Back to Basics

Troubleshooting
Process Control
Problems
Brian K. Sullivan Identifiying and troubleshooting
Lummus Technology
problematic control loops can significantly
improve plant performance.

T
he theoretically attainable performance of a plant controllers that do not work well, so this is a key indicator of
in the chemical process industries (CPI) is closely a problem. In some cases, primary controllers can remain in
aligned with the technology applied and the equip- automatic mode even though they have tracking or condi-
ment installed during its design and construction. Achieving tional status because the downstream secondary controller
that potential performance depends on proper equipment or function block is not in the preferred mode or position.
operation, which depends in part on the effectiveness of the Therefore, it is also necessary to look at the status of the
plant’s control system. For decades, the CPI have tried to controller to confirm whether it is in use.
close the gap between actual and potential plant performance Next, ask the following questions to help identify
through the application of advanced control. regulatory control loops with low service factors or
Numerous opportunities exist to improve plant opera- problematic performance:
tion through the rectification of the basic regulatory control • Do any controllers typically exhibit a large degree of
system without the application of advanced control. In fact, variability or oscillatory/cyclic behavior?
some of the benefit that is attributed to advanced control • Are any control loops periodically placed in manual
technology is often the result of correcting regulatory mode to handle large disturbances, implement setpoint
control problems during the implementation of an advanced changes, or squelch unstable control responses? This is com-
control project. mon, for example, during cracking heater swaps and dryer
Systematic methods can be used to identify problematic switches in ethylene plants.
regulatory control loops. Once identified, straightforward • Do any controllers typically require a very long time to
techniques are available to troubleshoot several typical con- reach their setpoint following a disturbance, process excur-
trol system problems. This article describes these methods sion, or setpoint change?
and techniques. • Do operators frequently change the setpoint of
some controllers?
Identifying problematic control loops Those questions can be answered easily by applying
While some control loop problems may be well known common statistical functions to data extracted from the
to plant personnel, others may be more obscure. Therefore, plant’s data historian or distributed control system (DCS).
the first step of correcting a regulatory process control sys- One simple approach is to import time series data (e.g.,
tem is identifying the problematic loops. one-minute data for a week) for the relevant control
Begin by identifying loops that are continuously oper- parameters of all the plant control loops into a spreadsheet
ated in manual mode. Operators quickly lose patience with where the analysis can be performed. While some effort

Copyright © 2020 American Institute of Chemical Engineers (AIChE). CEP May 2020 www.aiche.org/cep 55
Not for distribution without prior written permission.
Back to Basics

