Automatica, Vol. 32, No. 1. pp.
109-115, 19%
Pergamon 00051098(95)00096-8
Active Control of Compressor Stall Inception:
a Bifurcation-theoretic Approach*
DER-CHERNG LIAWt
Key Words-Compressors; stability; stabilizers; nonlinear control systems; control applications.
Abstract-Active control of the onset of stall instabilities in
axial flow compression systems is pursued using bifurcation
analysis of a dynamical model proposed by Moore and
Greitzer. The state variables of this model are the mass flow
rate, pressure rise, and the amplitude of the first-harmonic
mode of the asymmetric component of the flow. A
subcritical pitchfork bifurcation is found to occur at the
inception of stall, resulting in a large-amplitude instability
and associated hysteresis behavior. Using the throttle
opening of the compression system for actuation, it is found
that the eigenvalue that becomes zero at the onset of stall is
linearly uncontrollable. This, along with uncertainty of the
post-stall linearized model, motivates a consideration of
nonlinear feedback control laws for mitigation of the jump
and hysteresis behavior occurring at stall onset. Following
the bifurcation control calculations introduced by Abed and
Fu [Abed, E. H. and J. H. Fu (1986). Local feedback
stabiliiation and bifurcation control, I. Hopf bifurcation.
Syst. Control Lett., 7, 11-17. and Abed, E. H. and J. H. Fu
(1987). Local feedback stabilization and bifurcation control,
II. Stationary bifurcation. Syst. Control Letr., 8, 467-4731, it
is found that feedback incorporating a term quadratic in the
first-harmonic flow asymmetry variable renders the
pitchfork bifurcation supercritical. This introduces a new
stable equilibrium near the nominal equilibrium after the
nominal equilibrium itself has lost stability, thus eliminating
the undesirable jump and hysteresis behavior of the
uncontrolled system.
1. Introduction
Recently there has been significant interest in the dynamics
and control of stall phenomena in axial flow compression
systems (see e.g. Kerrebrock, 1992). This interest stems from
two facts. First, stall phenomena can occur when axial flow
compressors are operated near their maximum achievable
pressure rise. Second, high-pressure-rise operation is
associated with increased performance and efficiency. Thus
the desirable increase in performance is at the price of a
significantly reduced stability margin. Active control to
alleviate the compromised stability of an axial flow
compression system can be of great value. In modem gas
*Received 20 July 1992; revised 30 June 1993; revised 6
June 1994; received in final form 16 June 1995. The original
version of this paper was presented at the 2nd IFAC
Symposium on Nonlinear Control, which was held in
Bordeaux, France during 24-26 June 1992. The Published
Proceedings of this IFAC Meeting may be ordered from:
Elseiver Science Limited, The Boulevard, Langford Lane,
Kidlington, Oxford OX5 lGB, U.K. This paper was
recommended for publication in revised form by Associate
Editor Joe Chow under the direction of Editor Tamer Basar.
Corresponding author Professor Eyad H. Abed. Tel. +l 301
405 3631; Fax +l 301 314 9920; E-mail abed@eng.umd.edu.
t Department of Control Engineering, National Chiao
Tung University, Hsinchu, Taiwan, R.O.C.
$Department of Electrical Engineering and the Institute
for Systems Research, University of Maryland, College Park,
MD 20742, U.S.A.
109
Copyright 0 1995 Elsevier Science Ltd
Printed in Great Britain. All rights reserved
mm-10981% b5.00 + 0.00
Brief Paper
and EYAD H. ABEDS
turbine jet engines, for example, the ability to safely operate
an axial flow compressor at or near its maximum pressure
rise has important implications for reduced size and weight
requirements (Kerrebrock, 1992).
Two basic types of instability are known to occur in axial
flow compression systems. One, surge, is a low-frequency,
large-amplitude oscillation of the mean mass flow rate. The
second, rotating stall, corresponds to a traveling wave of gas
around the annulus of the compressor. This results in very
inefficient operation at constant mean mass flow rate and
pressure rise. These descriptions are for the fully developed
(i.e. post-transient) instabilities.
Greitzer (1976) employed a nondimensional fourth-order
compression system model and introduced a nondimensional
parameter B, which he found to be a determinant of the
nature of post-instability behavior. A global bifurcation of
periodic solutions and other bifurcations were found for this
model, and were used to explain the observed dependence of
the dynamical behavior on system parameters (Liaw et a/.,
1991; Abed et al., 1993).
