Chapter 6 Control

Download as pdf or txt
Download as pdf or txt
You are on page 1of 148
At a glance
Powered by AI
The document discusses various topics related to process control including classical control methods, the influence of process design on control, multivariable process systems, and estimating benefits from process control.

Some classical control methods discussed include single-loop control, cascade control, feedforward control, control of non-square process systems, and inferential control systems.

Factors that influence the design of a control system include process equipment in the feedback loop, process elements in the disturbance path, instrumentation elements in the feedback loop, process structure, and control performance goals.

Operability in process design

Chapter 6 Process Control

Operability in Process Design:


Achieving Safe, Profitable, and Robust
Process Operations
Chapter 6 Process Control

L
2

LAH
LAL

LC
1

FC
1

TAH

T
7

FSL
7

TC
2

TAH

F
4

P
2

TC
1
TC
20

fuel
T
10
T
12
T
13

Thomas Marlin
6-1

T
11

>
TY
15

Operability in process design

Chapter 6 Process Control

Process Control release 2.0 on December 2.0 2012

Copyright 2012 by Thomas Marlin


This document is copyrighted by Thomas Marlin. Content is not to be reproduced or
redistributed without the expressed consent of the author.
License for university use
A cost-free license is granted for use at not-for-profit universities. The material may be used for
classroom display, and students may store one copy in electronic or hard copy for their personal
use. No fee may be charged for distribution of copies beyond the cost of copying. Any use of
the material in part or in whole must include a citation of the source.
License for non-university use
For other use of the materials, including any commercial use, please contact T. Marlin at:
[email protected]

Acknowledgements

Peggy Hewitt for assisting in obtaining a control room picture.


Tariq Samad and Russ Rhinehart for assistance with permissions from the American
Control Conference
Everyone who posts materials with the Creative Commons license

Disclaimer

While care has been taken in the preparation of the information contained in this chapter, the author cannot
guarantee its accuracy or applicability for a specific application. Persons accessing and using this information
do so at their own risk and indemnify the author from any and all injury or damage arising from such use.

6-2

Operability in process design

Chapter 6 Process Control

Table of Contents
Section

Page
5
7

Symbols
Nomenclature
6.0
6.1
6.2
6.2.1
6.2.2
6.2.3
6.2.4
6.2.5
6.3
6.3.1
6.3.2
6.3.3
6.3.4
6.3.5
6.4
6.4.1
6.4.2
6.4.3
6.4.4
6.5.0
6.5.1
6.5.2
6.5.3
6.5.4
6.5.4.1
6.5.4.2
6.5.4.3
6.5.4.4
6.5.5
6.6
6.6.1
6.6.2
6.7

To the Student
Basics of Process Control
Classical Control Methods
Single-loop control
Cascade control
Feedforward control
Non-Square Process Systems
Inferential Control Systems
The Influence of Process Design on Control
Process equipment in the feedback loop
Process elements in the disturbance path
Instrumentation elements in the feedback loop
Process structure
Control performance goals
Multivariable Process Systems
Which Variables can be Controlled?
Operating Window
Multi-loop Systems: Interaction
Multi-loop Systems: Control Loop Pairings
Control System Design Procedure
Defining the Control Design Problem
Designing the Control System
Application to the Flash Process
Unique Features of Plant-wide Control
Production Rate and Inventory Control
Utilities Control
Control of Process with Recycle
Partial Control
Final Step Safety Review
Unit Operation Control
Flow Control
Heat Exchanger Control
Conclusions

115
116

Additional Learning Topics and Resources


References
(Continued on next page)

6-3

9
10
14
15
21
22
24
28
32
35
45
48
53
54
59
59
63
64
70
77
78
811
83
87
87
90
94
97
99
99
99
105
112

Operability in process design

Chapter 6 Process Control

Appendices
6.A
6.B
6.B.1
6.B.2
6.B.3
6.B.4
6.B.5
6.B.6

Understanding Frequency Response


Benefits from Process Control
Introduction
Benefits Calculation
Issues in the Benefits Calculation
Predicting Future Performance
Conclusions
Additional Learning Resources

6-4

119
125
125
126
133
140
146
147-8

Operability in process design

Chapter 6 Process Control

Symbols

6-5

Operability in process design

Chapter 6 Process Control

CENTRIFUGAL
FAN

FILTER AND DRAIN


CYCLONE

HEADER (STEAM OR
CONDENSATE)

