Process Control: Level control in communicating tanks

Kevin Peng, Yueyi Zhu, Penghao Zhang (Group 2T)

February 13, 2018
Abstract
The linearized transfer functions under different controller settings (e.g., P and I gains) and
configurations (e.g., open-loop and closed-loop) were derived to predict dynamic behaviors of non-
linear communicating tanks system. Experiments of the single-tank (first-order process) verified the
validity of the linearized transfer function models. The steady-state gain (�) was determined to be
. ± . s/mm , and the time constant ( ) was found as ± s. Influences of P and I
gains on the dynamic behaviors of close-loop systems were investigated. Larger P gain resulted in a
smaller offset and larger I gain resulted in aggressive oscillation. The optimized controller settings
were found as � �� = . ∙ − and �� = × −6 ∙ − , which resulted in minimized
oscillation and offset (< %). The controller performed great ability on controlling the system with a
sudden disturbance with optimized settings. Suggestions on further investigation of the linearized
transfer function model were discussed in the Conclusions section.

Introduction
Process control is essential in chemical engineering because it ensures the safety and
optimization in a process. Furthermore, process control is a useful tool to satisfy the environmental
protocol and the product quality requirements. For chemical engineers, process control is widely
applicable in gasoline refinery and textiles industry. In gasoline refinery, the process control
refractometers are used to monitor and to analyze the hydrocarbon composition of the gas streams of
crude oil. [1] In the processes of synthesizing textiles, the desizing and scouring are two major process
control steps used to maintain the consistent uniform quality of the material. [2] Moreover, process
control is utilized in thermostats and ventilation to maintain the indoor temperatures, which made
civil lives more convenient and comfortable. [3]
In the lab, single-tank and two-tank systems were considered (cf. Figure 1). The main
objectives of this lab are 1) to experimentally verify the validity of the linearization method that was
applied for inherently non-linear system, 2) to compare the response characteristics of first-order and
second-order systems, and 3) to investigate the effects of different controllers (P and PI controllers)
and different gains (P and I gains) determined by Direct Synthesis on robustness and
effectiveness of the closed-loop dynamics.

Open-loop system
For a single-tank system with one inlet and one gravity-fed drain, the overall mass balance
can be formulated as:
ℎ
vi - v0 = A , (1)
where A is the cross-sectional area of the tank, a is the cross-sectional area of drain, h is the liquid
level height in the tank, vi is the inlet flowrate, v0 is the outlet flowrate, and g is the gravity
acceleration. Then according to the Bernoulli’s equation with constant water density, the outlet
flowrate (v0) can be simplified to be:
v0 = �√ �ℎ . (2)
The non-linearity of equation (2) requires firstly linearizing the dynamic model to
approximate the non-linear behavior. This can be done by using the Taylor expansion for equation (1)
about the steady-state for h and vi. Hence, after linearizing the governing equations and using Laplace
transform, the transfer function can be obtained as:
ℎ̅
√
� � �
�� s
=� ̅
= , (3)
ℎ +
√ +
� �

1
 ̅
ℎ � ̅
ℎ
where � ≡ � √ , ≡ � √ , H is the deviation in liquid tank level, Vi is the deviation in the inlet
flowrate, κ is the process gain, and is the process time constant. Vi is the step change which can be
obtained from calibration of pump-drive voltage and corresponding liquid flowrate in the single tank.
Sufficiently linear ranges of κ and are determined from corresponding profiles as a function of
voltage, which dictates the required process parameters used in closed-loop system (See Figures B3
and B4 in Appendix B). In addition, the non-linearity has to be more carefully considered in higher-
order systems, which makes the linearization method an effective tool to approach complex dynamic
models.

