FEATURE SERIES: PRACTICAL PROCESS CONTROL

FEATURE SERIES PRACTICAL PROCESS CONTROL

11: Controlling Steam

Drum Level
Myke King continues his detailed series on process control, seeking to inspire
chemical engineers to exploit untapped opportunities for improvement

C
ONTROLLING the level in the steam drum shown responds and reduces boiler firing. The material in the drum
in Figure 1 might at first appear a relatively simple is not totally liquid but includes bubbles of water vapour. The
scheme, manipulating the flow of boiler feed transient increase in the drum pressure will compress these
water to maintain the level at setpoint. However, bubbles so that the water level falls. This is shrink (swell would
two issues conspire to make it difficult. be the result if the steam demand increases).
The first is that we require tight, rather than averaging The level controller responds to shrink by increasing the flow
control. It is important that the level does not become exces- of boiler feed water. However, falling steam demand requires
sively high and risk liquid entering the steam system. Similarly, less water; the level controller has initially made moves in
it must not become too low and risk starving the boiler tubes the wrong direction. As the pressure reverts to setpoint, there
of water. We therefore require a relatively large controller gain. will be surplus water in the drum that the level controller will
The second issue is the change of phase from liquid to vapour. then have to correct. If tightly tuned, the control response can
The drum contents do not behave as a simple liquid. This can become oscillatory. We must employ tuning closer to that used
require us to substantially reduce the controller gain below the for averaging control, potentially risking excessive deviations
desired value. from setpoint.

SHRINK AND SWELL LEVEL TRANSMITTER

Figure 2 shows a conventional level controller. Consider what Broadly, level transmitters operate by applying one of two
happens if there is a reduction in steam demand. Although the principles. One is based on a float and so will directly measure
steam header pressure will be under control, there will be a level. The other uses two pressure transmitters placed one
transient increase in pressure before the pressure controller above the other to measure the pressure difference (the head

Figure 1: Typical steam boiler arrangement Figure 2: Basic level control

STEAM
AIR PREHEATER
AIR
FI
WATER ECONOMISER PI

SUPERHEATER
STEAM
LC
STEAM
DRUM

FIREBOX

FC

FUEL WATER

MARCH 2024 | The Chemical Engineer | PAGE 58

Process Control 993.indd 58 22/02/2024 15:29:03

 FEATURE SERIES PRACTICAL PROCESS CONTROL

of liquid). Installing this type is a potential solution. Head is a

Level transmitters operate by
measure of mass, which is not subject to shrink or swell. It is
the change in drum content density that causes the change in applying one of two principles.
volume and hence liquid level. However, shrink and swell will One is based on a float and so will
also occur in the boiler tubes, displacing less or more liquid
into the drum and so changing the mass of liquid it contains. directly measure level. The other
Further, particularly on steam drums, engineering standards uses two pressure transmitters
can demand that the true liquid level is measured rather than
inferred from other measurements.
Signal conditioning offers a potential solution. Instead of that the level is below setpoint, the controller will increase the
the level controller using the measurement directly, we apply flow of water. On entering the drum this will cause a small
the correction reduction in drum temperature and so less of the contents will
be converted to steam and the volume of entrained vapour will
reduce. This causes the water level to fall – the opposite of what
we expect on increasing the flow of water. Figure 3 illustrates
The term k is determined empirically. At steady conditions we this behaviour when the flow of water is changed manually. If
switch level control to manual, then slightly change the drum the level controller were in automatic mode, it would respond
pressure setpoint and note the change in level. Repeating this by making a further increase in flow, which initially causes a
over a narrow range of pressures allows us to plot level against further reduction in level. If tightly tuned, the controller would
pressure. The slope of the resulting line is k. In principle, once almost certainly be unstable – again requiring tuning to be
implemented, the level measurement used by the controller more akin to averaging control.
will be almost immune to changes in pressure. However, this
addresses only one of the potential problems.
NOISE
The level measurement is prone to noise. The boiling water will
INVERSE RESPONSE have a turbulent surface. A well-engineered installation might
Inverse response is when the controlled variable responds to the include baffles or a stilling well but is unlikely to be sufficient to
manipulated variable by changing in the direction opposite completely eliminate noise. To avoid control valve damage, we
to what we expect at steady state. In principle, the boiler feed are likely to have to compromise on the requirement for a large
water, passing through the economiser, has recovered heat controller gain. We will address filtering, as a means of noise
from the flue gas to reach the bubble point of the water in the reduction, in the next article. However, filtering level measure-
drum. In practice, it will be somewhat cooler. So, if we imagine ments is generally counter-productive, particularly when tight
control is required. It introduces a lag into a process which has
virtually none – so requiring a substantial change in control-
ler tuning.

