Control of a Liquid-Level Process

Valve Electro-Pneum atic

Am plifier
Positioner Transducer
p eA
eE
xv Potentiom eter eR
K K
Bridge
eB hV
x
QM

hC
Float Level
Sensor
PU Rf QU
TANK
PUMP

Sensors & Actuators in Mechatronics K. Craig

Liquid Level Control 1
 Objective
Maintain tank level hC at the desired level hV in the face
of disturbances pressure PU(t) (psig) and volume flow
rate QU(t) (ft3/sec). Rf is a linearized flow resistance with
units psi/(ft3/sec).
Equilibrium Operating Point
All variables are steady
Inflow QM exactly matches the two outflows
hC = hV and eE = 0
When eE = 0, QM can be nonzero since the electropneumatic
transducer has a zero adjustment and the valve positioner has a
zero adjustment, e.g., p = 9 psig and the valve opening
corresponds to equilibrium flow QM.
We will deal with small perturbations in all variables away from
the initial steady state.
Sensors & Actuators in Mechatronics K. Craig
Liquid Level Control 2
 Assumptions and Equations of Motion
Tank Process Dynamics
Density of fluid is constant.

gh C PU dh C
QM QU = AT Conservation
Rf dt of
Rf 1 R Mass
( ps + 1) h C = g
QM +
g
PU f Q U
g
AT R f
p = process time constant
g

Sensors & Actuators in Mechatronics K. Craig

Liquid Level Control 3
 Float Level Sensor
Assume a zero-order dynamic model, i.e., the dynamics are
negligible relative to the process time constant p since the
cross-sectional area of the tank is assumed large.
Consider the actual dynamics to justify this assumption:

B (Unavoidable Friction)

Buoyant
Force F B

h x Wf FB = Wf + A f g ( h x )
Tank

A f = float cross-sectional area h-x

L f = float height Lf
W f = float weight
steady condition: h = x

Sensors & Actuators in Mechatronics K. Craig

Liquid Level Control 4
 Equation of Motion
dx d2x
FB Wf B = Mf 2
dt dt
d2x dx
Mf 2 + B = A f g ( h x )
dt dt
d2x dx
M f 2 + B + A f gx = A f gh
dt dt
D 2 2 D
2 + + 1 x = Kh n =
A f g
n n
Mf
d
D B
dt Differential =
2 A f M f g
d2 Operator
D 2
2
K = 1.0
dt
Sensors & Actuators in Mechatronics K. Craig
Liquid Level Control 5
 To measure rapid changes in h accurately, n must be
sufficiently large. The specific weight of the fluid (g)
is not a design variable, so strive for large values of
Af/Mf (i.e., hollow floats).
In our case, the tank has a large diameter and if the
inflow and outflow rates are modest, h cannot change
rapidly and so a zero-order model is justified.
Potentiometer Bridge and Electronic Amplifier
Obviously these two components are fast enough to be
treated as zero order in this system.

Sensors & Actuators in Mechatronics K. Craig

Liquid Level Control 6
 Electropneumatic Transducer
This device produces a pneumatic
output signal closely proportional (
5% nonlinearity) to an electrical
input ( 5V and 3-15 psig).

We are concerned with overall

dynamics from eA to p. The block
diagram shows a 4th-order closed-
loop differential equation.

However, experimental frequency

response tests show typically a flat
amplitude ratio out to about 5 Hz.
This response is very fast relative to
p so we model the electropneumatic
transducer as zero order.

Sensors & Actuators in Mechatronics K. Craig

Liquid Level Control 7
 Pneumatic Valve Positioner
We are only interested in the overall dynamics relating
xv to p. These are again quite fast relative to p, so we
model the component as zero order.
The valve positioner allows one to characterize the
static calibration curve between p and xv and thus
obtain desired linear or nonlinear relationships between
p and manipulated flowrate QM.
Relation between QM and xv
This relationship is assumed to be statically linear and
dynamically instantaneous and thus a zero-order model.
Although the dynamic response of QM to xv is not
instantaneous due to fluid inertia and compliance, the
response is much faster than the tank-filling dynamics.
Sensors & Actuators in Mechatronics K. Craig
Liquid Level Control 8
 Closed-Loop System Block Diagram and
Differential Equation P U 1 + - QU
Rf

eR eE eA p xv +
hv + R f / g hC
Kh Ka Kp Kx Kv
+
ps + 1
- QM
eB x
Kh 1.0
Rf
1 g
( K h h V K h h C ) K a K p K x K v + R PU Q U s + 1 = h C
f p
K 1 Rf
( ss + 1) h C = hV + PU QU
K +1 g ( K + 1) g ( K + 1)
p Closed-Loop 1
s = System
K + 1 Time Constant
K=
g
( K h K a K p K x K v R f ) System Loop
Gain

