Documente Academic
Documente Profesional
Documente Cultură
Level in a Gravity-Drained
Submitted: 03 May 2017
Accepted: 17 July 2017
Published: 20 July 2017
Abstract Keywords
• Liquid level
Dynamic models for liquid level in a four-tank system are derived in this work by • Gravity-drained tank
applying the principle of analogy to a single-tank case. In this system, there are two • Decoupling
top tanks and two bottom tanks. Each of the two top tanks receives liquid from a feed • MIMO
stream, while discharging liquid to the two bottom tanks by gravity. Each of the two • IMC
bottom tanks receives liquid from the two top tanks and discharges liquid by gravity
from the bottom of the tank. The process models and the disturbance models for the
levels of the two bottom tanks showing the effects of both feed streams are derived.
Relative gain array (RGA) based on the results of simulation from Loop-Pro’s multi-tank
process is used to predict the extent of loop interaction (or coupling). Feedback PID
control parameters are obtained by using internal model control (IMC) tuning rule. The
performance of the multiple-input/multiple-output (MIMO) feedback control system
with and without decoupling strategy is compared and analyzed.
INTRODUCTION liquid levels of the two bottom tanks, it is important to identify the
Liquid level control of all aspects remains one of the extent of loop interactions (i.e., coupling effect) and implement
most important case studies due to its widespread industrial proper strategies to eliminate potential loop interactions (i.e.,
applications. Mathematical models for the dynamic responses decoupling). This paper will outline the procedure of tuning
of liquid level are more easily perceived due to its simplicity individual feedback controllers as well as improving the
in physical setup. In the literature, open-loop and closed-loop controller performance by implementing decoupling strategy.
dynamic models as well as tuning rules are well developed for
single-tank systems [1,2][3(a)]. In this paper, the process model
for an open tank with liquid fed to the top and drained by gravity
from the bottom via a hole or valve of fixed opening is reviewed
(Figure 1). The transfer functions showing the effects of feed
rate on the liquid level and the draining rate are derived. This
system is then expanded to one that contains two top tanks and
two bottom tanks. Each of the two top tanks receives liquid from
one feed stream and discharges liquid to the two bottom tanks
by gravity via two valves with fixed openings. Each of the two
bottom tanks has two feed streams, one directly from the tank
above, and the other from the other top tank. The liquid is then
discharged from each of the two bottom tanks via one valve with
fixed opening (Figure 2). It is of interest to find the effects of the
two feed streams on the liquid levels in all four tanks as well as
the draining rates of the six streams leaving the four tanks. In this
work, dynamic models for the liquid levels in all four tanks are
derived based on the principle of analogy to the single-tank case.
Simulation data from a case study in Loop Pro (Control Station,
Inc.) are used to generate process models and disturbance Figure 1 Schematic diagram for a single open-tank, gravity-drained
models for the system. When a multiple-input/multiple-output system. The tank has a cross-sectional area of A and a valve with fixed
(MIMO) feedback control system is established to control the opening at the bottom.
Cite this article: Jang LK (2017) Feedback Control for Liquid Level in a Gravity-Drained Multi-Tank System. Chem Eng Process Tech 3(1): 1037.
Jang (2017)
Email:
Central
Figure 2 Schematic diagram of the feedback control loops for a four-tank system.
Central
Where Cij ‘s (i = 1,2; j = 1,2) are the discharge coefficients of τ 2,bot = A2,bot R2,bot (30)
the four valves below the two top tanks. Likewise, the gravity- where A’s are the cross-sectional areas of the four tanks. By
drained rates from the two bottom tanks are governed by their analogy to Eq. 11, we my write the transfer functions relating
liquid levels: liquid levels for the two top tanks to the two feed streams. For
simplicity, the symbol “(s)” for the Laplace domain is omitted
f1,out = C1 h1,bot (17) hereafter:
R1,top
f 2,out = C2 h2,bot (18) H1,top = F1,in (31)
τ 1,top s + 1
where Ci’s (i = 1,2) are the discharge coefficients of the two
valves below the two bottom tanks. If the linearization procedure R2,top
similar to Eq. 4 is employed, one may easily define the resistances H 2,top = F2,in (32)
of the six valves, four of which located below the two top tanks
τ 2,top s + 1
(Rij’s ) and the other two below the two bottom tanks (Ri,bot’s ): where
Central
Figure 3 Screen shot of simulation for response of liquid level to a doublet input by using Loop Pro algorithm.
