)
or any other text on turbomachinery,
i (m U1 ) = %h %h0cideal : (5.35)
0cideal + %hloss
where the %hloss -term is the sum of the friction and incidence losses from the
previous sections. Furthermore, the eciency will be corrected with losses in
the volute and the additional losses arising from clearance and back-ow. The
eciency is also dependent on the diusers ability to convert the kinetic energy of
the ow into pressure. Collection the various losses, the isentropic eciency from
equation (5.36) is adjusted to
i (m U1 ) = %h %h0cideal ; %bf ; %c ; %v ; %d (5.36)
0cideal + %hloss
where
%hloss = %hif + %hii + %hdf + %hdi : (5.37)
The additional losses are discussed below. In Figure 5.8, the eciency is plotted.
In the upper plot, the compressor is equipped with a vaned diuser and in the
lower plot with an annular diuser. As can be seen, the vaned diuser oer a
higher eciency, but a narrower range of ow compared with the annular diuser.
N = 20000 N = 35000 N = 50000

0.85
0.8
ivaned
0.75
0.7
0.65
0.6
0 0.1 0.2 0.3 0.4 0.5 0.6
mass ow 'kg/s]
0.85
0.8
iannular
0.75
0.7 N = 50000
0.65 N = 20000 N = 35000
0.6
0 0.1 0.2 0.3 0.4 0.5 0.6
mass ow 'kg/s]
Figure 5.8: Eciencies for compressor with vaned (upper plot) and annular (lower
plot) diuser.
Clearance
Pampreen (1973) found that the clearance loss of a centrifugal compressor can be
approximated by
%c = 0:3 lbcl (5.38)
where lcl is the axial clearance and b is the impeller tip width.
Backow
The back-ow loss occurs because the compressor has to reprocess the uid that
has been reinjected into the impeller due to pressure gradients existing in the
impeller tip region. Due to the lack of accurate modeling of this loss, Watson and
Janota (1982) suggest a loss of 3 points of eciency as typical:
%bf = 0:03: (5.39)
Volute
In the volute a loss will take place mainly due to the inability of the volute to use
the radial kinetic energy out of the diuser. Cumpsty (1989) assumes this loss to
lie within 2-5 point of eciency:
0:02 %v 0:05: (5.40)
This loss is likely to be higher for compressor with a vaned diuser than with an
annular diuser, as a larger part of the total kinetic energy at the outlet of the
vaned diuser is in the radial direction. A more comprehensive treatment of loss
in volutes can be found in Lorett and Gopalakrishnan (1986).
Diusion
The purpose of the diuser is decelerate the ow with high kinetic energy, and thus
convert this into pressure. This can be achieved more or less ecient depending
on the construction of the diuser. Due to inadequate diusion in the diuser
there will be a degradation %d in the eciency i . The eciency drop %d
is dependent on the pressure recovery coecient, see e.g. Cumpsty (1989) or
Watson and Janota (1982), but for simplicity %d will be considered constant
here. According to Watson and Janota (1982) vaned diusers oer a 2 to 7 points
increase in eciency compared to annular diusers.
5.4 Energy Transfer and Pressure Rise

Including the losses, the total specic energy transfer can be calculated by sub-
tracting (5.33), (5.20), (5.26) and (5.34) from (5.9):
%h0c (U1 m) = %h0cideal ; %hif ; %hii ; %hdf ; %hdi : (5.41)
%h0c is a second degree polynomial in m, and as opposed to the ideal case, we see
that energy transfer to the uid is varying with mass ow m. This is shown in
Figure 5.9.
To nd an expression for the pressure rise we now need a relation between pressure
rise and energy transfer. The pressure rise is modeled as
; 1
p2 = 1 + i (m UT1 )%c h0cideal p01 = #c (U1 m)p01 (5.42)
01 p
5.4 Energy Transfer and Pressure Rise 95
where the losses have been taken into account, and #c (U1 m) is the compressor
characteristic. It is shown in Appendix B how to arrive at (5.42). We now have
an expression for the pressure p2 needed in the model (5.1). Notice that for each
speed N , both the pressure rise (5.42) and the eciency (5.36) reach maximum
for the same value of mass ow m. Thus, the maximum eciency is reached on
the surge line, stressing the need for active control in order to be able to operate
safely in the neighborhood of the surge line. The inlet stagnation temperature T01 ,
specic heat capacity cp and are assumed constant. The ideal energy transfer
and the losses are shown for a compressor with annular diuser in gure 5.9. In
the case of vaned diuser, the curves look similar, but with a steeper slope due
to the incidence losses at the diuser. The curves are calculated for a compressor
speed of N = 35 000rpm.
4
x 10
5 %h0cideal = U22
specic energy 'J/kg]
4.8
4.6
%h0c %hf
4.4
4.2
%hi
4
3.8
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
mass ow 'kg/s]
4
x 10
2
specic energy 'J/kg]
1.5
1
%hf
0.5 %hi
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
mass ow 'kg/s]
Figure 5.9: Energy transfer for N = 35 000 rpm. Compressor with annular dif-
fuser.
The compressor pressure characteristic as calculated from equation (5.42) is showed

in Figure 5.10.
In gures 5.9 and 5.10 the numerical values for the compressor parameters is
taken from Fink et al. (1992). Comparing the compressor map in Figure 5.10 with
2.5 2.5
N = 50000
N = 50000
N = 45000
2
N = 45000 2
pressure rise
pressure rise
N = 40000
N = 40000
N = 35000 N = 35000
1.5 1.5
N = 30000 N = 30000
N = 25000 N = 25000
N = 20000 N = 20000
1 1
0 0.2 0.4 0.6 0.8 0 0.2 0.4 0.6 0.8
mass ow 'kg/s] mass ow 'kg/s]
Figure 5.10: Centrifugal compressor characteristic. The left plot is for an annular
diuser, and the right for a vaned diuser. The dashed lines to the right are choke
lines, also known as stone walls.
Figure 3 in Fink et al. (1992), which is based on physical measurements, we see

that they are almost similar.
The surge line is the line in the compressor map that divides the map into an area
of stable compressor operation and unstable (surge) operation. The line passes
through the local maxima of the constant speed lines in the map, and is drawn
with a solid line in gure 5.12.
5.5 Choking
When the ow reaches sonic velocity at some cross-section of the compression
system, the ow chokes. Assuming isentropic ow, Dixon (1978) calculated the
choking ow for the components most likely to choke in centrifugal compressors,
the impeller eye (the inducer) and at the entry of the diuser.
The eect of choking can be seen in Figure 5.10, where a choke line, also known
as a stone wall, has been drawn. In this paper, the eect of choking is treated in
a approximate manner. Due to sonic eects, the pressure rise would fall o more
5.6 Dynamic Model 97
gradually when approaching the stone wall than shown in Figure 5.10. Dixon
(1978) showed that the mass ows for which choking occurs are:
Choking in impeller
2 U1 2 3 2((+1)
;1)
2 + ( ; 1)
mchoke (U1 ) = A1 01 a01 64 +1
a 01 75 (5.43)
where p01 and a = pRT

01 = RT (5.44)
01 01
01
is the inlet stagnation density and inlet stagnation sonic velocity, respec-
tively. It is seen than the choking mass ow is dependent on blade speed
U1 . Thus the impeller can accept a greater limiting mass ow rate at higher
rotational speeds. By inserting U1 = 0 in (5.43) the expression for choked
ow through a nozzle is found. This is shown in Appendix B.
Choking in the diuser entry
2 2((+1)
1 + ( ; 1)imp aU012 2 ;1)
mchoke (U2 ) = A4 01 a01 r U2 2 + 1 (5.45)
1 + ( ; 1) a01
where A4 is the ow area of the diuser entry, and imp is the impeller
eciency.
5.6 Dynamic Model

To complete the dynamic model (5.1), an expression for the throttle mass ow is
needed. The mass ow mt through the throttle is modeled as
p
mt = kt pp ; p01 (5.46)
where kt is the throttle gain proportional to throttle opening and pp is the plenum
pressure. The momentum balance of the spool is
I !_ = t ; c : (5.47)
Using (5.3) it is seen that
_
! = 2DU1 ) !_ = 2DU1 (5.48)
1 1
and thus we get a dierential equation for U1 ,
U_ 1 = D2I1 (t ; c ) : (5.49)
The drive torque t may be delivered by a turbine, and will be used as a control
variable for speed control. The compressor and spool can only rotate in one
direction, and the speed is assumed upper bounded:
0 U1 (t) < Um : (5.50)
Using (5.12) and (5.49), and inserting equations (5.42) and (5.46) in (5.1), we get
the following dynamic model for the compression system
2 p
p_p = aV01 (m ; kt pp ; p01 )
p !
A i ( m U 1 )% h0 cideal
; 1
m_ = L 1 + T01 cp p01 ; pp (5.51)
U_ 1 = D2I1 (t ; c ) :
It is worth noticing that a time varying U is equivalent with a time varying B-
parameter Fink et al.q(1992). Greitzer's B-parameter as dened in Greitzer (1976a)
is given by B = 2Ua011 AV1 pLc where Vp is the plenum volume and Lc is the length
of the compressor and duct. Using (5.49), a nonlinear dierential equation for B
can be found.
5.7 Surge Control Idea

The reason for equilibria to the left of the surge line being unstable, and causing
the compressor to go into surge, is the positive slope of the characteristic in this
area. From Figure 5.9 it is seen that the positive slope is due to the incidence losses
at low mass ows. From the expression for the incidence loss, equation (5.20), it
is clear that a variable blade angle 1b would make it possible to minimize the
incidence losses over a range of mass ows. Thus variable inducer blades might be
used as a means of surge stabilization.
On the other hand, the maximum energy transfer and minimum incidence loss do
not occur for the same mass ow. This is due to the friction losses. The friction
shifts the point of maximum energy transfer, and consequently pressure rise, to
the left of the point of minimum incidence loss. From this, we conclude that the
friction losses in fact have a stabilizing eect, and introducing additional uid
friction would move the point of maximum energy transfer to the left. The result
of this is that the surge line will be shifted to the left, and the area of stable
compressor operation is expanded.
This motivates us to introduce a valve in series with compressor. The valve will
introduce a pressure drop into the system, and the characteristic of the valve will
have the same qualitative impact on the equivalent compressor characteristic as
introducing more uid friction. The pressure drop over this valve will serve as
the control variable, and it will be used to introduce additional friction at low
5.8 Controller Design and Stability Analysis 99
mass ows in order to avoid surge. The use of a close coupled valve for constant
speed centrifugal compressor surge control was studied by Dussourd et al. (1977),
Jungowski et al. (1996) and Pinsley et al. (1991).
5.8 Controller Design and Stability Analysis

