FOSTER WHEELER ENERGY LTD PEG/PEC/03

PROCESS PLANTS DIVISION PAGE 1 OF 16

LEVEL CONTROL PROCESS ENGINEERING GUIDELINE REV 0

REVISION ORIG. REV 1 REV 2 REV 3 REV 4 REV 5

DATE OCT 02
ORIG.BY HF RICKETTS
APP.BY

CONTENTS

SECTION SUBJECT

1. PURPOSE

2. SCOPE

3. REFERENCES

4. GUIDELINES

4.1 Introduction
4.2 Tight Level Control
4.3 Averaging Level Control
4.4 On-Off Control
4.5 Level-Flow Cascade Control
4.6 Feedforward Control
4.7 Dynamic Compensation
4.8 Two and Three Element Control

ATTACHMENTS

1 Derivation Of Maximum Gain For Tight Level Control

2 Derivation Of Minimum Gain For Averaging Level Control
3 Typical Volume Based Algorithm

U:\TWP\Level Control Guideline Rev 0.doc

PEG/PEC/03 FOSTER WHEELER ENERGY LTD
PAGE 2 PROCESS PLANTS DIVISION
REV 0 PROCESS ENGINEERING GUIDELINE LEVEL CONTROL

1. PURPOSE

The aim of this guideline is to discuss the various principles of level control
and to use this as a basis for the design and implementation of level control
systems.

2. SCOPE

This guideline looks at the principal aspects of level control that can
influence the specification of level control systems. The objective of level
control differs from the control of other parameters such as temperature,
flow or pressure in that stability of the manipulated variable is often more
important than the absolute level. This guideline does not cover the
mathematics of controller equations, which are covered in Reference 1. A
useful glossary of control system terminology appears in Reference 2.

3. REFERENCES

1. Process Level Instrumentation and Control, Cheremisinoff N., Dekker

1981

2. Measurement and Control of Liquid Level, Chun H. Cho, ISA 1982

3. Process Standard 508, Attachment A, 1981

4. GUIDELINES

4.1 Introduction

Level in a vessel is an example of a non-self regulating system i.e. it has no

natural equilibrium or steady state to which it will return following a
disturbance. Manual level control may be used to drain a KO drum or when
an overflow with lute is used, otherwise automatic control is normally
required to control liquid level in a system and hence make it a self regulating
process.

There are two main types of level control which are tuned for different
objectives:

• Tight level control is used where the actual level is important and must be

U:\TWP\Level Control Guideline Rev 0.doc

 FOSTER WHEELER ENERGY LTD PEG/PEC/03
PROCESS PLANTS DIVISION PAGE 3
LEVEL CONTROL PROCESS ENGINEERING GUIDELINE REV 0

maintained at a specific value.

• Averaging level control is used for situations where downstream flow rate
stability is more important than actual level in the vessel. The level is
allowed to fluctuate within limits and hence maximise use of the vessel
surge capacity to improve flow rate stability.

Unfortunately liquid level is often difficult to control due to unreliable or

Anoisy@ level measurements, inverse system response or variable process
gain.

Noise is defined as a random or periodic disturbance that is too fast to be

corrected by control action. It may be a function of a process, e.g. the level
of a boiling or splashing liquid is inherently noisy. Noise is common in flow or
level measurements and can also be caused by induced currents in signal
transmission or by genuine disturbances caused by poorly placed instrument
tappings.

Derivative action is rarely used in level controllers due to the normally noisy
measurement signal. The effect of noise must also be accounted for in
tuning proportional (P) and proportional-plus-integral (PI) level controllers to
avoid exceeding the high liquid level in a vessel.

Noise can be overcome by the use of filters which are normally dynamic (1st
order exponential) but have the disadvantage of slowing down the control
loop response. A non-linear control algorithm (e.g. error squared algorithm)
may also act as an efficient noise filter.

Inverse derivative action is another means of stabilization for processes with

noisy signals. The inverse derivative term is added to the controller output.
This means that the change in controller output will be dampened and will be
smaller for a rapidly changing signal than for a steady error of similar
magnitude. The net result is that the flow is more stable.

This guideline outlines several techniques for solving level control problems.