Moore and Greitzer (1986) extended the previous model
to describe surge and rotating stall phenomena in axial flow
compression sysems. This model incorporates non-
axisymmetric dynamics, whereas the model of Greitzer
(1976) reflected only axisymmetric flow dynamics, while
employing a nonaxisymmetric (i.e. measured) steady-state
compressor characteristic. Moore and Greitzer (1986) also
specialized the general model to a system of three nonlinear
ordinary differential equations. The state variables of this
dynamic model are the mass flow rate, pressure rise, and the
amplitude of the first harmonic mode of the asymmetric
component of the flow. For the case of a cubic axisymmetric
compressor characteristic, Moore and Greitzer found it
convenient to use the square of the amplitude of the first
harmonic mode rather than the amplitude itself as a state
variable. McCaughan (1989) performed a bifurcation analysis
of this model, observing a stationary bifurcation at stall
inception. This bifurcation entails the local emergence of a
new equilibrium point from the nominal one. This bifurcated
equilibrium point is not stable, and results in a jump effect
and thus hysteresis. Note that McCaughan (1989) also
observed a bifurcation to a large-amplitude periodic solution
for the model of Moore and Greitzer (1986). (Although
bifrucations of equilibria are discussed briefly in Section 2
and the Appendix, we refer the reader to texts such as Iooss
and Joseph (1990) for a detailed treatment.)
Several techniques have been proposed for active control
of install instabilities in axial flow compressors (see e.g.
Hosny et al., 1991; Badmus et al., 1991; Paduano et al., 1993).
From an analytical point of view, these methods employ
linear control to delay the occurrence of stall or to achieve
stall avoidance. Of course, the physical mechanisms for
controller implementation differ among the proposed active
control schemes.
The present paper begins with the recognition of the
importance of local bifurcations as determinants of the
nature of post-instability behavior of axial flow compression
systems. The philosophy of the control component of this
 Brief Papers
work is similar to that of Abed and Fu (1986, 1987). This
entails determining feedback control laws that ensure that a
given local bifurcation results in only stable bifurcated
solutions. Thus, even though the nominal equilibrium is not
stabilizable within the framework of linear theory, it may be
possible to stabilize a neighborhood of the nominal solution
for a range of parameter values including the stall value of
the disturbance parameter. It will be seen that an additional
outcome of the control laws proposed here is the elimination
of a hysteresis loop that occurs in the open-loop system
model. The results of this paper apply to models more
general than that stuided by McCaughan (1989). in the sense
that the axisymmetric part of the steady state compressor
characteristic is not required to be a cubic function. but
rather an arbitrary smooth single-valued function of mass
flow rate. We do employ a cubic model in demonstrating the
results for a particular compression system.
The paper proceeds as follows. In Section 2 bifurcation
theory is applied to study the stability of axial flow
compression systems in the vicinity of the stall point. A
pitchfork bifurcation is observed in the model at the stall
point. The dynamical behavior of the compression system
near this bifurcation point is found to be strongly dependent
on the axisymmetric compressor characteristic. This depen-
dence is exhibited through a formula showing that the
stability of bifurcated solutions is influenced by the
derivatives of the axisymmetric compressor characteristic at
the bifurcation point. In Section 3 a throttle opening control
law is given. This circumvents the uncontrollability of the
eigenvalue of the linearized model that vanishes at stall
onset. A purely quadratic state feedback using measurement
of the asymmetric component of the flow is given, and is
shown to result in local stabilization of the bifurcation
leading to stall. In Section 4 an illustrative example is
studied, and the results for this example are depicted using
bifurcation diagrams for the uncontrolled and controlled
bases. Concluding remarks are given in Section 5.
We employ the following notation:
e angle along circumference;
G, nondimensional axisymmetric compressor charac-
teristic (a function):
A amplitude of the first harmonic of asymmetric flow;
+k nondimensional compressor mass flow rate:
AP nondimensional plenum pressure rise;
F inverse function of nondimensional throttle pres-
sure rise:
Dx, h partial derivative of function h with respect to the
variables x, y, ; This result is useful in that it classifies the nominal
(.)’ differentiation.
2. Bifurcation analysis of stall inception
A third-order lumped parameter model, in terms of
nondimensional variables, for an axial flow compression
system has been introduced by Moore and Greitzer (1986).
