PID CONTROLLER DESIGN FOR CONSTRAINED OPTIMAL CONTROL OF INVENTORY LOOP

Moonyong Lee, Yeungnam University, Kyongsan, Korea Joonho Shin, JongKu Lee, LG Chemical, Daejun, Korea

Introduction
Liquid level loops are commonly encountered in process industries. Since the desired production rates and inventories are achieved through the proper control of flows and levels, level control is quite important for the successful operation of most chemical plants. The industrial importance of level loops has led to extensive research interest to achieve the enhanced control performance of the level loop1~8. In a level control loop, the manipulated flow is often situated upstream of a critical unit. In this case, controlling the behavior of the outlet flow is as important as that of the liquid level itself, in order to avoid rapid variations with a significant magnitude. Therefore, the level controller is required to provide non-aggressive and smooth control action as well minimizing the deviation of the level. Furthermore, a level loop normally has two important requirements : (1) the rate of change of the outlet flow should be kept below a specified allowable limit, (2) the deviation of the level should also be within a specified allowable limit. For this reason, level control problems can be considered as a typical constrained optimal control problem. In spite of its industrial and economic importance, the constrained optimal control strategy has rarely been employed in practical level loops. One of the main reasons for this is that most industrial level loops use simple P-only or PI controllers, which are generally accepted as being too simple to implement any sophisticated control strategy. The main obstacle to achieving optimal control is that it normally requires an optimization package, which most practitioners are not so familiar with, to find the optimal control action. Using a sophisticated advanced controller such as a model predictive controller might be a solution for the constraint control of the level loop, but it cannot be considered to be a practical approach for many reasons. In this study, we developed an analytical design method for an PI controller that enables the constrained optimal control of the liquid level loop. The constrained optimal level control problem is firstly formulated and then converted into a simple form with two independent variables by using a proper variable transformation. The Lagrangian multiplier method is applied to handle the constraint optimization problem and the optimal PI tuning rule is finally found from the analysis of the global optimum condition. The proposed method is shown to deal with the two major control specifications in level loops explicitly, while minimizing the optimal control performance measure.

Constrained Optimal Control of Inventory Loop

The liquid level control system presented in Figure 1 is simply described as 1 1 H (s) = Qi ( s ) Qo ( s ) As As Q 1 where Qo ( s ) = K L (1 + )( H ( s ) H set ( s )) and K L = K c o max H Is The closed-loop transfer functions for the level control system are then s I s +1 H (s)= H I Qi ( s ) + H set ( s ) 2 A H I s + I s + 1 H I s 2 + I s + 1 (1)

(2)

 I s +1 K s ( s + 1) set Qi ( s ) L H 2 I H (s) 2 H I s + I s + 1 H I s + I s + 1 A V ( H ) A Where H = and V = =

Qo ( s ) =

(3)

K L Kc Qo max The damping factor of the above closed-loop characteristic equation is expressed as 1 I = 2 H

(4)

Figure. 1. Schematic of a level control loop featuring manipulation of the outlet stream Regulatory control against a load change is the major concern in level control. The main disturbance in the level loop is the variation of the inlet flow. The control objective is to minimize both the rate of change of the outlet flow and the deviation of the level, while keeping (1) the rate of change of the outlet flow within a specified allowable limit and (2) the deviation of the level within a specified allowable limit for a given load variation. Therefore, the constrained optimal control problem of the level loop can be defined as finding the controller parameters that minimize the performance measure in Equation (5a) for a given step change in Qi , subject to the constraints in Equations (5b~c):

min =

Q ' (t ) H (t ) [ ] dt + (1 ) [ o ] dt 0 Q' H omax

(5a) (5b,c)

subject to Q 'o (t ) Q 'omax and H (t ) H max

By performing certain mathematical manipulations, the constrained optimal control problem can be expressed in terms of H and as follows :

1 1 1 + 2 H 4 subject to H L h( ) and H U g ( ) where 2 2 AH max Qi Qi 1 Qi , L = , = 2 , U = , = Qi Q 'o max 2 Q 'o max AH min ( H , ) = H 3 2 +

(6a)

(6b,c)

 3 tan 1 x 1 + x2 exp h( ) = 2 x

for 0 < < 0.5 ;

h( ) = 1

for 0.5

g ( ) =
where

tan 1 x exp x 1 + x2 2 1 2

for 0 < < 1 ; g ( ) == x=

tanh 1 x exp x 1 x2 2

for 1 >

x=

for 0 < < 1 ;

2 1 for > 1

It is preferable for a level controller to allow smaller Q 'o max and H max specifications. However, since the feasible region is surrounded by the two constraints, as smaller Q 'o max and/or H max specifications are applied, the feasible region bounded by these constraints also becomes smaller and eventually disappears. For a given Q 'o max ( or H max ) specification, the tightest H max ( or Q 'o max ) specification occurs at t = 0.4040 and thus is calculated by H max Q 'o max = h( t ) g ( t )( Qi2 / A) = 0.5206(Qi2 / A) (7)

