You are on page 1of 7

ARTICLE IN PRESS

Control Engineering Practice 16 (2008) 1391– 1397

Contents lists available at ScienceDirect

Control Engineering Practice


journal homepage: www.elsevier.com/locate/conengprac

Analytical design of a proportional-integral controller for constrained optimal


regulatory control of inventory loop
Joonho Shin a, Jongku Lee a, Seungyoung Park a, Kee-Kahb Koo b, Moonyong Lee c,
a
Corporate R&D, LG Chem, Moonji-dong, Yuseong-gu, Taejon 305-701, Republic of Korea
b
Department of Chemical and Biomolecular Engineering, Sogang University, Seoul 121-742, Republic of Korea
c
School of Chemical Engineering and Technology, Yeungnam University, Kyongsan, Kyongbuk 214-1, Republic of Korea

a r t i c l e in fo abstract

Article history: An analytical design method for a simple proportional-integral (PI) controller is developed for the
Received 31 October 2007 optimal control of a constrained inventory loop. The proposed method explicitly deals with the
Accepted 21 April 2008 important constraints in the inventory loop, such as the maximum allowable rate of change in the
Available online 10 June 2008
manipulated variable, the maximum allowable decay ratio and damping coefficient in the output
Keywords: response, as well as minimizing the optimal control specification. The simple and explicit form of the
Inventory control resulting tuning rule is clearly advantageous to practitioners.
PI (proportional-integral) controller tuning & 2008 Elsevier Ltd. All rights reserved.
Optimal control
Constraint control
Liquid level control

1. Introduction action for large errors and slow action for small errors in the liquid
level loops. MacDonald, McAvoy, and Tits (1986) proposed an
Inventory control loops are commonly encountered in the interesting method of deriving an averaging level control algorithm
process industry. Typical examples include accumulator and to minimize the maximum rate of change of the manipulated flow.
bottom level control in a distillation column and the inventory In practice, the operation of an inventory control system should
control of a tank (Yang, Seborg, & Mellichamp, 1994). Inventory be located somewhere between the two extreme situations: the
control is extremely important for the successful operation of first, referred to as tight inventory control, is where the level is very
most chemical plants, because it is through the proper control of important but any variation in the manipulated flow is not of great
the flows and levels that the desired production rates and importance; the second, referred to as averaging inventory control,
inventories are achieved (Marlin, 1995). occurs when some variation in the level is acceptable as long as the
Many studies have been conducted in an attempt to enhance the value remains within specified limits, but the manipulated flow
control performance of the inventory loop. Cheung and Luyben (1979) should not experience rapid variations of a significant magnitude.
studied the liquid level control system with P-only and PI feedback Thus, the control objective of inventory loops should consider
controllers. They proposed a procedure with a design chart for the variations not only in the controlled variable but also in the
tuning of a PI controller in response to a step change of the inlet flow manipulated variable. Furthermore, inventory loops often have
rate. However, it is quite complicated to determine the tuning several important constraints associated with both the controlled
parameters of a PI controller. Proportional-lag control (Luyben & and manipulated variables. This feature of the inventory loop often
Buckley, 1977) is a potentially good solution for liquid level control necessitates an optimal control strategy with constraint handling.
systems with feedforward compensation, but such feedforward However, since most inventory loops make use of a simple PI
control schemes (Luyben & Buckley, 1977; Wu, Yu, & Cheung, 2001) controller, constrained optimal control is rarely implemented in the
require an additional measurement which may be unavailable. Rivera, inventory loops. In this study, an analytical design method for PI
Morari, and Skogestad (1986) proposed the P-only controller using the controllers is developed for optimal regulatory control with
internal model control (IMC) principle for the critically damped explicitly handling the major specifications in the inventory loop.
closed-loop response of a liquid level control system. Buckley (1983)
discussed several nonlinear PI controllers to provide fast control
2. Liquid level control dynamics

