ChE curriculum

LEVEL CONTROL BY REGULATING

CONTROL VALVE AT THE BOTTOM
OF A GRAVITY-DRAINED TANK
Larry K. Jang

T
California State University • Long Beach, CA 90840-5103
ypical liquid-level control systems used in industry bottom of the tank, and k is a lumped constant.
are illustrated in Figures 1 (next page). The majority The initial steady-state volumetric balance (input flow rate
of process control textbooks cover control systems as = output flow rate) for this constant-density system is
shown in Figure 1(a) and (b). In the Unit Operations Labora-
tory of the author’s department, there is a level-control experi- d h ss
A = 0 = q in , ss − q ss (2)
ment with a direct-acting (or fail-close) control valve located dt
at the bottom of the tank [Figure 1(c)]. Therefore, the author
where t is time, A is the cross-sectional area of the open tank,
developed an instructional module that provides a thorough
subscript ss denotes the initial steady-state condition or null
analysis of the dynamic behavior of the control system in
operating condition, and qin is the incoming volumetric flow
Figure 1 (c). Transfer functions for the controller output and
rate. Assuming that at t = 0 both qin and q start to deviate from
the process variable in a feedback proportional-integral (PI)
their respective steady-state values, the response of the liquid
control system are derived for the servo problem (setpoint
level can be described by a transient-state balance equation:
tracking) and the regulatory problem (disturbance rejection).
dh
One main challenge to the development of dynamic mod- A = q in − q ( 3)
els for the case illustrated in Figure 1(c) emanates from the dt
fact that the discharge flow rate is governed not only by the When Eq. (2) is subtracted from Eq. (3), we have a differ-
liquid level (h, the process or controlled variable) but also by ential equation whose variables are expressed as deviation
the size, type, and valve stem position (x, the manipulated quantities:
variable) of the control valve. The following assumptions
are made in the derivation of the dynamic models: (1) the d ∆h
A = ∆q in − ∆q (4)
head loss of liquid in the discharge line is entirely due to the dt
control valve; (2) the control valve is of the direct-acting (or
where
fail-close) type and the valve trim is linear, i.e., the valve
characteristic function Cv(x) is linearly proportional to x; (3) ∆h = h − h ss (5)
the valve is never saturated (fully closed or fully open) during
the dynamic response, and; (4) the flow rate of liquid through ∆q in = q in − q in ,ss (6)
the control valve is proportional to (DPvalve)1/2, where DPvalve
is the pressure drop across the valve which can be taken as Larry K. Jang is a professor of chemical
(r g h) based on assumption one above, where r is the liquid engineering at California State University,

density and g the gravitational acceleration. We may then

Long Beach. He earned his B.S. and M.S.
from National Taiwan University and his
express the discharge volumetric flow rate q [m3/s] as Ph.D. from University of Southern Califor-
nia, all in chemical engineering. His area
∆Pvalve of research in recent years is remote au-
q = ( x,h ) = Cv ( x ) = kx h (1) tomatic control using LabVIEW technology.
S.G.
where x is the valve stem position or the extent of valve open-
ing, with x = 0 being fully closed and x = 1 fully open for a
direct-acting (or fail-close ) control valve; S.G. is the specific
gravity of the liquid, h is the liquid level measured from the © Copyright ChE Division of ASEE 2016

ChE errata
Due to a production error, there is one symbol missing in the paper “Level Control...” by Larry K. Jang, published in the Fall
2016 issue of CEE. The letter “f” is missing in the final print. Eq. (1) on Page 245 should appear like
Vol. 50, No. 4, Fall 2016 245
∆P
q = f ( x,h ) = Cv ( x ) valve = kx h (1)
S.G.
 we have a first-order differential equation in the standard
a form with the coefficient of the process variable Dh being 1:
d∆h
τp = R∆q in + K p ∆x − ∆h (9)
dt
where
2 h ss
R= (10)
k x ss
2h ss
K p = Process Gain= − (11)
x ss
b τ p = first-order time constant
=AR (12 )
By applying a Laplace transform with the initial condition
Dh(t = 0) = 0, we arrive at the open-loop transfer function
for the liquid level:
R Kp
∆h (s) = ∆q in (s) + ∆x (s) (13)
τ p s +1 τ p s +1