T
M

DAMPER
OR
BUTTERFLY VALVE

STEAM TRAP

STEAM
EJECTOR
MOTOR

DRAIN TO SEWER
FLARE STACK
POSITIVE
DISPLACEMENT
PUMP

SHELL AND TUBE


HEAT EXCHANGER
CONDENSER

FAN
OR
BLOWER

TURBINE FLOW
METER

FAN

ORIFICE PLATE

TURBINE USED TO
PROVIDE WORK
TO DRIVE
COMPRESSOR

IMPELLER

VAPORIZER
OR
REBOILER

6-6

Operability in process design

Chapter 6 Process Control

Nomenclature
A
B
BPCS
CP
CSTR
CV
D
D1
D0
E
Fj
h
HAZOP
I
Iv
IAE

Area for heat transfer


Benefits for the change in dynamic performance ($/y)
Basic process control system
Heat capacity
Continuous (flow) stirred tank reactor
Controlled variable
Disturbance
Distribution of the controlled variable under new control strategy
(histogram)
original (base case) distribution in controlled variable (histogram)
Error = SP - CV
Fraction of data in histogram in range j
Film heat transfer coefficient
Hazard and Operability Study
Initialization constant in PID control algorithm
(also termed bias)
Incremental value of change in process performance
Integral of absolute value of error
|

ISE

Integral of the squared error

K
KC
Kd
LOPA
M

Process gain
Controller gain
Disturbance gain
Layer of Protective Analysis
Correction factor in benefits calculation for other operating
conditions, e.g., production rate, (dimensionless)
Manipulated variable
Piping and instrumentation drawing
Proportional-integral-derivative controller
Process performance appearing in benefit calculation
Heat transferred
Service factor, the fraction of time that the control strategy is
improving the process performance (dimensionless)
Safety instrumented system
(sometimes referred to as safety interlock system)
Set point

MV
P&ID
PID
PR
Q
SF
SIS
SP

6-7

Operability in process design

T
T
Td
TI
U
V(*,*)

Chapter 6 Process Control

Time
Time when the control strategy should be in service (h/y)
Controller derivative time
Controller integral time
Overall heat transfer coefficient
Improved economic process performance at the base case operating
conditions ($/h), based on two arguments (D0, D)

Greek symbols

Relative gain array element


Dead time
Density
Time constant
frequency

6-8

Operability in process design

Chapter 6 Process Control

Chapter 6. Process Control


6.0 To the Student
Imagine that you are the operator of a complex chemical process without process control. You
introduce a change to the raw material feed rate to the process. In response, you would have to
implement changes to 10s to 100s of additional valve openings, affecting condenser cooling
and reboiler heating, solvent flows, boiler fuel and airflows, recycle flows, and many more. You
would not only have to introduce these changes in the correct magnitude and direction but also at
the correct times as the production rate change coursed through the process. This demanding
task would consume all of your time, would lead to periodic human error, would lead to frequent
large deviations in product quality, and could occasionally result in unsafe operations.
Fortunately, there is an alternative provided by process control! Process control is
essential for achieving safely, reliably, and desired process conditions as disturbances occur in
the plant. While engineers and operators understand the needed actions, only precise automation
through process control can implement actions as rapidly and reliably as required in demanding
chemical processes. In fact, the entire development of the technology of automatic control has
been in response to demands for controlling steam engines, airplanes, electronic circuits,
chemical plants, and many other complex systems.
Since you have already completed a course covering process control, this chapter
presents complementary materials that address some key design concepts. You may be surprised
(and pleased) to find that this coverage has little mathematical sophistication. This decision is
not made because dynamic modeling, simulation and controller calculations are unimportant but
because you have already mastered these topics in your process control course.
This is process control without the mathematics, but it requires deep understanding
and careful thought to successfully apply the topics to control design.
Little math Easy to do without preparation, skills, and knowledge!

This material emphasizes designing process and control structures. It explains process
characteristics that make process control challenging (or easy), and it presents control structures
to achieve the best dynamic performance possible given the process design and disturbances
occurring. Naturally, process control is realized with physical equipment for sensing, final
elements, signal transmission, human interfaces, and computation. Some of the key aspects of
control equipment are addressed in Appendix A.
This material supports all topics in operability, including product quality, safety,
reliability, and troubleshooting. So, lets learn some more process control!

6-9

Operability in process design

Chapter 6 Process Control

6.1 Basics of Process Control


Why is process control necessary in a chemical process?
following; can you think of others?