 Corresponding author. Tel.: +82 53 810 2512; fax: +82 53 811 3262. The liquid level control system presented in Fig. 1 is described
E-mail address: mynlee@yu.ac.kr (M. Lee). by the following differential equation with the nomenclature of

0967-0661/$ - see front matter & 2008 Elsevier Ltd. All rights reserved.
doi:10.1016/j.conengprac.2008.04.006
ARTICLE IN PRESS

1392 J. Shin et al. / Control Engineering Practice 16 (2008) 1391–1397

Nomenclature tI integral time constant


tH A/KL
Roman letters
tHmin minimum allowable tH
tv hold-up time of tank ðDHÞA=Q o max
A cross-sectional area of tank o weighting factor, 0ooo1
H liquid level deviation F objective function or performance measure for opti-
DH level transmitter span mal control
KL proportional gain
Kc dimensionless proportional gain
Superscripts
Qi inlet flow rate
Qo outlet flow rate
set set-point
Qomax maximum outlet flow rate through the control valve
y extreme point of the objective function
DQi magnitude of inlet flow rate change
*and** optimum point on the constraint z ¼ zmin and
Q0 o rate of change of the outlet flow rate
z ¼ zmax, respectively
Q0 omax maximum allowable rate of change of the outlet flow
rate
t time Acronyms

Greek letters P proportional-only control


PI proportional-integral control
PL proportional-lag control
z damping factor
DR decay ratio
zmin minimum allowable damping factor
DRmax maximum allowable decay ratio
zmax maximum allowable damping factor

Cheung and Luyben (1979): where


dH Q o max
A ¼ Qi  Qo (1) KL ¼ Kc . (4)
dt DH
where A is the constant cross-sectional area of the tank. The liquid
level is controlled by adjusting the outlet flow rate and is Note that the proportional term acts on the process value
disturbed by the inlet flow rate. itself and the integral term acts on the error. This type of a
The Laplace transform of (1) gives modified PI controller is available in most industrial DCS systems,
such as the PID controller with the type-C equation used in the
1 1 HoneywellTM TDC system. For the modified PI controller, the
HðsÞ ¼ Q ðsÞ  Q o ðsÞ (2)
As i As following closed-loop transfer functions are obtained for the level
The liquid level system is a well-known integrating process and control system:
a P-only controller leads to no steady-state offset in the case of a  
servo problem. However, a PI controller is normally used in liquid tH tI s
HðsÞ ¼ Q i ðsÞ
level loops, since a P-only controller gives a steady-state offset in A tH tI s 2 þ tI s þ 1
 
the case of a regulatory problem. In particular, the modified PI 1
þ Hset ðsÞ (5)
controller in (3) is widely accepted in process industries to avoid tH tI s2 þ tI s þ 1
the proportional kick for a step set-point change.
KL
Q o ðsÞ ¼ K L HðsÞ þ ðHðsÞ  Hset ðsÞÞ (3)  
tI s tI s þ 1
Q o ðsÞ ¼ 2
Q i ðsÞ
tH tI s þ tI s þ 1
 
K L tH s
 Hset ðsÞ (6)
tH tI s2 þ tI s þ 1

where

A tV
tH ¼ ¼ (7)
KL Kc

and

ðDHÞA
tV ¼ (8)
Q o max

The closed-loop characteristic equation is

ðtI tH Þs2 þ tI s þ 1 ¼ 0 (9)

Fig. 1. Schematic representation of a level control loop featuring manipulation of The tuning parameters Kc and tI determine the location
the outlet stream. of the two poles along with the damping factor that is expressed
ARTICLE IN PRESS

