You are on page 1of 5

An Experimental Evaluation of Optimal Control

Design for Coupled Tank system


Soumya Ranjan Mahapatro Bidyadhar Subudhi Subhojit Ghosh
Department of Electrical Engineering Department of Electrical Engineering Department of Electrical Engineering
National Institute of Technology, National Institute of Technology, National Institute of Technology,
Rourkela Rourkela Raipur
Odisha-769008, India Odisha-769008, India Chhatisgarh-492010, India
Mahapatro.soumya@gmail.com bidyadhar@nitrkl.ac.in

method fails to provide appropriate system response when


system uncertainties occur in the dynamics. To overcome
Abstract— This paper presents a Linear Quadratic this difficulty, various control techniques have been
Regulator (LQR) controller for the real-time control of a
proposed in literature these includes, an auto adjustable PI
coupled tank liquid level system. The model of the coupled
tank system has been developed based on system identification
Controller using MRAC technique [6], a standard 2 DOF
technique that employs least square error algorithm (LS) for PID with decoupling [7], an inverting decoupling techniques
parameter estimation. The proposed controller algorithm has [8].CRA (characteristics ratio assignment) [9], have been
been applied on the identified model. The performance of the applied to coupled tank system. This above communicated
projected control algorithm has been compared with that of a literature is inadequate to provide adequate response
Conventional PI controller. From both the simulation as well whenever certain physical constraint is assumed at both the
as the experimental results, it is observed that the performance control input and state of the dynamics. In present situation
of the proposed LQR control is more efficient then the widely the Linear Quadratic Regulator (LQR) controller is become
used conventional PI Controller. very popular in the control field application. The LQR is a
Keywords — Coupled Tank system, Linear Quadratic state space approach based optimal controller design
Regulator (LQR), Least Square Error Algorithm (LS), System technique, which operates the dynamics of a system at
Identification. minimum cost. Generally this type of controller techniques
provides best possible control action for a given set of
performance objectives [10]. Here the Performances are
I. INTRODUCTION
compared with conventional PI Controller in terms of time
Now a day the industrial application of liquid level control domain specification such as rise time, settling time, and
is tremendous especially in all process control industries[1] peak time.
and [2] Controlling liquid levels and flow in tanks of a
coupled tank system is considered as an important The rest of the paper is organized as follows. Section 2
benchmark control platform due to their wide spread provides development of a simplified mathematical model of
applications in the process control industries . The dynamics coupled tank system (CTS). Section 3 proposed the LQR
of the coupled tank system is nonlinear due to valve control algorithm. Section 4 presents both the simulation and
characteristics and also exhibits non-minimum phase experimental results. Finally conclusions are made in section
behavior[3]. The control objective in the coupled tank 5.
system is to maintain the liquid level at a desired level [1]. In II. MODELING OF COUPLED TANK
present scenario all process control industries present many Fig.1 gives sketch of the experimental set-up of the
challenging control problem since their nonlinear dynamic coupled tank system used in the present work. It consists
behavior, uncertain and time varying parameters, constraints four translucent tanks and each tank is fitted with an outlet
on manipulated variable, unmeasured disturbances, and pipe in order to transmit the over flow water to reservoir. In
dead-time. The coupled tank system dynamics has interred this process, fifth tank is used for water storage purposes i.e.
dependent chacteristics as it is a MIMO system. The as a reservoir. A level sensor is also attached at the base of
nonlinearity and the non-minimum phase behavior make the each tank in order to measure the water level of the
associated control problem very challenging. A number of PI corresponding tank. The output of the level sensor is
controllers have been extensively used in process control converted to 0-5 volt DC by the help of a signal
industries. These PI controllers exploit several tuning conditioning circuit. There are two pumps installed in the
methods for obtaining appropriate control parameters. reservoir in order to drive the water from bottom to top of
Among the several methods reported, Ziegler-Nichols based the tank. A scale is attached in front of all individual tanks
tuning method is widely used because of its simple structure for the purpose of monitoring the water level. By
[4] and [5]. However, this Ziegler-Nichols based tuning considering the mass balance equation, where the rate of
change of liquid volume in each tank is equals to the net ⎛ a1 ⎞ g
2
⎛ a1 ⎞ g
2

liquid flows into the tank and using Bernoulli’s equation Δh2 (t ) = ⎜ ⎟ . Δh1 (t ) - ⎜ ⎟ . Δh2 (t ) (5)
principle a set of nonlinear state equation can be obtained as ⎝ A ⎠ η u0 ⎝ A ⎠ η u0
follows.
Through linearization using Taylor series expansion, higher
order terms are omitted. Also some parameters of the
coupled tank system are not known perfectly .Henceforth
there is a need of an accurate model dynamics model of this
system. To overcome this above mentioned drawbacks this
paper considers system identification technique which is
accomplished by using black box model approach. In
identification pump control voltage is considered as the input
and level of the tank is considered as output.
u

y

Fig.2 A basic representation of Black box model Identification

In the current work a second order output error (OE221)


model has been considered for model identification since it
Fig.1 Schematic Diagram of a Coupled Tank System.
provides best fit to experimental data as compared to other
model such as ARMAX and ARX. The considered model is
given as follows,
dh1 a
= - 2 gh1 (t ) +η u (t ) (1)
dt A b
y (t ) = u (t ) (6)
f
dh2 a1 a
= 2 gh1 (t ) − 2 2 gh2 (t ) (2) In Eq. (6) the parameters b and f is obtained by using Least
dt A A Square Error estimation algorithm.

