Active surge control of

centrifugal compressors using drive torque

Jan Tommy Gravdahl , Olav Egeland and Svein Ove Vatland

Department of Engineering Cybernetics, NTNU, N-7491 Trondheim, Norway

ABB Corporate Research, Bergerveien 12, N-1375 Billingstad, Norway

Abstract
Flow in

A novel aproach to active surge control is presented. A

centrifugal compressor driven by an electrical motor is
studied, and the drive itself is used for surge control,
thus elliminating the need for additional actuators. It is
shown that by using the rotational speed of the motor as
control, previous unstable operating points to the left
of the surge line can be made globally exponentially
stable. It is also shown that using the torque of the
drive as control ensures exponential convergence. The
proposed method is simulated on a compressor model
using an approximation of a real compression system.

1 Background
Surge is an unstable operation mode of centrifugal compressors, which occurs when the operating point of the
compressor is located to the left of the surge line, which
is the stability limit in the compressor map. The phenomenon is manifested as oscillations of the mass flow,
pressure rise and rotational speed of the compressor.
Surge is highly undesired, and can cause severe damage
to the machine. Traditionally, surge has been avoided
using a surge avoidance scheme. Such schemes use various measures to keep the operating point of the compressor away from the surge line. Typically, a surge
control line is drawn at a distance from the surge line,
and the surge avoidance scheme ensures that the operating point does not cross this line. This method
restricts the operating range of the machine, and efficiency is limited. Usually a recycle line around the
compressor is used as actuation. Active surge control
is fundamentally dierent to surge avoidance in that the
unstable phenomenon is sought to be stabilized instead
of avoided. Thus the operating regime of the compressor is enlarged.
Active surge control of compressors was first introduced
by [1], and since then a number of results have been
published. Dierent actuators have been used, and examples include recycle, bleed and throttle valves, gas
injection, variable guide vanes and a number of others.
For an overview, consult [2] or [3]. In this work we
propose to use the electrical drive of the compressor as
a means of active surge control, as depicted in Figure
1. The advantage of this is that the drive is already
present, and no additional actuation device is required.
This means that the compressor can be operated at a
low flow without recycling, and there is a potential for
reduced energy consumption of the compressor.

Command
To drive
Electric drive
with control
system

Shaft
Compressor
Shaft speed
Flow out

usurge

Active Surge
Control law

uperfomance

Compressor
performance
control

Performance
Set point

Figure 1: The compression system considered cosists of

a centrifugal compressor driven by an electrical
motor.
Compressor

Plenum

Vp

Duct

p 01

p 02

p
Tp

mt (p )
Throttle

Figure 2: The compressor, plenum, throttle system of [4]

2 Model
2.1 Dynamics
A classical result in the field of compressor surge modeling is the model of Greitzer [4] who modelled a basic compression system consisting of a compressor, a
plenum volume, a throttle valve and in-between ducting as shown in Figure 2. In order to study the drive
torque as control variable for surge control, we need a
model that takes variable speed into account. In [5], the
Greitzer-model was further developed, and rotational
speed was included as a state in the model. A similar
model was derived in [6], using an approach based on
energy based analysis. Experimental results confirming
the model of [6] were presented in [7] and [8]. In this
paper we will employ the compressor model derived in
[6]. The model is derived by calculating the mass balance of the plenum volume, integrating the one dimensional Euler equation (the momentum balance) over the
length of the exit duct, and calculating the torque bal-

ance of the rotating shaft. The model is written

a201
(m mt )
Vp
A1
(c (m, )p01 p)
Lc
1
( d c ) ,
J

p =
m
=

(1)
(2)
(3)

where p is the plenum pressure, m is the compressor

mass flow, is the rotational velocity of the shaft,
c (m, ) is the compressor characteristic, mt is the
throttle flow, A1 is the throughflow area, Lc is the duct
length, Vp is the plenum volume, p01 is the ambient
pressure, a01 is the sonic velocity at ambient conditions,
J is the inertia of all rotating parts, and d and c is the
drive torque and compressor load torque, respectively.
The throttle flow is given by

mt = kt p p01 ,
where kt > 0 is a parameter proportional to throttle
opening. The compressor torque c is calculated as
c = |m|r2 U2 ,