J. Shin et al. / Control Engineering Practice 16 (2008) 1391–1397 1393

as follows: Multiplying (14) by tH/z gives


sffiffiffiffiffiffiffiffiffi   
rffiffiffiffiffi 1o ADH 2
1 tI 1 tI K c ðtH zÞ4 ¼ (15)
z¼ ¼ (10) 16o Q 0o max
2 tH 2 tV
The optimal tH for the unconstrained case can then be obtained
by substituting (15) into (13) and rearranging to give
rffiffiffiffiffiffiffiffiffiffiffiffiffi
3. Formulation of optimal regulatory control ADH 1o
ty2
H ¼ (16)
2Q 0o max o
Regulatory control in response to load changes is the major where the superscript y denotes the optimum in the uncon-
concern in inventory control systems. The PI controller for a liquid strained case. The optimal z can also be obtained as
level loop is often required to have a smoother, less aggressive control 1
action, even at the expense of less error minimization, which means zy ¼ pffiffiffi (17)
2
that the performance measure of the control system must include
minimizing not only the error in the controlled variable but also the It is noted that the optimal value of the damping coefficient
rate of change of the manipulated variable. At the same time, the is equal to p1ffiffi2 and is independent of the process dynamics
controller should be designed to meet all or some of the following and weighting factor. The optimal tuning values of K yc and tIy
typical specifications or constraints in the level loop: (i) the rate of for the unconstrained case can be simply calculated using (7)
change of the outlet flow should be under a maximum allowable and (10).
limit, (ii) the decay ratio in the response should be under a maximum The following simple relation can also be derived for the
allowable limit to avoid a severe oscillatory response, and (iii) the optimal tuning of the unconstrained case:
damping coefficient should also be less than a maximum allowable K yc tyI ¼ 2tV . (18)
limit in order to secure a suppression speed required.
Based on the operational goal and the three constraints listed Note that product of and Kcy tIy
is proportional to the hold-up
above, the optimal design problem of the PI controller in the time of a level tank. Seki and Ogawa (1998) also came to the same
inventory loop can be defined as finding the controller parameters conclusion as that described by (17) and (18) for the optimal
that minimize the performance measure in (11-1), subject to the control of a level loop.
constraints in (11-2)–(11-4)
Remark 1. For the optimal tuning, the product of Kcy and
1
Z  Z 1 0 2 tIy should be kept constant by doubling the hold-up time,
HðtÞ 2 Q o ðtÞ
minF ¼ o dt þ ð1  oÞ 0 dt (11-1) regardless of the weighting factor. When one of the PI para-
0 DH 0 Q o max
meters, K yc and tIy, is changed during tuning, the other para-
subject to meter should also be adjusted so that the product remains
 0  constant.
Q ðtÞpQ 0 (11-2)
o o max

DRpDRmax (11-3) 3.2. Optimal solution for constrained case

zpzmax . (11-4) When the constraints given in (11-2)–(11-4) are considered in


The first and second terms in (11-1) describe the normalized the controller design, the global optimum can be located either on
deviation of the liquid level and the normalized rate of change of the extreme point of the objective function or on the constraint
the outlet flow, respectively. boundary. Note that an extreme point denotes an extremum in the
absence of constraints. To deal with the constrained cases, all of
3.1. Optimal solution for unconstrained case the constraints have to be expressed in terms of the independent
variables, tH and z. The rate of change of the outlet flow rate for a
step disturbance can be determined by using the properties of the
Throughout this study, the regulatory problem to a step change
Laplace transform of the derivative of the outlet flow, which is
in the inlet flow rate (i.e., Qi(s) ¼ DQi/s) is considered. By
given by
performing certain mathematical manipulations, the objective
 
function F given in (11-1) can be expressed in terms of tH and z as dQ o ðtÞ
L ¼ sQ o ðsÞ  Q o ðtÞt¼0 (19)
follows (see Appendix A for details): dt
   