The equivalent compressor characteristic for compressor and close coupled valve
is dened as
#e (m U1 ) = #c(m U1 ) ; #v (m) (5.52)
where #v (m)p01 is the pressure drop across the CCV and
; 1
#c (m U1 ) = 1 + i (m UT1 )%c h0cideal : (5.53)
01 p
Assume p0 , m0 to be the equilibrium values of pressure and mass ow as dictated
by the intersection of the throttle and compressor characteristics, and Ud to be
the desired spool speed. Dene the following error variables
p^ = pp ; p0 m^ = m ; m0 U^ = U1 ; Ud: (5.54)
The equations of motion (5.51) are now transformed so that the origin becomes the
equilibrium under study. Notice that no assumptions are made about the numeric
values of m0 and p0 , so that the equilibrium can be on either side of the surge
line. The equilibrium values is calculated using the same method as described in
section 2.3.2. Dene
m^ t (^p) = mt (^p + p0 ) ; m0 (5.55)
#^ c (m
^ U^ ) = #c(m^ + m0 U^ + Ud) ; p0 (5.56)
^#v (m^ U^ ) = #v (m^ + m0 U^ + Ud ) ; p0 (5.57)
By including the CCV (5.52), and using (5.55)-(5.57), the model (5.51) can be
written in the form
2
p^_ = Va (m^ ; m^ t (^p))
p
A
m^_ = L #^ c (m ^ U^ ) ; #^ v (m^ ) p01 ; p^ (5.58)
U^_ 1 = D2I1 (^t ; ^c )
where a hat denotes transformation to the new coordinates (5.54), and (^p m^ U^ )T =
(0 0 0)T is the equilibrium. The expressions for the characteristics m^ t (^p), #^ c(m
^ U^ )
^
and #v (m^ ) are found in a similar manner as in Chapter 2. From (5.12) it is known
that 2 2
c = D
2D
2 jmjU = D2 sgn(m)mU :
1 2D 1 (5.59)
1 1
As in Nicklasson (1996), a scaled hyperbolic tangent function will be used to
mimic the signum function in the analysis in order to avoid additional diculties
regarding continuity:
2
c = D 2 tanh( m )mU
2D & 1 (5.60)
1
where & > 0 is a suciently large constant. The torque ^c is dened as
^c = c ; co (5.61)
and calculated as
2
c = D 2 tanh( m )(m
2D1
^
& ^ + m0 )(U + Ud)
2
= D 2 tanh( m )(m
2| D1
^ ^ d + m0 U^ )
& ^ + m{z0 )(m^ U + mU }
^c
2
+D 2 tanh( m )m U :
0 d (5.62)
| 1 {z &
2D }
c0
By choosing
^t = t ; co (5.63)
the last equation in (5.58) follows from the last equation in (5.51).
Theorem 5.1 The surge control law
#^ v = kv m
^ (5.64)
and the speed control law
^t = ;kp U^ ; ki I^
I_^ = U
^ (5.65)
where (
^ ^ U^ )
)
kp > 0 ki > 0 and kv > sup @ #c@(m
m^ + 1 (5.66)
^m
U ^
and 1 > 0, makes the origin of (5.58) semi-global exponentially stable. That is,
desired compressor speed Ud is achieved whether the equilibrium is to the right or
to the left of the original surge line. The integral term in (5.65) is added in order
to robustify the controller with respect to unmodeled disturbance torques.
Proof:
Dene
z =4 UI^^
2I
and P =4 D1
ki (5.67)
where > 0 and ki > 0 are design parameters. Consider the following Lyapunov
function candidate
V (^p m ^ I^) = 1 (Vp^ + Vm^ + Vspool )
^ U (5.68)
2
where
Vp^ = a2V p p^2 Vm^ = A L m^ 2 and Vspool = z T Pz: (5.69)
01 01 1 01
As all coecients in (5.68) are constant it follows that V is positive denite and
radially unbounded, provided that is chosen such that P > 0, that is
r 2Ik
< i
D1 : (5.70)
Calculating the time derivative of (5.68) along the solutions of (5.58) and account-
ing for (5.65) gives

V_ = m^ #^ (m
c ^ U^ ) ; #^ (m^ ) p01 ; 1 p^m
v 01 01 t ^ (^p) ; k U^ 2p
+U^ 2 ; k2iID1 I^2 ; k2pID1 U^ I^ ; U^ ^c ; D 1^

2I I ^c : (5.71)
The last term in (5.71) can be upper bounded as

; D 1 I^^ = ; D22 tanh( m )mU ^ ^
2I c 4I & ^ + m0 U I
D 2 Um m^ 2
4I 2 ^ 2 ^ ^
2 1 + 1 I + m0 U I (5.72)
using (5.12), (5.50) and Young's inequality. The parameter 1 > 0 can be chosen
freely. The U^ ^c -term can be upper bounded as
2
;U^ ^c = ; D2D2 tanh( m& ) (m^ + m0 )U^ + mU^ d U^
2
1
2 2
; 2D tanh( m& )mU^ 2 + D4D2 m^ + 2 (Ud U^ )2 : (5.73)
D 2
1 1 2
Now, (5.71) can be upper bounded as
2
^ U^ ) ; #^ v (m^ ) p 01 ; D
V_ m^ #^ c(m 2 tanh( m )mU^ 2
01 2D1 &
D 2 Um D 2 1
+ 8I2 + 4D2 m^ 2 ; p^m^ t (^p) ; zT Rz (5.74)
1 1 2 01
where 0 D22 Ud2 2
D22 m0
1
k ; ; 4I D1 kp ; 2
R=@
p 4D1
D22 m0 D22 Um 1
A: (5.75)
4I D1 kp ; 2 2I ki D1 ; 4
Demanding R > 0 gives the following conditions on kp , 1 and
2 2 2
kp > D24U d
D (5.76)
1
4ki D1
1 < D (5.77)
U 2 m
and
< min f1 2 g (5.78)
where 2 2 2
1 = kp ; D24U
D
d (5.79)
1
and 2 2
kp ; D24U
D1
d 2 ki D1 ; D2 U4 m 1
2 = ;
ki D1 ; D2 U4 m 1 + 81I D1 kp ; D22 m0 2
: (5.80)
It is assumed that m
^ t satises the sector condition
p^m^ t (^p) > 2 p^2 (5.81)
that is, the throttle is assumed passive as in Simon and Valavani (1991). As p^m^ t (^p)
is of order 32 in p^, (5.81) does not hold globally. However, for a given p^max such
that
jp^(t)j p^max 8 t > 0 (5.82)
it will always be possible to chose 2 small enough for (5.81) to hold for jp^(t)j
p^max . This is the same assumption as made in Assumption 3.1. Now, the CCV
pressure drop #v (m^ ) is to be chosen such that for the rst term in (5.74), the
condition
;m^ #^ c(m^ U^ ) ; #^ v (m^ ) p01 > 0 8U^ (5.83)
01
is satised. Since p0101 > 0, sucient conditions for (5.83) to hold is

^ U^ ) ; #^ v (m^ ) = 0
; #^ c(m (5.84)
m^ =0
and
@ ;#^ (m
^ ^
U ) + ^
# (m^ )

> 0: (5.85)
@ m^ c v
It can be recognized that
#^ c (0 U^ ) = #c(m0 Ud) ; #c(m0 Ud) = 0 (5.86)
#^ v (m^ ) = kv m
^ ) #^ v (0) = 0 (5.87)
and thus (5.84) is satised. From (5.85), we get
; @@m^ #^ c(m
^ U^ ) + kv > 0 (5.88)
and it follows that choosing kv according to
( ^ ^)
kv > sup @ #c@(m^ U)
m
^ (5.89)
^m
U ^
guarantees that (5.85), and thereby (5.83) being satised. Moreover, if kv is chosen
as ( ^ ^)
kv > sup @ #c@(m ^ U) +
m^ 1 (5.90)
^m
U ^
where 1 > 0, we get m^ #v (m^ ) > 1 m^ 2 , and (5.83) is modied to

^ U^ ) ; #^ v (m^ ) p01 > p 01 1 m^ 2 8U:
;m^ #^ c(m ^ (5.91)
01 01
Consequently, V_ can now be upper bounded as
2 2
V_ ; p 01 1 ; D 2 Um ; D2
8I 4D m^ 2 ; 2 p^2 ; 12 z T Rz 8 m
^ p^ z (5.92)
01 1 1 2
We now set out to compare the coecients of V and V_ . The cross terms in U^ and
I^ are upper bounded using Young's inequality,
^2 !
^ ^ U
U I 2 + 3 I ^2 (5.93)
3
; (4D1kp 2;I D2 m0 ) U^ I^ (4D1 kp 4;I D2 m0 )

