Вы находитесь на странице: 1из 5

Design of Radial Basis Function Neural Networks Controller Based on Sliding Surface for a Coupled Tanks System

M.Aliasghary
Control Engineering Department, Faculty of Electronics Engineering, Istanbul Technical University Istanbul, Turkey aliasghary@itu.edu.tr

H.Ghasemzadeh
Microelectronics Research Laboratory of Urmia University Urmia, Iran st_h.ghasemzadeh@urmia.ac.ir

A.Naderi
Electronic Engineering Department, Faculty of Electronics Engineering, Istanbul Technical University Istanbul, Turkey naderi@itu.edu.tr
Abstract In this paper, the level control of coupled tanks is investigated. We developed a radial basis function neural networks based on sliding mode for control of coupled tanks system. In this study we used sliding surface and generalized learning rule to eliminate jacobain problem in adaptive neural networks controllers. The simulation results show that the proposed controller is able to control coupled tanks and the chattering phenomenon of conventional switching type sliding mode control does not occur in this study. Keywords: Radial basis function, Sliding mode, Coupled Tanks system, Sliding surface, Neural networks

A.Pourazar
Electronic Engineering Department University of Tabriz Tabriz, Iran pourazar.alireza@gmail.com The paper is organized as follows. The mathematical model of the coupled tanks is presented in section II. The RBFSliding mode controller is proposed in detail in section III. Section IV discusses the simulation results of the proposed control schemes. Finally, the conclusion is given in Section V. II. MATHEMATICAL MODEL OF THE SYSTEM

Fig. 1 is a diagram of the coupled tanks system. The dynamic model of the system can be written as [6]
1 1 = (1 + )
2

I.

INTRODUCTION

Liquid tank systems play important role in industrial application such as in food processing, beverage, dairy, filtration, effluent treatment, pharmaceutical industry, water purification system, industrial chemical processing and spray coating [1]. In recent years, a lot of works have been reported in the literature for controlling liquid flow tanks. A nonlinear constrained predictive control applied to a coupled-tanks system is given in [2-3]. Other types of nonlinear controllers based on nonlinear methods have been reported in the literature [4-6]. In [6] the authors proposed a static sliding mode control scheme and two different dynamical sliding mode control schemes for a coupled tanks system. Control laws based on fuzzy control [7-8], and neural network techniques [9] have also been used to control coupled tanks systems. In this paper, we consider a coupled tanks system and propose a radial basis function (RBF) sliding mode controller for this system. The proposed approach combines the advantages of the adaptive, neural network and sliding mode control strategies.

and

Where 1() is the level in the first tank, 2() is the level in the second tank, 1 () is the flow rate from Tank 1 to Tank 2, 2() the flow rate out of Tank 2, () is the inlet flow rate, is gravitational constant, is the cross-section area of Tank 1 and Tank 2, 12 is the area of the coupling orifice and 2 is the area of the outlet orifice. Note that () cannot be negative because the pump can only pump water into the tank. for (3)

1 = 12 2(1 2 ) 2 = 2 22

= (1 2)
1

(1)
1 > 2

for

2 > 0

(2)

___________________________________ 978-1-4244-8624-3/11/$26.00 2011 IEEE

III.
1

RBF SLIDING MODE CONTROLLER

A. Radial basis function (RBF) neural networks


2

Fig. 1 Diagram of the coupled tanks system

Let

1 = 2 > 0, 2 = 1 2 > 0, = 1 = , 2 =

Radial basis functions were first introduced by Powell to solve the real multivariate interpolation problem [10]. In the field of neural networks, radial basis functions were first used by Broomhead and Lowe [11]. RBF neural networks are one of the more popular neural networks. Radial basis functions networks are embedded in a two layer neural networks, where each hidden unit implements a radial activated function. The output units implemented a weighted sum of hidden unit outputs. The input into an RBF network is nonlinear while the output is linear. Their excellent approximation capabilities have been studied in [12]. The output of the first layer for a RBF network is:
() =
2 22

and

2 2

122

The output of the linear layer is (4)


= 1, 2, 3, ,

Then we can write the dynamic model in Eq. (1) as following


1 = 1 1 + 2 2 = 1 2 = 1 1 22 2 +
1

= () = () = =1

= 1,2 ,3, ,

(10)

(11)

(5)