DQ i 2 3 2 DQ i 2 where Qo(t)t ¼ 0 ¼ 0.
FðtH ; zÞ ¼ 2o tH z þ ð1  oÞ 0 Throughout this study, it is assumed that the controller is
ADH Q o max
  designed to give a response with zX0.5 in order to avoid a severe
1 1
 1þ 2 (12) oscillatory response. In this case, since the largest value of the rate
2tH 4z
of change occurs at t ¼ 0, by applying the initial value theorem to
Thus, the unconstrained optimality conditions, i.e., the extre- (6), it is easily determined that
mum in the absence of constraints, can be found by solving the  
following equations simultaneously: dQ o ðtÞ DQ i
¼ (20)
     dt t¼0 tH
qF DQ i 2 2 2 1o DQ i 2
¼ 6o tH z  Therefore, the constraint given in (11-2) can be expressed in
qtH ADH 2 Q 0o max
!  terms of tH:
1 1
 2 1þ 2 ¼0 (13) DQ i
tH 4z tH X  tH min (21)
Q 0o max

       Furthermore, using the relation between the decay ratio and the
qF DQ i 2 3 1o DQ i 2 1 damping coefficient in the second-order process, the constraint
¼ 4o tH z  0 3
¼0 (14)
qz ADH 4 Q o max tH z imposed by (11-3) can be converted to the following inequality
ARTICLE IN PRESS

1394 J. Shin et al. / Control Engineering Practice 16 (2008) 1391–1397

condition in terms of z: Similarly, the local optimum value of tH on the boundary of


z ¼ zmax can be obtained as
1
zX qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi  zmin (22) !1=4   sffiffiffiffiffiffiffiffiffiffiffiffiffiffi !1=4
1 þ ð2p= ln DRmax Þ2 4z2max þ 1 1  o 1=4 ADH 4z2max þ 1

tH ¼ ¼ tyH
12z4max 4o Q 0o max 12z4max
Note that since all of the constraints are linear, the solution (25)
region is convex. The feasibility of the solution region should be
firstly checked for a given constraint set, which can be easily done It is clear from (23) that the local optimum z* is always less than
from (11-4) and (22). p1ffiffi (or zy) when tHyotHmin. Furthermore, since g(z) ¼ (4z2+1/
2
Once the constraints applied are confirmed as giving a feasible 12z p4 1/4
) ffiffiffi is a monotonically decreasing function in z with
solution region, the next step is to determine the global optimal gð1= 2Þ ¼ 1, tH* is less than tHy for zyozmin and tH** is greater
condition. The global optimum can exist either on the extreme than tHy for zy4zmax. These relationships indicate that the
point of the objective function or on the boundary of a constraint. contours of the objective function are left skewed in the ztH
The local optimum value of z on the boundary of tH ¼ tHmin can be plane.
calculated from (14) by replacing tH with tHmin: Based on the contour characteristics, the condition and
   sffiffiffiffiffiffiffiffiffiffiffiffiffiffi location of the global optimums of (z,tH) for every possible
1  o 1=4
y
1 1 ADH 1 t case are found as listed in Table 1. Fig. 2 shows five possible

z ¼ pffiffiffi 0 ¼ pffiffiffi H (23)
2 tH min 4o Q o max 2 tH min cases of a global optimum with the contours of the objec-
tive function and the constraints imposed by (11-2)–(11-4). Once
The local optimum value of tH on the boundary of z ¼ zmin can
the global optimum (zopt,tHopt) is obtained, the corresponding
be found from (13) by replacing z with zmin:
optimal PI parameters can be simply calculated from (7), (8),
!1=4   sffiffiffiffiffiffiffiffiffiffiffiffiffiffi !1=4 and (10) as
4z2min þ 1 1  o 1=4 ADH 4z2min þ 1
tH ¼ ¼ tyH
12z4min 4o Q 0o max 12z4min ðDHÞA
K opt
C ¼ (26)
(24) Q o max topt
H

Table 1
Global optimums of (x,tH) for the constrained case

Case Condition Location Solution

A (tHyXtH min and zmaxXzyXzmin) At the extremum (zy tHy) (zy tHy)
B (tHyotH min and zmaxXz*Xzmin) or (zy4zmax and tH**otH min) On the constraint tH ¼ tH min (z*,tH min)
C (zy4zmin and tH*XtH min) On the constraint z ¼ zmin (zmin,tH*)
D (zy4zmax and tH**XtH min) or (tHyotH min and z*4zmax) On the constraint z ¼ zmax (zmax,tH**)
E (tHyotH min and z*ozmin) or (zyozmin and tH*otH min) On the vertex by tH ¼ tH min and z ¼ zmin (zmin,tH min)