Some reasons are given in the

Plants are physically large, so that adjustments and data collection must be managed from
centralized locations
Materials can be hazardous and are often maintained at extreme conditions, e.g., high
pressures and temperatures
Equipment functions successfully without damage over only a limited range of
conditions, so that excursions outside of acceptable ranges must be avoided
Demands for product quality and safety require rapid and precise process adjustments
that are often beyond the capability of plant personnel
People must be relieved of high frequency decision making, so that they can perform
more complex analyses that are better performed by people

Process control involves a large and continuously expanding array of technology. Here, we will
address the technology that is implemented in a typical process control design. This technology
is based on one basic principle, feedback.
Feedback uses information from system outputs for deciding adjustments to system
inputs.
The use of system outputs requires measurements of process variables that are influenced
or caused by adjustable variables. The selection of output variables for measurement is critical
to success and will be addressed throughout the chapter. Manipulated input variables can be
adjusted by a person or computer; for example, a valve opening is an acceptable input variable.
In contrast, disturbance input variables cannot be adjusted; an example would be raw material
composition.
The schematic in Figure 6.1 shows a feedback control loop with limited detail, containing
the essential three elements of sensor, control calculation and final element. The loop requires
inputs from plant personnel in the form of controller tuning constants and the set point that
defines the desired value for the variable. Then, the controller functions essentially continuously
by adjusting the valve to bring the controlled variable to its set point. However, process control
does not result in a plant running on automatic pilot. Since control systems involve complex
equipment that can fail to operate properly, plant personnel monitor the performance of process
and control equipment and intervene when they diagnose a fault.

6-10

Operability in process design

Chapter 6 Process Control

Figure 6.1 Schematic of Feedback


control loop
Figure 6.1 pictures the concepts of control, but engineers need to design equipment to realize
these concepts. Lets begin with a single control loop shown in Figure 6.2 to determine the key
equipment common to all control systems. The sensor generates a signal that is proportional to
the measured variable; in the example, a thermocouple produces a millivolt signal that is
(approximately) proportional to the temperature. The sensor signal is converted in a transmitter
to an alternative signal that can be accurately transmitted longer distances and is compatible with
other equipment in the loop; the transmitted signal would typically be an analog current (4-20
mA) or a digital signal. The transmitted signal is converted for use in the calculation equipment;
this could be voltage for analog equipment or a decimal number for digital control. In addition
to the controller calculation, the measurement signal is used for display and historical storage

Figure 6.2. Typical equipment for a single-loop controller, with human-machine interface is not shown.
Note that signal transmission and control calculation control equipment can be based on analog (shown
here) or digital technology.

6-11

Operability in process design

Chapter 6 Process Control

and can be used for calculations. The controller output is converted for transmission to the final
element. Near the final element, the transmitted signal is converted to affect the final element. In
many process applications, the final element is a control valve, and the signal must determine the
force provided by compressed air. Thus, the transmitted signal is converted to an air pressure.
The pneumatic signal is applied to the final element that changes its stem position and opening
for flow. Every control loop in a plant has its own individual local equipment, sensor, pneumatic
converter, and final element. The control calculation is performed digitally in equipment
manufacturer in the last thirty years. (Control equipment has a long life, so you might encounter
analog control equipment that performs the PID calculations via an electronic circuit.) The
transmission until recently has been achieved using an individual wire carrying an amperage for
each signal. More recently, digital transmission has been introduced for transmission.
A large segment (or the entirety) of a process plant is controlled in a centralized location,
where a few people can observe all measurements and make adjustments throughout the plant.
The centralized control house enables coordinated actions, but it requires long transmission.
Fortunately, electronic and digital signals can be transmitted with essentially no delay. A picture
of a typical centralized control room is shown in Figure 6.3.
All sensors are located at the process equipment, while displays of the measurements can
be located either at the equipment or centrally, or both. Local displays of measurements are
essential for plant personnel who are performing maintenance and are monitoring the equipment.
For example, when an operator starts a pump, s/he wants to observe the outlet pressure and
perhaps, the flow as well, to ensure that the equipment is working properly. However,
coordinated analysis and control of the entire plant requires that most measurements be
transmitted and displayed in the centralized control room. Most of these will be recorded on
trend plots to provide a display of the recent dynamic behavior. When appropriate, a
measurement can be displayed both locally and remotely.

Figure 6.3 Picture of a typical centralized control room. (Photo courtesy of Worsley Alumina)
6-12

Operability in process design

Chapter 6 Process Control

Modern control equipment in the centralized facility includes a network of computers.


The network has the following advantages (over a single, centralized computer).

Parallel computation ensures minimal computing delay in control loops.


The behavior for a failure is superior. Even though the probability of failures is higher,
because of the increased number of computers, the impact of a failure is much lower
since only a small section of the plant would be affected. To further improve reliability,
most digital equipment is redundant with automatic switching upon failure detection.
Computer software and hardware can be tailored in each module for specific functions,
like process control, complex and flexible computations, history storage and display,
safety functions, and so forth.
The computing system can be designed to match a plant, without excess capacity, while
allowing subsequent modular expansion.