Closed-loop system
For closed-loop system, Direct Synthesis is used to verify the type and settings of the
controller. In this experiment, it is a reverse-acting controller since when the liquid level exceeds the
set-point, the controller reduces the inlet flowrate. Furthermore, P-only control results in offset since
the response never reaches a desired setpoint and the discrepancy always occurs. [5] For PI control, it
responds to the integral of error signal over time, and it is initially slow to respond to set-point
changes. Hence, PI control encounters oscillatory response, and the offset is eliminated due to
the integral control. [5] The closed-loop general transfer functions for P and PI controllers are:
= � for P controller, (4)
=� + for PI controller, (5)
�
� = , and = . (6)
� �
��
where P ≡ � , and I ≡ . Equation (6) indicates that increasing Kc and decreasing would cause a
�
more aggressive response (e.g., smaller ) to step-change in set-point. After applying the step
change M and taking the Laplace transform of equations (4) and (5), the transfer functions of closed-
loop single-tank system are:
ℎ′ = � � for P control, and (7)
−
ℎ′ = �� − � for PI control. (8)

Relevant background theories of the experiment are summarized as applying linearization

method to simplify transfer functions that are inherently non-linear, and applying Direct Synthesis
method to determine optimal type and settings of controller under P and PI control.

Experimental Procedures and Apparatus

To investigate the validity of the linearization theory on the non-linearized system, important
variables that are directly measured included the steady-state height (ℎ̅), the cross-sectional area of
the tank (A). Calculated variables include the cross-sectional area of the drain pipe (a), step change
(M), and linear regions of the process gain (κ) and the process time constant ( ). In the open-loop
system, M was calculated by calibrating the flow rate (u) versus voltages (5-10V) by timing how long
it takes for each tank to fill given volume (5 L and 2 gal for Tanks 3 and 4). Tank area, A, was
determined by measurements of heights under different volumes. To determine different ℎ̅ under
different voltages, dynamic responses of Tank 4 water level to step-change of 1 V in voltage were
measured by the PC LabVIEW. With ℎ̅ and u, drain pipe area (a) could be easily calculated by
equation (2), which was also checked with actual values. With all these variables determined, process
parameters, steady-state gain (�) and time constant ( ), were calculated by equation (3) for close-
loop system. In the lab, the controlled variables are water levels in Tanks 3 and 4 (h), and the
manipulated variables are inlet flow rates (u).
In the closed-loop system, to further model the dynamic behavior with P and PI controllers,
we varied P and I gains within the range determined by the Direct Synthesis with � and to find the
influence of the time responses after a step-change of 1 V in configuration 1. Experimental profiles
2
were discussed in the analysis with the theories. With the optimized P and I gains, a fixed set-point
of 150 mm was operated with a sudden disturbance to evaluate system’s ability to recover. In
addition, an open-loop system under configuration 2 was also performed to investigate the validity of
the linearization theory for second order process.
The equipment was composed by a water reservoir, four tanks with gravity drains, and two
water pumps that connected to the inlets of the tanks. Two configurations of this apparatus could be
switched by changing the setting of the valves (V). The liquid level of each tank was monitored and
controlled by the PC LabVIEW with measurements from the pressure transducer (P). The arrows
showed the direction where the water flow went during the experiment. Important experimental
variables, ℎ̅, a, A, and u were illustrated in Fig. 1.

Figure.1 Laboratory apparatus diagram of

the process control. Pumps were connected
to the motor controller (C) and the PC
LabView program. Paddle wheel flow
meters (F) at the pump outlets were used to
measure overall pump flow rates. The
configuration 1 was used for the single-
tank system (V1 and V2 closed, V3 and V4
opened). The configuration 2 was used for
the two-tank system (V1 and V2 opened,
V3 and V4 closed).

