You are on page 1of 4

A Comparison of Fuzzy, State Space with direct

eigenstructure assignment, and PID controller on


linearized MIMO plant Model
Davor Linaric, Todor Kostic and Vladimir Koroman
Brodarski Institute
Zagreb 10020, Croatia
tidavor, todor. kostic, vladimir. koroman)g@hrbi. hr

1. ABSTRACT Symbols and abbreviations.


DVT high pressure steam flow [kg/s]
-

Very often technical systems work very close to stationary 9 VT - high pressure steam valve position [p.u.]
working conditions, quasi stationary conditions. For D - intermediate pressure steam flow [kg/s]
example thermal power plant, steam turbine system mostly ST
working under quasi stationary working conditions. (PST intermediate pressure steam valve position [p.u.]
Bearing in mind that fact, the idea is to design control
structure optimal for that working conditions. For this e
purpose, State Space controller with direct eigenstructure pp -condenser pressure, [Pa]
assignment is designed and compared with Fuzzy and PID D l

[1] controller on linearized MIMO model of power plant, (DNT)1 - low pressure steam flow [kg/s],
steam turbine. 1. Controllability matrix
Key words: Eigenstructure Assignment, Fuzzy Control,
MIMO system Control, Steam-Turbine F=[B AB (A2)B (A3) B (A4) B] (1)
rank F=5=n.
2. MIJMO SYSTEM L-NEAR MODEL
2. Observability matrix

We consider multi-input time invariant system with linearized Q=[C CA ... CA'n] (2)
mathematical model [1,2] of steam turbine
x1=DVTI X2=Pe, X3=DSTI X4=DNT1, X5=Pp rank Q=5=n.
Y2= X5 Y1Y X2, U1= 9VT, U2= YST

2.1 MIMO SYSTEM CONTROL WITH DIRECT


-1 0 0 0 0 EIGENSTRUCTUREASIGMENT
12320 -0.03886 0 0 0
A= 0 0.0000274 -1 0 0 The state-space representation is given by the equations (3)
0 0 1 -1 0 for open and (4) for closed loop:
0 0 0 87184 -0.9541 xAx+Bu (3)
y=C x+D u
77.382 0 i =(A-BK). x (4)
0 -777240 y = C*x
B= 0 54.718 For characteristic equation
0 0 det(sI -A +BK)=0O (5)
0 0 and real and distinct closed-loop eigenvalues (6)
j
~~~~~diagA=[4i2...AZ]
0O 0 0 1-

1-4244-0020-1/05/$20.OO a2005 IEEE 1 of 4


Right eigenvectors for the system are (8) we could write
T=[v1 V2 .Vn] t
*x()= eAt .Ax(0)+ Je(tr) . B u()d (19)
Ar~ ~ ~ ~ ~ ~~~~~~~Xt e v(o + 9
~ =

Ar -v = V. (9) and system outputs are


Each closed-loop right eigenvector satisfies t
I- Ar]*Vi
=0 (10) y(t) = ( |C
+J )B u(r)dr (20)
K .I-A+B.K].v =0 (11) system dynamic matrix A can be represented by
Defining (12) A=VAV'-1= VAL (21)
qi = K vi
V is system eigenvector matrix
[2i I-AiB] {qj (13) e =V.e L=jve¾e lj
j=4
(22)
Then, for the n closed-loop eigenvalues
[q, q2 ...q1]=K[v1 V2 ...v.] Ai is the jth system eigenvalue, vj is jth column of V and Ij is jth
>, = -2,2 =-3,23 =-4,24 =-5,25 =-6 rowofL
Singular value decomposition (SVD) can be used to find a n t
basis for the null space [3] y(t) = C ve i(t)I. x(O) +
Ss-[AijIA B] B'] = Xi,k Vi,n+k ( C * .Bu(r)d(
q, k=1 y~~~~~~C*v1 `je2(t )u()J
1

K Q.T'T
= K=Q ' 15
~~~ ~ For~ ~~~~~(15)
i=1 0

B. u(t) =Z
k=1
bkuk (24)
-0.1348 -0.0000 -0.0444 -0.0407 0.0000 wher bk iS k column of B and uk kth system input
-0.1544 -0.0000 -0.0530 -0.5986 0.0005
That matrix K places eigenvalues and eigenvectors at values Output respectively to initial conditions and inputUk is given
by
specified by A , and v , respectively. n
y(t) = Z C -vieAj (.t)l x(0) +
2.2 ANALITICAL FORMULATION OF >' (25)
EIGENSTR UCTURE INFLUENCE ON Y v Fe T ,
SYSTEM DYNAMIC jj k Uk(F)dr
k

ith system output is


n
The eigenvalues and eigenvectors of system are related to
system dynamic response in the following way:
y(t)"'c'
y
ej (0 x(O)+
=1
j4 (26)
Taking Laplace- transforms of equation (3) gives n m .b-
1
;i 'vij Ij fbk e(
c Uk (T)dT
x(s) = [s I-A]' x(O) + [s I-A]' B u(s) (16) j4o
j=1 k=1
With initial conditions =0
n m t
Inverse Laplace transformation gives us yQ)=ZZR Uf (-r)d-r (27)
x(t) = {[s *I - A ] }- x(O) + (17) for Rijk=Ci vj lj bk
L' {[s * I - A B* u(s)} and for an impulsive input in the k input
ifreplace YQt) ZR IJA, .e<( (28)
L' {[s I1-A]' }= eA (18) j=1 k=l