Sensors & Actuators in Mechatronics K. Craig

Liquid Level Control 9
 Speed of Response
Response for a step input in hv (hold perturbations PU
and QU at zero) t

K s
hC = h Vs 1 e
K +1

Response for a step input in disturbances PU and QU
(hold hV = 0)
1 t
s
R f t
s

hC = PUs 1 e hC = Q Us 1 e
g ( K + 1)


g ( K + 1)

Increasing loop gain K increases the speed of response

p
s = = closed-loop system time constant
K +1
Sensors & Actuators in Mechatronics K. Craig
Liquid Level Control 10
 Steady-State Errors
A procedure generally useful for all types of systems
and inputs is to rewrite the closed-loop system
differential equation with system error (V-C), rather
than the controlled variable C, as the unknown.
In this case we have:
K 1 Rf
( ss + 1) h C = hV + PU QU
K +1 g ( K + 1) g ( K + 1)
hE = hV hC
K 1 Rf
( ss + 1)( h V h E ) = hV + PU QU
K +1 g ( K + 1) g ( K + 1)
1 1 Rf
( ss + 1) h E = ss + hV PU + QU
K +1 g ( K + 1) g ( K + 1)
Sensors & Actuators in Mechatronics K. Craig
Liquid Level Control 11
 For any chosen commands or disturbances, the steady-
state error will just be the particular solution of the
equation:
1 1 Rf
( s ) E s
s + 1 h = s + h
V P + QU
g ( K + 1) g ( K + 1)
U
K +1

We see that the steady-state error is improved if we

increase the loop gain K.
For any initial equilibrium condition we can trim the
system for zero error but subsequent steady commands
and/or disturbances must cause steady-state errors.
Ramp inputs would cause steady-state errors that increase
linearly with time, the rate of increase being proportional
to ramp slope and inversely proportional to K+1.
Sensors & Actuators in Mechatronics K. Craig
Liquid Level Control 12
 Stability
All aspects of system behavior are improved by
increasing loop gain up to a point! instability may
result, but our present model gives no warning of this.
Why?
We neglected dynamics in some components and a
general rule is:
If we want to make valid stability predictions we must
include enough dynamics in our system so that the
closed-loop system differential equation is at least third
order. The one exception is systems with dead times
where instability can occur even when dynamics are
zero, first, or second order.
Is our model then useless?
Sensors & Actuators in Mechatronics K. Craig
Liquid Level Control 13
 No! It does correctly predict system behavior as long
as the loop gain is not made too large. As K is
increased, the closed-loop system response gets faster
and faster. At some point, the neglected dynamics are
no longer negligible and the model becomes inaccurate.
We neglected dynamics relative to p, but in the closed-
loop system, response speed is determined by s.
Exercise:
Compare the responses of the following 3 systems for
K = 1, 5, 10 with 1 = 1.0, 2 = 0.1, and 3 = 0.05. The
input is a unit step.
Examine speed of response, steady-state error, and
stability predictions.

Sensors & Actuators in Mechatronics K. Craig

Liquid Level Control 14
 1
K C1
tau1.s+1
Step Input
System 1

1 1
K C2
tau1.s+1 tau2.s+1
Step Input System 2

1 1
K C3
tau1.s+1 tau2.s+1
Step Input
System 3
1
tau3.s+1

MatLab / Simulink Diagram

Sensors & Actuators in Mechatronics K. Craig
Liquid Level Control 15
 Response for K = 1
0.5
C3
0.45

0.4

0.35
C2
C1
response: C1 C2 C3

0.3

0.25

0.2

0.15

0.1

0.05

0
0 0.5 1 1.5 2 2.5 3
time (sec)