The model is
dA a 2n
C,,(t+ + WA sin 0) sin B de, (la)
dt - nW I<,
In
d+
dr= -AP+k C,,(& + WA sin 0) de, (lb)
i II
dAP 1 appearance of limit cycles in the dynamics, known as
dt = 482 [tic - F(ys AP)l~ (lc)
where W, a >O are two constants, Y is a parameter
associated with the throttle opening, and B is the
nondimensional parameter introduced by Greitzer (1976),
which is proportional to rotor speed. In the sequel, B > 0 and
F is a strictly increasing function with respect to each of the
variables Y and AP.
Taking the compressor characteristic C,, to be a cubic
function of tic, and using the squared amplitude A* (instead
of A. say) as a state variable, Moore and Greitzer (1986)
were able to explicitly evaluate the integrals appearing in (1).
giving a more convenient dynamic model. McCaughan (1989)
performed a bifurcation analysis of this model. One
conclusion of McCaughan is that a stationary bifurcation
from the nominal equilibrium point occurs as the throttle
opening parameter Y is varied. However, a similar analysis
for the more general case in which the axisymmetric
compressor characteristic is not necessarily a cubic function
of ti(- has not been reported. In this section we extend the
study of McCaughan (1989) to the case in which the
axisymmetric compressor characteristic is taken to be a
general smooth function of mass flow rate. Note that in
Section 4 we also employ a cubic axisymmetric compressor
characteristic, by way of illustration.
Suppose C,, is a smooth function of tic, and solve for the
equilibrium points of (1). By (la), it is easy to see that A = 0
always results in dA/dt =O. However, there may be
equilibrium points of (1) for which A ZO. First let us
consider equilibrium points in the case A = 0. Denote such
an equilibrium point, whose location depends on the
parameter Y. as x”(y) = [0 &O,(Y) Ap(y) The values
k?(y) and AP”(y) should then satisfy the relationships
ti:‘. = F(y. APO) and AP” = C&F). Under the assumption
A = 0, one such equilibrium point x0(y) is referred to as the
unstalled or nominal equilibrium point for the axial flow
compression model (1). Note that this is the normal
operating point of the system, whose location depends on the
throttle control parameter y. In the following, we consider
the stability of the unstalled equilibrium x0(y) as it depends
on Y. Viewing y as a bifurcation parameter, we study
possible bifurcations from x”(y) at parameter values for
which stability is lost. These bifurcations result in new
equilibrium points of the model. At these new equilibria.
A # 0. corresponding to stall inception.
Let X = [x, x2 ~~1’~denote the state variation of (1) near
the unstalled equilibrium point x(‘(y), where x1 = A,
x2 = ti, - r@.(y) and x3 = AP - APO(y). The linearization
of (1) at x*(y) is
with
/uC:,(t+?(Y)) 0 0
\
0 Cl,(+) -1
1 (3)
0 - ~$ DA&Y, AP")
4BZ
From (2) and (3), we readily obtain the following stability
result.
Lemmu I. The equilibrium point x”(y) is asymptotically
stable for the system (1) if C~,(r$.(y)) <O, while it is
unstable if C:,(&(Y)) > 0.
equilibrium as being stable or unstable, depending on the
value of the parameter y. Since we are interested in the
behavior of the compression system (1) for values of y for
which the nominal equilibrium is unstable, we are led to
study the behavior of (1) for values of y near the critical
value at which stability is first lost.
An equilibrium point of a dynamic system depending on a
single parameter typically loses stability either by an
eigenvalue of the system linearization vanishing as a
parameter approaches a critical value, or by a conjugate pair
of eigenvalues crossing the imaginary axis away from the
origin. The latter situation is associated with the local
Andronov-Hopf bifurcation. We shall not study this
(dynamical) bifurcation in any detail in this paper, since we
focus on a stationary bifurcation shown below to occur in the
compression system model at the initiation of stall. A
stationary bifurcation of an equilibrium of a dynamic system
typically occurs when an eigenvalue becomes zero as a
parameter is varied quasistatically. Stationary bifurcation
entails the appearance of new equilibrium points near the
nominal one as the critical parameter value is neared. The
generic stationary bifurcation entails the merging of the
nominal equilibrium with a saddle equilibrium and their
mutual annihilation at the critical parameter value. This
bifurcation. known as saddle node bifurcation, is not
encountered in our study. In this paper the nominal
equilibrium does not disappear, but only loses stability at the
bifurcation. In fact, we observe a particular type of

stationary bifurcation in our analysis, namely a pitchfork
bifurcation. In a pitchfork bifurcation, two new equilibrium
points emerge from a given one, and both occur locally on
the same side of criticality (in (1) this means that the new
equilibria either both occur for y > ye or for y < ya).