Model: Second Order Overdamped Loop-Pro: Design Tools File Name: Multi Tank G11 G21.txt
Manipulated Variable Process Variable
4.1
4.0
3.9
3.8
63.0
62.1
61.2
60.3
Gain (K) = 0.07867, 1st Time Constant (min) = 10.41, 2nd Time Constant (min) = 11.41
Goodness of Fit: R-Squared = 0.9972, SSE = 0.03101
Figure 4 Result of data fitting to the exact second-order model for the process model G11.
Central
R2,top R2,top
H 2,top R12
G22 =
R2,bot R22
F12 = = F (39) (50)
R12 τ 2,top s + 1 2,in (τ 2,botS + 1) (τ 2,topS + 1)
The resultant transfer functions can be expressed as linear
R2,top
combinations showing the effects of the liquid feed rates to
F22 =
H 2,top
= R22
F (40)
the two top tanks on the liquid levels in the two bottom tanks.
According to the control schematic diagram (Figure 2), the liquid
R22 τ 2,top s + 1 2,in levels of the two bottom tanks h1,bot and h2,bot would be controlled
where by regulating f1,in and f2,in, respectively. The transfer functions
G11 and G22 are then considered the process models showing
F11 = f11 − f 11 (41) effects of F1,in on H1,bot and F2,in on and H2,bot, respectively. On the
other hand, the transfer function G12 and G21 are considered the
F12 = f12 − f 12 (42) disturbance models showing the effects of F2,in on H1,bot and F1,in on
H2,bot, respectively. Again, by analogy to Eq. 12,
F21 = f 21 − f 21 (43)
H1,bot
F1,out = (51)
F22 = f 22 − f 22 (44) R1,bot
For the two bottom tanks, each tank receives two inlet
H 2,bot
streams. The combined flow rate of the two inlet streams F2,out = (52)
to bottom tank no. 1 on the left side is (f11 + f12). Likewise, the R2,bot
combined flow rate of the two inlet streams to bottom tank no. 2 Where
on the right side is (f21 + f22). By analogy to Eq. 11, one can derive
the following transfer functions for the liquid levels of the two F1,out = f1,out − f 1,out (53)
bottom tanks by substituting Eqs. 37-44 into Eqs. 45 and 46:
F2,out = f 2,out − f 2,out (54)
H1,bot = h1,bot − h1,bot
Effect of initial steady state on model parameters
R1,bot
= ( F11 + F12 ) (45) The initial steady-state condition of the four-tank system
τ 1,bot s + 1
depends on the feed rates f 1,in and f 2,in and the discharge
= G11 F1,in + G12 F2,in coefficients of the six valves below the four tanks. Since the sum
of the two outlet flow rates equals to the inlet flow rate for each
H 2,bot = h2,bot − h 2,bot
of the two top tanks at steady state, one may calculate the steady-
R2,bot
= ( F21 + F22 ) (46) state liquid levels h1,top and h 2,top by Eqs. 13-16 and Eqs.55-58:
τ 2,bot s + 1
= G21 F1,in + G22 F2,in f 1,in = f 11 + f 21
(55)
where =C
11 h1,top + C21 h1,top
R1,top
f 2,in = f 12 + f 22
(56)
G11 =
R1,bot R11 (47)
(τ 1,botS + 1) (τ 1,topS + 1)
=C h 2,top + C22 h 2,top
12
Or,
2
f 1,in
R2,top h1,top =
C11 + C21
(57)
G12 =
R1,bot R12 2
(48) f 2,in
(τ 1,botS + 1) (τ 2,topS + 1) h 2,top = (58)
C12 + C22
R1,top Once h1,top andh 2,top are calculated, one may calculate the
four discharge flow rates f ij ( i = 1, 2; j = 1, 2 ) from the two
G21 =
R2,bot R21
(49)
(τ 2,botS + 1) (τ 1,topS + 1)
top tanks at steady state according to Eqs. 13-16. In turn, one may
further calculate the steady-state liquid levels of the two bottom
Table 1: Summary of the time constants for the simulation results of the four-tank system from Loop-Pro.
τ1,top (min) τ1,bot (min) τ2,top (min) τ2,bot (min)
11.41 10.41 11.61 12.41
Central
4.0
3.9
3.8
63.0
62.1
61.2
60.3
Gain (K) = 0.04585, 1st Time Constant (min) = 11.41, 2nd Time Constant (min) = 11.11
Goodness of Fit: R-Squared = 0.9967, SSE = 0.01273
Figure 5 Result of data fitting to the exact second-order model for the disturbance model G21.
