Sei sulla pagina 1di 10

YSM Simamora, E. Suharno, M.

 Syaikhudin, “Active surge control of centrifugal 
compressor using self­tuning fuzzy approach”, Orbith vol. 5 No. 1 Mar. 2009, 
pp 222­229;  ISSN : 1858­2095

Active Surge Control of Centrifugal Compressor


using Self-Tuning Fuzzy Approach

Yohannes S.M. Simamora, Edy Suharno, Muhamad Syaikhudin


Mechanical Engineering Study Program – Politeknik Purbaya
Jl. Pancakarya No. 1 Talang, Kabupaten Tegal-Indonesia 52193
Phone.: (0283) 3447340; Fax.: (0283) 3447340; e-mail : simamora@me.purbaya.ac.id

Abstracts
Surge in centrifugal compressor is flow stable oscillation with a large-amplitude. Since it decreases performance
and even may damage the compressor, surge is highly undesired in compressor operation. To overcome the surge
problem, two common methods applied are surge avoidance and active surge control. The last mentioned method
seem more promising since it can extend the compressor operation range. In this paper, we present self-tuning fuzzy
(STF) controller as an approach on active surge control method. The controller consists of two level of fuzzy
controllers. The first level controller acts as the plant controller, while the second one provides parameter tuning of
the first level controller. By simulations, we demonstrate that STF can manage the compressor operation from
getting surge in several reference points in the surge region.
Keywords : “surge”, “compressor“, “self-tuning fuzzy”

1. Introduction
Surge is a large-amplitude stable oscillation of flow of centrifugal compressor, and is
characterized by limit cycle in compressor map. It occurs when the compression system is unable
to keep forward flow through the compressor due to an increasing pressure and flow reversal. In
a compressor map view, compressor undergoes surge if it operates at any point at the left side of
the surge line.
Surge is an unexpected operating condition of compressor, since it limits the compressor
operating range and may even damage the compressor. As an example, problems occur following
a long time surge are temperature rise, compressor load, and stress of compressor components,
all occur in every surge cycle (Brown, 2005) .
A classic approach to overcome compressor surge is surge avoidance or anti surge (Gravdahl,
1998), (Willems, 2000) and (van Helvoirt, 2007). The basic idea of surge avoidance is to manage
the compressor operating point stay at the right side of the surge line.
Still used in many applications, surge avoidance approach, however, has several drawbacks
(Gravdahl, 1998). It is worth concluding the drawbacks of surge avoidance approach lie in the
nature of the approach itself, that is to limit the compressor operating area.
As alternative to surge avoidance approach, active surge control is introduced. Unlike its passive
counterpart, active surge control is intended dealing with the surge, instead of avoiding it.
According to Gravdahl (1998), active surge control approach is stabilizing some part of the
unstable area in the compressor map using feedback.
The aim of active surge control, thus, is to keep the compressor operation stable, even at the left
side of the surge line. It means that compressor have an extended operating area. Hence, an
operating area extension becomes significant advantage of active surge control approach.
Introduced in 1989 by Epstein, Ffowcs Williams, and Greitzer (Gravdahl, 1998), research on
active surge control approach has become extensive ever since. Active surge control experiments
using feedback proportional controller, with variations in sensor and actuator were listed in
Willems (2000). Literature also reports active surge control approach using PI and adaptive
control (Gravdahl, 1998), fuzzy control (Al-Mawali and Zhang, 2006), and Linear Quadratic

1
YSM Simamora, E. Suharno, M. Syaikhudin, “Active surge control of centrifugal 
compressor using self­tuning fuzzy approach”, Orbith vol. 5 No. 1 Mar. 2009, 
pp 222­229;  ISSN : 1858­2095

Gaussian (van Helvoirt, 2007) with successful results.