(4)

where r2 is the impeller diameter, is the slip factor

and U2 is the impeller tip speed. The drive torque
d will be used as the control variable. For a detailed
derivation of the model, consult [6] and [8]. We will
study the dynamics around an equilibrium point. The
equilibrium values are denoted by ()0 , while deviations
from the equilibrium are denoted by (b). The deviations
from the equilibrium are written
= 0
m
= m m0 , p = p p0 ,
b
m
t = mt mt0 , c = c c0

where the equilibrium values must satisfy m0 = mt0

and p0 = c0 . The model (1)-(3) in new coordinates
(b) is written
p =
m
=

a201
(m
m
t)
Vp

A1 b
c p01 p ,
Lc
1
(b
d b
c) ,
J

(5)

while the dynamics of the plenum pressure and the mass

flow only are written
p =
m
=

a201
(m
m
t)
Vp

A1 b
c p01 p
Lc

(6)

For controller design we will assume that control variable is the angular velocity of the compressor shaft.
Then later we will study the eect of the fact that the
control variable is the electrical motor torque d , while
is controlled by an internal high gain loop.

2.2 Compressor map

2.2.1 Introduction: In order to simulate the
response of the compression system, a measured compressor map from a real centrifugal compressor will be
used. A number of operating points on each measured
constant speed line will be used to calculate a third order polynomial approximation to the speed lines. The
third order polynomial is accepted in the literature (see
e.g.[4]) as a suitable approximation of the speed lines.
The zero-mass flow point will also be used in the calculations of the speed lines, and the third order form also
gives the negative flow part of the characteristic.
2.2.2 Calculation of the zero-mass-flow
pressure rise: Due to centrifugal eects, a centrifugal compressor will produce a pressure rise even at zero
flow. This zero-mass-flow pressure rise can be found by
studying the rothalpy I, which is a conserved quantity,
[9]:
1
1
I = h + W 2 U 2,
2
2
where h is specific enthalpy, W is relative speed between
the fluid and the blades, and U is tangential impeller
speed. Since the rothalpy I is unchanged between the
inlet (subscript 1) and the outlet (subscript 2) of the
impeller, we have that
1 2

1 2
h = h2 h1 =
U2 U12
W2 W12 ,
2
2
where U1 and U2 are the tangential velocities at inlet
and outlet, and W1 and W2 are the relative velocities
between the moving fluid and the rotating blades at inlet and outlet. At zero mass flow, the relative velocities
W1 and W2 vanish, and we have
2N 2 2

1 2
U2 U12 =
D2 D12 . (7)
h|m=0 =
2
2
Assuming isentropic pressure rise,

p02
h 1
= 1+
, m > 0.
(8)
c (m, ) =
p01
cp T01
holds, for details see [8], and by combining (7) and (8),
we get at zero mass flow

2 N 2 (D22 D12 ) 1
c (0, N ) = o = 1 +
, (9)
2cp T01

where N = 2 is the rotational speed in rounds per

second. The zero-flow-pressure rise has been calculated
using (9) for five dierent speeds. These values, as well
as three points on each measured speed line (shown as
circles) are shown in Figure 3.
2.2.3 Polynomial approximation of compressor map: Also shown in Figure 3 are the third
order polynomial approximations of the speed lines at
five dierent speeds. The approximations are calculated using the Matlab function polyfit, and for the
five chosen speed lines, results are:

c (m, 300)
c (m, 340)
c (m, 400)
c (m, 460)
c (m, 500)

=
=
=
=
=

1.602 0.063m + 0.167m2 0.044m3 ,

1.829 0.097m + 0.183m2 0.030m3 ,
2.251 0.144m + 0.191m2 0.024m3 ,
2.809 0.166m + 0.178m2 0.018m3 ,
3.270 0.205m + 0.174m2 0.015m3 .

0.195

-0.01

0.19

5.5

-0.02

500/s

-0.03

c (N)

c (N)

0.185
0.18

0.175

460/s

-0.04
0.17

-0.05
300

3.5

350

400
N [1/s]

450

500

0.165
300

350

400
N [1/s]

450

500

350

400
N [1/s]

450

500

400/s
3

0.1

3.5

300/s

-0.1

2.5

c (N)

340/s

2.5

c (N)

Pressure ratio [1]

4.5

1.5

-0.2
1

-2

4
Mass flow [kg/s]

10

-0.3
300

350

400
N [1/s]

450

500

1.5
300

Figure 3: The measured speed lines (solid lines) and the

polynomial approximations (dashed lines).

Figure 4: The coecients ci as functions of speed N

(solid lines), and their polynomial approximations (dashed lines).