The Jacobian matrix of (1) at the nominal equilibrium
point is &,, given in (3). It is not difficult to see that for this
matrix to have a pair of pure imaginary eigenvalues,
C:.(rir’dy)) must be positive. (Simply examine the trace of
the lower right 2 X 2 submatrix of (3).) However, if this were
the case then the matrix Lo would have aCL(&(y)) as a
positive real eigenvalue, and the equilibrium would
therefore be unstable. This means that a stationnry
bifurcation must occur before an Andronov-Hopf bifurcation
for the nominal equilibrium x0 of the model (1). In the
remainder of this aper, therefore, we focus on a stationary
bifurcation from x !?and its stabilization.
From (3), the linearization of (1) has one zero eigenvalue
and two stable eigenvalues when C#$(y)) = 0. This
implies that a stationary bifurcation may occur from the
equilibrium point x0 for some value of y: y = y”. To analyze
the bifurcation behavior of the model, we employ a result
from bifurcation analysis recalled in the Appendix. This
result allows us to derive conditions for the existence and
stability of a bifurcation from the nominal equilibrium.
Let x0 be an equilibrium point at which C:&@(y)) = 0 for
some y = y”. The Taylor series expansion of (1) for (x, y)
near (x”, y”) is given by
$= LOX + QO(X, X) + C,(X, X, X) + (y - y’)L,X +. . .
Here Lo is as in (3), and
~C~(rir~)X,X*
:c:&@)(w*x: + 2x:)
QoK -V =
(
- & D~~P@(Y~, AP”)x: IT
++Yc:(mo,)(wzx: +4x,x3
C;(a;)($W*x:x, + ix:)
Co(X, x, X) =
- & Dw,9(y”, @)x:
( )I
0 0
@I G(rS)
L, = 0 41 G(r%) 0 (7)
i 0 0 -+2/4B2 1 ’
where
4i := D,F(y”, APO), I#J*:=D,,,F(y’, AP’-‘).
Set I= (LO, 0) and r = IT, the left and right eigenvectors
respectively corresponding to the zero eignevalue of L,,.
Next, we analyze the effect of small variations in y on the
dynamical behavior of (1) in the vicinity of the equilibrium
point x0.
To check the transversality condition iL,r f 0 (see the
Appendix), we compute
lL,r = a&C~&izO,).
We calculate the bifurcation stability coefficients using (A.3)
and (A.4) and the notation of the Appendix, obtaining
PI = lQo(r,4 = 0,
P2= 21[2QohVI+ Co(r,r, 41
= faW2{2D,&yo, AP”)[C&(tio,)]‘+ C::(&)}.
The next result follows readily from Lemma A.1 and the
discussion above. It is important to note that the stability
characteristics of the bifurcated equilibria determine whether
or not hysteresis and jump behavior result from the
bifurcation.
Theorem 1. Suppose that C&(&) # 0 and that F is strictly
Brief Papers 111
increasing in each of its variables. Suppose also that the
stability coefficient & given in (11) is nonzero. Then the
system (1) exhibits a pitchfork bifurcation with respect to
small variations of y at the point (x0, y”) where CA(&) = 0.
Moreover, if /J2 < 0 (respectively p2 > 0), the bifurcation is
supercritical (respectively subcritical), and the local bifur-
cated equilibria near x0 are asymptotically stable (respec-
tively unstable).
Note that the formula (11) shows that the stability of
bifurcation equilibria is influenced by the derivatives of the
axisymmetric compressor characteristic at the bifurcation
point. Theorem 1 quantities the local dependence of
post-stall behavior of the compression system on the
axisymmetric compressor characteristic (i.e. on the function
C,). Bifurcation diagrams for an axial flow compression
system with a particular form for C, are given in Section 4.
3. Control of stall inception
From Lemma 1, the unstalled equilibrium point becomes
unstable after the parameter y passes through the critical
value v”. The ooint (x0. r”) is therefore called the stall
incepfidn point. Moreover; according to Theorem 1, the local
bifurcated solutions near the stall inception point might not
be stable. In this case the bifurcation is said to be. subcritical.