Figure 3: Inverse response

STEAM DRUM LEVEL

PROCESS PARAMETER

BOILER FEED WATER FLOW

TIME

MARCH 2024 | The Chemical Engineer | PAGE 59

Process Control 993.indd 59 22/02/2024 15:29:04

 FEATURE SERIES PRACTICAL PROCESS CONTROL

and steam in pounds (lb/hr). The bias algorithm in modern control

Figure 4: Three-element level control
systems supports the inclusion of a scaling factor. In this example
STEAM that factor would be applied to the steam flow – multiplying it by
the reciprocal of water density (0.12 USG/lb).
The use of ratio feedforward would appear to offer that advan-
FI tage in that it will work whatever the units. Only the target
changes. In metric units it would be 1. In imperial units it would be
0.12. However, its use does have a disadvantage often mentioned in
BIAS the literature. We know that, at constant steam demand, the rate
STEAM DRUM LC OR RATIO
of change of level is proportional to the water flow

But the water flow is set by the ratio target (R) and the flow of
steam
FC

WATER
The level is the process variable (PV) of the controller. It
manipulates R, which is thus its manipulated variable (MV). In
general, for an integrating process
THREE-ELEMENT LEVEL CONTROL
Figure 4 shows the scheme designed to deal with the problems
of shrink/swell, inverse response, and noise. It is a very early
example of feedforward/feedback control. By measuring the steam Comparing equations shows that the process gain (Kp) varies
demand, we can keep the flow of water equal to the demand and
so, in theory, the water level will not change. This is the feedfor-
ward part of the scheme. The level controller relies on feedback to
correct any deviation from setpoint or, indeed, permit the operator Boilers can have a 4:1 turndown ratio and so Kp could poten-
to change the setpoint. tially change by a factor of four. In theory this could cause a
Most of the corrective action is taken by feedforward control. tuning problem. However, since the level controller is provid-
This does not stop shrink/swell or inverse response, but the flow of ing only a trim action, it is likely that it can be tuned to be
water is now changed in the correct direction. The level controller robust over the whole operating range.
needs now only take trim action correcting for minor deviations
caused, for example, by flow metering errors. Much slower tuning
can be implemented, so avoiding any oscillatory behaviour. And, NEXT ISSUE
with a much lower controller gain, measurement noise is unlikely Our next article will cover filtering as an example of signal
to cause a problem. conditioning. Filtering may be required to prevent measure-
Conventionally the scheme is configured using a bias algo- ment noise being transmitted to, and potentially damaging,
rithm. The level controller adds (positive or negative) corrective the control valve. We’ll explain the standard filter available
action to the measured steam demand. An alternative scheme in each of the leading distributed control systems (DCS). And
employs a ratio algorithm. This maintains the water flow in we’ll show how a custom filter enables derivative action to
proportion to the steam flow, with the level controller trimming be used without excessively amplifying the noise.
the proportion as necessary.
The bias scheme was first implemented some 80 years ago
when only pneumatic instrumentation existed. Summing pneu- Myke King CEng FIChemE is director of Whitehouse Consulting, an
matic pressures is readily achievable; multiplying them is not. independent advisor covering all aspects of process control. The
However, the disadvantage of the bias scheme is that the flowme- topics featured in this series are covered in greater detail in his book
ters must be calibrated in consistent units. In most of the world this Process Control – A Practical Approach, published by Wiley in 2016
is already likely to be the case. Steam demand is likely measured
in metric tonnes (per hour) and water flow in cubic metres (per
hour). The density of water is very conveniently 1 t/m3. However, Disclaimer: This article is provided for guidance alone. Expert
in the US, water is likely to be measured in US gallons (USG/hr) engineering advice should be sought before application.

MARCH 2024 | The Chemical Engineer | PAGE 60

Process Control 993.indd 60 22/02/2024 15:29:06