See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/267588171

Smith Predictor Position Control for the Inlet Guide Vane of Centrifugal Air
Compressors

Conference Paper · January 2006

DOI: 10.1115/IMECE2006-13101

CITATIONS READS
0 5,118

8 authors, including:

Angela Sodemann Yaoyu Li

Arizona State University University of Texas at Dallas
19 PUBLICATIONS 157 CITATIONS 142 PUBLICATIONS 2,230 CITATIONS

SEE PROFILE SEE PROFILE

Xin Wu Jay Lee

North China Electric Power University Foxconn Technology Group
16 PUBLICATIONS 402 CITATIONS 380 PUBLICATIONS 11,655 CITATIONS

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Industrial AI View project

Books on Industrial Big Data and CPS View project

All content following this page was uploaded by Jay Lee on 06 December 2014.

The user has requested enhancement of the downloaded file.

 IMECE2006-13101
Smith Predictor Based Position Control for the Inlet Guide Vane of Centrifugal
Air Compressor

Angela Sodemann, Yaoyu Li and Xin Wu

Department of Mechanical Engineering Rick Lancaster, Mark Rucker, Jeff Carnagie and
University of Wisconsin – Milwaukee Andy Inman
Facility Control Department
Jay Lee
Toyota Motors Manufacturing in Kentucky
Department of Mechanical, Industrial & Nuclear
Engineering, University of Cincinnati
ABSTRACT more efficient when operated close to the surge limit
Fast and accurate positioning of Inlet Guide line. A typical feedback surge control strategy to
Vane (IGV) is critical for flow regulation in regulate the air flow based on the detection of
centrifugal air compressors, especially for surge incipient surge. Therefore, a quick and accurate flow
avoidance control. Both transport delay and external regulation is a key to make this possible.
disturbance are present for IGV position control on a A typical valve used for centrifugal compressor
compressed air system. A Smith Predictor based is the so-called inlet guide vane (IGV). As shown in
control scheme with the process-model modification Figure 1, an IGV consists of several vanes positioned
is applied to this problem. The adopted control in a circular arrangement. A pneumatic actuator
scheme can both compensate for the time delay and called IP converter, which converts current input to
effectively reject step disturbances present in the IGV pressure activation, actuates the main shaft which in
system. Simulation was conducted using simple PI turn rotates the slave vanes via a ring-gear
controller and a PI controller with the Smith mechanism. Each vane can be rotated between 0 and
Predictor with process-model modification. It showed 90°, thus opening and closing the valve as shown in
that the proposed control scheme can achieve zero Figures 1 and 2. Currently not much work has been
steady-state error with the presence of external done to effectively control the positioning of the IGV
disturbance and the settling time was shortened. used in the industrial compressors. IGV positioning
Robust stability and performance of the system was has become a barrier for fast and accurate flow
discussed. regulation for surge avoidance control. Figure 3
Key Words: Smith Predictor, Inlet Guide Vane, shows the position command and measurement of a
Position Control, Delay, Robustness, Centrifugal new IGV during actual operation, which
Compressors demonstrates about 20% relative error in positioning
that primarily arises from aerodynamic disturbance.
I. INTRODUCTION The error will become larger and larger throughout
In a centrifugal air compression system, it is the IGV’s usage.
necessary to regulate the amount of flow through the There are two factors affecting the positioning of
compressor. Flow regulation is usually a requirement this valve: aerodynamic/frictional disturbance and
of load change, closely related to compressor surge. plant uncertainty. Uncertainty in the plant is a result
Surge is a phenomenon of large oscillating pressure of changes in the valve positioning mechanism that
and reverse flow that occurs when a compressor is occurs over time. Gears wear out causing variation in
operated at too low a flow rate for the given motion transfer as well as the change in net torque
discharge pressure. Surge can induce large vibration due to the changing friction. The IP converter
and may severely damage the compressor. For a mechanism is also subject to wear that occurs over
compressed air system, the surge induced cost may time and with use. Positioning error can be very
include the components and labor for fixing the large without recalibration. Another significant
compressors are costly, and the production down- source of positioning error is the aerodynamic
time cost for manufacturing plant, Therefore, it is disturbance that the vanes of the valve are exposed to
critical to operate the compressor away from the under normal operating conditions. This source of
high-pressure-low-flow surge zone. However, it is error is more difficult to compensate for with merely
also undesirable to operate the compressor with too calibration, as the disturbance changes in magnitude
high flow rate due to the actual load demand and with different loading conditions on the compressor.
efficiency consideration. Usually the compressor is