If such a condition occurs, the compression system may
exhibit a jump from the stable nominal equilibrium when the
parameter y crosses the critical value y”. This results in a
hysteresis loop in the dynamics of the system with respect to
the parameter y near the point of stall initiation. An example
of such behavior is given in Section 4.
In this section we seek feedback control laws governing the
(4) throttle setting near the point of stall onset that prevent the
occurrence of this hysteresis or jump behavior. The control
design begins by observing that jump behavior occurs
because there is no nearby stable equilibrium after the
nominal equilibrium loses stability. A low-energy throttle
(5) control is applied to the system, which is equivalent to
replacing the throttle parameter y in (1) by y + u, where IAis
a control signal that is of small-amplitude near the nominal
equilibrium. By examining the Jacobian matrix Lo of the
system as given in (3), we find that throttle control cannot
affect the system eigenvalue that passes through zero as y
(6) varies through the critical value y”. (This is simply the (1,l)
element of Lo, because of the block-triangular structure of
Lo.) Given these circumstances, we follow Abed and Fu
(1987) to seek nonlinear feedbacks that transform the
subcritical bifurcation at y” into a supercritical bifurcation,
i.e. to give rise to stable equilibrium points locally. These
equilibrium points serve as alternative steady states, so that a
jump to a distant equilibrium will no longer occur. Note also
that, even if the critical mode had been linearly controllable,
(8) a nonlinear feedback rendering the bifurcation supercritical
may be more robust to model uncertainty than a linear
stabilizing feedback.
Note that such a control law, although obtained through
local analysis, mitigates an undesirable global effect, namely
a large-amplitude jump and hysteresis. As in Section 2,
denote by y” the critical value of the throttle control
parameter at which stability of the nominal equilibrium is
first lost. We proceed to design a control law ensuring that
(9) the local bifurcated solutions near the stall inception point
are stable, i.e. that the pitchfork bifurcation is supercritical.
An important consequence of this is that the controlled
system will not exhibit jump or hysteresis phenomena near
the nominal equilibrium.
(10) Denote the stall inception point by (x0, y”), and let
y := y” + u, where u is the control input. Then we can
rewrite the system (1) as the nonlinear control system
(11)
.dA ..--.=-
a 2n
C,,(hc + WA sin 0) sin 13de, (I2a)
dt nW Io
2s
$= -Ap+& C,,(tic + WA sin 0) de,
(I2b)
I 0
dAP
-=- 1 .
F(y” + u, AP)].
df 482 ]mc -
112 Brief Papers

In this section these equations and their control are of the form
considered without specifying parameter values or the form
of the steady-state axisymmetric compressor and throttle ~1= y,A’ + qZA(tic - riz:‘-)+ q,A(AP - AP”) + qd(ljlc- - +I$)’
characteristics. In Section 4 an example is given where these
are specified. + y5(tic - ti$)(AP - AP”) + q6(AP - A/“))*, (13)
It is observed from (12a) that A = 0 is an invariant
manifold of (12) regardless of the choice of control input 11.
where the 4, are constants. We seek values of the control
implying that (12) is uncontrollable. Moreover, it is not
difficult to check that the system (12) possesses an gains q, such that, for the system (12) with u given by (13), /3Z
at the equilibrium x0 for y = y” will be rendered negative.
uncontrollable zero eigenvalue at the point of stall initiation.
This means that we cannot extend the stable region of the Denote by 07 and ps the two stability coefficients of the
controlled mode1 (12) at x” for y = y”. Formulas (A.3) and
nominal equilibrium point for a broader range of values of
the parameter y by using a linear state feedback. (A.4) are also valid for the controlled system. Using these. it
To make the selection of a control law tractable and is straightforward to express PT and Pt in terms of p, and
systematic, we begin by restricting the control to belong to a pz. the stability coefficients of the uncontrolled version of
(12). Indeed. we find that p1; = pi =0 and pt=pZ +
parameterized family of smooth feedback control laws. Next,
we further restrict the control by keeping only those terms in 2aq,D,F(y”, AP’i)C&(ti$). Note that PT = 0 regardless of
the choice of control gains. By Lemma A.1 in the appendix.