θˆLs = (ϕ T ϕ ) ϕ T y
−1
where (7)
h1 = water level in tank 1
h2 = water level in tank 2
a1= outlet area of tank 1 b1 ( q) = - 0.03402 q -1 + 0.03492 q -2
(8)
a2= outlet area of tank 2 f1 (q ) = 1-1.913 q -1 + 0.9131 q -2
A = cross-sectional area of tanks
g = gravitational constant III. PROPOSED CONTROL ALGORITHM
η = constant relating to the control voltage with the water The basic principle behind optimal control is that it
flow from the pump generally minimizes the cost which is associated with
generating control input [10] and [11]. The state feedback
For real time implementation, linear model is considered (LQR) is an optimal control method, which uses state space
for the controller design. Hence, the above linear model can approach to analyze a system. In this controller the main
be converted into a linear model by using Taylor series challenging task is that how to choose the value of Q and R,
expansion using two working points. by which it will match the design constraints.
2 Consider a LTI system given by,
1 ⎛ η uo A ⎞ ⎛ a1 ⎞
h10 = ⎜ ⎟ , h20 = ⎜ ⎟ h10 (3) x = A (t ) x + B (t )u
2 g ⎝ a1 ⎠ ⎝ a2 ⎠ (9)
y = Cx
For the above mentioned Eq. (9) the performance is readily
2
⎛a ⎞ g taken as
Δh1 (t) = - ⎜ 1 ⎟ . Δh1 (t) + η.Δu(t) (4) ∞
⎝ A ⎠ ηu 0 J = ∫ x(t )T Q x(t ) + u (t ) R u (t )
0
(10)

where
Q = Error weighted matrix TABLE 1.
SIMULATION PARAMETER
R = Control weighted matrix
Symbol Description Value
A Cross sectional area of the tanks 0.01389
a1 Tank 1 outlet area 0.1245
a2 Tank2 outlet area 0.1245
g Gravitational Constant 9.8
η Constant relating the control voltage 0.1194
with the water flow from the pump
d
dt Model linearisation plot
20

18 Tank1
d 16
Tank2

dt 14

12

L evel
10

8
Fig 3.Generalized Structure of the LQR Sate feedback controller 6

The control law is stated as follows 4

u = − Kx (11) 0
0 100 200 300
Time(sec)
400 500 600

Fig 4.Model Linearization Response


The optimal state x(t)* trajectory for the system is obtained
by solving the Algebraic Riccati Equation (ARE). 30
Identification Response of SISO model

20
AT P + PA - PBR-1 BT P = - Q (12) 10
Actual
Estimated

Then the optimal state feedback gain is obtained as -10


Level

-20

k = − R −1 BT P (13) -30

-40

Where -50

-60
0 100 200 300 400 500 600
P= Positive Semi-Definite Matrix Time(sec)

Fig 5.Identification Response of Tank


R= Positive scalar quantity
Response of input signal
Here the gain element matrix k is determines so as to 5
minimize the performance index, by which the feedback
gain i.e. u= -kx is optimal for any initial state x (0).In this 4
Supply Voltage

controller the error weighted matrix Q is always positive 3


semi-definite as to possess the error squared positive.
Generally the error weighted matrix (Q) is chosen as a 2

diagonal matrix on the other hand the control (R) is positive 1


definite matrix. In this optimal control the feedback gain K
can be produced by choosing a suitable value of Q and R 0
0 500 1000 1500 2000 2500 3000 3500
matrix. In this paper the value of Q and R is adapted by hit Sample
and trial approach and the value of Q and R is given as Eq. Fig.6 Response of input signal
(14).
Poles (x) and Zeros (o),
1
IV. RESULT AND DISCUSSION
0.5
The parameters of the CTS (Coupled tank system) as
given in Table 1 are used for simulation studies. Here a step Pole Zero
0
input is given to the system and the responses of both the
controllers are compared to each other. In the current work
-0.5
0.1s sampling time is considered for carrying out the
experiment.
-1
-1 -0.5 0 0.5 1 1.5