1
 The system model obtained from our dynamic the delay-free portion of the system transfer function,
measurement showed that the IGV is a first order while Pˆ0 ( s) represents an estimate of the delay-free
system with time-delay. An appropriate control portion of the system transfer function. C 0 ( s ) is the
strategy needs to be developed to obtain good
performance (sufficient response time and accuracy) controller transfer function.
and disturbance rejection for such delay systems. If the estimation of plant and the time delay
Smith Predictor has been an effective control scheme are both perfect, the Smith predictor effectively takes
for compensating the effect of time delay [1]. The the time-delay outside the loop, and then a traditional
Smith Predictor predicts the output of the system control method can be applied. However, if there is
“time delay” units of time in the future, and so any inaccuracy in either estimate, some modification
effectively taking the delay outside the loop [2]. The has to be made to ensure robustness.
Smith Predictor has been studied extensively to The closed-loop transfer function of the
determine the conditions for stability [3], improve inner-loop in Figure 4 can be written as
results in special cases such as a system with a pure C ( s ) Pˆ ( s ) (3)
T (s) =
integrator [4] [5], and develop the method for 1 + C ( s ) Pˆ ( s )
discrete-time systems [6] [7]. The Smith predictor
has been combined with other methods, such as the where C ( s ) = C 0 ( s ){1 + C 0 ( s )[ Pˆ ( s ) − Pˆ0 ( s )]}−1 .
digital “wait and see” method to attempt to improve Then the necessary and sufficient condition for robust
robustness [8] [9]. Much of the development in the stability and performance of the system with Smith
Smith predictor method has proved effective in delay predictor under time-delay mismatch is shown as [11]
compensation; however, it has not been successful W2T ∞ < 1 (4)
for rejecting disturbances. To achieve this, a process- 2 2
model modification was made to the Smith predictor ( W2T + W1 S )1 / 2 <1 (5)
∞
to achieve step disturbance rejection [10].
In this paper, an IGV position control where S is sensitivity function, and S = I − T .
scheme is proposed to reject aerodynamic The nominal plant model uncertainty can be
disturbance using the Smith Predictor with process- written as
model modification. The control design method will P ( s ) = Pˆ ( s )[1 + ∆ ( s )] (6a)
be described in next section. Simulation and where
experimental results are presented in Section III, and
P ( s) − Pˆ ( s ) (6b)
conclusions are given in Section IV. ∆( s) =
Pˆ ( s )
II. METHOD Several assumptions are made at this point.
A. Smith Predictor Method It is assumed that the plant is asymptotically stable
The Smith predictor method is a method of control and does not contain a pure differentiator, the
for systems with a time delay. A time delay system in controller C0(s) contains an integrator, and T(s) is
state-space form can be written as follows [2]: asymptotically stable. If these conditions are met, the
x = Ax (t ) + Bu (t − τ ) (1) Smith Predictor based system will have zero steady-
y (t ) = Cx (t ) state error for a step input.
where τ represents the time delay. The corresponding Given that the above assumptions are met, the
transfer function can be written as controller can be designed as
Y ( s) K
= Po ( s )e −τs (2) C 0 ( s) = Pˆ0 ( s) −1 (7)
U (s) s
Po ( s ) = C ( sI − A) −1 B There are two possibilities for the form of the
controller:
where P0(s) represents the delay-free part of the
system. The Smith predictor method is a method of kp
Pˆ0 ( s ) = (8a)
“feeding back” the time-delay estimate in order to τs + 1
properly account for the delay in calculating the kp
control signal, as shown in Figure 4. The output of Pˆ0 ( s ) = (8b)
the Smith Predictor effectively contains a prediction τ s + 2τξs + 1
2 2

of the output y(t) one time-delay amount of time in A rule of thumb is to set K as 3/τ.
the future. P(s) represents the actual system transfer
B. Process-Model Modification
function, while Pˆ ( s ) represents an estimate of the The Smith Predictor method is useful for
system transfer function. Again, P0 ( s ) represents compensating for a time delay; however, its
traditional design does not perform well when