CASE A
ζ = ζmin ζ = ζmax
τH

optimum
τH = τHmin

CASE B CASE C
ζ = ζmin ζ = ζmax ζ = ζmin
τH

τH

optimum τH = τHmin
optimum
τH = τHmin

ζ ζ

CASE D CASE E
ζ = ζmax ζ = ζmin
optimum τH = τHmin
τH
τH

τH = τHmin
optimum

ζ ζ

Fig. 2. Possible cases of a global optimum location: the shaded region denotes a feasible region.
ARTICLE IN PRESS

J. Shin et al. / Control Engineering Practice 16 (2008) 1391–1397 1395

topt
I ¼ 4ðzopt Þ2 topt
H (27) method. In order to provide a fair comparison, the closed loop
time constant in the IMC-PI method is adjusted so that both
controllers yield the same maximum peak level. As shown in
4. Illustrative examples Fig. 4, the PI controller using the proposed method gives a smaller
maximum rate of change of the outlet flow, as well as a faster
settling time in the level response. The value of the performance
Consider a liquid level system as follows: the liquid level of a
measure in (11-1) was also evaluated for each method, and the
tank with a cross-section area of 1 m2 and a working volume
proposed PI controller gave a smaller value of 0.2988 than that
(ADH) of 2 m3 is controlled by a PI controller. The maximum outlet
given by the IMC-PI tuning method of 0.3580.
flow (Qomax) 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 in the inlet flow (DQi) is 1 m3/min. Example 2. Optimal tuning: constrained case.
The maximum allowable rate of change of the outlet flow (Q0 omax)
is 1.5 m3/min2. When the control specifications given by (11-2)–(11-4) have to
Example 1. Optimal tuning: unconstrained case. be strictly satisfied, the optimal tuning values can be obtained by
categorizing the global optimum case from Table 1. Suppose that
Suppose that there is no hard control specification. Optimal the weighting factor is set to o ¼ 0.8 for the liquid level system
tuning can then bepcalculated
ffiffiffi based
pffiffiffion (16) and (17). For example, above. As an illustrative example, consider cases I, II, and III,
if o ¼ 0.5, tyH ¼ 1= 3 and zy ¼ 1= 2 are obtained. The optimal
pffiffiffi PI where the maximum allowable decay ratios are 0.0005, 0.1, and
parameters
pffiffiffi are calculated using (26) and (27) as K c ¼ 3 =2 and 0.1 and the maximum allowable damping coefficients are 1.0, 1.0,
tI ¼ 2= 3 min. and 0.4, respectively, while the maximum allowable rate of
The responses in the level and outlet flow rate for the proposed change of the outlet flow is 1.5 m3/min2 in all three cases. From
optimal tuning with various weighting factors are shown in Fig. 3. (21), tH min ¼ 0:6667 is obtained. The minimum allowable damp-
In the simulation, a step change of 1 m3/min in the inlet flow rate ing coefficients corresponding to the maximum decay ratios,
is introduced at 5 min and sequentially the level set-point calculated using (22), are 0.7708 for case I and 0.3441 for cases II
undergoes a 25% step increase at 20 min. As seen in the figure, and III. Therefore, cases I, II, and III correspond to cases E, B, and D
the lower the weighting factor, o, the more slowly the outlet flow in Table 1 and Fig. 2, respectively. The optimal PI parameters for
changes, but the higher the peak level becomes and the more the three cases are Kc ¼ 0.75, 0.75, and 0.5697 and tI ¼ 1.584, 1.0,
sluggishly the level is controlled. As o is increased, a smaller peak and 0.562, respectively.
can be obtained at the cost of a higher rate of change in the outlet Fig. 5 compares the responses for the level and rate of change
flow rate. In this manner, a clear tradeoff can be achieved between of the outlet flow for the three cases. In the simulation, a step
the tightness in the level control and the smoothness in the outlet change of 1 m3/min in the inlet flow rate is introduced at 1 min.
flow change with only the single tuning parameter, o. The global optimum is located at the vertex point by the two
To confirm the advantage of the proposed method, the closed constraints z ¼ zmin and tH ¼ tH min for case I, on the constraint
loop performance provided by the proposed PI controller is tH ¼ tH min for case II, and on the constraint z ¼ zmax for case III. As
compared with that afforded by the IMC-PI tuning method (Rivera seen in the figure, all of the responses strictly satisfy the given
et al., 1986). The weighting factor o is set to 0.5 in the proposed control specifications.