For simplicity, the symbol “(s)” that denotes the Laplace

c domain following each variable is omitted hereafter. In this
process, Dqin is considered the load or disturbance and Dx the
manipulated variable. Therefore, we may define the process
transfer function Gp and the disturbance transfer function Gd
in the Laplace domain:
Kp
Gp = (14 )
τ p s +1

Kd
Gd =
τ ps + 1
Figures 1. Typical liquid-level control systems.
R
= (15 )
It is noted that a nonlinear term [Dq = q – qss = kx(h) – kxss
1/2 τ ps + 1
(hss )1/2] is encountered, which needs to be linearized:
where the disturbance gain Kd = R and the disturbance time
∆q = q − q ss constant is identical to the process time constant τ p. We may
= f ( x,h ) − f ( x ss , h ss ) then express the feedback control system as a block diagram
 ∂f   ∂f  shown in Figure 2, assuming that the dynamics of both the
≈   ∆x +   ∆h actuator and the sensor are tentatively ignored and that the
 ∂x ss  ∂h ss actuator gain is lumped together with the controller gain.
k x ss
= k h ss ∆x + ∆h ( 7)
2 h ss

where
∆x = x − x ss (8)
If both sides of the resultant linearized differential equation
[Eqs. (4) and (7) combined] are multiplied with
2 h ss
k x ss
, Figure 2. Feedback control block diagram for the liquid-
level control system in Figure 1(c).

246 Chemical Engineering Education

TRANSFER FUNCTIONS FOR Dx, Dh, AND Dq transfer function for the control valve stem position x in this
IN THE SETPOINT TRACKING CASE instruction module and asked the students to perform their
own detailed derivations as a homework exercise. The detailed
Assume that a proportional-integral (PI) controller is used
derivation for the resultant equations below is available to
to adjust the control valve stem position according to error
any interested readers upon request. In the setpoint-tracking
ε and the transfer function Gc for the controller:
case, Dqin = 0. So, Dx Gp = Dh and Dx/Dhsp = Dh/(Gp Dhsp).
∆x  1  Therefore, by dividing Eq. (17) with Gp, we have
Gc = = K c 1+  (16)
ε  τ Is  ∆x Gc
=
∆h sp 1+ G c G p
where
K c = Proportional Gain of the PI Controller βs γ
= Kc + + ( 23)
τ 2 s 2 + 2ζτ s +1 τ 2 s 2 + 2ζτ s +1
τ I = Integral Time of the PI Controller
ε = Error = h sp − h = ∆h sp − ∆h where