4.2 Tight Level Control

Tight level control is used when the level in a vessel must be maintained
close to its setpoint at all times. This is the case in a boiler steam drum,
compressor KO drum, or where a constant level is critical. Proportional or
proportional-plus-integral action may be used for tight level control but
derivative action is not recommended. Note that it is important to check that
the controller is tuned aggressively enough so that expected load changes do
not drain or overflow the vessel.

U:\TWP\Level Control Guideline Rev 0.doc

PEG/PEC/03 FOSTER WHEELER ENERGY LTD
PAGE 4 PROCESS PLANTS DIVISION
REV 0 PROCESS ENGINEERING GUIDELINE LEVEL CONTROL

4.2.1 Proportional Control

The disadvantage with proportional only level controllers is that they will
always operate with an offset from setpoint, which can take up valuable
vessel surge capacity.

To tune a proportional level controller, the maximum gain that can be used
without causing oscillation is calculated as:

h. A
K max =
F .t

where:

Kmax = maximum controller gain

h = height between level tappings
A = cross sectional area of vessel
F = maximum flow through valve (100% open)
t = scan interval of controller

Note that the maximum controller scan interval must be not greater than
30% of the system time constant, T:

A.h.d
T=
100. f

where:

f = max flow disturbance normally expected

d = distance between setpoint and closest alarm limit (as
percentage of total range)

Note that this formula is based on a step disturbance in flow, which is a

fairly unrealistic process fluctuation. This particular tuning will result in over
tight control. The derivation of the term Kmax is given in Attachment 1.

4.2.2 Proportional-plus-Integral Control

Integral action can be added to a proportional controller to further stabilise

the control valve flow rate and to allow the level to return to its setpoint
after an upset.

To tune a proportional-plus-integral controller the following guidelines may be

used:

U:\TWP\Level Control Guideline Rev 0.doc

 FOSTER WHEELER ENERGY LTD PEG/PEC/03
PROCESS PLANTS DIVISION PAGE 5
LEVEL CONTROL PROCESS ENGINEERING GUIDELINE REV 0

K = 0.8 K max
A.h
Ti =
F

where:

Ti = controller integral time (time per repeat)

K = controller gain
Kmax, A, h, F are as defined in 4.2.1.

4.3 Averaging Level Control

Averaging level control is used whenever there is a need to take advantage

of a vessel’s surge capacity. Its main objective is to maintain a steady
outflow to downstream units. The level is prevented from reaching alarm
limits in the event of a normally expected flow disturbance. An averaging
level controller will slowly return level to its setpoint to retain maximum
surge capacity. This type of control can be achieved either through adjusting
the tuning of a proportional-plus-integral action controller or by applying a
non-linear control algorithm.

4.3.1 Proportional-plus-Integral Control

An averaging level controller is a variation of a proportional-plus-integral

action controller, tuned such that a substantial change in level takes place
before the outflow again equals the inflow condition. The change in outflow
is spread over a period of several minutes so that the downstream
disturbances are less severe than would be given by a tight level controller.
This is usually achieved by employing a controller with a low gain and very
slow integral action to maximise use of the vessel=s surge capacity.

Typical tuning that can be applied to an averaging proportional-plus-integral

level controller is as follows:

K = 0.8K min
A.h.d
Ti =
12.5 f

where:

100 f
K min =
d .F

U:\TWP\Level Control Guideline Rev 0.doc

PEG/PEC/03 FOSTER WHEELER ENERGY LTD
PAGE 6 PROCESS PLANTS DIVISION
REV 0 PROCESS ENGINEERING GUIDELINE LEVEL CONTROL

= the gain which will allow the level to just reach the
nearest alarm limit (Use only for Proportional control)

All other terms are as defined in 4.2.1. The derivation of the term Kmin is
given in Attachment 2.

4.3.2 Non-linear Control Algorithms

A non-linear control algorithm is one in which the change in controller output

for a given change in input variable is dependent on both the tuning and the
magnitude of the deviation from setpoint. Applied to level control it results in
minimal changes to the outflow from a vessel when the level deviates within
a safe range, yet acts rapidly to restore an acceptable level when the liquid
level approaches its limits. The level is slowly returned to the middle of the
vessel to maximise its surge capacity.