100
90
80
Level (%)

70
w = 0.2
60 w = 0.5
50 w = 0.7
w = 0.9
40
0 5 10 15 20 25 30 35
Time (min)
1.4
Outlet flow rate (m3/min)

1.2
1
0.8
0.6 w = 0.2
0.4 w = 0.5
w = 0.7
0.2 w = 0.9
0
0 5 10 15 20 25 30 35
Time (min)

Fig. 3. Level and outlet flow responses using the proposed PI controller (unconstrained case).
ARTICLE IN PRESS

1396 J. Shin et al. / Control Engineering Practice 16 (2008) 1391–1397

80
proposed
70 IMC-PI

Level (%)
60

50

40
0 5 10 15 20 25
Time (min)
1.5
proposed
dQo/dt (m3/min2)

1 IMC-PI

0.5

-0.5
0 5 10 15 20 25
Time (min)

Fig. 4. Comparison of responses by the proposed method and the IMC-PI tuning method (unconstrained case, w ¼ 0.5).

80
case I
70 case II
Level (%)

case III
60

50

40
0 5 10 15
Time (min)

2
case I
dQo/dt (m3/min2)

1.5 case II
1 case III

0.5

-0.5
0 5 10 15
Time (min)

Fig. 5. Level and rate of change of outlet flow responses using the proposed PI controller (constrained case): case I (DRmax ¼ 0.0005, zmax ¼ 1.0), case II (DRmax ¼ 0.1,
zmax ¼ 1.0), case III (DRmax ¼ 0.1, zmax ¼ 0.4).

Example 3. Robustness against modeling error in dead time and figure indicate that the realistic uncertainties in the dead time and
area. area have little effect on the control performance.

Level processes generally do not have a dead time except for


certain rare occasions such as level loops in which the control 5. Conclusions
valve has a dead zone. Therefore, the dead time in level processes
cannot be too large and has to be maintained at almost 10% of the An analytical design method for the optimal control of a liquid
hold-up time (Wu et al., 2001). level loop is developed by solving a constrained optimization
Fig. 6 shows the results assuming that the level process in problem. One of the main drawbacks of simple PID controllers is
Example 1 has a dead time of 20% of the hold-up time. The that they cannot handle the constraints explicitly. The proposed
weighting factor is set to 0.5. The issue of robustness is also design method explicitly deals with the important constraints in
studied by examining 720% errors in the cross-sectional area and the inventory loop, as well as minimizing the optimal control
the results are presented in Fig. 6. The responses shown in the specification. Several examples were presented to illustrate the
ARTICLE IN PRESS

J. Shin et al. / Control Engineering Practice 16 (2008) 1391–1397 1397

80 Therefore, the integral square error of the normalized liquid


level in (11-1) becomes
Nominal
75 +20% Error in Dead Time Z      2   
+20% Error in Area
1
HðtÞ 2 DQ i 2 1 2 1 1 1
o dt ¼ o  þ
70 –20% Error in Area 0 DH ADH r1  r2 r1 þ r2 2 r1 r2
 2
DQ i
65 ¼ 2o t3H z2 (A5)
ADH
Level (%)