is required to prepare the spreadsheets, they can be used range of the controller) of the setpoint over the duration of
repeatedly on a periodic basis to quickly identify new or the time series indicates such a controller.
developing problems. Statistics may not capture all of the issues, so conducting
• Service factor. Convert the mode and status discussions with several plant operators may identify addi-
states to a numerical value (e.g., IF mode = manual or tional problematic control loops. When doing so, it is best
status = conditional/tracking, THEN value = 0, ELSE to talk with operators from several different shifts because
value = 1) for each data point in the time series. The aver- different shifts can have different approaches to the same
age value across the entire time series represents the service operating and control issues.
factor for the controller (Figure 1). Service factors between Once you have compiled a shortlist of underperforming
0% and 50% are poor. Service factors between 50% and control loops, troubleshooting can begin. The two trouble-
90% are non-optimal, and the controller may be adversely shooting techniques addressed in this article are systematic
affected by particular disturbances. Service factors greater review of general control loop functionality and examination
than 90% are good. of specific control loop structures.
• Controller performance. Multiple calculations are
required to analyze controller performance. First, calculate Troubleshooting general control loop functionality
the difference between the setpoint and the process variable Many problems can be eliminated by methodically
of the controller for each available data point (the result can reviewing the general functionality of the problematic
be positive or negative). Then, calculate the standard devia- control loops.
tion for the full array of those calculated difference values. Tuning. Controller tuning is frequently blamed for regula-
The standard deviation value divided by the range of the tory process control problems, although it is often not the
controller (e.g., if the range is –50°F to 50°F, the range is culprit. Tuning procedures will not be addressed in this article
100°F) provides an excellent normalized indication of the because they are widely covered in the literature.
controller performance (Figure 2). Focus on the controllers Because control systems from different vendors use dif-
with the highest values to identify the controllers with the ferent control equations, be cognizant of whether the control
most significant performance issues. In the statistical analysis equations use proportional band or gain, interactive or inde-
example in Figure 2, Tag #2 should be addressed first. pendent gain, resets per minute (or second), or integral time
• Setpoint variance. Operators sometimes try to help a per reset when selecting your tuning constants. In addition, if
level, pressure, or temperature controller respond to a dis- a controller appears to work well in some scenarios and not
turbance by changing its setpoint to accelerate the control well in others, consider whether adaptive tuning is required.
response, instead of allowing the controller to do its job. A good example of an application requiring adaptive
A high normalized variance (i.e., variance divided by the tuning is the coil outlet temperature (COT) controller on
an ethylene plant’s cracking
heater. The COT control-
Tag #1 Tag #1 Service Tag #2 Tag #2 Service Tag #3 Tag #3 Service
Timestamp Mode State Status Mode State Status Mode State Status ler is typically tuned during
6/11/19 00:00 Auto Normal 1 Auto Track 0 Manual Track 0 normal cracking operation,
6/11/19 00:01 Auto Normal 1 Auto Track 0 Manual Track 0 when an endothermic crack-
6/11/19 00:02 Auto Normal 1 Auto Track 0 Manual Track 0 ing reaction is absorbing much
6/11/19 00:03 Auto Normal 1 Auto Track 0 Manual Track 0 of the fired duty. However,
6/11/19 00:04 Auto Normal 1 Auto Track 0 Manual Track 0 during decoking operations,
• • • • • • • • • • an exothermic combustion
• • • • • • • • • •
reaction occurs in the radiant
• • • • • • • • • •
coils. Therefore, the response
6/17/19 23:55 Auto Normal 1 Auto Normal 1 Manual Track 0
1 Auto 1 Track
of the COT controller to a
6/17/19 23:56 Auto Normal Normal Manual 0
6/17/19 23:57 Auto Normal 1 Auto Normal 1 Auto Track 0
step change in the fuel firing
6/17/19 23:58 Auto Normal 1 Auto Normal 1 Auto Normal 1 is quite different in those two
6/17/19 23:59 Manual Normal 0 Auto Normal 1 Auto Normal 1 operating modes. Adaptive
Service
90% 50% 20%
tuning can be used to auto-
Factor matically modify the gain of
the COT controller during
p Figure 1. Statistical analysis can help identify problematic control loops. In this example, the service factor
is calculated to determine which controller should be prioritized for troubleshooting. Tag #3 should be prioritized decoking operations to account
because it is only in service 20% of the time (assuming it is not a situational controller that is used only during for this change in response.
limited types of operation). Other cases in which adap-

56 www.aiche.org/cep May 2020 CEP Copyright © 2020 American Institute of Chemical Engineers (AIChE).
Not for distribution without prior written permission.
tive tuning might be applicable include:
• split-range controllers Process
• cascade loops with multiple secondary controllers Variable