a local series expansion that can readily be seen to influence
this implies that if pit # 0 then a pitchfork bifurcation must
the stability coefficients /3, and & at the stationary
also occur in the controlled system. From the expression
bifurcation point. In this way. we obtain a local control
design. Formulas (A.3) and (A.4) show that. in general, only above for /3T. it is apparent that only the quadratic
terms up to cubic order in the state affect the values of the component of the feedback, q1A2, contributes to the
determination of the sign of p$. Because of this, we obtain a
stability coefficients PI and & at a stationary bifurcation
point. This leads us to further limit the search to control very simple quadratic feedback controller for the system
(12). as summarized in the next theorem. We note, though,
laws containing only linear and quadratic terms in the state.
that other terms can be reintroduced in the control law if
Next, we notice that linear terms in the feedback control
desired, to improve or tune performance for nonlocal
might affect eigenvalues and eigenvectors of the system
excursions of state variables and parameters.
linearization at the bifurcation point. (However, recall that
the zero eigenvalue. being uncontrollable. would be
unaffected by linear feedback terms.) The analysis would 7lrroren1 2. Let C;l,(ti:‘.) # 0 at the point (x0, y”). Then the
become cumbersome if eigenvalues and eigenvectors were stationary bifurcation of (12) at (x”, y”) can be rendered a
not fixed, as can be seen from an examination of (A.3) and supcrcrittcal pitchfork bifurcation bv a purely quadratic
(A.4). Thus we arc led to investigate a feedback control law feedback control of the form 11= y,il! .

(4
Fig. I. Compression system open-loop operating points.
 Brief Papers 113
Pmsmre Rise AP MSS Flux tic
2.

Throttle opening 7 Throttle opening 7

00 (b)
Amplitude of Asymmetry Dynamics A Pr- F&e AP
0.75 2.75

Throttle opening -r

(4

Fig. 2. Compression system closed-loop operating points.

4. Illustrative example
In the foregoing, we have considered the analysis and
control of the initiation of stall and hysteresis behavior by
focusing on control of a stationary bifurcation occurring in
the third-order compression system model of Moore and
Greitxer (1986). In this section we illustrate the results for a
particular compression system model employing, for
simplicity, a cubic axisymmetric compressor characteristic.
As noted in Section 2, the results of this paper are applicable
to compression system models with noncubic compressor
characteristics as well.
The computations we report in this section were obtained
using the numerical continuation and bifurcation analysis
package AUTO (Doedel, 1981). These computations
illustrate the compression system behavior for the chosen
example. The behavior of both the uncontrolled and
controlled systems is considered. The compression system
example considered next involves the cubic axisymmetric
compressor characteristic given in (14) below, as well as the
throttle characteristic given in (15).
Let the axisymmetric compressor characteristic C&c) be
C&c) = 1.56 + 1.5(& - 1) - 0.5(& - l)3. (14)
Let F (the inverse of the throttle pressure rise map) be given
by
F(y, AP) = ym.
(15)
Choose parameter values TV= 0.4114, W = 1.0 and B = 0.5.
It is easy to check that there are two values of y for which
the compression system (1) has an equilibrium point
(A, kc, AF) with CL(titc) =O. Either using the analytical
results of the preceding sections or numerically, we can show
that these are pitchfork bifurcations. One pitchfork
bifurcation occurs for a large value of pressure rise AP, while
the other occurs for small pressure rise. The former pitchfork
bifurcation is found to be subcritical (i.e. it gives rise to
unstable equilibria) while the latter is supercritical (i.e. it
gives rise to stable equilibria). These observations are
depicted in Fig. 1. Solid lines indicate stable equilibria, while
dotted lines indicate unstable equhbria. It is clear from this
figure that the pitchfork bifurcation of most relevance in a
practical setting, i.e. the one occurring for a larger AP, is
subcritical. Thus hysteresis is expected to occur in the
compression system, and a hysteresis loop is discernible in
Fig. 1.
Each of the graphs in Fig. 1 shows the effects of the
pitchfork bifurcations occurring in the system on a system
state variable. Figure l(c) is perhaps the clearest. We see
from this figure that a subcritical pitchfork bifurcation occurs
from the nominal equilibrium as y is decreased. A second,
supercritical, pitchfork bifurcation occurs at y = 0. Note that
Figs l(a, b,d) also reflect these pitchfork bifurcations, but
symmetry in the other state variables results in their being
double-valued for the bifurcated solution. Hence these
figures seem to indicate only one bifurcated equilibrium,
rather than two.