A compressor map is also continuous in the rotational

speed, so in order to simulate the system, there is a need
for making the approximated map also continuous in
rotational speed. For this reason, the coecients of the
third order polynomials in are chosen to be functions
of rotational speed. The polynomial approximation for
each speed line can be written as

6
N=500/s

5.5

Pressure ratio [1]

4.5

c (m, N ) = c0 (N ) + c1 (N ) m + c2 (N ) m2 + c3 (N ) m3 ,
where the functions

3.5

2.5

ci (N ) = ci0 + ci1 N + ci2 N 2 + ci3 N 3

are calculated by using polynomial approximation yet

again. This approach was also taken by [10], using second order polynomials ci (N ) = ci0 + ci1 N + ci2 N 2 .
However, in our case it is necessary to use third order
polynomials. This comes clear when inspecting Figure
4, where the polynomial coecients of the five polynomials are plotted as a function of rotational speed. As
can be seen, a fairly good fit can be made with third
order. For c0 (N ), the zero-mass-flow pressure rise, it
is seen from the figure that a linear approximation is
sucient. The resulting approximated compressor map
which will be used in simulations is shown in Figure 5.

3 Controller design

1
-2

-1

3
4
Mass Flow [kg/s]

Figure 5: The approximated compressor map.

where the gain c is chosen according to
.
b c m

.
c>
b c

(11)

makes the origin of (6) globally exponentially stable.

3.1 Surge control

The new feature of the proposed method is that we let
the shaft velocity
be a function of the mass flow m,

thereby ensuring that the compressor can be operated

to the left of the surge line without going into surge.
We first assume that the shaft speed is our input
control variable, and we will later consider the case that
instead the drive torque d is the control. The following
theorem can now be stated:

Proof:

(10)

Consider the Lyapunov function candidate

V =

Vp 2 L
> 0, (m,
p) 6= (0, 0)
p + m
a201
A

(12)

The time derivative along the solutions of the nonlinear

system (6) is
V

Theorem 1 The control law

= cm,

N=300/s

1.5

a201 L
b c p01 p
m
= p(m
m
t) + m
pp + m

Vp
A
b c (m,
pm
t (
p, p0 ) + m

) p01 (13)
= V 1 + V 2 =
=

The throttle, or load, is assumed to be passive in the

sense that it consumes energy from the compressor,
which implies

180
160

(14)

= cm.

b c (m,
As

)
= 0, a sucient condition for
m=0

b c (m,

cm)
to be located in the 2nd and 4th
b c )-coordinate system is that
quadrant in the (m,

b c (m,

)
is monotonically decreasing, that is

=cm

b c
b c
bc
bc
b c (m,

d

)

=
+
=
c
<0
dm

which is satisfied
provided c is chosen according to (11).

b c (m,
As

)
now is monotonically decreasing

Mass flow m

for some kp > 0, where kp depends on the slope of the

throttle characteristic. This is illustrated in Figure 6.
In order to prove stability, we now have to show that
V 2 < 0. In open loop, the compressor characteristic
b c (m,

) is monotonically increasing in m
for m
< 0,
that is
b c (m,

)
> 0, m
< 0.
m

This is illustrated in Figure 7. We now chose

?p

m

140
120
100

! t p!
V% 1 = ?p! m

p

80
60
40
20
0

1.2

1.4

1.6

b c (m,
V 2 = m

cm)
p01 < 0, m
6= 0.

(15)

1.8
2
2.2
Pressure rise

2.4

2.6

2.8

Figure 6: Calculation of V 1 .

2.2

! ,C!
Hcm

2.0

Hc

! , ?cm
!
H c m

!
m

1.8

C! > 0

1.6

C! = 0

1.4

C! < 0

! H c m
! , ?cm
!
V% 2 = m

=cm

and passing through the origin, that is located in the

1st and 3rd quadrants, we have that

 t p
m

Pressure rise c

pm
t (
p, p0 ) < kp p2 < 0,
p 6= 0
V 1 =

200

1.2

25

50

75
125
100
Mass flow, m

150

175

200

Moreover, V 2 can always be bounded from above as

b c (m,

cm)
p01 < km m
2 , m
6= 0.
V 2 = m

(16)

Figure 7: Calculation of V 1 .

for a constant km > 0. From (16) it follows that

b c (m,
cm)
p01 < km m,
m
> 0.

(17)

b c (m,
As
cm)
p01 is monotonically decreasing and
b c (m,
passing through the origin,
cm)
p01 is also
bounded from above by the tangent through the origin, that is
b c (m,
d
cm)

b c (m,

m.

cm)
p01 < p01
dm

By choosing

km

b c (m,
d
cm)

= p01

dm

m=0

(17) and (16) follows. A similar argument can be made

for the case m
< 0. By (14) and (16), we now have that
2 < kV, (m,
p) 6= (0, 0)
V = V 1 + V 2 < kp p2 km m
and the result follows.