2
subjected to a disturbance. The process-model position variation, the process-model modification to
method is a modification to the Smith predictor that the Smith predictor was also incorporated.
effectively rejects step disturbances [10]. The The controller was designed using the Smith
process-model method is to substitute Pˆ0 ( s) with Predictor process-model method, the state-space
G1(s), as shown in Figure 5. The G1(s) can be parameters for P0(s) were found as A = − 0.43, B =1,
C = 0.45 and D = 0. These values were then
calculated by
τ substituted into Eq. (9), which leads to
G1 ( s) = Ce − Aτ ( sI − A) −1 B − ∫ Ce − At Bdt (9) − 0.094s + 0.4502
0 G1 ( s) = (13)
The variables in Eq. (9) are the same as defined in s + 0.43
Eq. (2). With this modification, the closed-loop Using Eqs. (10) and (11), the stability condition is
system will have zero steady-state error to a step satisfied with the following verification:
disturbance if and only if the impulse response of − 0.094s + 0.4502 0.45 −0.2 s
lim − e = 0 (14)
G1 ( s) − MG( s)e −τs (10) s →0 s + 0.43 s + 0.43
goes to zero as t goes to infinity, and as s goes to The Smith Predictor effectively removed the time
zero, respectively. delay to the outside of the loop, and then the PI
controller with higher gains was investigated. We
III. RESULTS adjust the PI controller to
An IGV experimental setup was installed at the 0.95(2.2s + 1)
GC ( s) = (15)
Mechatronics Research Lab at the University of s
Wisconsin – Milwaukee to test the response of an Air The closed-loop step response using the Smith
Relief® GS10CW-A2-P8 inlet guide vane, acquired Predictor with process-model control is shown in
by Toyota Motors Manufacturing in Kentucky, Inc. Figure 9. Notice that the settling time was
(TMMK). The actual opening of the IGV is measured approximately 2.6 seconds, while the settling time for
with a potentiometer mounted on the shaft of the the PI controller was about 4 seconds. Better
IGV. The IGV is positioned by an IP converter that performance was contributed to the higher PI gains
receives a 4-20 mA current input corresponding to an after delay compensation.
IGV positioning of 0-100% open. The actual The closed-loop system using process-model
position is read from the potentiometer also as a 4-20 Smith Predictor controller was then simulated by
mA current output. A swept-sine test was performed adding a step disturbance. An upper bound for the
on the IGV, and the resulting frequency response step disturbance was calculated by measuring the
plots are shown in Figure 6. The frequency response position error of the IGV under loading. The IGV
data are in solid lines, while the dashed lines indicate was opened and closed through a series of step
the response of the following estimated model: openings of different magnitude. The data from one
Y (s) 0.45 − 0.2 s (11) of the step tests is plotted in Figure 10. The
= e
U ( s ) s + 0.43 maximum position error of the IGV was found to be
where a time-delay value of 0.2 second was presented, approximately 4.36%. In order to provide for a
as shown in Figure 6(a). The uncertainty weight W2(s) margin of error, the IGV position error was bounded
to approximate the model uncertainty of IGV, as as 5%, corresponding to 0.8 mA command signal.
shown in Figure 6(b) [12] [13]. The simulation response is shown in Figure 11. In
A PI controller was obtained through root locus this simulation test, a step input from 12 to 16.8 mA
method, as shown in Figure 7. The controller was corresponding to a step change of 50% to 80% is
designed to provide the quick response without applied at t = 6 seconds. The upper bound of
overshoot. The step response of the IGV system with disturbance 0.8 mA corresponding to 5% position
PI controller is shown in Figure 8(a), with the control error is applied at t = 12 seconds. The disturbance is
input profile shown in Figure 8(b). Equation (12) is rejected, and the IGV settles at the 16.8 mA set point.
the transfer function of the selected PI controller. The robust stability and performance of the
0.61(2.2s + 1) controller are verified with Equation (4) (5). With the
GC ( s ) = (12)
s weights W = 1 , W = 1.08( s + 0.17) ,
1 2
The Smith Predictor method was then used to 5( s + 1)( s + 0.43) s + 0.43
achieve better control performance by compensating we obtain
for the time delay. PI controller was still adopted for W2T ∞ = 0.886 < 1 (16)
steady state disturbance rejection. In order to 2 2
compensate for the aerodynamic disturbances during ( W2T + W1 S )1/ 2 = 0.893 < 1 (17)
∞