2 2
^2
U =4 + 4 I^2 (5.94)
where 3 > 0 and 4 > 0 are constants that can be chosen freely. Using (5.93) and
(5.94) and comparing the coecients in (5.68) and (5.92), it can be recognized
that if the following inequalities are satised for some constant $ > 0:
2 2
1 ; D 2 Um 01 ; D2 01
8I p 4D p > $ A L (5.95)
1 01 1 2 01 01
V
2 > $ a2 p (5.96)
; D2 m0 ) > $ I +
01 01
2
kp ; ; (4D1 kp4I D1 23 (5.97)
4

ki ; (4D1 kp 4;I D2 m0 ) 4 > $ Dki + $2 3
2
(5.98)
1
the following holds
V_ ;$V ) V (t) V (0)e;$t : (5.99)
If 1 is chosen according to (5.77), and 1 is chosen according to
2 2
1 > D 2 Um 01 + D2 01
8I1 p01 4D1 2 p01 (5.100)
n
@ ^ c (m
o
^ U^ ) + , and is chosen accord-
and kv is chosen so that kv > supU ^m^ @ m^ 1
ing to (5.70) and (5.78), then the inequalities (5.95)-(5.98) are satised for some
$ > 0. By (5.99) the origin of (5.58) is exponentially stable. Due to assumption
(5.82), the stability result holds whenever jp^(0)j p^max, and thus the origin is
semi globally exponentially stable.
Notice that the parameter $ can be used to calculate a lower bound on the con-
vergence rate of the system.
Providing the parameters ki and with appropriate units, V is a energy-like
function of unit 'J ]. The term
VU^ = 12 z1 P11 z1 = 2DI 2 U^ 2 (5.101)

1
is recognized as the kinetic energy, expressed in the new coordinates, of the rotating
spool.
5.9 Simulations
In order to model the compressor pressure rise for negative mass ow (deep surge)
it is assumed that the pressure rise is proportional to the square of the mass ow
for m < 0, that is,
( cnm2 + c0(U1) m 0
#c(U1 m) = i (mU1 )h0cideal ; 1
1+ T01 cp m>0 (5.102)
where the choice

; 1
c0 (U1 ) = 1 + i (m UT1 )%c h0cideal (5.103)
01 p m=0
of the shut o value c0 ensures that #c(U1 m) is continuous in m. According to
Willems (1996) and Day (1994), the back-ow characteristic denes the resistance
which the rotating blades oer to ow in reversed direction. Day (1994) states
that, in reversed ow the compressor can be regarded as a throttling device with
a positive pressure bias.
A quadratic characteristic for reversed ow is also proposed by Hansen et al.
(1981) for centrifugal compressors, and by Mansoux et al. (1994), Willems (1996)
and Day (1994) for axial compressors. It is widely accepted in the literature (see
e.g. Greitzer (1981)) that the compressor characteristic has a negative slope for
negative mass ow. This slope depends on the choice of the constant cn .
The two cases of annular and vaned diuser are now simulated with and without
surge control.
5.9 Simulations 105
Annular Diuser
In Figure 5.11 the response (solid lines) of the compression system during surge is
showed. The set point for compressor speed was Ud = 150m=s ) N 50:000rpm
and the speed control law (5.65) parameters were set to kp = 0:1 and ki = 0:07.
The throttle gain was set to kt = 0:0003 which gives an unstable equilibrium to
the left of the surge line. Notice the oscillation in speed U1 . These variations in
spool speed during surge was rst described by Eveker and Nett (1991) for an
axial compressor. This simulation is also shown in Figure 5.12, where the pressure
rise has been plotted versus the mass ow in the compressor characteristic. As
can be seen, the compressor undergoes severe (deep) surge oscillations, and the
compressor speed oscillates around the desired value.
Now, the surge controller (5.64) is used with kv = 0:5. The speed set point, speed
controller parameters and throttle gain are as before. The results are shown in
Figure 5.11 (dashed lines). The desired speed is reached and the surge oscillations
are eliminated. As previously mentioned there is a loss associated with the CCV
control approach. The pressure drop over the valve is shown in the lower right
corner of Figure 5.11. At equilibrium the pressure drop for this particular case is at
ca. 5kPa. Compared to the pressure rise over the compressor at this equilibrium,
well over 200kPa this seems little when taken into account that the compressor now
is operating in an area of the compressor map previously not possible. The CCV
loss is dependent on the controller gain kv . In this simulation the gain was set to
kv = 0:5 to dominate the maximum positive slope of the compressor characteristic.
160 0.8
140 0.6
U1 'm/s]
m 'kg/s]
120 0.4
100 0.2
80 0
60 −0.2
0 2 4 6 0 2 4 6
time 's] time 's]
5 4
x 10 x 10
2.5 2
#v p01 'Pa]
1.5
2
pp 'Pa]
1.5
0.5
1 0
0 2 4 6 0 2 4 6
time 's] time 's]
Figure 5.11: Transient response of centrifugal compression system with annular
diuser. Without surge control, the compressor goes into surge, shown with solid
lines. The system response with the surge controller is shown with dashed lines.
2.6 N = 55000 const. speed lines

surge line
throttle
2.4 stone wall
N = 50000 respns wo/control
pressure rise p(t)=p01
2.2
respns w/control
2
N = 45000
N = 40000
1.8
1.6
N = 35000
1.4 N = 30000
1.2 N = 20000
1
−0.2 −0.1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
mass ow 'kg/s]
Figure 5.12: (m(t) p(t)=p01 )-trajectories plotted together with the compressor char-
acteristic. N is the compressor speed in rpm. In the case of no surge control, the
surge cycle is clearly visible, but with surge control the state converges to the in-
tersection of the throttle and the compressor characteristic.
Vaned Diuser
The response of the compression system with speed control only is shown in Fig-
ure 5.13 (solid lines). The speed control parameters were set to kp = 0:1 and
ki = 0:07, and the throttle gain was set at kT = 0:00075. As the vaned diuser
gives a steeper and more narrow characteristic, the amplitude of the pressure os-
cillations is larger than for the annular diuser. This is also the case for the speed
and mass ow oscillations.
When the surge control is in use, we get the response plotted with dashed lines in
Figure 5.13, and as can be seen the oscillations are avoided at the cost of a pressure
loss over the CCV. Since the positive slope of the compressor characteristic is larger
in this case compared to the annular diuser, this pressure drop is also larger.
However, a pressure drop of 35kPa over the valve is still less than the pressure rise
of 180kPa over the compressor.
By comparing the in-surge response of the two cases, it is seen that the frequency
of the surge oscillations is lower for the vaned diuser (3Hz), than for the annular
diuser (7Hz). This is in accordance with Greitzer (1981) and Willems (1996)
where it is shown that the surge frequency depends on the slope of the compressor
characteristic in such a way that a steeper slope leads to lower frequency, and a
less steep slope leads to higher frequency.
5.10 Conclusion 107
160 0.8
140 0.6
U1 'm/s]
m 'kg/s]
120 0.4
100 0.2
80 0
60 −0.2
0 2 4 6 0 2 4 6
5
time 's] 4
time 's]
x 10 x 10
2.5 5
#v p01 'Pa]
2
pp 'Pa]
2
1.5
1
1 0
0 2 4 6 0 2 4 6
time 's] time 's]