The object of the control is to regulate the output () = 1 () = 2() to a desired value H. We define the transformation = () such that
1 = 1 2 = 1 1 + 2 2 1 = 2

vector of the network, respectively, and = [1 , , ] is the hidden output vector. n is the number of hidden neurons, = [1 , , ] is the weights vector of the network, Where x R n and y R m are input vector and output parameters

ci

and

are centers and radii of the basis

functions, respectively. The adjustable parameters of RBF networks are W , ci and i . Since the networks output is linear in the weights, these weights can be established by least-square methods. The adaptation of the RBF parameters

(6)

ci and i

is a non-linear optimization problem that can be

So we can write the dynamic model in Eq. (5) as


2 = + =
1 2 1 2 2 2 2 2 1

solved by gradient-descent method. B. RBF sliding mode controller (7) In the design of sliding mode controller for coupled tanks system, the control objective is to drive the output () = 1 () = 2() to the desired value H. By defining the tracking error to be in the following form:
() = 1 ()
1

Where

= 1

2 +
1

1 2 2

2 2

(8) (9)

(12)

The sliding surface on the phase plane can be defined as:


(, ) = + = 11 (1) =0

(13)

In this case = 2:

The sliding variables, will be used as the single-input signal for establishing a RBF neural network model to calculate the control law, u. Then for the single-input and single-output case in this paper, the output of the controller based on RBF networks is:
= =1
2 2

(, ) = () + ()

(, ) = 1 1 + 2 2 + (1 )

(14)

Finally we can find updating rule as follow:


= . . .

(22)

(23)

from equation (15) we have:

Where n is the number of hidden layer neurons and u is the final closed-loop control input signal. In order to combine the advantages of sliding mode and adaptive control schemes into the RBFNN, an adaptive rule is introduced to adjust the weightings between hidden and output layers. Based on the Lyapunov theorem, the sliding surface reaching condition is:
= 2
1 2

(15)

It is clear that we do not need any identifier for coupled tanks system. For modify the control signal and obtain realistic results, the simulations are carried out using the 0 50 3 input constraint [6]. s
RBF Constraint Plant

= ()

(24)

If a control input u can be chosen to satisfy this reaching condition, the control system will converge to the origin of the phase plane. Adaptive law is used to adjust the weightings for searching the optimal weighting values and obtaining the stable convergence property. The adaptive law is derived from the steep descent rule to minimize the value of < 0 with respect to W. Then the updated equation of the weighting parameters is:
=
=

= < 0

(16)

Fig. 2 Block diagram of the modified RBF-sliding mode

IV.

SIMULATION RESULTS

or

(17)

= . .

And from equation (14) we have:


= + =
1 2 2

1 2 +
2 1

= 1 + 1
2 1 2

(18) (19)

2 2 +

(20)

2 2

+ (1 1 + 2 2 )

form equations (7) we can find that:


1

=0

In this section, simulation results are presented. The controllers that designed above are simulated using the MATLAB software. The areas of the orifices, 12 and 2 are 0.58 and 0.24 2 , respectively. The cross-section area of Tank 1 and Tank 2 are found to be 208.2 2 . The gravitational constant is 981 2. The desired value of the output of the system is taken to be = 6 [6]. The controller parameter is set to 0.1. Fig. 3 shows the simulation results when the static sliding mode control that proposed in [6] is used. Fig. 4 shows the simulation results when the RBF sliding mode controller without input constraint is used. It can be seen from Fig. 4.a that the output converges to its desired value H. Fig. 5 shows the simulation results when the modified RBF-Sliding mode controller (with input constrained) is used for control of coupled tanks system. It can be seen from Fig. 5.b that the control signal is smooth and without any chattering when we compared it to classical sliding mode control. In Fig. 6, we compare these results with static (classical) sliding mode controller. Our proposed method performs almost the same as classical sliding mode control but the chattering is completely avoided in our proposed method.

(21)

10

Sliding Mode Controller 6.5 6 5.5 40 5 60

Sliding mode control signal 0.05

sliding surface

50

-0.05

q(t) in cm3/sec

h (t) in cm

30

-0.1

4.5 4 3.5 3

-0.15 20 -0.2 10 -0.25 0

2.5

20

40

60

80 Time(s)

100

120

140

160

20

40

60

80 Time(s)

100

120

140

160

-0.3

20

40

60

80 Time(s)

100

120

140

160

(a)
Fig. 3 Static Sliding mode controller (a) Output
Sliding Mode Based on RBF Controller 6.5 6 5.5 5 100 90 80 70