The subscript sp denotes the level setpoint.  τ − KcKp τI 

β =  p  ( 24 )
Transfer Function for the Liquid Level  Kp 
With the process model Gp defined, the transfer function  1− K c K p 
for the closed-loop setpoint-tracking case Dh/Dhsp using a PI γ =   ( 25)
 Kp 
controller can be found in the literature [1(a),2]
∆h GcGp Therefore, if a step change in Dhsp with the magnitude a
= (17) is made (Dhsp = a/s), the resultant transfer function for the
∆h sp 1+ G c G p
valve stem position is
∆h τ Is +1 Kca
= (18) ∆x = + βaYim p + γ a Ystep ( 26)
∆h sp τ 2 s 2 + 2ζτs +1 s
where The time-domain equation Dx(t) is the sum of a step
change with the magnitude Kc a, the response of a second-
τp τI order model to an impulse input (baYimp), and the response
τ= (19)
KcKp of a second-order model to a step input (gaYstep), where
the standard functions Yimp and Ystep in the time domain are
1+ K c K p τI available in the literature.[2,3] Once the transfer functions and
ζ= ( 20)
2 τpKcKp the time-domain equations for Dh and Dx are obtained, the
change in discharge flow rate Dq is then expressed as a linear
Therefore, if a step change in Dhsp with the magnitude a combination as shown in Eq. (7).
is made,
a TRANSFER FUNCTIONS FOR Dx, Dh, AND Dq
∆h sp = ( 21) IN THE DISTURBANCE REJECTION CASE
s
Transfer Function for the Liquid Level
Then Eq. (18) becomes
The response of the liquid level to a change in disturbance
a τI a
∆h = + variable (incoming flow rate qin) in a closed-loop PI-control
τ s + 2ζτ s +1 s ( τ s + 2ζτ s +1)
2 2 2 2
system can be found in the literature.[1(b)]
= a τ I Yim p + aYstep ( 22) ∆h Gd
=
∆q in 1+ G c G p
where Yimp is the response of a standard second-order model
(with gain K = 1) to the unit impulse input and Ystep is that to K τ 
 d I  s
the unit step input. The time-domain equations for Yimp and  KcKp 
Ystep can be found in the literature.[2,3] = 2 2 ( 27)
τ s + 2ζτs +1
Transfer Function for Control Valve Stem Position
Therefore, for a step change in the disturbance variable with
Most textbooks do not provide derivations for the transfer
the magnitude b (Dqin = b/s), the transfer function for Dh is
function of the manipulated variable, which is the control
valve stem position x in this case. The author derived the

Vol. 50, No. 4, Fall 2016 247

 K τ  parameter; and up is the dead time of the process model Gp (up
 d I  b = 0 for a pure first-order model in this work).[4] For aggressive
 KcKp  control, a small value of τ f is chosen. Conversely, a large τ f
∆h = 2 2
τ s + 2ζτs +1 value will result in conservative control. By substituting Eqs.
K τ  (32) (with up = 0) and (33) into Eqs. (19) and (20), we have
=  d I  b Yim p ( 28)
 KcKp  τ = τp τf ( 34 )
τf + τp
Transfer Function for Control Valve Stem Position ζ= ( 35)
2 τf τp
Most textbooks do not provide derivations for the transfer
function of the manipulated variable in the disturbance re-
jection case. Again, the author derived it for the instruction SIMULATION
manual and asked students to perform their own derivations The following operating conditions are chosen for simulation:
as a homework exercise. The author would like to make the
A = cross-sectional area of the open tank = 0.1 m2
detailed derivation for the resultant equations below available
to any interested readers upon request. In the disturbance qin,ss = qss = steady-state incoming and discharge flow
rejection case (Dhsp = 0), rates = 0.001 m3 / sec
∆x = εG c k = lumped valve constant = 0.002 m2.5 / sec
= ( ∆h sp − ∆h ) G c xss = steady-state valve stem position = 0.5
Under these conditions, we have
= −∆hG c ( 29)
hss = steady-state liquid height = 1.0 m [Eq. (1)]
By combining Eqs. (27) and (29), we have Kp = Process gain = - 4 m [Eq. (11)]
∆x −G d G c τ p = 200 sec [Eqs. (10) and (12)]
=
∆q in 1+ G c G p Kd = 2.0 3 103 sec/m2 [Eqs. (10) and (15)]
 −K τ   −K  Since most level-control systems in industry deal with
 d I  s +  d 
 Kp   Kp  regulatory problems, simulation is done here by assuming
= ( 30) that a step change of 0.0001 m3 /sec is made in the incoming
τ 2 s 2 + 2ζτs +1
flow rate of liquid (i.e., b = 0.0001 m3/sec, or 10% change
If a step change in the disturbance variable with the mag- of the initial steady-state condition) at t = 0. A feedback PI
nitude b (Dqin = b/s) is made, the transfer function for Dx controller is used to reject the disturbance caused by a change
becomes in the incoming flow rate, while attempting to maintain the
liquid level at the setpoint. Table 1 lists the proportional gain
 −K τ b   −K b 
 d I   d  Kc at various values of the desired closed-loop time constant
 K p   Kp  τ f, as well as the time constant τ and damping factor z of
∆x = 2 2 +
τ s + 2ζτs +1 s ( τ s + 2ζτs +1)
2 2 the closed-loop second-order characteristic equation with the
integral time τ I set at τ p (= 200 sec). Using the VBA created
 −K τ b   −K b  by the author for Yimp and Ystep for various ranges of z values,
=  d I  Yim p +  d  Ystep ( 31)
 Kp   Kp  the responses of liquid level, control valve stem position,
and discharge flow rate, all expressed in deviation quantities,
Again, the change in discharge flow rate Dq is simply a are calculated and plotted in Figures 3-5. It is noted that the
linear combination of Dx and Dh as expressed in Eq. (7). IMC tuning rule results in a critically damped or overdamped
closed-loop characteristic equation with the damping factor
IMC TUNING RULE z ≥ 1 (Table 1).
If the internal model control (IMC) tuning method is used, From Figure 4, the smaller the τ f value (the greater the
the tuning parameters of a PI-controller for a first-order-plus- Kc value), the faster the response of the control valve, as
dead-time process model are expected. As a result, the liquid level returns to the setpoint
1 τp value and the discharge flow rate reaches the new steady-state
Kc = ( 32) value more quickly (Figures 3 and 5). In other words, a more
K p (θ p + τ f )
aggressive control action or tighter control favors both the
τI = τp ( 33) performance of level control and the response of the discharge
flow rate in the disturbance rejection case.
where τ f is the desired closed-loop time constant, an adjustable Interested readers may wish to explore the dynamics of
248 Chemical Engineering Education
 Interested readers may wish to
explore the dynamics of level,
control valve stem position,
and discharge flow rate in the
setpoint-tracking case using the
equations developed in this work.