Figure 5.13: Transient response of centrifugal compression system with vaned dif-
fuser. Without surge control, the compressor goes into surge. This is plotted with a
solid line. The dashed lines is the system response when the CCV surge controller
is in use.
5.10 Conclusion
In this chapter, a dynamic model of a centrifugal compression system with non-
constant compressor speed was presented. The compressor characteristic was
derived by calculating the energy transfer end losses in the components of the
compressor. Incidence and friction losses in the impeller and the diuser were
considered in addition to other losses. Both vaned and annular diusers were
considered.
Control laws for surge and speed of the centrifugal compression system were de-
veloped. A close coupled valve was chosen as an actuator for the control of surge.
Using Lyapunov's method, the systems equilibrium was showed to be semi-global
exponentially stable. Through simulations it was conrmed that the compressor
can operate stable
and reach desired speed in the previous unstable area to the right of the surge line
in the compressor map.
From a surge control point of view, the main dierence between the annular and
vaned diusers are the steeper slope of the compressor characteristic when a vaned
diuser is used. A consequence of this is that if a close coupled valve is used to
control surge, a greater pressure drop must be accepted over the valve in the case
of a vaned diuser than in the case of an annular diuser.
Chapter 6
Conclusions
In this thesis modeling and control of surge and rotating stall in axial and cen-
trifugal compressors have been studied.
First, surge and stall controllers for a close coupled valve in series with a compressor
were developed. Using backstepping, a surge control law for the close coupled valve
was derived. Global asymptotic stability was proven. A more complicated surge
control law was derived for the case of both pressure disturbances and mass ow
disturbances. Global uniform boundedness and convergence was proven. In order
to stabilize the compression system in the presence of constant disturbances, or
biases, in mass ow and pressure, an adaptive version of the surge controller was
derived. This controller ensures global asymptotic stability. Then, controllers
for rotating stall were considered. The close coupled valve was incooperated into
the Moore-Greitzer model, and controllers were derived that enables stabilization
of rotating stall beyond the surge line. Without disturbances, an asymptotically
stable equilibrium is ensured, and in the presence of pressure disturbances uniform
boundedness was proven.
Then, the passivity properties of the Greitzer model was used to derive a surge
control law for a close coupled valve. This resulted in a simple proportional control
law, that was capable of stabilizing the compression system in the presence of both
mass ow disturbances as well as pressure disturbances.
A multi mode Moore-Greitzer axial compressor model with spool dynamics was
derived. This resulted in a model with time varying B -parameter. Through sim-
ulations it was demonstrated that the model was capable of demonstrating both
rotating stall and surge, and that the type of instability depended on the compres-
sor speed. In the original Moore Greitzer model only the rst mode of rotating
stall is included. The simulations in this chapter show that during stall inception,
higher order modes can dominate the rst mode. This is in accordance with known
results, and is shown here to be valid also for variable speed compressors.
A dynamic model of a centrifugal compression system with non-constant compres-
sor speed was presented. The compressor characteristic was derived by calculating
110 Conclusions
the energy transfer and losses in the components of the compressor. Incidence and
friction losses in the impeller and the diuser were considered in addition to other
losses. Both vaned and annular diusers were considered. Control laws for surge
and speed of the centrifugal compression system were developed. A close cou-
pled valve was chosen as an actuator for the control of surge. Using Lyapunov's
method, the systems equilibrium was showed to be semi-global exponentially sta-
ble. Through simulations it was conrmed that the compressor can operate stable
and reach desired speed in the previous unstable area to the right of the surge line
in the compressor map.
Bibliography
Abed, E.H., P.K. Houpt and W.M. Hosny (1993). Bifurcation analysis of surge and
rotating stall in axial ow compressors. Journal of Turbomachinery 115, 817{
824.
Adomatis, R.A. and E.H. Abed (1993). Local nonlinear control of stall inception
in axial ow compressors. In: Proceedings of the AIAA/SAE/ASME/ASEE
Joint Propulsion Conference. Monterey, CA.
Badmus, O.O., S. Chowdhury and C.N. Nett (1996). Nonlinear control of surge in
axial compression systems. Automatica 32(1), 59{70.
Badmus, O.O., S. Chowdhury, K.M. Eveker and C.N. Nett (1995). Control-
oriented high-frequency turbomachinery modelling: Single-stage compression
system one-dimensional modell. Journal of Turbomachinery 117, 47{61.
Balchen, J.G and K.I. Mumme (1988). Process control : Structures and applica-
tions. Van Nostrand Reinhold Company. New York.
Balje, O.E. (1952). A contribution to the problem of designing radial turboma-
chines. Transactions of the ASME 74, 451{472.
Banaszuk, A., H.A. Hauksson and I. Mezic (1997). A backstepping controller for
a partial dierential equation model of compression system instabilities. In:
Proceedings of the 36th Conference on Decision and Control. San Diego, CA.
pp. 1050{1055.
Bazanella, A.S., P.V. Kokotovic and A.S. e Silva (1997). On the control of dynamic
systems with unknown operating point. In: Proceedings of the 1997 European
Control Conference. Brussels, Belgium.
Behnken, R.L. and R.M. Murray (1997). Combined air injection control of rotating
stall and bleed valve control of surge. In: Proceedings of the 1997 American
Control Conference. Albuquerque, NM.
Brown, D., L. Rachul and K. Williams (1997). NASA Researching Engine Air-
ow Controls to Improve Performance, Fuel Eciency. NASA press release,
downloaded from Usenet newsgroup sci.space.news. Release: 97-183.
Cohen, H., G.F.C. Rogers and H.I.H Saravanamuttoo (1996). Gas turbine theory.
4th ed.. Longman. Essex.
112 BIBLIOGRAPHY
Covert, E.E. (1995). Evolution of the commercial airliner. Scientic American
273(3), 82{85.
Cumpsty, N.A. (1989). Compressor Aerodynamics. Longman.
Day, I.J. (1993). Active suppression of rotating stall and surge in axial compressors.
Journal of Turbomachinery 115, 40{47.
Day, I.J. (1994). Axial compressor performance during surge. Journal of propulsion
and power 10(3), 329{336.
de Jager, B. (1995). Rotating stall and surge control : A survey. In: Proceedings
of the 35th Conference on Decision and Control. New Orleans, LA. pp. 1857{
1862.
DeLaat, J.C., R.D. Southwick and G.W. Gallops (1996). High stability engine
control (HISTEC). In: Proceedings of AIAA/SAE/ASME/ASEE 32nd Joint
Propulsion Conference. Lake Buena Vista, FL. NASA TM 107272, AIAA-96-
2586.
Dixon, S.L. (1978). Thermodynamics of turbomachinery. 3rd ed.. Pergamon Press.
Dussourd, J.L., G.W. Pfannebecker and S.K. Singhania (1976). Considerations
for the control of surge in dynamic compressors using close coupled resis-
tances. In: Centrifugal compressor and pump stability, stall and surge (P.C.
Tramm and R.C. Dean, Jr., Eds.). pp. 1{28. ASME. New Orleans, Louisiana.
Presented at the 1976 Joint gas turbine and uids engineering divisions' Con-
ference.
Dussourd, J.L., G.W. Pfannebecker and S.K. Singhania (1977). An experimental
investigation of the control of surge in radial compressors using close coupled
resistances. Journal of Fluids Engineering 99, 64{76.
Eastop, T.D. and A.McConkey (1986). Applied thermodynamics for engineering
technologists. 4th ed.. Longman. New York.
Emmons, H.W., C.E. Pearson and H.P. Grant (1955). Compressor surge and stall
propagation. Transactions of the ASME 77, 455{469.
Epstein, A.H., J.E. Ffowcs Williams and E.M. Greitzer (1989). Active suppres-
sion of aerodynamic instabilities in turbomachines. Journal of Propulsion and
Power 5(2), 204{211.
Erskine, J.B. and N. Hensman (1975). Vibration induced by pump instability and
surging. In: Vibrations and noise in pump, fan, and compressor installations.
The Institution of Mechanical Engineers. Mechanical Engineering Publica-
tions limited. Southamton, UK. pp. 87{98.
Eveker, K.M. and C.N. Nett (1991). Model development for active surge control
rotating stall avoidance in aircraft gas turbine engines. In: Proceedings of the
1991 American Control Conference. pp. 3166{3172.
BIBLIOGRAPHY 113
Ferguson, T.B. (1963). The centrifugal compressor stage. Butterworths. London.
Ffowcs Williams, J.E. and X.Y. Huang (1989). Active stabilization of compressor
surge. Journal of Fluid Mech. 204, 245{262.
Ffowcs Williams, J.E., M.F.L. Harper and D.J. Allwright (1993). Active stabi-
lization of compressor instability and surge in a working engine. Journal of
Turbomachinery 115, 68{75.
Fink, D.A., N.A. Cumpsty and E.M. Greitzer (1992). Surge dynamics in a free-
spool centrifugal compressor system. Journal of Turbomachinery 114, 321{
332.
Godhavn, J.-M. (1997). Topics in nonlinear motion control: nonholonomic, under-
actuated and hybrid systems. PhD thesis. Norwegian University of Science
and Technology. Dept of Engineering Cybernetics.
Gravdahl, J.T. (1998). Modeling and Close Coupled Valve Control of Compressor
Surge and Rotating Stall. Advances in Industrial Control. Springer-Verlag.
London. Accepted for publication, to appear.
Gravdahl, J.T. and O. Egeland (1997a). A Moore-Greitzer axial compressor model
with spool dynamics. In: Proceedings of the 36th IEEE Conference on Deci-
sion and Control. San Diego, CA. pp. 4714{4719.
Gravdahl, J.T. and O. Egeland (1997b). Centrifugal compressor surge and speed
control. Submitted to IEEE Transactions on Control Systems Technology.
Gravdahl, J.T. and O. Egeland (1997c). Compressor surge control using a close-
coupled valve and backstepping. In: Proceedings of the 1997 American Control
Conference. Albuquerque, NM.
Gravdahl, J.T. and O. Egeland (1997d). Control of the three-state Moore-Greitzer
compressor model using a close-coupled valve. In: Proceedings of the 1997
European Control Conference. Brussels, Belgium.
Gravdahl, J.T. and O. Egeland (1997e). Modeling of rotating stall and surge in a
free spool axial compression system. Submitted to Journal of Turbomachinery.
Gravdahl, J.T. and O. Egeland (1997f). Passivity based compressor surge control
using a close-coupled valve. In: Proceedings of the 1997 COSY Workshop
on Control of Nonlinear and Uncertain Systems (A. Isidori and F. Allg-ower,
Eds.). Zurich, Switzerland. pp. 139{143.
Gravdahl, J.T. and O. Egeland (1997g). Speed and surge control for a low or-
der centrifugal compressor model. In: Proceedings of the 1997 International
Conference on Control Applications. Hartford, CT. pp. 344{349.
Gravdahl, J.T. and O. Egeland (1998a). Close Coupled Valve control of surge and
rotating stall in compressors. Submitted to Automatica.
114 BIBLIOGRAPHY
Gravdahl, J.T. and O. Egeland (1998b). Speed and surge control for a low order