It is seen from (10) that the gain c has a lower bound

given by the ratio between the sensitivity of the characteristic c with respect to the mass flow m, and the
sensitivity of c with respect to the shaft speed .
This result is related to another well known result in
compressor control (see e.g. [11] or [6]): When using
a close coupled valve (CCV) to stabilize a centrifugal
compressor, the control law gain must be greater than
the sensitivity of the characteristic c with respect to
the mass flow m. The CCV-approach aims at using
the pressure drop over the valve to create a combined
compressor and valve characteristic with negative slope
in the equilibrium and thereby ensuring stability. The
approach in this study achieves the same eect without
imposing an unwanted pressure drop in the compression
system. The proposed controller ensures that in closed
loop the compressor characteristic has a negative slope
in the equilibrium. This can bee seen from the plot of
b c (m,

cm)
in Figure 7.
Also worth noticing is that, according to Chetaevs instability theorem, the equilibrium is unstable if the
weaker sector nonlinearity (15) does not hold. This
agrees with the well-known result that if is kept con-

stant, that is
= 0, then the compressor is unstable to
the left of the surge line.
If we take the shaft dynamics (3) into consideration and
follow the same procedure as in the proof of Theorem 1,
the time derivative of the Lyapunov function candidate
(12) along the trajectories of (6) is,
V < kp p2 km m
2 + m(t),

(18)

e c := c (m, ) c (m, d ) is
e c and
where = A/L
the error in the compressor pressure c due to the shaft
dynamics related to the convergence of to the desired
value d = 0 cm.
It is seen that the system will converge exponentially towards the equilibrium whenever
km |m|
> |(t)| .

Disturbances may be treated in a similar way. Assume

that there are disturbances to the system so the dynamics are
=

m
=

a201
(m
m
t) + p
Vp

A1 b
c p01 pp + m
Lc

By using the analysis presented in equations (18) to

(19), it can be concluded that applying a drive torque
according to (21) ensures that the states of the system
(5) converges exponentially to a small region around
the origin.

(19)

Therefore, if the shaft dynamics are suciently fast so

that (t) converges quickly to zero, the system will converge to a small area around the equilibrium with exponential rate of convergence.

p p

where the feedback gain for the mass flow is K2 = K1 c.

In practical implementations we propose the use of integralRaction to generate the term 0 . The integral term
(t0 )dt0 is included in order to keep the compresKI
sor at the desired speed, and can be regarded as part
of the performance control system, see Figure 1. This
gives the control law
Z
(t0 )dt0 .
= K1
K2 m
KI
(21)

(20)

Then the time derivative of the Lyapunov function candidate becomes

V < kp p2 km m
2 + p p + m
m,

4 Simulations
4.1 Surge
In this section it is illustrated that the model is capable of simulating surge, the instability which will be
stabilized in the next section. Results are presented
from simulations of the compressor system when it is
driven into surge by a drop in mass flow. The compressor is initially operating in a stable operating point
close to m 5 kg/s, when a throttle change induces a
drop in mass flow of 20%, consequently driving the compressor over the surge line down to about m 4 kg/s,
and ultimately ending up in a deep surge condition.
The compressor response to this disturbance is shown
in Figure 8. A constant drive torque is used. The compressor undergoes deep surge with oscillations in mass
flow, pressure rise and shaft speed.

which shows that V will decrease as long as

2

kp p + km m
> p p + m
m.

5.5
5

In fact the state vector x

= (
p m)
T will converge exponentially to the region
max ( p , m )
.
min (kp , km )

Note that this is valid for all disturbances without any

upper bound on ( p , m ). The scalar kp is given by the
slope of the throttle characteristic, while km is given
by the mass flow gain c. It is seen that min (kp , km )
has an upper bound, which is kp . It will make sense
to select the gain c so that km is of the same magnitude as kp . Satisfactory results will then be obtained
provided that the disturbances ( p , m ) are suciently
small so that k
xk < gives acceptable operation of the
compressor around the equilibrium. This depends on
the disturbance specifications and the slope kp of the
throttle curve.
3.2 Velocity control
Let the electrical motor torque be generated by d =
+ 0 where 0 = c0 is the torque required in the
equilibrium point, and = K1 (d ) is the feedback
control law that is used to obtain the desired shaft speed
d = 0 c
. The resulting control law is
d = K1
K2 m
+ 0

Pressure ratio [1]

k
xk < :=

4.5
4
3.5
3
2.5
2
1.5
1
-2

-1

2
3
4
Mass Flow [kg/s]

Figure 8: Compressor response when the operating point

is driven over the surge line.

