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

net/publication/221399290

Fuzzy Control for a Liquid Level System

Conference Paper · January 2003

Source: DBLP

CITATIONS READS

4 254

3 authors, including:

Basil M. Al-Hadithi
Universidad Politécnica de Madrid
136 PUBLICATIONS   583 CITATIONS   

SEE PROFILE

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

SSR antenna analyser under operating conditions View project

All content following this page was uploaded by Basil M. Al-Hadithi on 19 May 2014.

The user has requested enhancement of the downloaded file.

 Fuzzy Control for a Liquid Level System

Basil M. Al-Hadithi Fernando Matı́a Agustı́n Jiménez

Departamento de Electrónica y Sistemas Departamento de Automática, Departamento de Automática,
Universidad Alfonso X El Sabio Ingenierı́a Electrónica Ingenierı́a Electrónica
Villanueva de la Cañada, e Informática Industrial e Informática Industrial
Av/de la Universidad,1. Universidad Politécnica de Madrid Universidad Politécnica de Madrid
28691-Madrid, Spain J.Gutierrez Abascal, 2. J.Gutierrez Abascal, 2.
[email protected] 28006-Madrid, Spain 28006-Madrid, Spain
[email protected] [email protected]

Abstract In the past, the design of controllers based on a lin-

earized model of real control systems. In many cases
This paper deals with the design of a fuzzy a good response of complex and highly non-linear real
controller (FC) based on pole assignment process is difficult to obtain by applying conventional
method for the control of a liquid level control techniques which often employ linear math-
system. The problem is to control level ematical models of the process. One reason for this
set point changes by adjusting the flowrate lack of a satisfactory performance is the fact that lin-
of liquid entering the tank through a feed earization of a non-linear system might be valid only
pump [2], [5]. The behaviour of the closed as an approximation to the real system around a deter-
loop fuzzy system controlled by the fuzzy mined operating point [1], [6].
controller is identical to the linear system However, fuzzy controllers are basically non-linear,
whose state transition matrix is the desired
and effective enough to provide the desired non-linear
one.
control actions by carefully adjusting their parameters
The main objective of this paper is the use [4], [8], [9].
of the affine T-S fuzzy model to represent
In this paper, the non-linear system is represented by
both the controlled system and the fuzzy
affine T-S model, where the consequent part of each
controller taking into consideration the ef-
rule represents an affine model of the original sys-
fect of the constant term. Most of the re-
tem in a certain operating point. The final fuzzy sys-
search works analyzed the T-S model as-
tem can be obtained by blending of these affine mod-
suming that the non-linear system is lin-
els. The control is carried out based on the fuzzy
earized with respect to the origin in each
model via the so-called parallel distributed compensa-
IF-THEN rule, which means that the con-
tion scheme [7]. The idea is that for each local affine
sequent part of each rule is a linear function
model, an affine linear feedback control is designed.
with zero constant term. This will in turn
the resulting overall controller, which is also a non-
reduce the accuracy of approximating non-
linear one, is again a blending of each individual affine
linear systems [3].
linear controller.
Keywords: Fuzzy control, Affine Takagi-
Sugeno model, Liquid level control.
2 Takagi and Sugeno’s fuzzy model

1 Introduction The continuous fuzzy dynamic model proposed by

Takagi and Sugeno is described by fuzzy IF-THEN
Up to now, the fuzzy controller has been the most suc- rules, where the consequent part is an affine input-
cessful application field for fuzzy logic. Many appli- output relation. The main feature of this model is to
cations show that the fuzzy controllers yield results express the local dynamics of each fuzzy implication
superior to those obtained by the conventional control by an affine model. The final fuzzy system is resulted
algorithms. from blending of the affine linear system models.
The IF-THEN rules are as follows:

R i1





in  : If x is M1i1 and ẋ is M2i2 and   

and x n 1  is Mnin
i i
then ẋ  ao 1  n   A i1 in x  B i1 in u
(1)
 Figure 1: Fuzzy control system
where

M1i1 i1  1 2    r1
are fuzzy sets for x, M2i2
i2  1 2    r2
are fuzzy sets for ẋ and 
Mnin in 
1 2    rn
are fuzzy sets for x n 1 
. x t
is the state and x n 1  is Mnin
vector, u t
is the input vector. Therefore the whole i i
then ẋ  ao 1  n   A i1 in x  B i1 in u
fuzzy system consists of r1 r2     rn rules. The expres-
sion for A i1 in  is (2)
 
0 1 0  0 
where the membership functions are given by µMi1 x
,
 .. .. 
  1
 0 0 1 . . 
µMi2 ẋ
and µMnin x n 1 
, respectively as shown in fig-
 
 .. .. .. ..  2
 . . . . 0  ure 2.
 