4.0
3.9
3.8
63.0
62.1
61.2
60.3
Gain (K) = 0.04007, 1st Time Constant (min) = 11.41, 2nd Time Constant (min) = 12.61
Goodness of Fit: R-Squared = 0.9965, SSE = 0.009501
Figure 6 Result of data fitting to the exact second-order model for the disturbance
tuning model
methodG12for
. single-input-single-output (SISO) [3(b)][5]:
1 τ1 + τ 2
may conclude that the extent of loop interaction to be severe and/ K c ≡ Proportional Gain = (69)
or the process variables and the manipulated variables may be Kp τc
paired incorrectly [4(a)]. When severe loop interaction exists, we τ I ≡ Integral Time = τ 1 + τ 2 (70)
may need to implement strategies such as decoupling or revising
manipulated variable/process variable pairs [3(d)][4(b)]. ττ
τ D ≡ Derivative Time = 1 2 (71)
IMC TUNING PARAMETERS τ1 + τ 2
Therefore, by using the model parameters from Eq. 64,
If the internal model control (IMC) tuning method is used, PID tuning parameters may be calculated by Eqs. 69-71with
the PID (proportional-integral-derivative) tuning parameters for the expected closed-loop time constant for both loops chosen
both loops can be calculated by using process model parameters arbitrarily at τc = 12.0 min.
(from G11 and G22) with the expected closed-loop time constant
τc as the adjustable parameter. For a general exact second-order Loop 1
process model Kc = 23.11 % /m
Kp τ I = 21.82 min.
Gp = (68)
(τ 1 + 1)(τ 2 + 1) τD = 5.44 min.
PID tuning parameters can be determined by using the IMC Loop 2
Central
Kc = 23.689 %/m “PID with Decoupler” for both loops in the Loop Pro’s multi-tank
case study, the same PID control parameters above are entered
τ I = 24.02 min.
and data sample time is maintained at 6.0 seconds. The decoupler
τD = 6.00 min. D1 in Figure 10 is essentially a feed forward controller that would
reject the disturbance (or interference) from the controller of
Since the extent of loop interaction is mild in this case, tuning
Loop 2 on process variable of Loop 1 (i.e., h1,bot). Likewise, the
rule based on SISO may yield satisfactory tuning parameters for
decoupler D2 in Figure 10 is a feed forward controller that would
multiple-input-multiple-output (MIMO) systems. However, in the
reject the disturbance (or interference) from the controller of
presence of significant loop interactions, the tuning parameters
Loop 1 on process variable of Loop 2 (i.e., h2,bot). The decouplers
based on the SISO must be detuned to suit MIMO cases. The
used in this simulation are
procedures of detuning control parameters are recommended in
−0.04585
the literature [6][7][3(e)][4(c)][8].
−G12 (10.41s + 1) (11.61s + 1)
FEEDBACK CONTROL WITH AND WITHOUT D1 = = (72)
DECOUPLING G11 0.07867
The control block diagram for the feedback control of single
(11.41s + 1) (10.41s + 1)
loops showing the effect of loop interactions in a MIMO system is
given in Figure 8. With the above PID tuning parameters entered −0.04007
to the PID controllers for both loops and the data sample time
−G21 (11.41s + 1) (12.41s + 1)
chosen at 6.0 seconds, the performance of the control system in D2 = = (73)
tracking level setpoints in both loops are shown in Figure 9. Level G22 0.08452
setpoint for the bottom tank on the left side (h1,bot,sp) is changed (11.61s + 1) (12.41s + 1)
from 3.96 meters to 4.5 meters and back to 3.96 meters, while
maintaining the level setpoint of the bottom tank on the right side Similar setpoint-tracking simulations are done as in the
(h2, bot, sp) at 3.92 meters. Similar simulation is done by changing case without decoupling; the results are shown in Figure 11. By
h2, bot, sp from 3.92 meters to 4.5 meters and back to 3.92 meters comparing (Figure 9 and 11), it appears that the controllers move
while maintaining h1,bot,sp at 3.96 meters (Figure 9). It appears more aggressively if the decoupling strategy is implemented.