60 Now, consider the integral square error of the normalized rate of


change of the outlet flow rate in (11-1).
55 From (6), with Qi(s) ¼ DQi/s and Hset(s) ¼ 0, the Laplace trans-
form of the outlet flow rate is
50
 
tI s þ 1 DQ i
Q o ðsÞ ¼ (A6)
45 tH tI s2 þ tI s þ 1 s

40 Therefore, using the property of the Laplace transform for a


0 5 10 15 derivative, i.e., Q0 o(s)L[dQo(t)/dt] ¼ sQo(s), the rate of change of
Time (min) the outlet flow rate is obtained by
 2 r2 t 
Fig. 6. Effect of modeling error in dead time and area on the control performance. r 2 e  r 21 er1 t
Q 0o ðtÞ ¼ DQ i for r 1 ar 2 (A7)
r1  r2

Therefore, the integral square error of the normalized rate of


change of the outlet flow rate becomes
effectiveness of the proposed design method. The results also
showed that the proposed method gives satisfactory responses, Z  2  2  2
1
Q 0o ðtÞ DQ i 1
not only in the nominal condition but also under uncertainties in ð1  oÞ dt ¼ ð1  oÞ
0 Q 0o max Q 0o max r1  r2
the dead time and cross-sectional area.  3 
r r3 2r 21 r 22
  2 1þ
2 2 r1 þ r2
   2  
Acknowledgment DQ i 2 1 1
¼ ð1  oÞ 0 
Q o max r1  r2 2
This research was supported by the Yeungnam University  2 2 
4r 1 r2
research grants in 2007.  ðr 1 þ r 2 Þ3  3r 1 r 2 ðr 1 þ r 2 Þ 
r1 þ r2
   
DQ i 2 1 1
¼ ð1  oÞ 0 1þ 2 (A8)
Appendix A. Derivation of the objective function U in (12) Q o max 2tH 4z

From (5) with Qi(s) ¼ DQi/s and Hset(s) ¼ 0, the liquid level
response is obtained as
  References
DQ i er1 t  er2 t
HðtÞ ¼ for r 1 ar 2 (A1)
A r1  r2
Buckley, P. (1983). Recent advances in averaging level control. In Productivity
where r1 and r2 are the roots of the characteristic equation s2+(1/ through control technology (pp. 18–21), Houston.
Cheung, T., & Luyben, W. (1979). Liquid-level control in single tanks and cascades
tH)s+(1/tItH) ¼ 0. of tanks with P-only and PI feedback controllers. Industrial and Engineering
Thus, Chemistry Fundamentals, 18(1), 15–21.
Luyben, W., & Buckley, P. S. (1977). A proportional-lag controller. Instrumentation
1 Technology, 24(12), 65–68.
r1 r2 ¼ (A2)
tI tH MacDonald, K., McAvoy, T., & Tits, A. (1986). Optimal averaging level control. AIChE
Journal, 32, 75–86.
Marlin, T. E. (1995). Process control. Mcgraw-Hill 581.
Rivera, D. E., Morari, M., & Skogestad, S. (1986). Internal model control, 4: PID
1 controller design. Industrial & Engineering Chemistry Process Design and
r1 þ r2 ¼  (A3)
tH Development, 25(1), 252–265.
Seki, H., & Ogawa, M. (1998). Japan Patent # 2811041.
Wu, K., Yu, C., & Cheung, Y. (2001). A two degree of freedom level control. Journal of
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi sffiffiffiffiffiffiffiffiffiffiffiffiffiffi Process Control, 11, 311–319.
1 4tH 1 z2  1 Yang, D. R., Seborg, D. E., & Mellichamp, D. A. (1994). The influence of inventory
r1  r2 ¼ 1 ¼ (A4) control dynamics on distillation composition control. Control Engineering
tH tI tH z2 Practice, 2(6), 27–32.

You might also like