Sensors & Actuators in Mechatronics K. Craig

Liquid Level Control 16
 Response for K = 5
1

0.9
C3
0.8

0.7
C2
response: C1 C2 C3

0.6
C1
0.5

0.4

0.3

0.2

0.1

0
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
time (sec)
Sensors & Actuators in Mechatronics K. Craig
Liquid Level Control 17
 Response for K = 10
1.4

1.2

C3
1
response: C1 C2 C3

0.8

C2
0.6

C1
0.4

0.2

0
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
time (sec)

Sensors & Actuators in Mechatronics K. Craig

Liquid Level Control 18
 Response of C1 for K = 1, 5, 10
1

0.9

0.8

0.7
K=10
0.6
response: C1

K=5
0.5

0.4

0.3

0.2

0.1
K=1
0
0 0.5 1 1.5 2 2.5
time (sec)

Sensors & Actuators in Mechatronics K. Craig

Liquid Level Control 19
 Response of C2 for K = 1, 5, 10
1.4

1.2

K=10
1
response: C2

0.8

K=5
0.6

0.4

0.2

K=1
0
0 0.5 1 1.5 2 2.5
time (sec)

Sensors & Actuators in Mechatronics K. Craig

Liquid Level Control 20
 Response of C3 for K = 1, 5, 10
1.4

1.2

K=10
1
response: C3

0.8

0.6
K=5

0.4

K=1
0.2

0
0 0.5 1 1.5 2 2.5
time (sec)

Sensors & Actuators in Mechatronics K. Craig

Liquid Level Control 21
 Note that as loop gain K is increased, the speed of
response is increased and the steady-state error is reduced.
For what value of loop gain K will any of these systems
go unstable?
Lets look at the closed-loop system transfer functions and
characteristic equations:
Transfer Functions
C1 K (The characteristic equation
= is obtained by setting the
V 1s + 1 + K
denominator polynomial
C2 K equal to zero.)
=
V 12s 2 + ( 1 + 2 ) s + 1 + K
C3 K ( 3 + 1)
=
V 12 3s3 + ( 12 + 2 3 + 13 ) s 2 + ( 1 + 2 + 3 ) s + 1 + K

Sensors & Actuators in Mechatronics K. Craig

Liquid Level Control 22
 C1 K
=
V s +1+ K
C2 K
=
V 0.1s 2 + 1.1s + 1 + K
C3 K ( 0.05s + 1)
=
V 0.005s3 + 0.155s 2 + 1.15s + 1 + K

The only system which will go unstable as the loop

gain K is increased is the third system; its characteristic
equation is third order. The first two systems will
continue to show improved speed of response and
reduction of steady-state error without any hint of
instability!

Sensors & Actuators in Mechatronics K. Craig

Liquid Level Control 23
 Lets apply the three methods of determining closed-
loop system stability Routh, Nyquist, and Root-Locus
- to the third system and determine the value of K for
which this system becomes marginally stable.
Routh Stability Criterion
Closed-Loop System Characteristic Equation

0.005s3 + 0.155s 2 + 1.15s + 1 + K = 0

Routh Array 0.005 1.15
0.155 1+ K
(0.155)(1.15) (1 + K)(0.005)
0
0.155
1+ K 0

Sensors & Actuators in Mechatronics K. Craig

Liquid Level Control 24
 For stability we see that:

(0.155)(1.15) (1 + K)(0.005) > 0

1+ K > 0
This leads to the result that for absolute stability:

1 < K < 34.65

A simulation with the loop gain set to K = 34.65 should verify
this result. The value of gain K = -1 will give the closed-loop
system characteristic equation a root at the origin but that value
is of less interest, since we rarely use negative gain values.
Note that at the loop gain value of 34.65, only system 3 is
marginally stable. Systems 1 and 2 show no signs of
instability, only improved speed of response and reduced
steady-state error.

Sensors & Actuators in Mechatronics K. Craig

Liquid Level Control 25
 Response for K = 34.65
2.5

C3
1.5
response: C1 C2 C3

C2
1

0.5
C1
0

-0.5
0 0.5 1 1.5
time (sec)

Sensors & Actuators in Mechatronics K. Craig

Liquid Level Control 26
 Root-Locus Interpretation of Stability
The open-loop transfer function is:

K
( 1s + 1)( 2s + 1)( 3s + 1)
K
=
12 3s3 + ( 12 + 2 3 + 13 ) s 2 + ( 1 + 2 + 3 ) s + 1
K
=
0.005s3 + 0.155s 2 + 1.15s + 1

The root-locus plot shows that when K = 34.65, the system is

marginally stable. For that value of K, the closed-loop poles
are at: -31 and 15.1658i.