• controllers in nonlinear processes in which the process

gain changes significantly as the plant load changes (i.e., Control
100% capacity operation vs. turndown at 70%, etc.) Output
• controllers that can normally operate across their entire
control valve range. Time
Instrument reliability, range, and calibration. For control- p Figure 3. Valve stiction occurs when the valve does not move
lers that operate mainly in manual mode (or tracking/condi- after a change request is sent to its actuator from the controller. As a
tional status), check whether the measured process variable consequence, the controller’s output signal continues to change until
is reliable. Trend the measured process variable while the the actuator generates enough force to move the valve. When the
valve finally moves, it creates a large step change in the measured
controller is in manual mode and the associated control valve process variable.
is at a constant opening. Look for the following characteris-
tics as an indication of an instrumentation problem: of blow-through [i.e., the actual level is lower than the read-
• the value is constantly frozen at the low or high end-of- ing] or carryover [the actual level is higher than the reading]).
scale value (the instrument might not be scaled properly or Troubleshoot the instrumentation’s installation before
might be installed incorrectly) moving on to the control loop configuration.
• the value exhibits high-frequency noise with a large Final control elements. If a controller exhibits oscillatory
amplitude or highly variable performance when in automatic mode,
• the value appears to be relatively constant and then check the control valve and its associated hardware. In par-
exhibits large jumps in value. ticular, if the controller output appears as a sawtooth pattern
These measurement behaviors may be due to instrumen- and the process variable exhibits a square-wave response (as
tation problems, such as: in Figure 3), then valve stiction could be the problem.
• installation mistakes (inversion of a flow orifice, insuf- To quickly test for valve stiction, place the controller in
ficient space between flow element and control valve, etc.) manual mode and maintain a constant valve opening. If the
• mismatch between the installed thermocouple type and measured variable stabilizes, then control valve stiction could
the defined transmitter type be the source of the problem. The valve could be sticking
• incorrect calibration, such as calibrating a level mea- because the packing is too tight or because there is friction
surement for a density that does not match that of the liquid between the valve seat and disc combined with an under­
in the vessel (this can produce a significant error in the form powered actuator. Alternatively, the final control element (i.e.,
actuator, positioner, etc.) could
have a deadband that does not
Tag #1 Tag #1 Tag #1 Tag #2 Tag #2 Tag #2 Tag #3 Tag #3 Tag #3
Timestamp SP PV Error SP PV Error SP PV Error initiate a valve movement until
6/11/19 00:00 100 102 –2 50 50 0 800 801 –1 a threshold value is reached.
6/11/19 00:01 100 101 –1 50 51 –1 800 802 –2 This can be easily adjusted,
6/11/19 00:02 100 99 1 50 52 –2 800 801 –1 but is occasionally set too wide
6/11/19 00:03 100 98 2 50 55 –5 800 800 0 by well-intentioned instrument
6/11/19 00:04 100 98 2 50 53 –3 800 800 0 engineers who are trying to
• • •
• • • • • • • extend the valve life by mini-
• • • • • • • • • •
mizing thrashing of the valve.
• • • • • • • • • •
0 50 1
A stroke test is often used
6/17/19 23:55 100 100 49 810 812 –2
–1 50 3 811
when troubleshooting this
6/17/19 23:56 100 101 47 810 –1
6/17/19 23:57 100 102 –2 50 48 2 810 810 0 issue. However, large changes
6/17/19 23:58 100 101 –1 50 50 0 810 809 1 in output during a stroke test
6/17/19 23:59 100 100 0 50 52 –2 810 810 0 usually apply enough force to
Standard
1.40 2.28 0.92
move even the stickiest valve,
Deviation (SD) which will mask the problem.
SD/Range Range = 200 0.007 Range = 100 0.023 Range = 300 0.003
Small incremental changes
p Figure 2. The normalized standard deviation is another statistical analysis tool for evaluating controller in output should be used in
performance. In this example, Tag #2 has the highest normalized standard deviation and thus should be such cases.
prioritized for troubleshooting first. Mismatch between the

Copyright © 2020 American Institute of Chemical Engineers (AIChE). CEP May 2020 www.aiche.org/cep 57
Not for distribution without prior written permission.
Back to Basics