centrifugal compressor model. Modeling, Identication and Control 19(1), 13{
29.
Gravdahl, J.T. and O. Egeland (1998c). Two results on compressor surge con-
trol with disturbance rejection. Submitted to the 37th IEEE Conference on
Decision and Control.
Greitzer, E.M. (1976a). Surge and Rotating stall in axial ow compressors, Part
I: Theoretical compression system model. Journal of Engineering for Power
98, 190{198.
Greitzer, E.M. (1976b). Surge and rotating stall in axial ow compressors, Part
II: Experimental results and comparison with theory. Journal of Engineering
for Power 98, 199{217.
Greitzer, E.M. (1977). Coupled compressor-diuser ow instability. Journal of
aircraft 14(3), 233{238.
Greitzer, E.M. (1980). REVIEW{Axial compressor stall phenomena. Journal of
Fluids Engineering 102, 134{151.
Greitzer, E.M. (1981). The stability of pumping systems { The 1980 Freeman
scholar lecture. Journal of Fluids Engineering 103, 193{242.
Greitzer, E.M. and F.K. Moore (1986). A theory of post-stall transients in a axial
compressor systems: Part II{Application. Journal of Engineering for Gas
Turbines and Power 108, 231{239.
Greitzer, E.M. and H.R. Griswold (1976). Compressor-diuser interaction with
circumferential ow distortion. The journal of mechanical engineering science
18(1), 25{43.
Gu, G., S. Banda and A. Sparks (1996). An overview of rotating stall and surge
control for axial ow compressors. In: Proceedings of the 35th Conference on
Decision and Control. Kobe, Japan.
Gysling, D.L., J. Dugundji, E.M. Greitzer and A.H. Epstein (1991). Dynamic
control of centrifugal compressors surge using tailored structures. Journal of
Turbomachinery 113, 710{722.
Haddad, W.M., J.L. Fausz, V.-S. Chellaboina and A. Leonessa (1997). Nonlinear
robust disturbance rejection controllers for rotating stall and surge in axial
ow compressors. In: Proceedings of the 1997 International Conference on
Control Applications. Hartford, CT. pp. 767{772.
Hansen, K.E., P. Jrgensen and P.S. Larsen (1981). Experimental and theoretical
study of surge in a small centrifugal compressor. Journal of Fluids Engineering
103, 391{394.
Hendricks, G.J. and D.L. Gysling (1994). Theoretical study of sensor-actuator
schemes for rotating stall control. Journal of Propulsion and Power
10(1), 101{109.
BIBLIOGRAPHY 115
Hendrickson, E.S. and A.G. Sparks (1997). On the suitability of bifurcation sta-
bilization control laws for high order Moore-Greitzer compressor models. In:
Proceedings of the 1997 American Control Conference. Albuquerque, NM.
Horlock, J.H. (1958). Axial ow compressors: Fluid mechanics and Thermody-
namics. Butterworths scientic publications. London.
Humbert, J.S. and A.J. Krener (1997). Analysis of higher order Moore-Greitzer
compressor models. In: Proceedings of the 1997 International Conference on
Control Applications. Hartford, CT. pp. 651{656.
Hynes, J.M., G.J. Hendricks and A.H. Epstein (1994). Active stabilization of ro-
tating stall in a three-stage axial compressor. Journal of Turbomachinery
116, 226{239.
Hynes, T.P. and E.M. Greitzer (1987). A method for assessing eects of circum-
ferential ow distortion on compressor stability. Journal of Turbomachinery
109, 371{379.
Johnston, J.P. and R.C. Dean, Jr. (1966). Losses in vaneless diusers of centrifugal
compressors and pumps. Analysis, experiment and design. Journal of Engi-
neering for Power pp. 49{62.
Jungowski, W.M., M.H. Weiss and G.R. Price (1996). Pressure oscillations occur-
ring in a centrifugal compressor system with and without passive and active
surge control. Journal of Turbomachinery 118, 29{40.
Ko, S.G. and E.M. Greitzer (1986). Axisymmetrically stalled ow performance
for multistage axial compressors. Journal of Turbomachinery 108, 216{249.
Krstic, M. (1996). Invariant manifolds and asymptotic properties of adaptive non-
linear stabilizers. IEEE Transactions on automatic control 41(6), 817{829.
Krstic, M. and P.V. Kokotovic (1995). Lean backstepping designs for jet engine
compressor model. In: Proceedings of the 4th IEEE Conference on Control
Applications. pp. 1047{1052.
Krstic, M., I. Kanellakopoulos and P.V. Kokotovic (1995a). Nonlinear and Adap-
tive Control Design. John Wiley & Sons Inc.
Krstic, M., J.M. Protz, J.D. Paduano and P.V. Kokotovic (1995b). Backstepping
designs for jet engine stall and surge control. In: Proceedings of the 35th
Conference on Decision and Control. New Orleans, LA. pp. 3049{3055.
Leonessa, A., V.-S. Chellaboina and W.M. Haddad (1997a). Globally stabiliz-
ing controllers for multi-mode axial ow compressor models via equilibria-
dependent lyapunov functions. In: Proceedings of the 1997 International Con-
ference on Control Applications. Hartford, CT. pp. 63{68.
Leonessa, A., V.-S. Chellaboina and W.M. Haddad (1997b). Robust stabilization
of axial ow compressors with uncertain pressure-ow maps. In: Proceedings
of the 1997 International Conference on Control Applications. Hartford, CT.
pp. 671{676.
116 BIBLIOGRAPHY
Liaw, D.-C. and E.H. Abed (1992). Stability analysis and control of rotating stall.
In: Proceedings of NOLCOS'92: Nonlinear Control System Design Sympo-
sium (M. Fliess, Ed.). Bourdaux, France. pp. 88{93.
Liaw, D.-C. and E.H. Abed (1996). Active control of compressor stall inception: a
bifurcation-theoretic approach. Automatica 32(1), 109{115.
Lorett, J.A. and S. Gopalakrishnan (1986). Interaction between impeller and vo-
lute of pumps at o-design conditions. Journal of uids engineering 108, 12{
18.
Mansoux, C.A., D.L. Gysling, J.D. Setiawan and J.D. Paduano (1994). Distributed
nonlinear modeling and stability analysis of axial compressor stall and surge.
In: Proceedings of the 1994 American Control Conference. Baltimore, Mary-
land. pp. 2305{2316.
Mattingly, J.D. (1996). Elements of gas turbine propulsion. McGraw-Hill series in
aeronautical and aerospace engineering. McGraw-Hill.
McCaughan, F.E. (1989). Application of Bifurcation theory to axial ow compres-
sor instability. Journal of Turbomachinery 111, 426{433.
Moore, F.K. (1984a). A theory of rotating stall of multistage axial compres-
sors: Part I{Small disturbances. Journal of Engineering for Gas Turbines
and Power 106, 313{320.
Moore, F.K. (1984b). A theory of rotating stall of multistage axial compressors:
Parts I,II,III. J. of Engineering for Gas Turbines and Power 106, 313{336.
Moore, F.K. and E.M. Greitzer (1986). A theory of post-stall transients in a axial
compressor systems: Part I{Development of equations. Journal of Engineer-
ing for Gas Turbines and Power 108, 68{76.
Murray, S. Yeung R.M. (1997). Reduction of bleed valve rate requirements for
control of rotating stall using continuous air injection. In: Proceedings of
the 1997 International Conference on Control Applications. Hartford, CT.
pp. 683{690.
Nicklasson, P.J. (1996). Passivity-based control of electric machines. PhD thesis.
Norwegian Institute of Technology. Dept of Engineering Cybernetics.
Nisenfeld, A.E. (1982). Centrifugal compressors: principles of operation and con-
trol. Instrument society of America.
Osborn, W.M. and J.M. Wagner (1970). Eect of simulated downstream ow
blockage doors on the performance of an axial-ow fan rotor. Technical Report
TN D-6071. NASA.
Paduano, J.D., A.H. Epstein, L. Valavani, J.P. Longley, E.M. Greitzer and G.R.
Guenette (1993). Active control of rotating stall in a low-speed axial com-
pressor. Journal of Turbomachinery 115, 48{56.
BIBLIOGRAPHY 117
Paduano, J.D., A.H. Epstein, L. Valavani, J.P. Longley, E.M. Greitzer and
G.R. Guenette (1995). Modelling for control of rotating stall. Automatica
30(9), 1357{1373.
Pampreen, R.C. (1973). Small turbomachinery compressor and fan aerodynamics.
Journal of engineering for power 95, 251{256.
Perko, L. (1991). Dierential equations and dynamical systems. Vol. 7 of Texts in
Applied Mathematics. Springer-Verlag.
Pinsley, J.E., G.R. Guenette, A.H. Epstein and E.M. Greitzer (1991). Active stabi-
lization of centrifugal compressor surge. Journal of Turbomachinery 113, 723{
732.
Protz, J.M. and J.D. Paduano (1997). Rotating stall and surge: Alternate model-
ing and control concepts. In: Proceedings of the 1997 International Conference
on Control Applications. Hartford, CT. pp. 866{873.
Rues, P.C (1996). Innovation in aero engines. The Aeronautical Journal
100(1000), 473{484.
Sepulchre, R. and P. Kokotovic (1996). Shape signiers for throttle control of a
low-order compressor model.
Simon, J.S. and L. Valavani (1991). A Lyapunov based nonlinear control scheme
for stabilizing a basic compression system using a close-coupled control valve.
In: Proceedings of the 1991 American Control Conference. pp. 2398{2406.
Simon, J.S., L. Valavani, A.H. Epstein and E.M. Greitzer (1993). Evaluation of ap-
proaches to active compressor surge stabilization. Journal of Turbomachinery
115, 57{67.
Simon, S. (1993). Feedback stabilization of compression systems. PhD thesis. MIT.
Stenning, A.H. (1980). Rotating stall and surge. Journal of uids engineering
102, 14{21.
van de Wal, M. and F. Willems (1996). Selection of actuators and sensors for
compressor control. Technical Report WFW 96.155. Eindhoven University of
Technology.
van de Wal, M., F. Willems and B. de Jager (1997). Selection of actuators and
sensors for active surge control. In: Proceedings of the 1997 International
van der Schaft, A.J. (1996). L2 -gain and passivity techniques in nonlinear con-
trol. Vol. 218 of Lecture notes in Control and Information Sciences. Springer-
Verlag. Heidelberg.
Wang, H.-H. and M. Krstic (1997a). Control of Deep-Hysteresis aeroengine
compressors{Part I : A Moore-Greitzer type model. In: Proceedings of the
1997 American Control Conference. Albuquerque, NM.
118 BIBLIOGRAPHY
Wang, H.-H. and M. Krstic (1997b). Control of deep hysteresis compressors un-
der limited actuator bandwidth. In: Proceedings of the 1997 International
Watson, N. and M.S. Janota (1982). Turbocharging the internal combustion engine.
MacMillan.
Weigl, H.J. and J.D. Paduano (1997). Application of H1 control with eigenvalue
perturbations to stabilize a transonic compressor. In: Proceedings of the 1997
International Conference on Control Applications. Hartford, CT. pp. 691{698.
White, F.M. (1986). Fluid mechanics. 2nd ed.. McGraw-Hill. New York.
Whiteld, A. and F.J. Wallace (1973). Study of incidence loss models in radial
and mixed-ow turbomachinery. In: Proceedings of the conference on heat
and uid ow in steam and gas turbine plant. The instn. of Mech. Engnrs..
U. of Warwick, Coventry. pp. 122{128.
Whiteld, A. and F.J. Wallace (1975). Performance prediction for automotive tur-
bocharger compressors. Proceedings of the institution of mechanical engineers
189(12), 59{67.
Willems, F. (1996). Modeling and control of compressor ow instabilities. Technical
Report WFW 96.151. Eindhoven University of Technology.
Appendix A
Nomenclature
A.1 Acronyms
CCV Close Coupled Valve
CLF Control Lyapunov Function
IGV Inlet Guide vanes
ODE Ordinary Dierential Equation
PDE Partial Dierential Equation
PDF Positive Denite Function
GAS Globally Asymptotically Stable
GES Globally Exponentially Stable
MG Moore-Greitzer
A.2 Subscripts
0 Equilibrium value
c Compressor
d Disturbance or desired value
x a Axial
p Plenum
Partial derivative with respect to , or
A.3 Superscripts
0 ^0 is a disturbance velocity potential