While linear approaches were applied successfully in active surge control, the use of nonlinear
approaches seem more promising, since surge problem itself is a nonlinear one (Al-Mawali and
Zhang, 2006). A linear approach is usually accurate only in a small range operation that
considered linear, while nonlinear counterparts may handle the nonlinearities in a larger one
(Slotine and Li, 1991).
The aim of this paper is to present self-tuning fuzzy (STF) controller as a nonlinear approach on
active surge control of centrifugal compressor. The controller consists of two level fuzzy
systems. The first level acts as a plant controller, while the second controller parameterize both
input and output of the first one. As a nonlinear approach, it is expected that the controller
succeeds dealing with surge nonlinearities.
This paper is organized as follows. Section 2 presents compressor model used in this paper.
Section 3 presents self-tuning fuzzy controller design. Section 4 presents simulation results and
discussions. Conclusions of this paper is presented in section 5.
2. Compressor Model
Compression system under study consists of centrifugal compressor, plenum, throttle and control
valve (Willems, 2000) and (Al-Mawali and Zhang, 2006) as described in Figure 1. Dynamic
equations of the system are given by (Al-Mawali and Zhang, 2006) :
1
̇ =   c  −  (1)
lc
1
̇=   − T  − CV   (2)
4B2 l c
where  is nondimensional annulus averaged mass flow coefficient, l c is effective flow
passages length of the compressor and its duct,  c is Moore-Greitzer compressor characteristic
(Gravdahl, 1998),  is nondimensional plenum pressure,  T  is throttle mass characteristic
and  CV  is control valve mass characteristic.

Figure 1. Compression system with additional control valve (Al-Mawali and Zhang, 2006)

2
YSM Simamora, E. Suharno, M. Syaikhudin, “Active surge control of centrifugal 
compressor using self­tuning fuzzy approach”, Orbith vol. 5 No. 1 Mar. 2009, 
pp 222­229;  ISSN : 1858­2095

Constant B in (2) is Greitzer B-parameter (Gravdahl, 1998) that is :

B=
U
2 as  Vp
Ac L c
where U is compressor constant speed, a s is sound velocity,
(3)

V p is plenum volume, Ac is
flow area, and L c is duct and compressor length.
It is worth noticing ̇ and ̇ in (1) and (2) are first derivations of  and , respectively,
against nondimensional time, that are :
̇
̇ = (4)

̇
̇= (5)

where  is non-dimensional time. Quantity  is given by (Gravdahl, 1998) :
Ut
= (6)
R
where t is actual time, and R average compressor radius.
Moore-Greitzer compressor characteristic in (1) is a cubic equation, that is (Gravdahl, 1998),
(Willems, 2000), and (van Helvoirt 2007) :

[ ]
3
3  1 
c =c0 H 1  −1−  −1 (7)
2 W 2 W
where c0 0 is shut-off value, while H and W are compressor characteristic semi height and
semi width, respectively.
Characteristic curve in a nondimensional compressor map view is shown in Figure 2.

Figure 2. Compressor characteristic and surge line in a compressor map.


The presence of surge line in the above map is obtained by a transformation to the local
maximum of (7) (Gravdahl, 1998), that is :
  max , max  = 2 W ,  c02 H  (8)
Throttle mass characteristic and control valve characteristic in (2) are given by (Al-Mawali and
Zhang, 2007) and (Willems, 2000):
3
YSM Simamora, E. Suharno, M. Syaikhudin, “Active surge control of centrifugal 
compressor using self­tuning fuzzy approach”, Orbith vol. 5 No. 1 Mar. 2009, 
pp 222­229;  ISSN : 1858­2095

 T =c T u T  =T   (9)
 CV =c CV uCV   (10)
where c T is is a measure for the capacity of the fully opened throttle, u T is throttle position,
T is throttle gain with range [0 1], c CV is position of control valve, also with range [0 1], and
u CV is is a measure for the capacity of the fully opened control valve.
Similar to Al-Mawali and Zhang (2006), T was set to become reference point of the system,
c CV acts as actuator, while  was assigned as system feedback. It is worth noticing any value
of T corresponds to a particular point at compressor map, that is the real and positive solution
of crossing points of (7) and (9).

3. Controller Design
3.1 Controller Operating Range
All numerical values of compressor parameters were taken from Appendix D of Gravdahl
(1998). The controller was designed dealing with surge at the range of 0.2≤ T ≤0.61 .
Finding the real and positive solution for crossing point of (7) and (9), T =0.61 corresponds to
 =0.4955 and  =0.6599 at the compressor map. This is the point where compressor start
surging, since local maximum of (7), that is the surge line, is  =0.66 at  =0.5. On the other
hand, design value T =0.2 corresponds to  =0.1185 and  =0.3511.
3.2 Self-Tuning Fuzzy Controller
Basic idea and structure of self-tuning fuzzy (STF) controller were taken from (Tian, et.al.,
2007), as described in in figure 3.