controller output signal and the actual valve position can the field rather than in a common location (e.g., the DCS),
result in a loss of control range for the valve. If the problem a variety of possible control equation structures may be
continues to reoccur after recalibration, then installation of a employed at a single plant site due to the different smart
smart positioner might be warranted. device vendors using different control equations. It is there-
In addition, each valve has a particular type of trim that fore important to confirm which control equations are used
provides for more reliable response (i.e., linear) in a particu- in which devices.
lar region of the valve opening (Figure 4). If the valve has the Control action and valve failure mode. If a controller
wrong trim for the application or if the process can operate in is always in manual mode or is unstable when it is not in
multiple regions of the valve opening, then the valve may not manual mode, check that the control action is configured
be operating in its optimal region and the controller perfor- properly. Controllers are either specified as direct or reverse,
mance can degrade. If the trim cannot be changed, adaptive which defines whether the controller output increases
tuning may solve the problem. (direct) or decreases (reverse) when the measured process
Control equation. Within a vendor’s control system, it variable increases. If the controller is designated with the
may be possible to select one of several control equations. wrong control action, it will become unstable almost imme-
Although these equations have the same structure, they con- diately upon activation — within minutes in most cases.
tain subtle differences that affect the control loop response. Also check the valve failure mode in combination with
Depending on the equation selected for a particular applica- the control output processing. Output signals from the
tion, each of the control terms — i.e., the proportional (Kp), control system to the valves in the field are normally dis-
integral (Ki), and derivative (Kd) terms — can act on either played such that 100% represents a fully open valve and 0%
a function of the controller error (i.e., setpoint minus the represents a fully closed valve. For valves that use a signal
measured value) or a function of the process variable. to close (i.e., fail-open valves), the inversion of the control-
In most situations, the proportional and integral terms ler output is usually performed in a manner that is invisible
should act on the error and the derivative term should act to the operator. The inversion commonly takes place in the
on the process variable for two reasons. First, if the pro- positioner or is accomplished by configuration of an output
portional term acts on the process variable instead of the processing block in the control system (Figure 5). If this
error, then there will not be a proportional kick whenever inversion is not implemented properly, the effect on control
the operator changes the setpoint. This can produce a very performance is the same as if the control action was config-
sluggish response to setpoint changes, which some opera- ured incorrectly — i.e., the control response will be unstable.
tors interpret as a reason to maintain the controller in manual Setpoint tracking and initialization. If a controller
mode. Second, if the derivative term acts on the error, then exhibits a significant disturbance when it is initially placed
the derivative action (which responds to rate of change) can in service but eventually stabilizes, then it is possible that the
overreact to setpoint changes and cause a rapid rate of change controller’s setpoint tracking and/or initialization was not
in the error. configured properly.
With new smart devices that perform control logic in Initialization is primarily an issue with cascade and other
multilevel control strategies. In these cases, the output of
100
the primary (i.e., higher-level) controller is set equal to the
Flow, % of Rated Flow

Quick
Opening SP Signal
PV Inversion
FC FY
Performed In
Linear Output
Reverse-Acting Processing
Controller Block
Equal
Percentage FI

0
0 Valve Opening, % 100
Fail-Open Valve
p Figure 4. The control valve characteristic curves are generated by
plotting the flow (percent of rated flow) against valve opening. Each p Figure 5. A signal output inversion is usually invisible to the
control valve requires the correct trim to produce the appropriate operator. Here, it is configured in an output processing block in the
response for a particular process control application. control system.

58 www.aiche.org/cep May 2020 CEP Copyright © 2020 American Institute of Chemical Engineers (AIChE).
Not for distribution without prior written permission.
equivalent value of the setpoint of the secondary (i.e., lower- example, if an override controller is placed back in service
level) controller when the secondary controller is not in the with its setpoint equal to the process variable value, then the
mode that allows it to receive a setpoint (Figure 6). Most controller will immediately override its primary controller
control systems automatically provide this type of initializa- and perform poorly.
tion when control functions are connected to each other, Input filtering and variable clamping. If a controlled
although there are some special cases in which this does variable is highly oscillatory or extremely sluggish, it
not happen. could be due to improper input filtering. A process vari-
The more common problem is improper setpoint track- able exhibiting high-frequency signal noise (Figure 7) can
ing. Setpoint tracking refers to whether or not the value of produce unstable control by causing the proportional action
the controller’s setpoint is automatically adjusted to equal of the associated controller to overreact and rapidly make
the value of the process variable when the control algorithm large changes to the controller output. Filtering can improve
is not in service (i.e., manual mode or tracking status). If this situation, but it should only be used to alleviate high-
setpoint tracking is configured, then the initial error (i.e., frequency noise. Use of filtering to attenuate low-frequency
setpoint minus process variable) will be zero when the oscillation will elongate the oscillation and make it more
controller initializes and the controller output will initialize difficult to control.
smoothly (i.e., the control output will not bump when the On the other hand, if the control response is sluggish,
control algorithm first executes). Most controllers are config- watch out for double dipping on the filtering. Some transmit-
ured for setpoint tracking, but there are some special cases in ters have filtering applied in the field. If the signal is filtered
which setpoint tracking should not be used, such as: again in the DCS or programmable logic controller (PLC),
• override controllers (e.g., a high-level override that the signal may be made too sluggish. A first-order filtering
prevents carryover of liquid into a vapor stream) constant of 1–5 sec is normally sufficient to squelch high-
• relief controllers (e.g., a high-pressure relief controller frequency noise, regardless of where it is employed.
on a tower that sends the overhead stream to a flare if the
pressure is too high) Troubleshooting specific control loop structures
• equipment protection controllers (e.g., a minimum-flow While problems can occur in any loop, control loops
protection controller on a pump). with certain features are more prone to non-optimal imple-
In each of these three cases, the correct setpoint for the mentation that can lead to poor performance. When you are
controller is based on the safety of the equipment, which trying to identify problematic loops, pay special attention to:
in turn depends on its design. Therefore, once determined, • primary controllers with multiple cascaded secondary
the setpoint should never change in those special cases. controllers
If setpoint tracking is used on these types of controllers, • control loops with overrides
then the setpoint will reset to the current process variable • split-range controllers
value when the controller is placed in manual mode. For • level controllers with gap action
• controllers that use calculated input values
• controllers that use inputs from gas chromatog-
raphy (GC) analyzers.
Primary Controller If a problematic loop falls into one of these special
SP1 Primary When Tracking: SP1 = PV1
TC Controller
categories, you should check the controller for common
When Initializing: CO1 = SP2
configuration errors.
CO1 Secondary Controller
SP2 When Tracking: SP2 = PV2