( ^0 ) is an approximation of ^0
120 Nomenclature
A.4 Mathematical Symbols and Denitions
(_) Total time derivative, dtd or dd

@ Partial derivative with respect to variable x
@x
();1 Inverse operator
(^) Deviation from equilibrium, transformed coordinates
k k1 The L1 -norm of the function
f : IIR+ ! IIR is dened as kf k1 = supt
0 jf (t)j
jj Magnitude of vector
()T Transpose operator
8 For all
9 Exists
IIR Real numbers
IIR+ Nonnegative real numbers fx 2 IIR : x 0g
IIRn Linear space of n-tuples in IIR
r Divergence operator. r f (x1 x2 : : :) = @x @f1 + @f2 + : : :
1 @x2
K Class of functions. A function f : '0 a) ! IIR+ is said to
belong to class K if it is strictly increasing and f (0) = 0.
K1 Class of functions. A function f : '0 a) ! IIR+ is said
to belong to class K1 if it belongs to class K,
a = 1 and f (r) ! 1 as r ! 1.
The function f g is the composition of
the functions f and g
7! Mapping
L2 Space of square integrable
R 1 functions. f : IIR+ ! IIR
belongs to L2 i 0 jf (t)j dt < 1
2
fT (t) f (t) of f 0to'0t<T ].T
The truncation
fT (t) = 0 tT
L2e The extension of L2 . f 2 L2e i fT 2 L2
G Input-output mapping R
hu yiT Inner product on L2e . hu yiT = 0T u(t)y(t)dt
kuk2T Truncated norm. kuk2T = hu uiT
sgn() Signum function
tanh() Hyperbolic tangent
sup() Supremum, lower upper bound
A.5 Symbols 121
A.5 Symbols
The equation numbers are references to where the symbol was dened or rst used.
Latin Uppercase
Ac Flow area (2.3)

Acl Closed loop Jacobian (2.193)
A1 Reference ow area. Area of impeller eye. (5.1)
An Amplitude of mode number n of rotating stall (4.48)
Az (z d) Nonlinear part of error dynamics (2.140)
A Area of attraction (2.154)
B Greitzer B-parameter (2.3)
C1 C2 Tangential uid velocity at the rotor entrance
and exit (4.13)
Cx Ca1 Ca2 Axial uid velocity (4.21)
C1 Fluid velocity at inducer (5.2)
C2 Fluid velocity at diuser entry, Figure 5.5
Ch Surface friction loss coecient (5.27)
C2 C3 C4 Functions used to calculate upper bound on c3
in the proof of Theorem 2.5
C1 C3 C4 Functions used to calculate lower bound on c3
in the proof of Theorem 2.5
D Hydraulic diameter (5.27)
D1 r1 Average inducer diameter and radius (5.4)
D2 r2 Diameter and radius at impeller tip, Figure 5.5
Dt1 Diameter at inducer tip (5.4)
Dh1 Diameter at impeller hub casing (5.4)
F( ) Pressure rise coecient in blade passage (4.20)
G1 ( ) G2 ( ) Functions used in the proof of Theorem 2.5
G1 G2 Input-output mappings (3.13), (3.14)
H Compressor characteristic semi height (2.4)
I Spool and rotor moment of inertia (4.8)
I^ Integral of U^ (5.65)
J Squared amplitude of rotating stall (2.1)
Ji Mode i of squared amplitude of rotating stall (4.72)
Jne Equilibrium value of Jn (4.74)
Jmax Maximum value of squared rotating
stall amplitude (2.150)
122 Nomenclature
Latin Uppercase
K ( ^) Function, used in passivity analysis (3.24)

KG Entrance recovery coecient (4.30)
Lc Length of ducts and compressor (2.3)
LI Length of inlet duct (4.4)
LE Length of exit duct (4.4)
M1 M2 M3 Moments calculated in Galerkin approximation (4.66)
Ns Number of compressor stages (2.12)
N Number of revolutions per second (5.3)
N () Number of rotating stall modes (4.50)
P Positive denite constant matrix (5.67)
Pz Constant skew symmetric matrix (2.140)
P( ) Matrix used in proof of Theorem 2.5
Q Supplied heat
Rn Residue nr. n used in Galerkin approximation (4.61)
R Mean compressor radius (2.5)
R Positive denite constant matrix (5.75)
Re Reynolds number (5.29)
R R1 R2 R3 Residual sets (2.72), (2.87), (2.116), (2.202)
S (z ) Storage function (2.36)
T01 Inlet stagnation temperature
U (z1 z2 J ) PDF (2.185)
U Tangential speed of rotor
U1 Tangential speed of rotor, at diameter D1 (5.3)
U2 Tangential impeller tip speed, Figure 5.5
Ud Desired tangential speed of rotor (4.3)
Um Upper bound on U1 (5.50)
Vp Plenum volume (2.3)
V1 V 2 LFCs
V_2 i Term number i in V_2 (2.166)
W Compressor characteristic semi width (2.4)
W (z1 z2 ) PDF
W1 Fluid velocity relative to moving impeller blades (5.5)
W1b Component of W1 in blade direction (5.32)
W1 Tangential component of W1 (5.18)
W_ c Compressor power (5.7)
A.5 Symbols 123
Latin lowercase
a Reciprocal time lag of the compressor passage (2.2)

a Compressor characteristic slope (2.53)
as Sonic velocity
am Max positive slope of compressor characteristic (2.53)
b Constant. U = bB (4.7)
c2 c3 Surge/stall controller gains
c Constant. Used in convergence proof (2.90)
cspeed Speed controller gain (4.76)
cn Slope of backow characteristic (5.102)
cmin
3 c3
max Lower and upper bound on c3 (2.189), (2.190)
cp Specic heat capacity at constant pressure
cv Specic heat capacity at constant volume
d d Mass ow and pressure biases (2.37)
d1 d2 Surge/stall controller damping coecients
d d Estimates of biases, Theorem 2.4
d~ d~ Estimation errors (2.126), (2.134)
d~ (d~ d~ )T
f Friction factor (5.29)
g( ) Disturbance of axial ow coecient (4.28)
h( ) Circumferential velocity coecient (4.28)
h1 h2 h3 Weight functions used in Galerkin approximation (4.63)
h Specic enthalpy
k1 k2 k3 Compressor characteristic coecients (2.24)
kv Surge control law parameter (5.64)
kp ki Proportional and integral gain of speed controller (5.65)
lc Nondimensional compressor length (2.2)
li Nondimensional length of inlet duct (2.2)
le Nondimensional length of exit duct (2.2)
l Mean ow length (5.27)
m mc Compressor mass ow
m Compressor duct ow parameter (4.45)
mchoke Choking mass ow (2.45)
p Pressure
pp Plenum pressure (5.1)
p^max Upper bound on p^ used to calculate 2 (5.82)
p0 m0 Equilibrium value of pp and m (5.54)
rn Phase angle of mode number n of rotating stall (4.59)
q Dimension of state space in Chapter 4
s( z ) Signal. Used in convergence proof (2.90)
s Specic entropy
u Control variable
v Specic volume
z1 z2 Error variables
z State vector. z = (z1 z2 )T
124 Nomenclature
Greek uppercase
%bf Loss in eciency due to backow (5.39)

%c Loss in eciency due to clearance (5.38)
%v Loss in eciency in the volute (5.40)
%d Loss in eciency due to incomplete diusion
% Approximation error % = 0 ; apprx (2.64)
%h0c Ideal specic enthalpy delivered to uid (5.9)
%h0cideal Total specic enthalpy delivered to uid (5.54)
%hii %hid Incidence losses in impeller and diuser (5.21), (5.26)
%hfi %hfd Friction losses in impeller and diuser (5.27), (5.34)
;t Nondimensional turbine (drive) torque (4.10)
;c Nondimensional compressor torque (4.10)
; Positive denite constant matrix (2.133)
+1 +2 Constants in MG model (4.12), (4.46)
G1 G2 Feedback interconnection of G1 and G2 , Theorem 3.1
Axial mass ow coecient, annulus averaged (2.1)
T ( ) Throttle mass ow coecient (2.1)
^ d Time varying mass ow disturbance, section 2.2.5
#d( ) Monotonically decreasing non negative function,
Corollary 2.1
d ( ) Monotonically decreasing non negative functions,
Corollary 2.2
# Pressure coecient (2.1)
#T ( ) Throttle characteristic (2.9)
#v ( ] CCV characteristic (2.11)
#c ( ) Compressor characteristic (2.7)
#e ( ) Equivalent compressor characteristic (2.10)
#s ( ) In-stall characteristic (2.17)
#es ( ) Equivalent in-stall characteristic (2.16)
#^ d Time varying pressure disturbance, section 2.2.5
A.5 Symbols 125
Greek lowercase
Virtual control used in backstepping