Sensors & Actuators in Mechatronics K. Craig

Liquid Level Control 27
 30

Root-Locus Plot for K > 0

20

K = 34.65
10
Imag Axis

-10

Open-Loop Poles:
-20 -20, -10, -1

-30
-25 -20 -15 -10 -5 0 5
Real Axis

Sensors & Actuators in Mechatronics K. Craig

Liquid Level Control 28
 Nyquist Stability Criterion
A polar plot of the open-loop transfer function for the gain K =
34.65 goes through the point 1, indicating marginal stability
of the closed-loop system.
A polar plot of the open-loop transfer function for the gain K =
10 shows a gain margin = 3.46.
The Bode plots for a gain K = 10 show a gain margin = 10.794
dB = 3.46 and a phase margin = 40.5 degrees.

Sensors & Actuators in Mechatronics K. Craig

Liquid Level Control 29
 Nyquist Diagrams
From: U(1)
25

20

15
K = 34.65
10
Imaginary Axis

= -
5
= 0-
To: Y(1)

= 0+
= +
-5

-10

-15

-20

-25
-5 0 5 10 15 20 25 30 35

Real Axis

Sensors & Actuators in Mechatronics K. Craig

Liquid Level Control 30
 Nyquist Diagrams
From: U(1)
6

2
K = 10
Imaginary Axis

To: Y(1)

-0.289
-2
Gain Margin = 1/.289 = 3.46
-4

-6
-2 0 2 4 6 8 10

Real Axis

Sensors & Actuators in Mechatronics K. Craig

Liquid Level Control 31
 Bode Diagrams
K = 10 Gm=10.794 dB (at 15.166 rad/sec), Pm=40.539 deg. (at 7.4486 rad/sec)
20

0
Phase (deg); Magnitude (dB)

-20
GM = 10.8 dB
-40 PM = 40.5 deg
-60

-100

-200

-300
10-1 100 101 102

Frequency (rad/sec)

Sensors & Actuators in Mechatronics K. Craig

Liquid Level Control 32
 Saturation
The amplifier, electropneumatic transducer, and valve
positioner all exhibit saturation, limiting their output
when the input becomes too large.
All real systems must exhibit such power limitations
and one of the consequences is that the closed-loop
response speed improvement will not be realized for
large signals.
Exercise:
Simulate the following two systems for gains of K = 1,
10 and a unit step input.
Examine speed of response and steady-state error.

Sensors & Actuators in Mechatronics K. Craig

Liquid Level Control 33
 1
K C1
s+1
Step Input

M1

M2

1
K C2
s+1
Step Input Saturation
Upper Limit: +1
Lower Limit: -1

MatLab / Simulink Diagram

Sensors & Actuators in Mechatronics K. Craig

Liquid Level Control 34
 Unit Step Response for C1 and C2 with K = 1
0.5

0.45

0.4

0.35
C1 and C2 responses are identical.
response: C1 C2

0.3

0.25

0.2

0.15

0.1

0.05

0
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
time (sec)

Sensors & Actuators in Mechatronics K. Craig

Liquid Level Control 35
 Unit Step Response for C1 and C2 with K = 10
1

0.9

0.8
C1
0.7
response: C1 C2

0.6
C2
0.5

0.4
Speed of response is
0.3
affected, but steady-state
0.2
error is not.
0.1

0
0 0.5 1 1.5 2 2.5 3 3.5 4
time (sec)

Sensors & Actuators in Mechatronics K. Craig

Liquid Level Control 36
 Step Response of Magnitude 1.5 for C1 and C2 with K = 10
1.4

1.2

C1
1
response: C1 C2

0.8

C2
0.6

Both speed of response

0.4
and steady-state error are
0.2
affected.