Results and Discussions

Open-loop transfer functions provide sensible predictions of tank level’s dynamic response
after a step-change in input (e.g., pump-drive voltage, or flow rate). The main objective for open-
loop operation was to experimentally verify the legitimacy of linearization method used for
determining the open-loop transfer functions of non-linear systems. Theoretically, the linearization
method that was used to determine open-loop transfer functions (See equations A.1.1 to A.1.7 in
Appendix A) dictated the linearity of the transfer function, thus leading to a first order time response.
Two-tank configuration was predicted to take longer to reach steady-state than single-tank, because
it was an interacting process that delayed dynamic responses of water level in the final tank.[4]
To examine whether open-loop single-tank and two-tank system resulted in first order time
response, Figure 2 presents the predicted and measured Tank 4 water level’s time responses of
single-tank and two-tank configurations to a step-change in the voltage from 5 V to 6 V. Measured
time responses agreed qualitatively and quantitatively with predictions, as exponential patterns in Fig.
2 indicated first order behavior. The highly consistent trends of experimental results and predictions
verified that it was legitimate to regard the single-tank system as a linear system in the range of 5-6
V. Applying the same comparison to full experimental voltage range confirmed that the
linearization method was valid and the Bernoulli’s equation did predict tank draining correctly
over the entire measured range of 5-10 V.
For two-tank system, results in Fig. 2 demonstrated that the two-tank system took 70 s longer
to reach steady-state than the single-tank system, which excellently agreed with predictions. The
more sluggish time response of two-tank system resulted from the fact that the first tank in series

3
delayed the response of the second tank to voltage step-change by decreasing the actual inlet flow
rate to the rate of water draining. Notably, the two-tank system showed similar time response to the
single-tank system, because the operating conditions and source of change in input did not vary.

Figure 2. Measured and theoretical

(solid line) open-loop time responses of
Tank 4 water level after a step-change
in the pump-drive voltage from 5 V to 6
V for both single-tank system (red circle
markers) and two-tank system (yellow
triangle markers). Dashed lines marked
the time when systems reached steady
state. Results presented were analogous
to Tank 3, and series of Tank 1 and
Tank 3. For clarity, error bars were not
shown in the figure (See detailed error
calculations in Appendix C).

To determine valid regions and values of process constants where closed-loop operation
would be conducted, the process gain, �, of Tank 4 was determined to be . ± . s/mm ,
and the time constant, , of Tank 4 was determined to be ± s (See Figures B3 and B4 in
Appendix B). The cross-sectional area of the drain pipe, a, for Tank 3 was found as ± ,
and that of Tank 4 was found as ± (See Tables B1 and B2 in Appendix B). Both values
agreed with actual drain size.
With process constants and sufficiently linear range determined by open-loop operation,
closed-loop operation provided generalized modeling for actual processes in industry. The main
objective for closed-loop operation was to study the dynamic behaviors of different types of
controller with various controller settings. Theoretically, for P-only control, an offset always exists,
because the new steady-state value after a step-change in set point is different than the desired value.
[4]
Equation (7) suggests that offset can be suppressed by increasing the controller gain, � .
However, the possibility of instability must be considered. [4] For PI control, offsets will be
eliminated, because the integral action renders the error signal automatically change to the value
required to make offset zero. [5]
To determine dynamic behaviors and trends of closed-loop single-tank system under P
control, Figure 3 presents the measured closed-loop time responses of Tank 3 water level to set-point
change for different P gains. The offset for � = . was found as ± , and the offset for � =
. was found as ± . The reduction in offset for larger P gains agreed excellently with the
prediction that larger controller gain (� or P) resulted in smaller offset. Results in Fig. 3 also
indicated that aggressive control (e.g., larger P value) had greater system sensitivity, leading to less
overshoot, shorter time to reach steady-state, and more oscillation around the set-point. Larger P
value resulted in oscillatory or unstable responses due to the effect of additional lags and time delays
that have been neglected in the present analysis.
To study the trends of how P and I gains affect the performance of the single-tank system, it
was experimentally found that under PI control for single-tank system, overshoot and time to reach
steady-state were minimized by large � (large P) and small (large I), but oscillation was
minimized by moderate � (moderate P) and large (small I). Measured results implied that system
performances could not be optimized simultaneously by varying controller settings. To obtain desired
first order response using Direct Synthesis under two-tank configuration, the optimal controller
settings were found as � = . ∙ − and = × −6 ∙ − . Measured time response
under optimal controller settings gave desired first order behavior, which verified the validity of using
Direct Synthesis to adjust P and I gains.
4
 hsp = 176 mm
Figure 3. Measured closed-loop time
responses of Tank 3 water level to
change in set-point from 136 mm to 176
mm for P-only controlled single-tank
system with controller gains � = .
(red “*” markers; conservative) and � =
P = 1.1 P = 0.5 . (blue circle markers; aggressive).
Time scale was normalized to present the
entire profiles. Grey dashed lines
indicated the new steady-state values.
Results presented were analogous to
Tank 4. For clarity, error bars were not
shown in the figure (See detailed error
calculations in Appendix C).