Figure 3. Self-Tuning Fuzzy Controller


As previously mentioned, the first level self-tuning fuzzy controller (FLFC) acts as plant
controller, while the second-level fuzzy controller (SLFC) is a tuning mechanism for parameters
of FLFC.
Both of FLFC and SLFC use two inputs and one output. In a discrete form, the controller inputs
are given by :
e [ k ] =ref − [ k ] (11)
ė [ k ] =e [ k ]−e [ k −1 ] (12)
where e and ė are error and change-in- error, respectively, while  ref is reference point
corresponded by T . On the other hand, tuned parameters of FLFC are error gain, change-in-
error gain, and control action gain, represented by G E , G CE , and G U , respectively.

4
YSM Simamora, E. Suharno, M. Syaikhudin, “Active surge control of centrifugal 
compressor using self­tuning fuzzy approach”, Orbith vol. 5 No. 1 Mar. 2009, 
pp 222­229;  ISSN : 1858­2095

To construct the fuzzy system, we use triangular membership function as fuzzification method,
Mamdani implication, fuzzy rule base, and finally center of gravitation as defuzzification
methods. When it necessary, we also apply scaling factor such that all crisp point mapped to the
fuzzy system.The constructed fuzzy system is in a standard form that can be found in many
literatures about fuzzy, for examples, in Jager (1995) and Wang (1997), and Jantzen (2007).
3.3 First-Level Fuzzy Controller
The FLFC consists of two inputs and one output. Each of the inputs are fuzzified by seven
membership functions as shown in figure 4. Since in active surge control case the reference
point is smaller than initial condition, universe of discourse of error range was set to [-6 0].
Conversely, universe of discourse of change-in-error was set to [0 6]. This design choice was
inspired by rule base table in (Al-Mawali and Zhang, 2006), where all negative values of flow
change led to zero control action.

(a) (b)
Figure 4. Membership functions of first level fuzzy controller : (a) error ; (b) change in
error.
A large range universe of discourses were chosen for design convenience only. Actually, both
error range and change-in-error range are very small compared to their respective universe of
discourse. Hence, scaling factors were applied on the both inputs such that their values mapped
properly.
3.4 Rule Base Design of FLFC
The Rule base of FLFC was design by applying some control knowledges on the valve as the
actuator. The valve opening positions c CV was divided into seven opening positions: 0 (closed),
1/6, 2/6, 3/6, 4/6, 5/6, and 6/6 (fully open). This design choice corresponds to number of
membership functions used in each input.
According to our design experience, applying incorrect valve opening position give two results
in contrast. If it is smaller than it necessary, system undergoes surge. Conversely, if it is larger
than it has to be, the system may not achieve the reference point at all. These control knowledges
were then applied to arrange a set of fuzzy rule base as presented in Table 1.
While the arranged rule base did satisfy our design requirements, we disclaim that our rule base
is an optimal arrangement. Hence, we assume that it is still possible to arrange other rule bases
which may give better results .

5
YSM Simamora, E. Suharno, M. Syaikhudin, “Active surge control of centrifugal 
compressor using self­tuning fuzzy approach”, Orbith vol. 5 No. 1 Mar. 2009, 
pp 222­229;  ISSN : 1858­2095

Table 1. Fuzzy rule base for first-level fuzzy controller


Change in Error

CE0 CE1 CE2 CE3 CE4 CE5 CE6

E0 C C C C C C V6

E1 C C C C C V5 V6

E2 C C C C C V5 V6
Error

E3 C C C C C V5 V5

E4 C C C C C V4 V5

E5 C C V1 V2 V3 V4 V5

E6 C V1 V1 V2 V3 V4 V5

3.5 Second-Level Fuzzy Controller


The SLFC controller consists of two inputs and three outputs. Each of the input are fuzzified by
five membership functions as shown in figure 5. Since the structure of membership functions for
error and change of error are similar, the difference between them lie in their scaling factors.
Designing a three outputs fuzzy controller seems increasing an extreme computational
complexity. Therefore, a single output fuzzy system was still used instead. Inspired by Al-Nemar
(1996), the single output of second-level fuzzy controller was assigned for GE , GCE , and GU
tuning by applying some parameterization onto it.