Secondary High-
PC Frequency
Controller
CO2 Noise
PV2

Low-
Frequency
Oscillation
p Figure 6. A cascade control scheme performs slightly differently
during initialization than during standard operation. During initialization, Time
the output of the primary controller tracks the setpoint of the secondary
controller (i.e., CO1 is set equal to SP2). However, when the secondary p Figure 7. A process variable exhibiting high-frequency signal noise can
controller is in manual mode (i.e., tracking), its setpoint tracks its produce unstable control by causing the proportional action to overreact
process variable (i.e., SP2 is set equal to PV2). and make large changes.
Article continues on next page

Copyright © 2020 American Institute of Chemical Engineers (AIChE). CEP May 2020 www.aiche.org/cep 59
Not for distribution without prior written permission.
Back to Basics

Primary Primary controllers with multiple cascaded secondary

PC Controller controllers. A standard cascade control loop has one primary
controller and one secondary controller, and the output of
CO
the primary controller sets the setpoint of the secondary
controller (Figure 6).
However, when a cascade structure has one primary
SP SP controller and multiple secondary controllers, such as a tower
PV
Secondary
PV
Secondary pressure controller adjusting flow to multiple condensers
Controller with Controller with
FC
Internal Bias
FC
Internal Bias (Figure 8), the initialization mechanism is nonstandard and
requires customization. So, it is more likely to be imple-
mented incorrectly. If the two secondary controllers are
operating in automatic mode with two different setpoints,
then the output of the primary controller cannot initialize to
p Figure 8. A single tower pressure controller adjusting flow to both of the secondary controller setpoints simultaneously
multiple condensers, an example of a primary controller with multiple (since it can only be set to a single value). In this case, when
secondary controllers, has a nonstandard initialization mechanism and
requires customization. it initializes, it will send the same output to both secondary
controllers, which will change one of the setpoint values
(assuming they did not both start with the same value); this is
often referred to as bumping.
High-Level If such a cascade structure exhibits an initialization prob-
Override
Controller lem, the problem can be overcome with the use of internal
biases in the secondary controllers. The internal biases would
Vaporizer LC be applied to the setpoints of each secondary controller to
compensate for the difference between the primary control-
< FC ler output value and the individual secondary controller
setpoint value.
Note also that the gain of the primary controller will
change, depending on how many secondary controllers are in
service. In the example shown in Figure 8, when the primary
pressure controller output signal changes by 1%, flow to the
p Figure 9. In this example of an override control strategy, the process condensers will change by 1% when both of the condenser
liquid into the vaporizer is controlled at a specific flowrate unless the liquid
level in the vaporizer reaches its high-level override setpoint. If that occurs, flow controllers are in service (i.e., both secondary flow
the control output signal from the high-level override controller will become controllers are in cascade mode). However, it will change by
less than the control output signal from the flow controller, which will cause only 0.5% if one of the two condenser controllers is not in
the valve opening to be reduced (thereby reducing the flow) and prevent the cascade with the primary pressure controller. Therefore, the
level from rising higher.
gain of the control response is significantly different depend-
ing on how many of the secondary flow controllers are in
cascade mode. This can be addressed by applying adaptive
tuning logic (also called gain scheduling) to the primary
TC Primary pressure controller based on the number of secondary flow
XXXX Controller
controllers that are in cascade mode.
Override Control loops with overrides. A simple override control
SP Selector < loop has one primary controller and one override controller,
Secondary FC FY PC Override plus a signal selector that selects the higher or lower of the
Controller XXXX XXXX XXXX Controller
two controller output signals, depending on the objective of
the override (Figure 9). An override controller normally oper-
External ates with an offset between its setpoint and measured vari-
Feedback able. Therefore, unless the system is configured with external
feedback from the selector output to each of the controllers in
p Figure 10. The controllers in an override strategy must be configured the override loop, the nonselected controller typically contin-
with external feedback to update the value of the nonselected control ues to change its output (i.e., called windup) and reaches its
output so that it does not wind up (i.e., continue to change its output). minimum or maximum output value (Figure 10). Some DCS