Now, following Theorem 2 of Section 3, we choose the
control input u = l.OA*, which is quadratic in the asymmetric
flow amplitude A. In Fig. 2 bifurcation diagrams for the
system with this control law in effect are given. It is clear
from this figure that the hysteresis loop of the stable system
equilibria of Fig. 1 has been eliminated. The system will tend
to operate at the stabilized bifurcated equilibria after the stall
inception point, thus maintaining performance near the
maximum pressure rise. Note that although the figures depict
behavior with respect to large parameter variations, the
114 Brief Papers
theory of this paper applies only in a local neighborhood of
the point of stall initiation. Thus in specific applications our
results must be combined with numerical calculations aimed
at ensuring a design that is effective for system parameter
ranges that are of interest.
5. Conclusions
The control of axial flow compression systems at the
initiation of stall has been studied using the model of Moore
and Greitzer (1986). The state variables of this model are
pressure rise, mass flow rate, and the first harmonic of the
asymmetric component of the Row. It has been found that
control based solely on the system linearization cannot
ensure stable operation past the stall point, owing to
uncontrollability of the system eigenvalue. which causes
instability as the throttle parameter is varied. The
bifurcation behavior of the model has been studied and used
as a basis for nonlinear control design. A simple bifurcation
analysis showed that quadratic feedback of the first harmonic
of the asymmetric component of the flow could render the
bifurcation supercritical. The effect of this for the
compression system is the replacement of the global jump
and hysteresis behavior occurring in the uncontrolled system
with a new, stable. behavior near the original operating
point.
Acknowledgements-The authors are grateful to the referees
for their helpful comments and careful reading of the paper.
This work was supported in part by the General Electric
Comoanv. bv the AFOSR under URI Grant AFOSRXI-
0015: by- the’ NSF under Grant ECS-86-57561, by the NSF
Engineering Research Centers Program NSFD CDR-88-
03012. During the preparation of this paper, the authors were
supported in part by the National Science Council, Taiwan.
R.O.C. under Contracts NSC 81-0401-E-009-637 and NSC
82-0424-E-009-324, and by the AFOSR under Grant
F4%20-93-l-0186.
References
Abed, E. H. and _I. H. Fu (1986). Local feedback
stabilization and bifurcation control, I. Hopf bifurcation.
Syst. Control Lert., 7, 11-l 7.
Abed, E. H. and J. H. Fu (1987). Local feedback
stabilization and bifurcation control, II. Stationary
bifurcation. Syst. Conrrol Letr., 8, 467-473.
Abed, E. H., P. K. Houpt and W. M. Hosny (1993).
Bifurcation analysis of surge and rotating stall in axial flow
compressors. ASME J. Turbomachinery, l&817-824.
Badmus, 0. O., C. N. Nett and F. J. Schork (1991). An
integrated, full-range surge control/rotating stall avoid-
ance compressor control system. In froc. American
Control Conf., Boston, pp. 3173-3179.
Chow, S. N. and J. K. Hale (1982). Methods of Bifurcation
Theory. Springer-Verlag, New York.
Davis, M. W., Jr and W. F. O’Brien (1991). Stage-by-stage
poststall compression system modeling technique. A IAA J.
Propulsion and Power, 7,997-1005.
Doedel, E. .I. (1981). AUTO: A program for the automatic
bifrucation analysis of autonomous systems. Congressus
Numerantium, 3& 265-284.
Greitzer, E. M. (1976). Surge and rotating stall in axial flow
compressors. Part I: theoretical compression system
model. ASME J. Enging for Power, 98, 190-198.
Hosnv. W. M.. L. Leventhal and W. G. Steenken (19911.
A&e stab&ation of multistage axial-compressor aerody-
namic system instabilities. In Proc. ASME Gas Turbine
Conf, Orlando.
Iooss, G. and D. D. Joseph (1990). Elementary Stability and
Bifurcation Theory, 2nd ed., Springer-Verlag, New York.
Kerrebrock, J. L. (1992). Aircrafr Engines and Gas Turbines,
2nd ed., MIT Press, Cambridge, MA.
Liaw, D. C., R. A. Adomaitis and E. H. Abed (1991).
Two-parameter bifurcation analysis of axial flow compres-
sor dynamics. In Proc. American Control Conf, Boston,
pp. 2955-2960.