1 2 1 2 Fluid angles
1b 2b 1b 2b Constant blade angles
i Angle of incidence (5.15)
1 2 3 Class K1 functions (2.70), (2.71)
T Throttle gain, parameter proportional
to throttle opening (2.9)
CCV gain, parameter prop. to valve opening (2.11)
1 2 Constants (5.66), (5.81)
1 2 Constants (5.78)
n 4 n ; rn ( ) (4.63)
n =
Nondimensional x-coordinate
0 1 2 3 4 Constants. Used in Young's inequality
i Isentropic eciency (5.35)
Angular coordinate
#1 #2 Adaption gains, Theorem 2.4
1 2 Constants used in passivity proofs (3.9), (3.21)
Ratio of specic heats, = ccpv
Viscosity (4.49)
1 2 Constants (2.114)
$ Convergence rate (5.99)
Nondimensional time (2.5)
% Constant in the MG model (2.4)
Density
01 Inlet stagnation density
Slip factor (5.8)
& Constant (5.60)
t Turbine (drive) torque (4.10)
c Compressor torque (4.10)
c; c+ Compressor torque for neg. and pos. ow (5.10)
df Disc friction torque (5.13)
Local axial mass ow coecient (4.27)
0 Equilibrium value of (2.18)
apprx Approximation of 0 (2.64)
choke Upper bound on (2.151)
m Lower bound on (2.153)
~ Velocity potential (4.33)
~0 Disturbance velocity potential (2.34)
0 Equilibrium value of (2.18)
c0 Shut o value of compressor characteristic (2.7)
!H Helmholtz frequency
! Rotational velocity of spool (4.8)
126 Nomenclature
Appendix B
Some Thermodynamic and
Fluid Mechanical Relations
B.1 Flow and Pressure Coecients
In this section various nondimensional quantities related to compressors are pre-
sented.
Flow Coecient
The ow coecient is dened as
= CUx (B.1)
where Cx is the axial velocity and U is the blade speed. Using the ideal gas law it
is found that
p p p
= 01pRT01 CUx = RT
U
01 m RT01 = RT01 F
Ap U (B.2)
01 01
where F is known as the (nondimensional) ow function. Mention must be made
to another form of nondimensional ow, the corrected mass ow dened as
p
mcorr = m p T=p
01 =Tref
: (B.3)
01 ref
The reference state is usually taken as sea level static. It can be shown that the
corrected mass ow is related to the ow coecient as
pT =T
= 01p ref mcorr (B.4)
RUA
ref
that is, for constant U , is proportional to mcorr .
128 Some Thermodynamic and Fluid Mechanical Relations
Pressure Coecient
The total to static pressure rise coecient is dened as
; p01 :
# = p1 U (B.5)
2
In (Cohen et al. 1996) it is shown that

# = i cp % T0s %h0s
U 2 = i U 2 (B.6)
where i is the isentropic eciency, %T0s is the temperature rise in the stage and
h0s is the enthalpy rise. The fraction cp %T0s =U 2 is known as the temperature
coecient.
Speed Coecient
Blade speed U can be nondimensionalized by dividing with inlet stagnation sonic
velocity a01 ,
U U DN D
a = RT = RT = RT Ncorr
01 01 01 ref
(B.7)
where Ncorr
p
= N= T =T0 ref is known as the corrected speed. Note that by intro-
ducing the Greitzer B-parameter it is found that
s
U = D N = 2B Ac Lc : (B.8)
a01 RTref corr Vp
Thus, B can be used as nondimensional compressor speed, and it is connected to
Ncorr as s
Ncorr = 2B AVc Lc RT ref : (B.9)
p D
B.2 Isentropic Processes

Let s denote specic entropy, u denote specic internal energy, v = V=m = ;1
denote specic volume and h = u + pv denote specic enthalpy. It can be shown
(Eastop and A.McConkey 1986), that
ds = dQ
T (B.10)
where
dQ = du + pdv (B.11)
is known as the dierential form of the non-ow energy equation. It follows that
Tds = du + pdv
Tds = dh ; vdp: (B.12)
B.3 Mass Balance of the Plenum 129
With constant specic heat capacities cp and cv ,
Tds = cv dT + pdv
Tds = cp dT ; vdp: (B.13)
For isentropic processes, that is reversible and adiabatic, ds = 0, and accordingly
dT = ; cp dv
v
dT = cv dp (B.14)
p
which in turn implies
dp = ; dv (B.15)
p v
where = ccpv is the ratio of specic heats.
B.3 Mass Balance of the Plenum

The plenum process is assumed to be isentropic, which means that the dierential
form (B.15) of the isentropic relation is valid, so that
dpp = ; dvp = d p (B.16)
pp vp p
where it has been used that
= v1 ) d dv
=; v : (B.17)
It follows that
p_p = p p
_p = RTp _p : (B.18)
p
The mass balance of the plenum is
_p = V1 (m ; mt ) (B.19)
p
and it follows that
p_p = RTp (m ; m ) = a2s (m ; m ) (B.20)
V t V t
p p
p
where as = RTp is the plenum sonic velocity. As velocities in the plenum are
assumed negligible, a0 can be used in (B.20). The sound speed in the plenum will
wary as both temperature and pressure in the plenum varies with time. In (Simon
et al. 1993), a time mean speed of sound in the plenum, as was used. Another
approach to avoid using the plenum sonic velocity was taken in (Greitzer 1976a).
It was recognized that by assuming that the temperature ratios of the compression
system are near unity, the quantity p =pp = RTp is not appreciably dierent from
01 =p01. Thus, the speed of sound at ambient conditions can be used in (B.20).
This approach is also taken in this thesis.
B.4 Flow through a Nozzle
The stagnation temperature is
T0 = T + 21c Cx2 (B.21)
p
where T is the static temperature and Cx is the velocity of the ow. The Mach
number is
M = Ca x (B.22)
s
where p
as = RT (B.23)
is the sonic velocity. Then Cx2 = M 2 RT and it follows that
T0 = 1 + M 2 R = 1 + ; 1 M 2 : (B.24)
T 2cp 2
Using the isentropic relation
T0 = p
(; 1)
T p0 (B.25)
it is found that ; 1
p0 = 1 + ; 1 M 2 : (B.26)
p 2
The mass ow is
m = ACx (B.27)
where A is the ow crossectional area. Alternative forms are
p p
m = AMas = A RTM = A p p M
r p RT
p p
= A p 0 p T0 M T (B.28)
RT0 0
where it is used that = p=RT . Then, by the use of (B.25) and (B.26), it is found
that ; 2(+1;1)
m = Ap p 0 p M 1 + 2 M ; 1 2 : (B.29)
RT0
The critical mass ow, or choked ow, is found by inserting M = 1 and using the
ideal gas law in the form RT0 = 0 =p0 . This gives
2(+1;1)
mchoke = Ap 0 p0 +2 1 (B.30)
2(+1;1)
= A 0 a0 +2 1 : (B.31)
B.5 Compressor Pressure Rise 131
B.5 Compressor Pressure Rise
Here equation (5.42) for the compressor pressure rise is derived. The compressor
total to static isentropic eciency can be written
i (m U1 ) = hh2s ; h01
;h (B.32)
02 01
where h2s is the outlet static enthalpy obtained with isentropic compression, h01
is the inlet stagnation, or total, enthalpy and h02 is the outlet stagnation, or total,
enthalpy.
Considering a perfect gas, we have that h = Tcp, where cp is the specic heat
capacity at constant pressure. For a perfect gas, cp is constant. The eciency
now is
i (m U1 ) = TT2s ; T01 = TT201s ; 1 : (B.33)
02 ; T01 T02 ; 1
T01
Using the relation for isentropic compression (B.25),
p2 ; 1 p2 ; 1
; 1 T01 cp p01 ;1
i (m U1 ) = p01T02 = cp (T02 ; T01) : (B.34)
T01 ; 1
For a radially vaned impeller (Watson and Janota 1982):
p2 ; 1
T01 cp p01 ;1
i (m U1 ) = %h0cideal (B.35)
which can be rearranged to

p2 i (m U1 )%h0cideal
; 1
p01 = 1 + T01 cp : (B.36)
Appendix C
Including a Close Coupled
Valve in the Moore-Greitzer
Model
Equation numbers starting with MG are references to the paper Moore and Gre-
itzer (1986).
Equation (MG5), which gives the pressure rise over the compressor is modied to
pE ; p1 = N F ( ) ; 1 2 @ + @ ;# ( )

s v (C.1)
| U {z 2a @ @ }
2
Equation (5) in Moore and Greitzer (1986)
where #v ( ) is the pressure drop across the CCV, and pE now is the pressure at
the exit of the CCV. Using equation (MG23), the pressure rise over the equivalent
compressor is written
#e ( ) = Ns F ( ) ; 21 2 ;#v ( ): (C.2)
| {z }
c ()
Using this, the local (in ) and annulus averaged momentum balances (equations
(MG42) and (MG43)), are modied to
#( ) + lc dd = c( ; Y ) ; v ( ; Y ) ; mY + 21a (2Y + Y ) (C.3)
and
d 1 Z 2
#( ) + lc d = 2 f c ( ; Y ) ; v ( ; Y )g d: (C.4)
0
In the case of pure surge, that is Y 0, the two momentum balances are the same
and is reduced to
_ = l1 (#c () ; #v () ; #) (C.5)
c
134 Including a Close Coupled Valve in the Moore-Greitzer Model
which is the same expression as used in Simon (1993). The CCV has a character-
istic given by
#v ( ) = 12 2 (C.6)
where > 0 is proportional to the valve opening. Y is now represented by a single
harmonic approximation
Y = WA( )sin( ; r( )) = WA( )sin( ) (C.7)
where A( ) is the time varying stall amplitude. A residue R, is dened as
R=4 Y ; Y : (C.8)

The Galerkin approximation is calculated using the weight functions
h1 = 1 h1 = sin h2 = cos (C.9)
and the inner product
1 Z 2
hR hi i = 2 R( )hi ( )d: (C.10)
0
Calculating hR hi i = 0 for i = 1 2 3, we get
Z 2
M1 = 2 1 e ( + WA sin )d = # + lc dd (C.11)
Z 0
2
M2 = 21 e ( + WA sin ) sin d = (m + a1 ) dAd (C.12)
Z 2
0
M3 = 21 e ( + WA sin ) cos d = ;( drd ( m + 1 ) ; 1 )A:(C.13)
a 2a
0
By using (C.2) and (2.7) the moments Mi are calculated to be
M1 = 2 4 0 ; 2H 3 ; 2W 2 A2 ; 3H A2 ; 42
W3 2 W 2
2
+ 6H + 3HA2
W2 (C.14)
2 3 2AW 6AH
M2 = 2 ; 3HW2 A ; 3HA
4 ; 2 + W (C.15)
M3 = 0: (C.16)
By combining (C.11) to (C.13) with (C.14) to (C.16) and rearranging, the following
dierential equations, which correspond to (MG60) and (MG61), for and J = A2
are found: 3
_ = H ; # ; 0 ; 1 ; 1 + 1 (C.17)
lc H 2 W
3
J
1 W 2J 2 !
+ 2 W ; 1 1 ; 2 ; 2 2H + H (C.18)
2 !
_J = J 1 ; ; 1 ; J ; 12 4W : (C.19)
W 4 3H
135
The dierential equations for and r are left unchanged by the introduction of
the CCV.
136 Including a Close Coupled Valve in the Moore-Greitzer Model
Appendix D
Numerical Values Used in
Simulations
Symbol value Symbol value Symbol value
R 0:1m 1:15 mkg3 as 340 ms
lE 8 lI 2 Lc 3m
Vp 1:5m3 Ac 0:01m2 a 0:3
H 0:18 W 0:25 c0 0:3
I 0:03kgm2 m 1:75
Table D.1: Numerical values used in the simulations of Chapters 2, 3 and 4.
Symbol value Symbol value Symbol value