that the PID controller implemented according to the procedure This is obvious due to the additional feedforward action from
developed in this work provides satisfactory performance of the decouplers. One may observe a striking contrast in the
setpoint tracking for both liquid levels. However, while the level response of both liquid levels. In the case without decoupling
for the left bottom tank (h1) is responding to a change in h1, setpoint, (Figure 9), when h1,bot and h2,bot are responding to their respective
the level for the right bottom tank (h2) deviates from h2, setpoint due setpoint changes, the level of h2,bot and h1,bot , respectively, are
to the interference from Loop 1. The reverse is also true. The disturbed somewhat from their original setpoints. However, such
simulation results suggest that the loop interactions cannot be disturbances are almost fully eliminated when the decoupling
eliminated effectively by two individual PID feedback loops. strategy is implemented (Figure 11). When h1,bot is responding
to step changes in its setpoint, h2,bot pretty much stays very near
The control block diagram for the feedback control of its setpoint value. The reverse is also true. Evidently, the control
single loops using two-way decoupling strategy to eliminate or strategy developed in this work not only successfully identifies
minimize loop interactions is shown in Figure 10. By choosing model and model parameters, but also develops an effective
4.1
4.0
3.9
3.8
3.7
Manipulated Variable
63.0
62.1
61.2
60.3
Gain (K) = 0.08452, 1st Time Constant (min) = 11.11, 2nd Time Constant (min) = 12.61
Goodness of Fit: R-Squared = 0.9966, SSE = 0.04246
Figure 7 Result of data fitting to the exact second-order model for process model G22.
Central
Figure 8 Block diagram for 2x2 MIMO feedback control scheme without using the decoupling strategy.
Figure 9 Setpoint tracking for the two bottom tanks using PID settings for both feedback controllers without using decoupling strategy.
decoupling strategy to eliminate loop interactions. combinations of the effects of the two feed streams to the system.
Simulation results for the four-tank system in Loop Pro show
If one examines Eqs. 72 and 73, it is evident that the time
that the dynamic responses of process variables to the changes
constants involved in this system are very close to each other.
in controller outputs fit the expected overdamped second-order
Therefore, one may ignore the dynamic part of the decouplers
behaviors. With the process models and disturbance models
and simply use static decouplers D1 ~ -0.04585/0.07867 and
D2 ~ -0.04007/0.08452. The results are very similar to those in clearly developed and model parameters obtained, one may
Figure 11 and not demonstrated here. identify the extent of loop interactions using relative gain array.
Model-based controller tuning method such as IMC provides
CONCLUSION adequate PID tuning parameters for the two feedback controllers.
The transfer functions for a four-tank system illustrated However, the system encountered in this work exhibits certain
in this work can be derived by using the principle of analogy degree of loop interaction by using two individual PID feedback
to the single-tank case, with resistances of the six valves controllers. With decoupling strategy applied to both loops, loop
below the four tanks and first-order time constants of the four interactions are almost eliminated entirely.
tanks clearly defined. The final results show that the transfer
REFERENCES
functions of the liquid levels of the two bottom tanks are linear
1. Jang LK. “Level Control by Regulating Control Valve at the Bottom of A
Central
Figure 10 Block diagram for 2x2 MIMO feedback control scheme with decoupling strategy.
Figure 11 Setpoint tracking for the two bottom tanks using PID settings for both feedback controllers with decoupling strategy implemented. The
legends are the same as those of Figure 9.
multivariable systems.” Ind Eng Chem Process Des Dev. 1986; 654–
Gravity-drained Tank”. Chem Eng Educ. 2016; 50: 245-250. 660.
2. Lee M, Shin J. “Constrained Optimal Control of Liquid Level Loop 6. 6. Malwatkar GM, Khandekar AA, Asutkar VG, Waghmare LM. “Design
Using a Conventional Proportional-Integral Controller,” Chem. Eng. of Centralized PI/PID Controller: Interaction Measure Approach,”
Commun. 2009; 196: 729–745. in 2008 IEEE Region 10 and the Third international Conference on
Industrial and Information Systems. 2008; 1–6.
3. Riggs JB, Karim MN. Chemical and Bio-Process Control, 4th edition.
2016; 182–183. 7. Lengare MJ, Chile RH, Waghmare LM. “Design of decentralized
controllers for MIMO processes,” Comput Electr Eng. 2012; 140–147.
4. Smith CA, Corripio AB. Principles and Practices of Automatic Process
Control, 3 edition. Hoboken, NJ: Wiley. 2005; 441-442. 8. Gatzke EP, Meadows ES, Wang C, Doyle FJ. “Model based control of a
four-tank system.” Comput Chem Eng. 2000; 1503–1509.
5. Luyben WL. “Simple method for tuning SISO controllers in