0 0  0 1 
i i i i i i i i
 a1 1  n   a2 1  n    1  n 
an  1 an 1  n 


while

 
0
 .. 
i i
a0 1  n 

 
. 

0 
i i
a0 1  n 

xT   x ẋ  x n 1  

The final output of the fuzzy system can be repre-

sented as Figure 2: Membership functions of the fuzzy system
    
ẋ  a0 x
 A x
x  B x
u x
The j1    jn
th fuzzy controller rule is:

C j1  jn 
: If x is N1j1 and ẋ is N2j2 and   
3 Fuzzy Controller based on Pole
Assignment Method and x n 1  is Nnjn
j j
Let us consider the problem of designing a FC using then u  r  ko 1  n   K j1  jn 
x (3)
the pole assignment method. The fuzzy model is de- j  j
scribed by T-S model. Figure 1 represents the fuzzy where N1 1 j1  1 2    s1
are fuzzy sets for x, N2 2
 j 
control system. j2  1 2    s2
are fuzzy sets for ẋ and Nn n jn 
 1 2    sn
are fuzzy sets for x n 1  as shown in figure
Let the i1    in
th rule of T-S fuzzy model be repre-
3. h j1  jn   k0 j1  jn   K j1  jn x is r-dimensional
sented by:
feedback vector, the reference input vector and r is r-
dimensional and K j1  jn  is r n matrix. The fuzzy
S i1 in  : If x is M1i1 and ẋ is M2i2 and    controller will be
 r r s s
∑i11 1  ∑inn 1 ∑ j11 1  ∑ jnn 1 w i1 in  j1  jn  x A i1 in x
ẋ  
r r s s
∑i11 1  ∑inn 1 ∑ j11 1  ∑ jnn 1 w i1 in  j1  jn  x 
r r s s
∑i11 1  ∑inn 1 ∑ j11 1  ∑ jnn 1 w i1 in  j1  jn  x B i1 in  K  j1  jn x
r1 rn s s
∑i1 1  ∑in 1 ∑ j11 1  ∑ jnn 1 w i1 in  j1  jn  x 
(8)

 j j j j j j
where the elements ki1 1  n  ki2 1  n     kin 1  n 
of
the feedback vector h j1  jn   k0 j1  jn   K j1  jn x
are computed from the following matrix, where its
eigen values are the desired closed loop poles:

Figure 3: Membership functions of the fuzzy con- A i1 in   B i1 in K j1  jn 

(9)
troller
j1  jn 
and the constant term k0 can be calculated from
(6)

u  r  h x
(4) The behaviour of the fuzzy system represented by (2)
and controlled by fuzzy controller in (3) is the same
Substituting (3) in (2), we get the feedback system. as the linear system which contains the desired closed
loop poles (9).

SC i1 in j1  jn 
: If x is M1i1 and N1j1
and The fuzzy controller implemented in this paper is the
 one shown in (3) modified by introducing the condi-
j
ẋ is M2i2 and N2 2
   and x n 1  is tion given by (6) which can be written as,

Mnin and Nnjn

s s 

then ẋ  a0 i1 in 
 A i1 in x  

∑ j11  1  ∑ jnn
1w
j1  jn  x K j1  jn x


ũ x
 r s1 sn 

∑ j1 1    ∑ jn 1 w j1  jn  x

B i1 in  r  k0 j1  jn 

 K j1  jn 
x  (5) (10)
with
j1  jn 
Suppose that we choose k0 such that

i1 in   K j1  jn x
a0  Bi1 in k0 j1  jn  0 (6)  
x
 ẋ 
then (5) will be: j j j j  
 
ki1 1  n   kin 1  n 

 ..  (11)
. 

 x n 1 
SC i1 in j1  jn 
: If x is M1i1 and N1j1
and

ẋ is M2i2 and N2j2
   and x n 1  is It can be also written in another form,

Mnin and Nnjn
 
j j 
i1 in  i1 in  ũ1 1  n  x

then ẋ  A x B r j1  jn  

.. 
j1  jn 
ũ x
 
.