the symbol Dh stays the same.

In the author’s CHE 460 (Chemical Process
Control) class, all three control systems il-
lustrated in Figure 1 are covered. Since very
few textbooks cover the details of the system
illustrated in Figure 1(c), the author developed
this module for students to learn how to derive
process and disturbance transfer functions by
applying the linearization skill (to tackle the
nonlinear term of discharge flow rate through
Figure 3. the control valve at the bottom of the tank). The
performance of the feedback PI-controller is
evaluated not just for the liquid level, but also
for the control valve stem position and the
discharge flow rate. The IMC control strategy
is stressed when selecting control parameters
Kc and τ I.
The first-order differential equation is analo-
gous to that for a simple RC-circuit in which a
voltage source is applied to a series combina-
tion of a resistance R and a capacitance C. This
circuit has a dynamic first-order time constant
of RC.[5] Therefore, we may consider the vari-
able R [Eq. (10)] as the linearized resistance
and the cross-sectional area of the tank, A, as
the capacitance of the level process.
If the expression for time constant τ p [Eq.
(12)] is manipulated in the following manner,
it may provide further insight into the nature
of the time constant:

Figure 4.
TABLE 1
Proportional gain Kc of the PI controller at various values of the desired
level, control valve stem position, and closed-loop time constant τf . By choosing integral time τI = τp = 200 s , the
discharge flow rate in the setpoint-tracking resultant second-order time constant τ and damping factor ζ of the closed-
case using the equations developed in this loop characteristic function are calculated and used in simulation.
work. τf (sec) 25 50 100 150 200 250
Kc (1/m)
DISCUSSION (Eq. 32)
-2.00 -1.00 -0.500 -0.333 -0.250 -0.200

If the control valve is located L (m) τ (sec) 70.7 100 141 173 200 223
below the bottom, the models developed (Eq. 34)
in this work can be easily modified with h ζ
1.59 1.25 1.06 1.01 1.00 1.01
replaced by (h + L) and hss by (hss + L). But (Eq. 35)