4.2 Simulation of active surge stabilization

Simulations of the proposed active surge control approach will now be presented. The idea is to control
the compressor speed with feedback from the mass flow
so that the compressor can operate in a stable mode
even to the left of the surge line and thereby avoiding

the unstable operation demonstrated in the simulations

above.
New throttle line

3.5
Pressure rise [1]

Mass flow [ks/s]

5
4.8
4.6
4.4
4.2

5
4

15

x 10

10
Time [s]

15

20

10
Time [s]

15

20

400

2.4

2.3995

2.399
0

3.4

3.35
0

20

Drive torque [Nm]

Shaft speed [rpm]

2.4005

10
Time [s]

3.45

10
Time [s]

15

20

300
200
100
0

Figure 9: Active surge control using drive torque.

5 Conclusion
Active surge controllers for a centrifugal compressor
driven by an electrical motor have been designed. This
is a new approach to the active surge control problem.
The use of the rotational speed as control variable renders the equilibrium globally exponentially stable, and
the use of the drive torque as control ensures exponential convergence. The control manipulates the compressor map in such a way that the compressor sees a
negative compressor characteristic slope even to the left
of the surge line. Simulations confirmed the theory.

Initial throttle line

3.6
Pressure ratio [1]

The controller is implemented with the input torque

given by (21), with a desired shaft speed of d =
400/s = 24000 rpm. In this simulation, the controller is
active at all times, and as the same mass flow drop as in
the surge simulation above is introduced at t = 5 s, the
compressor remains stable. This can bee seen in Figure 9, where the mass flow, pressure rise, shaft speed
and drive torque are plotted as a function of time. The
mass flow is reduced from 4.9 kg/s to 4.2 kg/s, a value
that in open loop would cause the compressor to surge.
In this controlled case it remains stable. The peak in
the applied torque from the drive as seen in the lower
left plot of Figure 9 is expected to be within the performance envelope of the drive. The reason for the slightly
increased speed as the torque is reduced is that the compressor torque is reduced with reduced mass flow as can
be seen from equation (4). From simulations it is found
that the speed control by using drive torque has quite
good performance. As can be seen from Figures 9 and
10, the desired speed of 24000 rpm is not quite reached.
This deviation lies within the small region around the
equilibrium, as concluded in section 3.2. As can be seen
in Figure 10, the new stable operating point is located
to the left of the surge line, illustrating the capability
of the control system to achieve active surge control.

3.7

3.5

New, stabilized
operating point
Initial operating point

3.4
System trajector
3.3
24000 rpm
3.2

3.1
4

4.2

4.4
4.6
Mass Flow [kg/s]

4.8

Figure 10: The simulation plotted in the compressor map.

References
[1] A. Epstein, J. F. Williams, and E. Greitzer, Active suppression of aerodynamic instabilities in turbomachines, Journal of Propulsion and Power, vol. 5,
no. 2, pp. 204211, 1989.
[2] J. Gravdahl and O. Egeland, Compressor surge
and rotating stall: modeling and control. Advances in
Industrial Control, London: Springer-Verlag, 1999.
[3] F. Willems and B. de Jager, Modeling and control of compressor flow instabilities, IEEE Control systems, vol. 19, no. 5, pp. 818, 1999.
[4] E. Greitzer, Surge and Rotating stall in axial
flow compressors, Part I: Theoretical compression system model, Journal of Engineering for Power, vol. 98,
pp. 190198, 1976.
[5] D. Fink, N. Cumpsty, and E. Greitzer, Surge
dynamics in a free-spool centrifugal compressor system, Journal of Turbomachinery, vol. 114, pp. 321
332, 1992.
[6] J. Gravdahl and O. Egeland, Centrifugal compressor surge and speed control, IEEE Transactions
on Control Systems Technology, vol. 7, no. 5, 1999.
[7] J. Gravdahl, F. Willems, B. de Jager, and O. Egeland, Modeling for surge control of centrifugal compressors: Comparison with experiment, in Proceedings
of the 39th IEEE Conference on Decision and Control,
(Sydney, Australia), 2000.
[8] J. Gravdahl, F. Willems, B. de Jager, and O. Egeland, Modeling for surge control of centrifugal compressors: Experimental validation. Submitted to IEEE
Transactions on control systems technology, 2000.
[9] A. Whitfield and N. Baines, Design of Radial Turbomachines. Harlow: Longman, 1990.
[10] F. Willems, Modeling and Bounded Feedback Stablization of Centrifugal Compressor Surge. PhD thesis,
Fac. of Mechanical Engineering, Eindhoven University
of Technology, 2000.
[11] J. Simon and L. Valavani, 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. 23982406, 1991.