0
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
time (sec)

Sensors & Actuators in Mechatronics K. Craig

Liquid Level Control 37
 Lets make the system model more realistic by
modeling the pneumatic valve positioner as a first-
order system:
xv Kx
(D) =
p vp D + 1

The open-loop system is now second order. The closed-

loop system differential equation is now:
D 2 2 D K vp D + 1 R f ( vp D + 1)
2 + + 1 h C = hV + PU QU
n n K +1 g ( K + 1) g ( K + 1)
K +1 p + vp 1
n = = K= ( K h K a K p K x K v R f )
p vp 2 p vp ( K + 1) g

Sensors & Actuators in Mechatronics K. Craig

Liquid Level Control 38
 To get fast response (large n) for given lags p and vp,
we must increase loop gain K. How does K affect ?
If p = 60 sec and vp = 1.0 sec and we desire = 0.6,
what is K? What is n ? Is absolute instability possible
with this model? What does the Nyquist plot show as
K is increased? What does the root-locus plot show as
K is increased?
Consider Gain Distribution
How does gain distribution affect stability and dynamic
response of the closed-loop system?
Are steady-state errors for disturbances sensitive to
gain distribution?
Should one optimize the distribution of gain so as to
minimize steady-state errors?
Sensors & Actuators in Mechatronics K. Craig
Liquid Level Control 39
 K2n
G(s) = 2
s + 2n s + 2n
Location of Poles
s1,2 = n in 1 2 Of
Transfer Function
s1,2 = id

y (t ) = 1 e t
cos d t + sin d t
d
1.8
tr rise time
n
4.6
ts settling time
n

General All-Pole
Mp = e 1 2
( 0 < 1) overshoot 2nd-Order
Step Response
1 ( 0 0.6 )
0.6
Sensors & Actuators in Mechatronics K. Craig
Liquid Level Control 40
 1.8 4.6
n
tr ts

0.6 (1 M p ) 0 0.6

Time-Response Specifications vs. Pole-Location Specifications

Sensors & Actuators in Mechatronics K. Craig
Liquid Level Control 41
 Integral Control
Consider integral control of this liquid-level process.
Replace the amplifier block Ka by KI/s.
The closed-loop system differential equation is:
Rf
Kh KIKpK x K v 1 g
( h V h C ) + PU Q U = hC
s Rf ps + 1
1 Rf
( p D + D + K ) h C = Kh V + g DPU g DQU
2

1 Rf
( p D + D + K ) h E = ( p D + D ) h V + g DPU g DQU
2 2

K h K I K p K x K v R f loop system
K= gain h E = h V h C error
g
Sensors & Actuators in Mechatronics K. Craig
Liquid Level Control 42
 Note the following:
Step changes (constant values) of hV, PU, and/or QU
give zero steady-state errors.
For ramp inputs, we now have constant, nonzero
steady-state errors whose magnitudes can be reduced
by increasing K.
The characteristic equation is second order, so define:

K 1
n = =
p 2 Kp

If we take p as unavailable for change, we see that an

increase in K to gain response speed or decrease ramp
steady-state errors will be limited by loss of relative
stability (low ).
Sensors & Actuators in Mechatronics K. Craig
Liquid Level Control 43
 If we design for a desired , the needed K is easily
found and n is then fixed.
Absolute instability is not predicted; the model is too
simple.
See the comparison between proportional control and
integral control: root locus plots and Nyquist plots.
Note the destabilizing effects of integral control.

Sensors & Actuators in Mechatronics K. Craig

Liquid Level Control 44
 V + K C V + K C
ps + 1 s ( p s + 1)
- -

Proportional Integral
Control Control
Root
Loci

x x x
1 1

p p

-1 -1

Nyquist
Plots
Sensors & Actuators in Mechatronics K. Craig
Liquid Level Control 45
 What if we change the controlled process by
closing off the pipe on the left side of the tank.
This not only deletes PU as a disturbance but also
causes a significant change in process dynamics.
Take Rf = . The from Conservation of Mass we
have: 1
hC = ( QM QU )
ATs
The original tank process had self-regulation.
If one changes QM and/or QU the tank will itself in time
find a new equilibrium level since the flow through Rf
varies with level.

Sensors & Actuators in Mechatronics K. Craig

Liquid Level Control 46
 With Rf not present, the tiniest difference between QM
and QU will cause the tank to completely drain or
overflow since it is now an integrator and has lost its
self-regulation.
Even with proportional control, the integrating effect in
the process gives zero steady-state error for step
commands, but not for disturbances.
If we substitute integral control to eliminate the QU
error, the system becomes absolutely unstable.

Sensors & Actuators in Mechatronics K. Craig

Liquid Level Control 47