Vol. 50, No. 4, Fall 2016 249

 τ = AR 3. By using the IMC turning rule, integral time τ I of
the feedback PI controller is set at τ p. The propor-
2 h ss tional gain Kc can be determined at various values
=A
k x ss of the desired closed-loop time constant τ f. The
A h ss resultant second-order characteristic equation for
=2
k x ss h ss the closed-loop system is critically or over-damped
with z ≥ 1.
Steady-state Volume of Liquid in the Tank
=2 ( 36) 4. With τ I = τ p, aggressive control (small τ f and large
Steady-state Discharge Flow Rate [Eq. (1)]
Kc) favors both the level control and the response of
It is interesting to find that the time constant τ p is twice discharge flow rate in the disturbance rejection case.
as much as the “holding time” (the time needed to drain the ACKNOWLEDGMENTS
liquid from the tank at the steady-state discharge flow rate).
The author would like to thank Dr. Roger C. Lo and Dr.
Furthermore, the fact that the time constant τ p, the process
Gregory Smith, professors of chemical engineering at Cali-
gain Kp, and the disturbance gain Kd, all depend on the steady-
fornia State University Long Beach (CSULB), for their com-
state or operating condition, further illustrates the nonlinear
ments. This work is partially supported by Faculty Sabbatical
nature of the dynamic models in this system.
Leave award from CSULB.
One merit of using the IMC method to tune the PI-controller
for a pure first-order system is that the closed-loop dynamics REFERENCES
is always critically or over-damped (z ≥ 1 ). Therefore, the risk 1. Seborg, D.E., T.F. Edgar, D.A. Mellichamp, and F.J. Doyle III, in
of the final control element becoming saturated due to over- Process Dynamics and Control, 3rd ed., Hoboken, N.J: Wiley, pp. (a)
187–188; (b) 192–193 (2010)
shooting (in the event of under-damped response when large 2. Marlin, T.E., in Process Control: Designing Processes and Control
| Kc | and small τ I are used) may be minimized. As seen in Fig- Systems For Dynamic Performance, New York: McGraw-Hill Com-
ure 4, the maximum response of the control valve is Dx = + 0.05 panies, pp. 254–255 (1995)
from the initial steady-state condition of xss = 0.5, or x (final) = 3. Jang, L.K., and R.C. Lo, “Developing a Straightforward Tuning Method
for Weak Acid or Weak Base Neutralization Control System,” Chem.
0.55. Since the response pattern never overshoots even when tight Eng. Process Tech., 2(1), 1023 (2014)
control (with τ I = τ p = 200 sec and “aggressive” tuning parameter 4. Riggs, J.B., and M.N. Karim, in Chemical and Bio-Process Control,
τ f = 25 sec) is implemented, the valve stem position always stays 3rd ed., Lubbock, TX: Ferret Pub, p. 315 (2007)
within the saturation limits (between x = 0 and x = 1). Care must 5. Coughanowr, D., Process Systems Analysis and Control, 2nd ed., New
York: McGraw-Hill Science/Engineering/Math, pp. 71-72 (1991)
be taken, however, when the unit is operating at the initial valve 6. Lee, M., and J. Shin, “Constrained Optimal Control of Liquid Level
stem position closer to x = 0 or x = 1 and/or the integral time τ I Loop Using a Conventional Proportional-Integral Controller,” Chem.
is set at smaller values that may lead to an underdamped response. Eng. Commun., 196(6), 729 (2009) p
In the literature, optimization for level control
in the pumped tank case [Figure 1(b)] has been
explored.[6] The performance of a level-control
system does not only consider the response of level
alone, but also takes into account the dynamics of
the discharge flow rate. Interested workers may
develop a similar optimization scheme for the
gravity-drained case in Figure 1(c).

CONCLUSIONS
1. The process model and the distur-
bance model for the level control sys-
tem with the control valve located at
the bottom of the tank can be derived
after the governing equation of the
discharge flow rate is linearized.
2. Parameters for the first-order process
model and disturbance model can
be determined from the steady-state
condition (or the null operating condi-
tion) using the open-loop transfer
functions derived in this work. Figure 5.
250 Chemical Engineering Education