r K x
 i1 in  j1  jn  
B K x (7) ũr
j1  jn 
x

Then the final output of the fuzzy closed loop system

j1  jn  
is ũi x

  
x (head). The capacitance of the tank is equal to its
 ẋ 
  cross-sectional area. qi and qo represent the inflow
 r  ki1j1  jn 


j j
kin 1  n 

 ..  (12) rate and outflow rate. If the flow through the restric-
. 

x n 1  tion is turbulent, the relation between qo and h can be

represented as:
4 Liquid level control system


qo t
 k h t
(15)
The process applied in this paper is a level control sys-
tem which consists of a pump unit which is consid- where k is related to the restriction and the valve at the
ered to be a second order linear system. It pumps out output of the tank. Substituting (15) in (14) we obtain,
the liquid as an inlet flow to reach a predetermined
level inside a tank with a uniform horizontal cross- dh
 
section, which discharges liquid through a restriction A k h t
 qi t
(16)
dt
in its base as shown in figure 4.
Linearizing the resulted non-linear system (16) in var-
ious operating points gives:

dh 
k    
A  ḣ0   h t
h0
 qi t
 qi0 (17)
dt 2 h0

Substituting (17) in (19) and rearranging the equation

gives:
Figure 4: Liquid level system scheme
kT T
The feed pump can be described by the following A T1 T2  1 2 2 A  k T1 T2 
d3 h 2 h0  d h 2 h0  dh
equation: dt 3   AT1 T2 dt 2   AT1 T2 dt
k dh
 A dt0  k h0 qi0
2 h0 2 h0  Kp u t 
 h t
   (18)
 AT T
Qi s
Kp 1 2 AT1 T2 AT1 T2
    (13)
U s
1  T1 s
1  T2 s

5 Control System Design and Experiments

where K p represents the steady state gain and T1 and
T2 represent the time constants of the pump. The val- The pump unit has been approximated to be a second
ues of K p , T1 and T2 are evaluated using the data ob- order linear system with the following transfer func-
tained from the experimental tests carried out to deter- tion:
mine the relationship between input voltage applied to

the pump and the input flowrate to the tank. Qi s
1 3
    (19)
The tank is a non-linear system with the output U s
1  0 36s
1  0 71s

flowrate proportional

to the square root of the decreas-
ing level h1 2 t
. The differential equation can be ob- The tank was simulated depending upon various ex-
tained as follows: Since the inflow minus outflow dur- periments looking for a suitable model that coincide
ing small interval dt is equal to the additional amount with the non-linear model given by (16). The area of
stored in the tank, the tank A is equal to 10 cm2 The controlled process
is linearized in three operating points and a fuzzy con-
troller is designed which meets the requirements of
dh  
A  qi t
 qo t
(14) small overshoot in the transient response and a well
dt
damped oscilations with no steady state error.
where A is the capacitance of the tank, which is de-
Recalling (18),
fined to be the change of the quantity of stored liq-
uid necessary to cause a unit change in the potential Let
 xT dx d2 x (20)
 x dt dt 2

k01 
1
0 45 k11 
1
0 9 k21 
1
78 k31  199
2
Where x denotes the head of the liquid, dx d x
dt and dt 2 are k02 
2
0 39 k11 
2
2 8 k21 
2
45 6 k31  101 8
the first and second derivative of the head respectively.
Firstly, a T-S affine fuzzy model is used to represent k03 
3
0 37 k11 
3
2 4 k21 
3
34 k31  74
the controlled process. The process is linearized in
three operating points as shown in Figure 5. In order The transient response in figure 6 shows a good re-
to simplify the simulation, it is assumed that the mem- sponse near the top and the bottom of the modeled
bership functions of the fuzzy controller are the same level system.
as those of the fuzzy system.