Figure 5. Membership function of second-level fuzzy controller.


The basic idea of such parameterization was creating an equation such that the values of GU ,
GCE , and GU varying near around their respective initial value. It was chosen that :
G E=12  (13)
G CE=1 (14)
1
GU= (15)
1max
where ∣∣1 is the output of second-level fuzzy controller. It was chosen that ∣∣0.5 . In

6
YSM Simamora, E. Suharno, M. Syaikhudin, “Active surge control of centrifugal 
compressor using self­tuning fuzzy approach”, Orbith vol. 5 No. 1 Mar. 2009, 
pp 222­229;  ISSN : 1858­2095

(15), however, the tuning parameter is restricted to stay in the output range of first-stage fuzzy
system , that is the range of c CV .
Since the actual output remains single, a two dimension fuzzy rule base is still apply to the
controller, as shown in Table 2. This is a typical rule base arrangement, and left by default
throughout the design.
Table 2. Fuzzy rule base for second-level fuzzy controller
Change in Error

NB NS Z PS PB

NB NB NB NB NS Z

NS NB NB NS Z PS

Z NB NS Z PS PB
Error

PS NS Z PS PB PB

PB Z PS PB PB PB

3.6 Simulation Set Up


Simulations were carried out using fourth-step Runge-Kutta method in Matlab environment. The
step size h was set to 0.2. Throughout the simulation, u CV was set to 0.45, while initial flow
mass and pressure rise were set to 0.8 and 0.1, respectively. The last three numerical values were
taken from (Al-Mawali and Zhang, 2006).

4. Simulation Results
4.1 Open Loop Simulation
The aim of open loop simulation is to demonstrate centrifugal compressor behavior without
controller. Thus, c CV was set to 0 throughout this simulation. The throttle gain T was set to
0.61. This gain corresponds to  =0.4955 and  =0.6599, which is the nearest point from the
left side of the surge line.
As it is expected, the simulation setting lead the compressor to a surge. It is confirmed by the
presence of a limit cycle in compressor map view, as shown in Figure 6. It can also surge can
also be viewed as stable oscillation of  and  , both against  as plotted as dotted lines in
Figure 7.

Figure 6. Surge in compressor map view as T was set to 0.61.


7
YSM Simamora, E. Suharno, M. Syaikhudin, “Active surge control of centrifugal 
compressor using self­tuning fuzzy approach”, Orbith vol. 5 No. 1 Mar. 2009, 
pp 222­229;  ISSN : 1858­2095

The open-loop simulation shows that small change of  from the surge line can drastically
decrease the compressor performance. It also indicates that compressor operations at the surge
line can be very risky. The last mentioned may confirm the importance of a surge margin in surge
avoidance methods.
4.2 Closed Loop Simulations
In closed loop simulation, STF controller is inserted to the compression system and c CV is
activated as well. To demonstrate how STF dealing with surge, four different close-loop
simulations were scheduled.
The first closed-loop simulation is arranged for T =0.61. This is the similar reference used in
the open loop simulation. As results, compressor dynamic responses against  are plotted as
solid lines in figure .
In contrast to the dotted ones, solid lines in figure 7 indicate that STF controller can stabilize the
compressor. That figure also shows that compressor operates in a steady state as  ∞.

Figure 7. Compressor dynamic responses against  when T =0.61, for the open-loop (the
dotted lines) and the closed-loop (the solid lines).
The same simulation also provides / plot in figure 8. This compressor map view indicates
that a stable compressor is in the similar trajectory with c . However, since in this figure surge
line and T were hard to distinguish, there is no adequate information about surge range that
STF dealing with.