3
So the system satisfies the stability and performance Control, Vol. 44, No.8, pp.1597-1603, 1999
robustness. The poles of the system are all in the left [5] K. J. Astrom, C. C. Hang, and B. C. Lim, “A
half-plane. We also make use of Nyquist stability New Smith Predictor for Controlling a Process
criterion to verify the stability of the system, as with an Integrator and Long Dead-Time,” IEEE
shown in Figure 12. The system stability is Transactions on Automatic Control, Col. 39, No.
maintained, as the plot does not circle the (−1, j 0) 2, pp. 343-345, 1994.
point. [6] Z. J. Palmor and Y. Halevi, “Robustness
Properties of Sampled-data Systems with Dead
IV. CONCLUSIONS Time Compensators, ”Automatica, Vol. 26, No.
Smith Predictor based control of IGV position 3, pp. 637-640. 1990
was investigated in this study. The inlet guide vane [7] I. D. Landau, “Robust Digital Control of
was tested and determined to be a first-order system Systems with Time Delay (the Smith Predictor
with time delay. Process model Smith Predictor was Revisited),” Proceedings of IEEE Conference on
applied to obtain good control performance in Decision and Control, Vol. 1, pp.865-869, 1994.
disturbance rejection in spite of delay nature of IGV. [8] R. Whalley and Z. Zeng, “A Robust Smith
Simulation showed that the proposed control method Predictor,” Trans IChemE, Vol. 74, Part A,
can successfully compensated for the time delay, pp.21-29, 1996.
allowing higher gains to be used, zero steady-state [9] J. E. Normey-Rico and E. F. Comancho, “Smith
error was obtained with a 2.6 second settling time. In Predictor and Modifications: A Comparative
comparison, simply using PI controller could only Study,” Proceedings of the European Control
achieve up to 4 second settling time. Further Conference, 1999.
simulation showed the ability of disturbance rejection [10] K. Watanabe and M. Ito, “A Process-Model
under the disturbance even much larger in magnitude Control for Linear Systems with Delay,” IEEE
than that observed from actual compressor operation. Transactions on Automatic Control, Vol. 26,
This demonstrates that using Smith Predictor based No. 6, pp.1261-1269, 1981.
PI control is promising in achieving good IGV [11] J.-C. Doyle, B. A. Francis, Tannenbaum, A.
position control performance regardless of large Feedback Theory, Macmillan Pulishing Co.,
disturbance, and it will help improve the surge 1990, pp.40-52
avoidance control to be developed. And the robust [12] Gjerrit Meinsma and Hans Zwart, “On H ∞
stability and performance of the system are satisfied. Control for Dead-Time System,” IEEE
In near future, the closed-loop control experiments Transactions on Automatic Control, Vol. 45,
will be conducted for the IGV on the actual No.2, 2000.
compressor under different flow conditions. The [13] Q.-C. Zhong, “ H ∞ Control of dead-time System
calibration of flow regulation for different IGV
opening will also be carried out. based on a transformation,” Automatica, 39,
361-366, 2003
References
[1] M. R. Stojic, M. S. Matijevic, and L. S.
Draganovic, “A Robust Smith Predictor
Modified by Internal Models for Integrating
Process with Dead Time,” IEEE Transactions on
Automatic Control, Vol. 46, No. 8, pp.1293-
1298, 2001.
[2] Z. J. Palmor, “Time-Delay Compensation –
Smith Predictor and its Modifications,” The Figure 1: Inlet guide vane (IGV) in the closed
Control Handbook, pp.224-236. Edited by position
William S. Levine, CRC Press, Boca Raton, FL,
1996.
[3] T. H. Lee, Q. G. Wang, and K. K. Tan, “Robust
Smith-Predictor Controller for Uncertain Delay
Systems,” AIChE Journal, Vol. 42, No. 4,
pp.1033-1040, 1996
[4] J. E. Normey-Rico and E. F. Camancho, “Robust
Tuning of Dead-Time Compensators for
Processes with an Integrator and Long Dead-
Time,” IEEE Transactions on Automatic