Figure 5: Membership functions of the fuzzy system


R1 : I f h t
is M11
then

d3h d2h dh  
  
65  13 5 
8 7h t
 0 09  0 2u t

dt 3 dt 2 dt Figure 6: Transient response of the liquid level system

(21)
 
1 1   1 1 
C : I f h t
is N1
then u t
 r ko  K h t
 We can improve the transient response of the control
 system as shown in figure 7 by changing the position
R2 : I f h t
is M12
then of the closed poles. For example if we choose the
d3h d2h dh  
desired poles to be situated in 3 65  7 48 4 95,
  
53  8 2 
3 9h t
 0 2  0 5u t
then the controller parameter are:
dt 3 dt 2 dt
(22)
 
2 2   2 2 
C : I f h t
is N1
then u t
 r ko  K h t

 k01 
1
0 45 k11 
1
24 k21 
1
398 k31  1452
R3 : I f h t
is M13
then

d3h d2h  dh    k02 

2
0 39 k11 
2
14 k21 
2
193 k31  678
 
5 04 7 09 2 9h t
 0 26  0 68u t

dt 3 dt 2 dt
(23) k03 
3
0 37 k11 
3
10 k21 
3
140 k31  485 7
  
C3 : I f h t
is N13
then u t
 r  ko3  K3 h t


Using the pole asignment design method explained

above and by choosing the desired closed loop poles
to be situated in 1 95  4 1 2 65, the feedback
controller parameters given by the feedback vector
K j1  jn  can be calcualted from (9) and the con-
stant parameters k0 j1  jn  can be determined from (6).
Thus the controller parameters are:
 rules1 In Fuzzy Sets ans Systems, volume 71,
pages 369-387, 1995.

[2] B. E. Postlethwaite. Building a model-based

fuzzy controller. In Fuzzy Sets ans Systems, vol-
ume 79, pages 3-13, Abril 1996.

[3] B. M. Al-Hadithi and F. Matı́a and A. Jiménez.

Slide-Mode Fuzzy Controller for Highly Non-
Linear Systems. In FUZZ-IEEE 2001, The 10th
International Conference on Fuzzy System, Mel-
bourne, Australia, 2-5 December 2001.

[4] Bin-Da Liu and Chun-Yueh Huang. Design and

implementation of the tree based fuzzy logic
Figure 7: Transient response of the liquid level system controller. In Transactions on Systems Man and
Cybernetics-Part B: Cybernetics, volume 27,
No.3, pages 475-487, June 1997.
6 Conclusion
[5] B. P. Graham. Fuzzy identification and control of
The paper presented the design of a T-S fuzzy con- a liquid rig. In Fuzzy Sets and Systems, volume
troller for the control of liquid level system repre- 26, pages 255-273, 1988.
sented by the affine Takagi-Sugeno (T-S) fuzzy model.
[6] Young-Moon Park and Un-chul Moon and
The problem was to control level set point changes
Kwang Y. Lee. A self- organizing fuzzy
by adjusting the flowrate of liquid entering the tank
logic controller for dynamic systems using a
through a feed pump. A fuzzy controller has been de-
fuzzy auto-regressive moving average (FARMA)
signed based on pole assignment method. The main
model. In Transactions on Fuzzy System, vol-
advantage of using the pole assignment method is that
ume 3, No. 1, pages 75-82, february 1995.
behaviour of the closed loop fuzzy system controlled
by the suggested fuzzy controller is identical to the [7] Hua O. Wang and Kazua Tanaka and Michael
linear system whose closed loop poles are the desired F. Griffin. An approach to fuzzy control of non-
ones and that the fuzzy controller designed with the linear systems: stability and design issues. In
pole assignment method guarantees the stability of the Transactions on Fuzzy System, volume 4, No.
controlled system. 1, pages 14-23, february 1996.
The results obtained in this paper have shown a fast re- [8] L.X. Wang. Stable adaptive fuzzy control of non-
sponse with small overshoot in the transient response linear systems. In Transactions on Fuzzy Sys-
and a well damped oscilations with zero steady state tem, volume 1, pages 146-155, 1993.
error.
[9] Won Chul Kim and Sang Chul Ahn and Wook
Acknowledgements Hyun Kwon. Stability analysis and stabilization
of fuzzy state space models. In Fuzzy Sets ans
This work is funded by Spanish Ministry of Sci- Systems, volume 71, pages 131-142, 1995.
ence and Technology (URBANO project DPI2001-
3652C0201) and by European Union 5th R+D Frame-
work Program (WebFAIR project IST-2000-29456).

References

[1] Andreas Bastian. Handling the nonlinearity of a

fuzzy logic controller at the transition between

View publication stats