Figure 8. Compressor dynamic responses behavior in a stable condition compressor


map view.
The second close-loop simulation demonstrates how the controller dealing with a smaller
reference point. To achieve that goal, T was set to 0.4 (  =0.2285). Compared to the previous

8
YSM Simamora, E. Suharno, M. Syaikhudin, “Active surge control of centrifugal 
compressor using self­tuning fuzzy approach”, Orbith vol. 5 No. 1 Mar. 2009, 
pp 222­229;  ISSN : 1858­2095

reference point, this new reference point is farther from the left side of surge line. As results, a
/ plot is presented in figure 9.
Unlikely in figure 8, the surge line is now appear properly in figure 9. This appearance gives an
emphasis, that STF controller can manage the compressor operation in the surge region, and in a
fair distance point from the surge line.

Figure 9. Compressor performance when tracking  =0.2285.


The third closed-loop simulation demonstrates how STF controller dealing with changing
reference point during the operation. Here, T was initially set to 0.5 (  =0.3929), and then to
0.2 (  =0.1185) as  >1500. As results, the controller STFC can successfully track both
references, as shown in Figure 10.

Figure 10. Compressor response to an on line changing of reference point.


Since STF controller consists of two level fuzzy controllers, it is also of interest to demonstrate
behavior the SLF controller. Taken from the latest simulation, SLFC is shown in Figure 11.

Figure 11. SLFC action addressed to G E , G CE , and G U tuning.

9
YSM Simamora, E. Suharno, M. Syaikhudin, “Active surge control of centrifugal 
compressor using self­tuning fuzzy approach”, Orbith vol. 5 No. 1 Mar. 2009, 
pp 222­229;  ISSN : 1858­2095

As shown in figure 11, the SLFC tune the gains during the transient time. As the step response
goes to steady, the tuning parameters back to their initial conditions.

5. Conclusions
Self-tuning fuzzy approach on active surge control of centrifugal compressor is presented.
Several simulations demonstrates that STF controller can successfully deal with surge from
initial surge point at  =0.4955, to  =0.1185. Thus, we demonstrate that the STF controller
can handle the surge far better than the surge avoidance in the sense of operating point.
Since the STF controller presented here and the previous fuzzy approach (Al-Mawali and Zhang
(2006) are both applied to centrifugal compressor with constant speed, we suggest that the future
research of fuzzy based approach on active surge control will be addressed to centrifugal
compressor in non constant speed. We also suggest that in the future research, current fuzzy
based approaches will be carried on to experimental stage.

References
● Al-Mawali, S.H.T. & J. Zhang (2006), “A fuzzy approach to active surge control of
centrifugal compressors”, Proc. Int. Conf. Control, Glasgow, Scotland , UK
● Al-Nemar, T.M.T, (1996), “Fuzzy Logic-Based PID Self-Tuning”, Master Thesis, King Fahd
University of Petroleum and Minerals, Dhahran, Saudi Arabia
● Brown, R.N. (1997), “Compressor Selection and Sizing. Second Edition”, Houston, TX: Gulf
Professional Publishing
● Gravdahl, J.T. (1998), “Modeling and Control of Surge and Stall in Compressors”, Dr.ing.
Thesis, Norwegian University of Science and Technology, Trondheim
● Jager, R. (1995), “Fuzzy Logic in Control”, Ph. D dissertation, Technical University of Delft
● Jantzen, J. (2007), “Foundation of Fuzzy Control”, West Sussex, England: John Wiley &
Sons Ltd
● Slotine J-J.E and W. Li (1991),”Applied Nonlinear Control”, Englewood Cliffs, NJ :
Prentice-Hall, Inc.
● Tian, X., X. Wang, and Y. Cheng (2007),”A Self-tuning Fuzzy Controller for Networked
Control System”,IJCSNS,vol.7 No.1, pp 97-102
● Wang, L.X. (1997), ”A Course in Fuzzy System and Control”, Upper-Saddle River, NJ :
Prentice-Hall, Inc.
● Willems, F.P.T. (2000), “Modeling and Bounded Feedback Stabilization of Centrifugal
Compressor Surge”, Ph. D dissertation, Technical University of Eindhoven
● van Helvoirt, J. (2007), ”Centrifugal Compressor Surge, Modeling and Identification for
Control”, Ph. D dissertation, Technical University of Eindhoven

10

Potrebbero piacerti anche