D2 0:128m Dt1 0:074m Dh1 0:032m
0:9 p01 105Pa 1 1:15 mkg3
T01 303K 1 =2 1b 0:61
Re 100000 D 0:02m 1:4
cp J
1005 kgK Cm 0:01 a01 340 ms
Vp 0:21m3 Lc 1:253m J :001kgm2
Table D.2: Numerical values used in the simulations of Chapter 5.
138 Numerical Values Used in Simulations
Appendix E
Bounds on the Controller
Parameters of Theorem 2.5
In this appendix the upper and lower bounds on the controller gain c3 dened in
Theorem 2.5 are calculated. The numerical values used are
Sym. value Sym. value Sym. value Sym. value
Jmax 4 W 0:25 H 0:18 co 0:3006
choke 0:8 m 0:2
Table E.1: Numerical values
The equilibrium value of the mass ow coecient 0 is found by solving equation
(2.63) with respect to 0 . The lower bound C 1 dened in (2.178) is
2
C 1 = 43WH2 4W0 + W + 0 ; c2 : (E.1)
Using (2.184), the upper bound C 2 is calculated as

!
C 2 = m ; 0 43W
2
( 0 ; m )#^ c( 0 ; m );c2( 0 ; m )2 ; m2 ; 2 m +c2 (E.2)
m H W W
With the help of symbolic toolbox in MATLAB, the following expressions are
found for C 3 , C 4 , C 3 and C 4 :
C 3 (c2 choke ) = ; 2TV0 4+J T1 (E.3)
max
C 3 (c2 choke ) = ; 2TV0 4;J T1 (E.4)
max
where
T0 = ;4 2choke ; 8 choke 0 ; 4 0 + 3 HWJmax choke 0
2
140 Bounds on the Controller Parameters of Theorem 2.5
;3 W 2HJmax choke ; 3 W 2 HJmax 0 + 3 HWJmax 2choke
+2 W 4J
pp max c2
T1 = ;2; 2 T2
T2 = ; 3 HWJmax choke ; 2 choke ; 4 choke 0
2 2
;3 W 2HJmax choke + 3 HWJmax choke 0
+2 W 4Jmax c2 ; 2 20 ; 3 W 2HJmax 0 ( choke + 0 )2
and
C 4 (c2 m ) = ; 2TV04+J T1 (E.5)
max
T 0 ;
C 4 (c2 m ) = ; 2V 4 J T1 (E.6)
max
where
T0 = ;4 (; m)2 ; 8 (; m ) 0 ; 4 20 + 3 HWJmax (; m ) 0
;3 W 2HJmax (; m ) ; 3 W 2 HJmax 0 + 3 HWJmax (; m )2
+2 W 4
p pJmax c2
T1 = ;2; 2 T2
T2 = ; 3 HWJmax (; m )2 ; 2 (; m )2 ; 4 (; m ) 0
;3 W 2HJmax (; m ) + 3 HWJmax (; m ) 0
+2 W 4Jmax c2 ; 2 20 ; 3 W 2HJmax 0 ((; m ) + 0 )2
As the value of 0 depends on the choice of c2 and c3 , the gains have to be
chosen rst, then 0 is calculated, and nally the bounds cmin max
3 and c3 have to
be calculated and checked against c3 . After some iterations of this procedure the
following gains are used: c2 = 1 and c3 = 0:55, 0 is calculated to 0 = 0:34 and
the bounds are calculated to be
C 1 (c2 0 ) = 0:52
C 2 (c2 0 m ) = 2:24
C 3 (c2 0 choke ) = 0:33
C 3 (c2 0 choke ) = 419:34
C 4 (c2 0 m ) = 0:0011
C 4 (c2 0 m ) = 6:83:
Now,
cmin
3 = max fC 1 C 3 C 4 g
= max f0:52 0:33 0:0011g
= 0:52 (E.7)
and
cmax

= min C 2 (c2 m ) C 3 (c2 choke ) C 4 (c2 m )

3
= min f2:24 419:34 6:83g
= 2:24: (E.8)
141
As can be seen the choice c3 = 0:55 now satises
cmin max
3 = 0:52 < c3 = 0:55 < c3 = 2:24: (E.9)

Centrifugal Compressors

Uploaded by

Copyright:

Available Formats

Centrifugal Compressors

Uploaded by

Document Information

Original Description:

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Centrifugal Compressors

Uploaded by

Copyright:

Available Formats

Modeling and Control of Surge and

Rotating Stall in Compressors

Jan Tommy Gravdahl

Jan Tommy Gravdahl

1.3 Stability of Compression Systems

1.3.2 Rotating Stall

In-stall char. comp. char.

1.3.3 Stability Analysis

1.4 Modeling of Compression Systems

1.5 Control of Surge and Rotating Stall

1.5.2 Active Surge/Stall Control

1.6 Contributions of this Thesis

1.7 Outline of the Thesis

The throttle mass ow T (#) is given by the throttle characteristic

2.2.2 Close Coupled Valve

Figure 2.3: Compression system with CCV

2.2.4 Change of Variables

Figure 2.5: Change of variables, principal drawing

To prepare for the analysis of the system, it is desirable to perform a change of

The system characteristics in the new coordinates are de ned as

and in the case of pure surge

2.3.1 Undisturbed Case

Proof: The backstepping methodology of Krstic et al. (1995a) will be employed

Step 1. Two error variables are de ned as z1 = ^ and z2 = ^ ; . The control

Step 2.The derivative of z2 is

Assuming that only an upper bound am on the positive slope of #^ c ( ^) is known,

From (2.68) it follows that V_ 2 is negative whenever W (z ) > 0 (% )2 . Combining

Proof: The controller will be derived using backstepping.

Step 2. The derivative of z2 is

It is now shown that an additional assumption on the disturbance ensures that

Since lim!1 #d2 (=2) = 0 it follows that

Proof: The backstepping procedure is as follows:

The nal upper bound for V2 can now be written as

Now lim!1 #d 2 (=2) = 0 and lim!1 d2 (=2) = 0 implies that

2.3.4 Adaption of Constant Disturbances

Proof: Backstepping is used

2.4 Control of Rotating Stall

The following assumption on the mass ow coecient is also needed:

2.4.1 Undisturbed Case

Step 1. Let the error variables z1 and z2 be chosen as

Step 2. The derivative of z2 is

Choosing c3 according to (2.186) ensures that

The requirements (2.172), (2.178), (2.184) and (2.186) are summarized as

2.4.2 Disturbed Case

where z = (z1  z2  J )T . This implies that z( ) is uniformly ultimately bounded

Remark 2.5 is also applicable to this case.

2.5.1 Surge Control

Surge control will now be demonstrated. A Greitzer parameter of B = 1:8 is used

( 0  0 ) = (0:6 0:6): (2.204)

A simulation of surge induced by a constant disturbance is showed in Figure 2.10.

0 50 100 150 200 250 300 350 400

2.5.2 Rotating Stall Control

Figure 2.12: Same simulation as in Figure 2.11, superimposed on the compressor

Of particular interest here, will be the use of input/output methods to design

Then, it is evident that

3.3.2 Passivity of Pressure Dynamics

Remark 3.1 In Simon and Valavani (1991), the incremental energy

3.3.3 Control Law

Figure 3.1: The closed loop system G1 G2

The throttle mass ow T (#) is given by the throttle characteristic

The system characteristics in the new coordinates are dened as

Proof: The backstepping methodology of Krstic et al. (1995a) will be employed

Step 1. Two error variables are dened as z1 = ^ and z2 = ^ ; . The control

From (2.68) it follows that V_ 2 is negative whenever W (z ) > 0 (% )2 . Combining

Since lim!1 #d2 (=2) = 0 it follows that

Now lim!1 #d 2 (=2) = 0 and lim!1 d2 (=2) = 0 implies that

The following assumption on the mass ow coecient is also needed:

where z = (z1 z2 J )T . This implies that z( ) is uniformly ultimately bounded

( 0 0 ) = (0:6 0:6): (2.204)

Figure 3.1: The closed loop system G1 G2

_^ = 4B12 l ( ^ + ^ d ( ) ; ^ T ( ^)) (3.31)

where the term ( ~ )0 is due to unsteadiness in and g. A straight inlet duct of

; 21a 2( ~0 )0 + UU ( ~0 )0 (4.46)

= 2 @@ + 2( ~0 )0 + ( ~0 )0 (4.47)

; 21a 2( ~0 ) + UU ( ~0 ) ; ( ~0 )

; m 1(B ) #( ) + lc(B ) dd + lE Ud;+ b

where n = 1 : : : N (), Jn is dened as the square of the stall amplitude An

4.4.1 Unstable Equilibrium, = 0:5

( # B J1 J2 J3 )0 = (0:55 0:65 0:1 0:1 0:05 0:01): (4.78)

Figure 5.3: Diagrammatic sketch of centrifugal compressor tted with a volute.

Figure 5.7: Incidence angles at diuser.

W1b = sinC1 : (5.32)

where p01 and a = pRT

+U^ 2 ; k2iID1 I^2 ; k2pID1 U^ I^ ; U^ ^c ; D 1^

; (4D1kp 2;I D2 m0 ) U^ I^ (4D1 kp 4;I D2 m0 )

(_) Total time derivative, dtd or dd

%bf Loss in eciency due to backow (5.39)

Virtual control used in backstepping