4
 Frequency Response Measured from Encoder With Approximated Fit Curve
10
actual frequency response
approximated fit curve
0

-10

Magnitude [dB]
-20

-30

-40

-50
-2 -1 0 1
10 10 10 10
Frequency [Hz]

(a): Frequency response measurement for IGV and

Figure 2: IGV in the open position the plant approximation
Model Uncertainty Measured from Encoder With Approximated Fit Curve
20 5
Output
19.5
Command
19 0

18.5
IGV (%)

18 -5
Model Uncertainty

Magnitude [dB]
17.5 W2

17 -10

16.5
-15
16
15.5
-20
15
5 10 15 20
Time (second) 25 30
-25
-2 -1 0 1
Figure 3: IGV position command and measurement 10 10 10 10
Frequency [Hz]
during actual operation
(b): Fitting the model uncertainty weight W2
r(t) + Controller Plant y(t) Figure 6 Plant and uncertainty modeling of the IGV
+
C0 (s) P(s)(1+∆(s)) Controller Plant y(t)
r(t) +
− C0 (s) P(s)
-
Pˆ0 (s) − Pˆ(s)
Delay Compensation Figure 7: Traditional control block diagram
13
Figure 4: The idea of Smith Predictor 12

11
+ d(t) y (t )
Set Points
r(t) + Controller Plant 10 Step Response
+
Step Response (mA)

GC (s) G ( s ) e − τs
+ 9

− − 8

G1(s) − MG(s)e−τs 7

6
M
5

Figure 5: Block diagram of process-model Smith 4

Predictor control
3
0 2 4 6 8 10 12 14 16 18 20
Time (sec)

(a) Step Response

5
 13
85
Set Points
IGV Feedback
Control Input
12 80 IGV Command

75
11
Control Input (mA)

70
10

% Open
65

9
60

8 55

50
7
0 2 4 6 8 10 12 14 16 18 20
Time (sec)
45
0 10 20 30 40 50 60 70 80
(b) Control Input Time (sec)
Figure 8: Simulation of traditional PI control for the
IGV system Figure 10: IGV step test response while compressor
is in operation, showing position error magnitude.
13
18
12

16
11
Set Points
10 Step Response 14
Step Response (mA)

Step Response (mA) Set Points

Step Response
9
12

8
10
7

8
6

5 6

4 4
0 2 4 6 8 10 12 14 16 18 20 0 2 4 6 8 10 12 14 16 18 20
Time (sec) Time (sec)

(a) Step Response Figure 11: Simulated response of IGV to step input
20
Set Points
and upper bound of estimated step disturbance
Control Input
18 Nyquist Diagram
1
2 dB 0 dB -2 dB -4 dB
4 dB
16 0.8
-6 dB
Control Input (mA)

0.6 6 dB
14
0.4 10 dB -10 dB

12 0.2
Imaginary Axis

20 dB -20 dB

0
10

-0.2
8
-0.4

6 -0.6
0 2 4 6 8 10 12 14 16 18 20
Time (sec)
-0.8

(b) Control Input -1

-1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1
Figure 9: IGV closed-loop step response with the Real Axis

Smith Predictor with process-model modification Figure 12: Nyquist plot of IGV with process-model
controller

View publication stats