Figure 2. Typical contours and constraints with four possible cases of optimum location

Figure 2 shows several possible instances of a global optimum with the contours of the objective function and the constraints. Four cases are possible with respect to the global optimum location: (1) the global optimum is in the interior of the constraint set; (2) the global optimum is on the constraint

H = L h( ) ; (3) the global optimum is on the constraint H =

U ; (4) the global optimum is located g ( )

on the right vertex point formed by the two constraints. Based on these characteristics of the optimum point location, the global optimums of ( , H ) are found as follows if the feasible region is bounded (i.e., L U ) :
Case A . The extreme point ( , H ) is in the interior of the constraint set and the global optimum is :

1 4 = ; = (8) 2 * Case B : The global optimum is located on the constraint H = L h( ) and denoted by ( * , H ):
H
* H = L h( * ) ; * =

H 1 * 2 L h( )

(9)

Case C : The global optimum is located on the constraint H =

** H =

U ** and denoted by ( ** , H ) : g ( )
(10a)

U g ( ** )

and ** can be calculated by a simple root-finding method from the following equation.

g ( ** )5 1 3 **2 ** 4 2 g ( ) 3 ( ) 1 g g '( ** ) = 0 + U **3 **2 2 U U 4

3 U ** **

(10b)

Case D. The global optimum denoted by ( , H ) is located on the vertex point formed by

H =

U and H = L h( ) . The value of at the global optimum point can be found by a simple g ( )
U L
(11)

root finding method from g ( )h( ) =

The value of H at the global optimum point is

H = L h ( )

(12)

U , then ( , H ) is the global optimum. g ( ) * * * (ii) If H < L h( ) , then calculate ( * , H ) and ( , H ) . If , then ( , H ) is the global optimum ; If , then ( , H ) is the global optimum.
If L h( ) H

In summary, the procedure for finding the global optimum is as follows : (i) Calculate the unconstrained extreme point ( , H ).

U ** ** , then calculate ( ** , H , then ( ** , H ) and ( , ) is the global H ) . If g ( ) optimum ; If , then ( , H ) is the global optimum.
(iii) If H >

If the feasible region is unbounded, no vertex point formed by the two constraints exists for > 0.4040 , and accordingly case D does not exist. Also, the inequality conditions and do not need to be evaluated for cases B and C, respectively. Once the global optimum is obtained in terms of and H , the corresponding optimal PI parameters can be directly calculated as ( H ) A ; I = 4 2 H (13) KC = Qo max H

Illustrative Example
The liquid level of a drum with a cross section area of 1 m 2 and a working volume AH of 2 m3 is controlled by a PI controller. The maximum outlet flow rate Qo max is 4 m3 /min. The initial steady-state level is 50% and the nominal flow rates of the inlet and outlet are both 1 m3 /min. The maximum expected change of the inlet flow rate Qi is 1 m3 /min. The weighting factor for optimal control is set to = 0.8 . Consider the following four cases : Case 1 ( Q 'o max is 3 m3 /min2 and H max is 1.0 m) ; Case 2 ( Q 'o max is 1.2 m3 /min2 and H max is 1.0 m) ; Case 3 ( Q 'o max is 4 m3 /min2 and H max is 0.2 m) ; Case 4 ( Q 'o max is 1.5 m3 /min2 and H max is 0.4 m)

Figure 3. Responses of level and rate of change of outlet flow for cases 1,2,3, and 4

Figure 3 compares the responses for the liquid level and the rate of change of the outlet flow rate for each case. A step change of 1 m3 /min is made in the inlet flow at t=0 in the simulation. As seen in the figure, the PI controllers designed by the proposed method give the optimal responses while strictly satisfying the given H max and Q 'o max specifications.

Acknowledgment
This research was supported by Yeungnam University research grants in 2007.

References
1. 2. 3. 4. 5. 6. 7. 8. Buckley, P. (1983). Recent Advances in averaging level control, in: Productivity through Control Technology, 18-21, Houston. Cheung, T. and Luyben, W. (1979). Liquid-level control in single tanks and cascades of tanks with P-only and PI feedback controllers, Ind. Eng. Chem. Fund., 1, 15-21. Dennis, J.B. (1959). Mathematical Programs and Electron Networks, Jhon Wiley & Sons, New York. Luyben, W. and Buckley, P.S. (1977). A proportional-lag controller, Instrum. Tech., 24, 65-68. MacDonald, K., McAvoy, T., and Tits, A. (1986). Optimal averaging level control, AIChE J., 32, 75-86. Rivera, D.E., Morari, M., and Skogestad, S. (1986). Internal model control. 4. PID controller design, Ind. Eng. Chem. Process Des. Dev., 25, 252-265. Seki, H. and Ogawa, M. (1998). Japan Patent # 2811041. Wu, K., Yu, C., and Cheung, Y. (2001). A two degree of freedom level control, J. of Process Control, 11 311-319.