Non-linear controllers can be either conventional PID-type controllers with

more than one gain setting (an extension of on/off controllers) or use a non-
linear PID algorithm such as error-squared. Predictive volume computational
algorithms can also be used.

Non-linearity may also be a problem when measuring level in a small

horizontal cylindrical vessel due to the effect of curvature. This needs to be
addressed when selecting the appropriate level control algorithm.

4.3.2.1 Error Squared Algorithms

The most common algorithm developed specifically for level control is the
error squared algorithm and this has the advantage of accommodating both
large and small process upsets. The error-squared algorithm is similar to a
PID algorithm but the error term is replaced by the square of the error
(retaining the sign of the error). The controller makes small changes when
the level is close to its setpoint, but will make exponentially increasing
changes as the error increases. The valve position affected by a given
change in level is least sensitive at the midrange. Integral action must be
added to the controller, otherwise the desirable flat portion of the curve
would correspond with the actual valve position only when the valve is 50%
open. This algorithm has the added advantage of the ability to cope with
noise.

Most DCS vendors have their own implementation of error squared

algorithms. Before using a vendor algorithm to calculate initial tuning
parameters, it is important to understand the implementation structure.

U:\TWP\Level Control Guideline Rev 0.doc

 FOSTER WHEELER ENERGY LTD PEG/PEC/03
PROCESS PLANTS DIVISION PAGE 7
LEVEL CONTROL PROCESS ENGINEERING GUIDELINE REV 0

4.3.2.2 Volume-based Algorithms

Volume based non-linear level controllers have been configured for certain
applications (e.g. LNG plants) to eliminate downstream flow deviations. The
imbalance between inlet and outlet flows is calculated and the outlet flow is
adjusted so that the two flows will be in balance by the time the level
reaches a limit. The flow is then adjusted to return the liquid level to its
setpoint within a given Ahorizon@ time. If a major plant upset occurs and the
vessel level is forced outside its normal surge limits then the algorithm makes
a more drastic change in outlet flow to return the liquid level to within the
surge limit. A typical volume based algorithm is given in Attachment 3. This
type of level control can be used successfully in systems such as gas-oil
separators, where slugs of liquid can occur travelling through the system.

4.4 On-Off Level Control

The simplest level controller is the on-off controller. It consists of a controller

that is either at its maximum or minimum output depending on whether the
vessel level is above or below its setpoint. For example, a tank level can be
controlled by a valve on its outlet. If the vessel level rises above its setpoint,
the valve will fully open and if the level falls below its setpoint the valve will
close. A proportional-only controller with a high gain creates this cyclic
action. An illustration of the response is shown in Figure 1.

Level

Output

Figure 1 On-Off Control

This type of controller has the disadvantage that the control valve (or pump)
is frequently opening and closing, causing mechanical wear.

On-off control may also be implemented through logic rather than a PID
controller with a large gain. In this case the valve or pump switching occurs
at high and low levels. This type of control is most often used where there is
insufficient liquid to merit continuous removal, e.g. KO drums and drain
sumps.

U:\TWP\Level Control Guideline Rev 0.doc

PEG/PEC/03 FOSTER WHEELER ENERGY LTD
PAGE 8 PROCESS PLANTS DIVISION
REV 0 PROCESS ENGINEERING GUIDELINE LEVEL CONTROL

An improvement on the simple on-off controller is a neutral zone or gap

controller where no action is taken around the setpoint deadband, but normal
PID control is performed outside this band. More complex gap controllers
may have multiple regions with different tuning within each band. Most
algorithms allow the definition of upper and lower bands. This type of control
is then more like the averaging level control described in 4.3.

4.5 Level-Flow Cascade Control

In many instances the liquid outlet from a vessel is the feed to a downstream
unit or piece of equipment. With direct control the level controller output
directly manipulates the outlet control valve. However if there are changes in
pressure at either the source or destination, or changes in the system
pressure drop between the source and destination, then flow through the
valve will change. Eventually this will manifest itself as a change in level and
the controller will take action to restore the level. This can result in an
unsteady flow out of the vessel. By replacing the direct control with a level-
flow cascade, the flow controller rejects any disturbances affecting the flow
rate, resulting in more stable flow out of the vessel and improved level
control. Level-flow cascade is strongly recommended wherever the
destination unit/equipment is sensitive to feed rate changes, e.g. distillation
columns and reactors.