60 www.aiche.org/cep May 2020 CEP Copyright © 2020 American Institute of Chemical Engineers (AIChE).
Not for distribution without prior written permission.
and PLC systems automatically incorporate this logic when in the response of the split-range controller in the transi-
an override control function is configured, whereas others tion region between the two valves. In this case, it may be
require the configuration engineer to specify it. In the latter necessary to overlap the action of the two valves (i.e., start
case, when it is not done properly, the ensuing perfor- opening Valve B at a controller output value of 45% and
mance issue often motivates operators to place the override do not fully close Valve A until the controller output value
controller in manual mode and set its output to either 0% or reaches 55%).
100% to avoid any interference with the primary loop. This Level controllers with gap action. A gap-action
essentially negates the existence of the override. level controller responds differently when the differ-
Split-range controllers. A split-range controller sequen- ence between the setpoint and process variable (referred
tially adjusts more than one downstream controller or valve to as controller error) is less than the specified gap than
in series based on the value of its control output signal. For when the difference is greater than the specified gap. A
example, as shown in Figure 11, the output of a split-range gap-action controller is often used to allow the level in a
temperature controller can be sent to two control valves. vessel to float between high and low limits (i.e., a gap)
As the temperature increases, Valve A is closed from 100% without taking any control action to change the flow to the
open to 0% open as the controller output value changes downstream equipment.
from 0% to 50%. Then, Valve B opens from 0% open to This type of strategy is often applied in a manner that
100% open as the controller output value changes from increases instability in the system. Applying gap-action
50% to 100%. level control with a zero gain inside the gap subjects the
Split-range controllers are notorious for working well process to an integrating effect. For example, a step change
some of the time, but not always. There are several poten- in the feed flowrate to the vessel will eventually overflow
tial causes (and remedies) for this issue. or empty the vessel if no adjustment is made to the outlet
If the process response of the two valves is different — flow (Figure 12). As a result, the level may bounce between
i.e., changing one of the valves by 1% changes the value the gap limits and create large changes in the flow to the
or dynamics of the process variable differently than a 1% downstream equipment when crossing the gap limits. Using
change in the other valve — then a single set of tuning con- a smaller, nonzero gain inside the gap may minimize the
stants may not be optimal for performance across the entire effect on downstream equipment and eliminate the continu-
range of operation. In this case, adaptive tuning, gain, or ous oscillation caused by frequently bouncing between the
scheduling may be required. two gap limits. A larger gain would be used outside of the
If the installed valves have very little sensitivity at gap limits to handle severe disturbances that occur much
the ends of their valve opening range — i.e., if there is less frequently.
very little change in flow between a valve opening of 0% Controllers that use calculated process variables. The
to 5% or 95% to 100% — then there will be a deadband power of the DCS has encouraged the implementation of

FC1 If LC is not in service, then the level

(B) measurement will integrate if there
is a step change in FC1 flow.

100
LC
Response

100 LC

Low-Pressure FC2
Valve Position

Steam FC2
TC
(A) (B) FC1
(A) Valve 0
Valve Time

0 Control Output 100

p Figure 12. This example shows the integrating effect of a simple
level control response. Unlike typical process control responses, levels
do not achieve a new steady-state value when a disturbance is imposed.
p Figure 11. A split-range controller adjusts more than one controller or If the liquid feed rate (i.e., FC1) to the vessel increases while the liquid
valve in series. Here, the temperature controller closes the steam valve (A) outlet flow (i.e., FC2) remains constant, then the liquid level (i.e., LC) will
first and then opens the bypass valve (B) to reduce the temperature. continue to increase until the liquid overflows the vessel.