To obtain dynamic responses of closed-loop single-tank system under PI control to set-

point and disturbance changes, Figure 4 demonstrates the comparisons between theoretical time
responses determined by Direct Synthesis with = , , and measured set-point and
disturbance change for single-tank system. As increased, corresponding time response became
more sluggish. Under optimal controller settings, the closed-loop time response reached a new
steady-state after a set-point change in 100 s with minimal offset and oscillation. Figure 4 also
demonstrates the system’s ability to recover to its original set point within 50 s from a
disturbance introduced at approximately 110 s. The ability to recover related compactly to the
robustness of the system. Such effective and robust response further verified the optimal
controller settings determined experimentally. Furthermore, a combination of Figures 3 and 4
showed that PI control reduced the offset to less than 5%. Optimal P and I gains that were
determined experimentally, even though not perfect, were within acceptable range for processes
that have low sensitivity to offset (e.g., industrial-scale liquid storage systems that aim at
avoiding overflow).

Figure 4. Measured (blue circle

markers) closed-loop time response
of Tank 3 water level to set-point
(from 122 mm to 150 mm) and
disturbance change, and theoretical
(solid lines) closed-loop time
responses of Tank 3 water level to
such set-point change with =
, , under PI control. For
clarity, error bars were not shown in
the figure (See detailed error
calculations in Appendix C).

5
Conclusions
For open-loop operation, measured results agreed excellently with predictions determined
by linearized mass balance for single-tank and two-tank systems, which further confirmed the
legitimacy of using linearization method to predict inherently non-linear systems. The
experimental linear regions were determined to be � = . ± . s/mm and = ±
s. Noteworthy applications of the linearization method include the exponential dependence of
reaction rate on temperature, the nonlinear behavior of pH with flow rate of an acid or a base
stream, and the asymmetric responses of distillate and bottoms compositions in a distillation
column to increases and decreases in feed flow. [6] Besides, results from two-tank system verified
the prediction that previous tank delayed the time response of the final tank to set-point change.
Therefore, open-loop operation was concluded to provide important operating range and process
constants for closed-loop operation, which enhanced the understanding of system behavior. The
measurement and control of industrial process level parameter is of great importance in industrial
field. The level of liquid may affect both the pressure and the rate of flow in and out of the tank.
Notably, two-tank system took approximately 70 s longer to reach steady-state than single-tank
system, since the inherently interacting process decreased the robustness of the system. The
detainment of time to reach steady-state resulted from that water level in the final tank was not as
sensitive as that under single-tank configuration to a step-change in set-point. The lower
sensitivity might be due to the fact that inlet flow rate to final tank was reduced to the rate of
water draining from the previous tank. The series structure propagated and accumulated the
buffer effect introduced by the previous tank. Transfer functions for time responses of water level
in ith tank in series to a step-change were presented in equations (A.3.14-20) in Appendix A.
For closed-loop operation of single-tank system, measured results under P control
demonstrated that more aggressive control (smaller ) led to less offset, shorter time to reach
steady-state, and more oscillation around the set-point. Under PI control, measured results
indicated that system’s performances could not be optimized simultaneously, since all response
specifications depended on the controller gain. For instance, it is impossible to reduce the rise
time and maximize overshoot simultaneously. Thus, different controller settings could be selected
accordingly to emphasize on different purposes or constraints of the system. Optimal controller
settings of single-tank system were found as � = . ∙ − and = × −6 ∙ − ,
where desired first order response was achieved. It concluded the feasibility of using Direct
Synthesis to model the system behavior. In addition, the system’s ability to recover from a
disturbance proved the effectiveness and robustness of the system under optimal controller
settings. Therefore, for systems that can bear moderate offset, the measured optimal controller
settings could result in much faster first order responses.
To further improve the experiment, cross-sectional area of drain pipe can be set to larger
value so that the data acquisition will be more complete and the tank will not overflow as
approaching to high voltage change (e.g., 9-10 V). Larger drain pipe area will lower the steady-
state height in a systematic fashion since larger amount of water leaves the tank in a fixed time.
Besides, to further study the validity and effectiveness of applying the linearization method on
higher-order system, the experimental apparatus can be updated to include more tanks in series.
Therefore, dynamic behaviors of water levels in a series of tanks can be measured and compared.