Cascade control is more expensive than direct control since it requires an

additional flow measurement and controller. However in many instances a
flow measurement is already available, so there is minimal additional cost in
implementing the cascade.

4.6 Feedforward Control

Feedforward control is used when level itself is an important variable (tight

level control). An example of its use is in boiler drum level control - see
Figure 2. To maintain stable operation the mass flow of feed water entering
the drum should equal the mass flow of steam leaving the drum. This is
achieved by using a steam flow measurement as the principal means of
setting the feed water flow. The level control then provides a trim function
to correct any mismatch. This concept can be applied to any situation where
significant supply or demand disturbances can not be handled by feedback
control alone.

Steam drum level control is treated more fully in Reference 3.

U:\TWP\Level Control Guideline Rev 0.doc

 FOSTER WHEELER ENERGY LTD PEG/PEC/03
PROCESS PLANTS DIVISION PAGE 9
LEVEL CONTROL PROCESS ENGINEERING GUIDELINE REV 0

Steam

FI

LIC

Feed Water

Figure 2 Feedforward Control

4.7 Dynamic Compensation

PID controllers are designed on the basis that the process they are controlling
is linear with first order characteristics. If the actual process characteristics
deviate from this assumption significantly then control is likely to be poor. To
overcome this dynamic compensation can be used to account for
characteristics such as inverse response and deadtime.

Consider as an example the case of a boiler steam drum. If there is an

increase in steam demand then there is a slight fall in steam drum pressure.
Since the liquid in the drum is at its boiling point the reduction in pressure
causes a temporary increase in vaporisation of the water. The creation of
bubbles in the liquid causes the volume of liquid to “swell” and give an
increase in liquid level. Once equilibrium is reestablished between the vapour
and liquid the level falls with the higher rate of steam withdrawal. This
inverse response would cause level control on its own to decrease the feed
water flow initially, and then have to rapidly increase feed water flow once
the level drops.

Figure 3 shows drum pressure used as dynamic compensation to counteract

the natural inverse response of the system.

U:\TWP\Level Control Guideline Rev 0.doc

PEG/PEC/03 FOSTER WHEELER ENERGY LTD
PAGE 10 PROCESS PLANTS DIVISION
REV 0 PROCESS ENGINEERING GUIDELINE LEVEL CONTROL

Steam

FI
PI

LIC
LT

Feed Water

Figure 3 Dynamic Compensation

Dynamic compensation in this example consists of adding the derivative of

drum pressure to the drum level measurement. Under steady state conditions
the level controller sees the output of the level transmitter alone, but on a fall
in drum pressure, the signal to the level controller is decreased by the
compensating signal. This represents approximately the amount that the
liquid swells.

Dynamic compensation is rarely used on its own and often involves the
addition of a deadtime lead/lag compensation algorithm.

4.8 Two and Three Element Control

Both feedforward and feedback controllers have some disadvantages. A

feedback controller only acts when the controlled variable deviates from its
setpoint and a feedforward controller cannot compensate for unmeasured
process disturbances. Feedforward controllers are therefore often used in
conjunction with feedback controllers and this is known as two-element
control. Figure 2 above is a two-element drum level control scheme.
However variations in drum pressure or feed water supply pressure will
cause variations in level. Adding a third element, a feed water flow controller
as shown in Figure 4, will stabilise feed water flow and further improve
control stability. This is the classical three-element drum level control
scheme.

U:\TWP\Level Control Guideline Rev 0.doc

 FOSTER WHEELER ENERGY LTD PEG/PEC/03
PROCESS PLANTS DIVISION PAGE 11
LEVEL CONTROL PROCESS ENGINEERING GUIDELINE REV 0

Steam

FI

LIC

FIC

Feed Water

Figure 4 Three Element Drum Level Control

U:\TWP\Level Control Guideline Rev 0.doc

PEG/PEC/03 FOSTER WHEELER ENERGY LTD
PAGE 12 PROCESS PLANTS DIVISION
REV 0 PROCESS ENGINEERING GUIDELINE LEVEL CONTROL

Attachment 1 - Derivation Of Maximum Gain For Tight Level Control

For a proportional-only controller with the level at setpoint at previous scan

interval:

∆mn = K .en

where:

∆mn = change in controller output at scan n

K = controller gain
en = error (setpoint – measured variable) at scan n

For a given change in inlet flow of f, then the error after one scan interval is
given by:

f .t
en =
h. A

where:

t = scan interval
h = height between level tappings
A = vessel cross-sectional area

To make the change in outlet flow equal to the change in inlet flow in one
scan interval:

f
∆mn =
F

where:

F = maximum flow through outlet valve

Hence the controller gain is given by:

∆mn f h. A
K max = = .
en F f .t
h. A
=
F .t

U:\TWP\Level Control Guideline Rev 0.doc

 FOSTER WHEELER ENERGY LTD PEG/PEC/03
PROCESS PLANTS DIVISION PAGE 13
LEVEL CONTROL PROCESS ENGINEERING GUIDELINE REV 0

Attachment 2 - Derivation Of Minimum Gain For Averaging Level Control

For a proportional-only controller the change in output is given by:

∆m = mn − mn −1 = K .(en − en −1 )

where:

mn = controller output at scan n

mn-1 = controller output at previous scan
en = error at scan n
en-1 = error at previous scan

If the level is allowed to just reach the nearest alarm limit, then:

malarm − minitial = K min .(ealarm − einitial )

where:

malarm = controller output when level is at alarm setting

minitial = controller output at initial level
ealarm = error when level is at alarm setting
einitial = error at initial level

For a given change in inlet flow of f, then if outlet flow is changed by the
same amount when the level reaches the alarm setting:

f
malarm − minitial =
F

where:

F = maximum flow through outlet valve

If the level is initially at setpoint, then:

einitial = 0
d
ealarm =
100

where:

d = distance between alarm setting and setpoint as percentage of

U:\TWP\Level Control Guideline Rev 0.doc

PEG/PEC/03 FOSTER WHEELER ENERGY LTD
PAGE 14 PROCESS PLANTS DIVISION
REV 0 PROCESS ENGINEERING GUIDELINE LEVEL CONTROL

range
Hence:

malarm − minitial
K min =
ealarm

100 f
=
d .F

U:\TWP\Level Control Guideline Rev 0.doc

 FOSTER WHEELER ENERGY LTD PEG/PEC/03
PROCESS PLANTS DIVISION PAGE 15
LEVEL CONTROL PROCESS ENGINEERING GUIDELINE REV 0

Attachment 3 - Typical Volume Based Algorithm

F i

Upper surge limit V H

SP
LIC FIC
Lower surge limit
V L

At any time t the imbalance between inlet and outlet flows can be calculated
by:

(Vt − Vt −1 )
I=
t

where:

I = imbalance between inlet and outlet flows

Vt = liquid volume this execution scan
Vt-1 = liquid volume last scan
t = scan time interval

If the average imbalance is assumed to be half of the current imbalance, then

the number of scans, N, in which the volume will reach its upper limit, VH, is
given by:

2.(VH − Vt )
N=
I .t

If the upper volume limit is not to be exceeded, then after N scans the
increase in outlet flow, F, must be equal to the imbalance, I. Using a simple
average the incremental change in outlet flow, ∆F, on each scan is given by:

I
∆F =
int ( N + 1)

Once the inlet and outlet flows are in balance, the controller should return
the level to setpoint in a specified horizon time. The incremental change in

U:\TWP\Level Control Guideline Rev 0.doc

PEG/PEC/03 FOSTER WHEELER ENERGY LTD
PAGE 16 PROCESS PLANTS DIVISION
REV 0 PROCESS ENGINEERING GUIDELINE LEVEL CONTROL

outlet flow, ∆Fbal, to achieve this is calculated from:

 3.(Vs − Vt ) 
0.96 2.I − 
 th 
∆Fbal =
( N h 2)
where:

Vs = setpoint volume
th = horizon time
Nh = number of scans until horizon time

On a major plant upset (large rate of change of flow into the vessel) a shorter
upset horizon is used, and the change in outlet flow, ∆Fupset, is calculated
from:

 3.(Vs − Vt ) 
0.962.I − 
 tu 
∆Fupset =
(N u 2)
where:

tu = upset horizon time

Nu = number of scans until upset horizon time

U:\TWP\Level Control Guideline Rev 0.doc