Copyright © 2020 American Institute of Chemical Engineers (AIChE). CEP May 2020 www.aiche.org/cep 61
Not for distribution without prior written permission.
Back to Basics

many loops that use a calculated value as their process vari- with the update frequency of the analyzer input, several
able. The basic calculation and control is normally quite problems can occur.
straightforward, but these loops are vulnerable to a few If the controller that is using the analyzer input is run-
hidden pitfalls. ning continuously (i.e., executing once a second), it will be
Consider the simple example of a flow controller that subject to windup, which will produce oscillatory behavior.
is compensated for temperature and pressure shown in Alternatively, if the controller has been detuned to compen-
Figure 13. The pitfall occurs when the quality of the non- sate for the mismatch in execution timing, it may exhibit a
critical variable (i.e., temperature or pressure) signal drifts sluggish response.
out of its normal range or changes to a bad status due to If the controller runs intermittently, on the same fre-
instrumentation error. Depending on how the configuration quency as the analyzer, other problems can occur:
was designed, one of several things could happen: • if the signals are not synchronized, additional dead
• the calculated flow reading is inaccurate time will be introduced into the loop, which will cause
• the calculated flow value experiences a bump that sluggish response
causes the controller to respond erroneously • if the analyzer freezes (i.e., stops sending an updated
• the controller is automatically changed to manual value), the controller output will wind up — its output will
mode operation. continue to change and drive the controller output and the
These problems can be mitigated by applying the fol- process variable to an out-of-range value.
lowing logic: To avoid these problems, it is best to trigger the execu-
• clamp the noncritical inputs within reasonable bounds tion of the controller whenever a new analyzer value is
to protect against signal drift sent. This can be done directly if the analyzer also sends
• install logic to continue to use the last good value of a a digital bit indicating that a new value has been sent or
noncritical input if its status becomes bad (which allows the indirectly by continually monitoring the analyzer output for
controller to continue to operate based on the value of the a change in value (to several significant digits) and generat-
critical input). ing a digital bit trigger.
A heat duty controller is another example of a controller If the DCS or PLC functionality does not allow such
with a calculated input. triggering of controllers, then watchdog timer logic should
Controllers with measured process variables from GC be used to determine if the analyzer value has frozen and, if
analyzers. The input from a GC analyzer requires special so, change the controller to manual mode.
processing; such processing steps may have been neglected
during controller implementation. A GC analyzer sends Closing thoughts
a new analysis value on an intermittent basis (e.g., once Achieving the full potential performance of a petro-
every 3–5 min, or longer if the analyzer is multistreamed). chemical plant depends on proper operation of the process
If the controller that uses that signal is not synchronized equipment, which is closely aligned with the effectiveness
of the plant’s control system. Prior to investing in advanced
Fuel Gas Flow
process control (APC) technologies to close the gap between
Compensation actual and potential plant performance, a rectification of
Factor the basic regulatory control system may identify several
X SP opportunities for improved performance at very low cost.
FY FY PV FC Moreover, periodic reviews of the basic regulatory control
101A 101B 101 system will identify more opportunities to maintain peak
performance throughout the lifecycle of the plant. CEP

PI TI FI
101 101 101
BRIAN K. SULLIVAN is Director of Aftermarket Services at Lummus Technol-
ogy, where he specializes in technical and operational support (includ-
ing troubleshooting process control issues) for petrochemical plants
throughout the world. He has nearly 40 years of experience in the
hydrocarbon process industries and has been with Lummus Technology
for the past 20 years. Previously, he was Director of Advanced Control
p Figure 13. This controller uses calculated process variables. The for ABB Simcon, where he specialized in implementation of process
control projects for hydrocarbon processing plants. Sullivan holds a BS
fuel gas flow is compensated for temperature and pressure. When the
in chemical engineering from Rensselaer Polytechnic Institute and an
noncritical variable (temperature or pressure) is out of range, the flow MBA from Montclair State Univ.
calculation may be inaccurate.

62 www.aiche.org/cep May 2020 CEP Copyright © 2020 American Institute of Chemical Engineers (AIChE).
Not for distribution without prior written permission.