(b)
(b) Control signal (c) Sliding surface

(c)

Control signal 0.05 0 -0.05 -0.1

sliding surface

q(t) in Cm3/sec

60 50 40 30 -0.15 -0.2 -0.25 -0.3 -0.35 -0.4

h (t) in cm

4.5 4 3.5 3 2.5

20 10 0 0 20 40 60 80 Time(s) 100 120 140 160 0 20 40 60 80 Time(s) 100 120 140 160

20

40

60

80 Time(s)

100

120

140

160

(a)

(b)
(b) Control signal
0.05 0 50 -0.05 40 -0.1 -0.15 -0.2 20 -0.25

(c)
(c) Sliding surface
sliding surface

Fig. 4 RBF-Sliding mode controller without constraint on control signal (a) Output
Sliding Mode Based on RBF Controller 6.5 6 5.5 5 60 Control signal

q(t) in Cm3/sec

h (t) in cm

30

4.5 4 3.5 3

10

-0.3 -0.35 -0.4

0 2.5 0 20 40 60 80 Time(s) 100 120 140 160 0 20 40 60 80 Time(s) 100 120 140 160

20

40

60

80 Time(s)

100

120

140

160

(a)

(b)

(c)

Fig. 5 Modified RBF-Sliding mode controller with constraint on control signal (a) Output (b) Control signal (c) Sliding surface

6.5 6 5.5 5

h (t) in cm

4.5 4 3.5 3 2.5 RBF-Sliding Mode Control Static Sliding Mode Control

20

40

60

80 Time(s)

100

120

140

160

Fig.6 Comparison result with static sliding mode controller

1 1

V.

CONCLUSION

In this paper, a new RBF-sliding mode control method for coupled tanks system is proposed, which combines the merits of adaptive neural network and sliding mode control. Based on the Lyapunov stability theory, a RBF-sliding mode controller is designed for stabilization of coupled tanks system at the desired point in the state space. Simulation results show that the proposed controller is able to control coupled tanks system and the chattering phenomenon of conventional switching type sliding control does not occur in this study.

ACKNOWLEDGMENT First Author gratefully acknowledges that this work was partially supported by the Mechatronics Education and Research Center of Istanbul Technical University. REFERENCES
[1] M.S. Ramli, RM.T. Raja Ismail, M.A. Ahmad, S. Mohamad Nawi and M.A. Mat Hussin, Improved Coupled Tank Liquid Levels System Based on Swarm Adaptive Tuning of Hybrid Proportional-Integral Neural Network Controller, American J. of Engineering and Applied Sciences 2 (4): 669-675, 2009

Poulsen NK, Kouvaritakis B, Cannon M. Constrained predictive control and its application to a coupled-tanks apparatus. Int J Control;74:55264, 2001. [3] Poulsen NK, Kouvaritakis B, Cannon M. Nonlinear constrained predictive control applied to a coupled-tanks apparatus. IEE Proc- Control Theory Appl;148:1724, 2001. [4] Grega W. Minimum-time liquid level control. Presented at third IEEE conference on control applications, 1994. [5] R. Benayache, L. Chrifi-Alaoui, P. Bussy and J.M. Castelain, Design and implementation of sliding mode controller with varying boundary layer for a coupled tanks system. 17th Mediterranean Conference on Control & Automation, p:12151220 2009. [6] N.B. Almutairi, M. Zribi, Sliding mode control of coupled tanks, Mechatronics 16 pp.427441, 2006. [7] Korba P, Babuska R, Verbruggen HB, Frank PM. Fuzzy gain scheduling: controller and observer design based on Lyapunov method and convex optimization. IEEE Trans Fuzzy Syst;11: 28598, 2003. [8] Korba P, Frank PM. An applied optimization-based gainscheduled fuzzy control. Presented at American control conference, 2000. [9] Evans JT, Gomm JB, Williams D, Lisboa PJG. Implementation and performance evaluation of an on-line neural network control scheme. Presented at international conference on control, 1994. [10] Powell, M.J.D., Restart Procedures for the Conjjugate Gradiant Method, Mathematical Programming, vol. 12, pp. 241-254,1977. [11] Broomhead, D.S, Lowe, D., Multivariable Functional Interpolation and Adaptive Networks, Complex Systems, vol. 2,pp. 321-355, 1988. [12] J. Park, J. W. Sandberg, Universal approximation using radial basis functions network, Neural Computation, vol. 3, pp. 246-256, 1991.

[2]

1 2