6
 References

1. Young, Robert E. “Petroleum refining process control and real-Time optimization.” Petroleum
refining process control and real-Time optimization - IEEE Journals & Magazine, Vol 6,
pg 74. IEEE Control Systems, 2006.
2. Yu, Ruey‐Fang, et al. “Control of the Fenton process for textile wastewater treatment using
artificial neural networks.” Journal of Chemical Technology and Biotechnology, Vol 85,
pg 270. John Wiley & Sons, Ltd., 2009.
3. McKernan J.L. and Ellenbecker M.J. “Ventilation equations for improved exothermic process
control.” Ann Occup Hyg, Vol 3, pg 269, 2007
4. “Spring 2006 Process Dynamics, Operations, and Control.” OpenCourseware MIT, MIT, 2006.
5. Seborg, D. E., Edgar, T. F., Mellichamp, D. A., Doyle, F .J., Process Dynamics and Control,
4th Edition, New York: John Wiley and Sons, Inc., 2017.
6. Henson, M. A. and Seborg, D. E., Non-linear Process Control, Chapter 4, section 4.1, pg 149.
Prentice Hall, 1997.

7
 Appendix A – Transfer Functions

1. Single-tank open loop transfer function:

Overall mass balance on one tank is:
ℎ
�� − � =� . (A.1.1)
Bernoulli’s equation for determining the relationship between flowrate and liquid level is:
� + �� + ��ℎ = � + �� + ��ℎ . (A.1.2)
Eliminated terms, it resulted in:
� = �ℎ . (A.1.3)
Linearization on the non-linear system to solve the Ordinary differential equation with the
substitution of
ℎ
�� − a√ �ℎ = �
ℎ′
� = �� − a√ �ℎ

� �
= | . ℎ′ + | . �� ′
�ℎ ����

= −�√
ℎ
ℎ′ + �� ′ , (A.1.4)
̅ �
where the deviation variables are defined ℎ′ ≡ ℎ − ℎ, ��′ ≡ �� − ��′ , s. s ≡ steady state.
Therefore, the simplified transfer function is:
ℎ̅
√
� � �
= � ℎ̅
=
+
, (A.1.5)
√ +
� �
̅
ℎ � ̅
ℎ
where � = √ ; = √ .
� �
�
Let �� = , where M is the slope of the calibration curve between height and voltage. After
using the Laplace transform table:
� �
� ̅
ℎ − √
� ̅̅̅̅
ℎ
ℎ = √ − (A.1.6)
�
�
A is the cross-sectional area: �=̅ . m
a is the drain pipe area (assuming steady state):
�� = a√ �ℎ̅ (A.1.7)
υi
a= ̅
= . mm
√ ℎ

2. Single Tank System in closed loop system

The Direct Synthesis method:
(�⁄� )
�
= ̃[ ]. (A.2.8)
−(�⁄� )
�
The single tank system is a first order process:
�
��
= . (A.2.9)
� +
The process has its own transfer function:

8
 ̃= . (A.2.10)
� +
Substitute Eqn.A.2.10 into Eqn.A.2.8, we eventually will get:
Gc s = � + , (A.2.11)
�
where � = and = .
� �
Based on the definition of PI controller and Eqn.A.2.11,
�= , (A.2.12)
� �

= , (A.2.13)
� �

3. Multi-Tank System transfer function

The transfer function in an open loop system for multiple tank system is given as:
�
= ∏�= � , (A.3.14)
+ �
n is the number of the tanks.
After the derivation, the transfer function is determined as:
= , (A.3.15)
� � +
�
̃= , (A.3.16)
+ +
Then plugging the Eqn.A.2.15 and Eqn.A.2.16 into the Direct Synthesis, we get:
+
Gc s = + + , (A.3.17)
� � + +
In the end, we have:
+
=� + + ,� =
� �
, = + , = +
.
The controller settings will be:
+
�= , (A.3.18)
� �
= , (A.3.19)
� �
�=� , (A.3.20)
�

9
 Appendix B – Raw Data

1. Calibration Curves
Table B1. Measured calibration data for relating the inlet flow rate of Tank 3 and the pump-
drive voltage.
H0(L) H(L) t(s) Voltage(V) Flowrate(L/min)
6 10 47 5 5.106382979
6 10 35 6 6.857142857
6 10 27 7 8.888888889
6 10 25 8 9.6
6 10 22 9 10.90909091
6 10 18 10 13.33333333
16

14

12
Flow rate [L/min]

10

8
y = 1.5429x - 2.456
R² = 0.981
6

4
5 6 7 8 9 10
Pump drive voltage [V]

Figure B1. Calibration curve of Tank 3 with error bars shown (See error analysis in Appendix
C). The magnitude of step-change in input, M, was found as 1.5429 for Tank 3.

Table B2. Measured calibration data for relating the inlet flow rate of Tank 4 and the pump-
drive voltage.
H0(L) H(L) t(s) Voltage(V) Flowrate(L/min)
7.57082 11.35623 41 5 5.53962439
7.57082 11.35623 33 6 6.882563636
7.57082 11.35623 26 7 8.735561538
7.57082 11.35623 23 8 9.874982609
7.57082 11.35623 19 9 11.95392632
7.57082 11.35623 17 10 13.36027059

10
 16

14

Flow rate [L/min] 12

10

8 y = 1.5845x - 2.4924
R² = 0.996

4
5 6 7 8 9 10
Pump drive voltage [V]

Figure B2. Calibration curve of Tank 4 with error bars shown (See error analysis in Appendix
C). The magnitude of step-change in input, M, was found as 1.5845 for Tank 4.

2. Process gain and characteristic time profiles

Figure B3. Solely based on the plot, the linear region is approximately from s.s. height = 300
mm to 1000 mm. Accordingly, the linear range was found as . ± . . The
∙
experimental range of s.s. height was from 100 mm to 250mm, which was technically out of the
linear range. However, the experimental range became linear locally if zoomed in the plot.

11
Figure B4. Solely based on the plot, the linear region is approximately from s.s. height = 300
mm to 1000 mm. Accordingly, the linear range was found as ± . The experimental range
of s.s. height was from 100 mm to 250mm, which was technically out of the linear range.
However, the experimental range became linear locally if zoomed in the plot.

3. Open-loop time responses of water level under two-tank configuration

Figure B5. Measured open-loop time responses of Tank 3 water level to step-change in set-point
(from 104 mm to 146 mm) under two-tank configuration.

12
 Appendix C - Error Analysis

The errors occurred in the measurements included the uncertainty of the liquid level ℎ̅ and the
uncertainty of the flowrate, �� . The errors were given as ± for the ℎ̅ and ± ⁄ .
Known that, the area of the drain pipe was calculated by the Eqn.A.1.7.
�
�=
√ �ℎ̅
The errors then could be further calculated by the formula of the error propagation method.
�� ��
�� = √ � �
+ ̅
�ℎ̅ (C.1.1)
� � �ℎ
Hence, the uncertainty of the a was determined as:
�� = . mm

13