Fig.7 pole-Zero analysis of system


Correlation Analysis From fig.6 it is clearly observed that the coupled
1.5
tank system is a non-minimum phase system as one zero is
X= 0
Y= 1
present in right hand side (RHS) and auto correlation
1 analysis Fig.8 it is observed that all lags (which is the time
Correlation

difference (in samples) between the signals at which the


correlation is lies with in 90% of the confidence interval
0.5
(where the 90% confidence level means, the region around
zero represents the range of residual values that have a 90%
0 probability of being statistical insignificant).hence from the
-2 0 2 4 6 8 10
Lags obtained response plots it is envisaged that the selected
Fig 8.Model Validation Response by Correlation Analysis model is the suitable one for the controller design.
Simulation Response of LQR
30 It is clearly observed that from the simulation
25 results shown Fig.9and Fig.10 as well as from the
20
Desired Level
experimental response results Fig.11 & Fig.12 that there is
less overshoot occurs and the rise time, settling time, peak
Level

15
time yielded by the proposed LQR controller are less in
10
comparison the corresponding values obtained with the
5 Ziegler-Nichols based PI controller. The gain parameters
0
obtained from the proposed algorithm i.e. LQR controller is
5 10 15 20
Time(sec)
25 30 35 40
given as Eq. (15).
Fig 9.Simulation Response of proposed( LQR) controller

⎡1 0 ⎤
⎥ , R = [1]
Simulation Response of PID using ZN method
40 Q=⎢ (14)
30
⎣0 0 ⎦
20
K = [ -0.9998 -1.3262 ] (15)
Level

10
TABLE II
0
RESPONSE ANALYSIS BY TIME DOMAIN SPECIFICATION
-10

-20
Performance (Sec) Tuning Based On ZN Proposed LQR
0 100 200 300 400 500 600
Time(sec) Method Approach
Fig 10.Simulation Response of PI Using Ziegler-Nicholas Approach Rise Time 100 3.25

Experimental Response of LQR Peak Time 38.8 29.12


25
Settling Time 245 10.07
20
Experimental
15 Desired

10
V. CONCLUSIONS
Level

0 In this paper a Linear Quadratic Regulator (LQR)


-5 controller is proposed for the controlling of liquid levels in a
-10 coupled tank system. it has been observed from the results
-15
0 50 100 150 200 250
that, when the conventional PI controller is applied to system
Time(sec)
a large overshoot occurs, also it takes more time to settled at
Fig 11.Experimental Response of Proposed Controller algorithm the set point and as well as it takes large settling time to
Experimental Response of PI using ZN approach
30 reach the steady state. On the other hand by the proposed
LQR controller both rise times, settling time is minimizes
20
Experimental
and also less over shoot occur. Thus, the proposed control
10 Desired algorithm provides better liquid level control performance.
Level

0
REFERENCES
-10 [1] “Coupled Tank System: Control Experiment Manual”, Feedback
Instruments Ltd
-20
0 50 100 150 200 250 [2] B, Wayne Bequette, “Process Control Modeling design and simulation”,
Time(sec) PHI Publication, 2003
Fig 12.Experimental Response of PI tuning by Ziegler-Nicolas Approach [3] K.H.Johansson, “The quadruple-tank process: a multivariable
laboratory process with an adjustable zero”,IEEE Transaction on
Control system technologies,vol.8, no.3,pp.456-465,2000
[4] K. J. Astrom, R.M.Murry, “Feedback systems: An Introduction for
Scientists and Engineers”, N. Princeton University Press 2008.
[5] G. Yun Li, G.Chong, “PID control system analysis, design, and
technology”, IEEE Transaction on Control System Technologies,
vol.13, no.4, pp.559-576, 2005.
[6] S.Kangwanrat,V.Tipsuwannapom,A.Numsomran ,“ Design of PI
controller using MRAC techniques for coupled-Tanks Process”,
International Conference on Control Automation and Systems
(ICCAS),IEEE,pp.485-499,2010.
[7] A. Numsomran, T.Suksri and M.Thumma, “Design of 2 DOF PI
Controller with decoupling techniques for Coupled Tank Process”,
International Conference on Control, Automation and systems,
pp.339-344, 2007.
[8] A.Numsomran, V.Tipsuwanpom and K.Tirasesth, “Design of PID
Controller for the Modified Quadruple-Tank Process using Inverted
Decoupling Technique”, International Conference on Control,
Automation and systems (ICCAS), pp.1363-1368, 2011.
[9] J.Chaoraingern,A.Numsomranana,T.Suesut,T.Trisuwannawat,“PID
Controller Design Using Characteristics Ratio Assignment Method For
Coupled Tank Process”, IEEE Conference on Computational
Intelligence for Modeling Control and Automation,vol.1,pp.590-
594,2005.
[10] D.S. Naidu, Optimal Control System”, CRC press, FL, 2003.
[11] Ashish Tewari, “Modern control design with Matlab and Simulink”,
John Wiley & sons, 2003.

You might also like