2 of 4
Last equations show the importance of eigenvalues and
eigenvectors influence on system behavior, dynamic. The _ sko dobro virok
eigenvalues determine the natural frequency and damping of
each mode. The system eigenvectors determine the residues. =\
The residues are an indicator of how much each mode of the
system contributes to a given output. [3, 4]
3. FUZZY CONTROLERB 0.4-
02
s

The fuzzy controller used in this example is Mamdani type


[5,6] and structure of controller is shown in Figure 1 a,b Inside regoduz
the block Fuzzy in Figure 1 fuzzyfication of input signals
inference mechanism and defuzzyfication are included.
Inference rules and parameters of the fuzzy controller are
presented in [1]. The Figures 2,3 show membership functions atvori ztofimaIo bezpromjene
and Figures 4,5 output surface, for valve 1 and 2. This
Mamdani fuzzy controller is used to the control non-linear
MIMO plant [1] (the steam turbine plant). Fuzzy controller
combined with PID algorithm is used as steam turbine 1
governing system. In this example pressure control of steam
turbine was performed.

Ref-rntna vrijednost
Tubn k e uao 114146 G41 411 01 11 11 O6 118
r REGULAIOR I ~~~~~Figure 3. Membership functions output
R
P-nn -iadano .t3 N

(3)

11E1ZR-1T1 ~ ~ ~~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ pouta - 1re~u

RE.11,IT.R

~~~~~~Figure 4. Outputrshiurfc,tighsouprsurevav

parnira.n

-0

§y"" _ 9W 2 MMM - of 4

Figure I.a,b Fuzzy control system structure _--


2 -
1_t X \ X _\ X
E~~~~~~~~~~~~~
(3)
PtotunakVOITAM (5)~~~~~~~~~~~prtulalad
5x55 - - rg0u
Figure 5. Output surface, intermediate pressure valve The lower value of functional J have system with lower control
error [7].
4. SIMlVULA ZTIONREULT
.5 CONCL USION
For an impulsive input control simulation result for state space
(with direct eigenstructure assigment), the fuzzy control
algorithm and PID algorithm are presented. The responses Of In this paper, we presented an eigenstructure assignment
the compared systems are shown in approach control for dynamic system and influence on system
Figure 6. and Figure 7. All control algorithms are compared on behavior, dynamic. On the basis of the simulation result of
the same linear MIMO model used for synthesis of state space linear MIMO stem turbine system with impulse input, state
controller. The simulation was performed with parameters space, fuzzy and PID controller are compared. With applied
detailed describe in [1]. integral quadratic transition analysis method it can be said that
state space controller show the best characteristics. The further
i investigation is engaged with reconfigurable control of the
P-D steam turbine with fuzzy controller which have the better
16 _ REG-STANJA
NEIZFZAZIT] performance in transient regime and state space with define
14-
12
_ eigestructure for the quasi stationary working conditions [1]. In
10
_ addition, as one of the main problems, this approach would
have to resolve state /error residual generation and estimation
*11 \i as expert system for detection of system working regime
~~change.
4 2
6. REFERENCES
0 5 10 15 2D 25 30 35 48 415 50
H [ 1] Linaric D., Steam turbine control system analysis, Master thesis FER,
Figure 6. Simulation result outputl University Zagreb, 2002.
1 [ 2] A. W. Ordys, A. W. Pike, M.A. Johnson, Modeling and Simulation of
_PD
Power Generation Plants, Springer-Verlag 1996
FRZEG-STANJA
6 0 /\
NEIZRAZITI
[ 3] Guy 0. Beale, Dominick Andrisani, Multivariable Control Placement
of Eigenvectors, Electrical and Computer Engineering department
3-
George Mason University Fairfax, Virginia 1998

2I f * [4] John B. Davidson, Dominick Andrisani, NASA Technical


Memorandum 109130, Langley Research Center Hampton, Virginia,
1 ~~~~~~~~~~~~~~~~~~~~1994.
[5] Earl Cox, The Fuzzy system Handbook, Morgan Kaufmann, 1999
-1I
[] [ 6] Roland R. Yager, Dimitar P. Filev, Essentials of Fuzzy modeling and
control, John Wiley and Sons, 1994
Figure 7. Simulation result output2
[7] Novakovic B., Metode vodenja tehnickih sistema, Fakultet
Quantitative comparison between responses: strojarsva i brodogradnje Zagreb, 1990
00

0
2
Ed (t) Control error
JPID =7.601.102
JFuzzy = 5.767 .1 02
STATESPACE = 5.162 * 10

From the result shown in Figure 6 and Figure 7 it could be seen


that state space controller have the lowest value of functional J.

4 of 4

You might also like