McCaughan, F. E. (1989). Application of bifurcation theory
to axial flow compressor instability. ASME J.
Turbomnchinery, 111,426-433.
Moore, F. K. and E. M. Greitzer (1986). A theory of
post-stall transients in axial compression systems. Part
I-development of equations. ASME J. Engng for Gas
Turbines and Power, 108,68-76.
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 compressor. ASME J.
Turbomachinery, l&48-56.
Appendix-bifurcation stability coeficienrs
In this appendix we recall a result on pitchfork bifurcation
of one-parameter families of nonlinear systems.
Consider a one-parameter family of autonomous systems
.f = f(x. CL). (A.1)
where x E BB”,p E R and the vector field f is sufficiently
smooth in x and CL.Suppose that (A.l) possesses a nominal
equilibrium point that depends smoothly on p for all values
of p in some interval of interest. (This means we are
considering bifurcation from known solutions (Chow and
Hale, 1982).) For simplicity of notation, suppose f(0, p) = 0
for all CL.Let the Jacobian matrix D,f(O, 0) possess a simple
zero eigenvalue. with all remaining eigenvalues in the open
left-half complex plane. The parameter p is referred to as the
hifrucation parameter.
The formulas and results given below remain valid if the
nominal equilibrium is not the origin and the critical
parameter value is not p = 0; simply evaluate all quantities
(including partial derivatives) at the adjusted equilibrium and
parameter value.
The equilibrium points of system (A.l) are the solutions
of f(x, II) =O, and thus depend on the value of the
parameter CL.With the assumption that D,f(O, 0) is singular,
the system (A.l) may possess several equilibrium paths .X(P)
emanating from the origin for p near 0. Such a situation is
known as a stationary (or static) bifurcation, and the point
I = 0. p = 0 is a bifurcation point. The solutions x(p) # 0 of
(A.1) emanating from the origin for p near 0 are called
bifurcated equilibria.
Following the notation of Abed and Fu (1987). we write
the Taylor series expansion of (A.l) at (0,O) as
.i = f(x. CL)
= L,,x + Qo(x. x) + C{,(X, x, I) +.
+ CL(L,X+Q,(~,~)+...)+CLI(L~X+...)+ __.,
(A.2)
where
1
Q,(x, x) := pjj DfiArxf(x> CL),
C&x, x, x) := & D,r-,,f(x, CL) etc.,
are the quadratic terms, cubic terms etc. off (x, p). Here the
quadratic and higher-order terms in the expansion are
written as symmetric forms. For instance, Qk(x, y) =
Qk(y, x) for each k 20 and any x, y E R”. Moreover. in
(A.2). Lk := (l/k!)D,ti,f(O, 0) for k 2 1, and the Jacobian
matrix L,, is assumed to have only a simple zero eigenvalue,
with the remaining eigenvalues stable.
Denote by 1 and r the left (row) and right (column)
eigenvectors respectively of the matrix Lo corresponding to
the simple zero eigenvalue. Set the first component of r equal
to 1 (possibly after a reordering of state variables) and
choose the left eigenvector 1 so that lr = 1. Denote
B, := lQo(r. r). (A.3)
PZ : = 21[2Q”(r. u) + C&. r, r)], (A.4)
where u is the unique solution to
&,u = -Qo(r. r). (A.5)
lu = 0. (A.6)
We are now in a position to recall the following stability
criterion for the system (A.2). This criterion applies only in
the case PI = 0, and is useful for studying pitchfork
bifurcation. Thus this result is sufficient for the present
 Brief Papers 115

paper. If it happens that pi ZO then another type of
stationary bifurcation, a transcritical bifurcation, would be
expected to occur. For details, see Abed and Fu 61.1987).
Lemma A.l. Let the assumptions above hold, and assume
the transversality condition lL,r #O. Then (A.2) undergoes
a stationary bifurcation from x = 0 at p = 0. If /3i = 0 and
/3s < 0, the bifurcated equilibrium points of (A.2) for p near
0 are asymptotically stable. If, on the other hand, pi = 0 and
& > 0 then the bifurcated solutions of (A.2) for p near 0 are
unstable. In the former case the bifurcation is said to be
supercritical; in the latter, it is subcritical. In either case the
bifurcation is a pitchfork bifurcation, and the bifurcated
equilibria locally occur only for p(IL,r)/& CO.