It is important to recognize that control systems have many preprogrammed functions, so that
plant engineers do not program PID control algorithms, details of graphical displays, and so
forth. Most control systems require configuring calculations, displays and history storage
using existing functions.
We conclude this section with a brief discussion of drawings that are used to document
designs. We have to recognize that complex designs could not be documented using written
descriptions alone. Drawings are widely used as a basis for construction, and there are many
forms of drawings, including Block Flow, Mechanical Detail, Piping and Instrumentation
Drawing (P&ID), and Isometric (3-D) Layout. A clear explanation of process drawings is
provided with examples by Turton et. al. (2012). Here, we will concentrate on the P&ID, whose
major characteristics are given in Table 6.1. The P&IDs are used during day-to-day operation
and for safety studies; therefore, the P&ID must be maintained up-to-date as changes are made to
the original design and construction.
Various levels of detail are presented in a P&ID. Limited information is available during
the preliminary design of the process, so that the P&ID does not include as much detail
concerning the sensors and control implementation. The drawings in this educational material
will tend to follow the preliminary P&ID level of detail. Preparing P&IDs is facilitated by
special-purpose software that includes a library of process-related symbols. Perhaps the best,
low-cost software for use by university students is MS Visio.
Prior to addressing control technology, we need to refresh our understanding of two
aspects of basic process control. The first aspect involves classical control methods, which are
used widely in chemical processes; this aspect is addressed in the next section. The second
aspect is process control equipment, i.e., instrumentation; the reader is referred to Appendix A
for a review of instrumentation. This information provides the platform for designing process
and control structures.

6-13

Operability in process design

Chapter 6 Process Control

Table 6.1. Typical features of Piping and Instrumentation Drawings (P&ID)*


P&ID contain the following

P&ID does not contain the following

All piping and equipment connections

An approximate location for connections


(e.g., top or bottom of tank, tray location,
etc.)
Equipment identification (numbers)

The size of piping

All sensors (whether locally or remotely


displayed and recorded) and whether used
for an alarm, with priority
All valves (whether automated or manual),
including failure position if remotely
operated valves
Control strategies, as much detail as
possible graphically. These can be
regulatory and safety-related
Whether signals and control calculations
are implemented using analog or digital
equipment

The distance between objects. The


drawing is not to scale.
The vertical or horizontal (or 3D) position
of objects
The sizes of objects(e.g., vessel), not even
the relative size
The exact design for piping connections,
including those to vessels
Sensor details such as physical principle
(e.g., thermocouple) and measurement
range
Details of the control calculations when
involving complex logic and/or
calculations
Any detail about the human interface
display or the type of historical data
Operating policy (which appears in a
separate operations manual)

* Various levels of detail are presented in a P&ID, depending on the status of the design (preliminary to definitive)
and company practices.

6.2 Classical Control Methods


Classical control refers to a collection of technology that was developed over many decades and
applied successfully in the process industries. Because of limited computing during the early
decades of the twentieth century, classical control is founded on the PID feedback algorithm.
While powerful and flexible, simple single-loop feedback is not alone adequate to achieved
required dynamic performance. Therefore, a number of enhancements were developed that
complemented single-loop PID; the enhancements addressed in this section are cascade control,
feedforward control, non-square system control with PIDs, and inferential control. The methods
in this section are required to design industrial strength classical control. The classical control
methods are introduced briefly in this section, since most of the material refreshes a typical
process control course. (Even though it was a great course, we do not want to review when we
can learn new topics!) The topics on non-square systems and inferential control are presented in
more detail, because they are not usually included in the undergraduate course and are essential
for industrial control designs.

6-14

Operability in process design

Chapter 6 Process Control

6.2.1 Single-loop Control


Single-loop control involves controlling one measured process variable using a controller that
adjusts one final control element. Process control can be achieved through the application of
many single-loop controllers, one for each important controlled variable. Generally, each control
calculation shares no information with any other control calculation; it has blinders on.
Single-loop control has been the paradigm for process control for over two hundred years,
because it provided adequate dynamic performance with simple real-time calculations, and it
continues to be the dominant approach to industrial control. Later in the chapter, enhancements
to the single-loop paradigm will be introduced. These enhancements are essential for the
challenges posed by complex processes.
Nearly all single-loop feedback control is implemented with the Proportional-Integral
Derivative (PID) algorithm. This algorithm was developed originally to give good performance
while requiring only simple computation. Remember that process control was needed and was
applied long before digital computation became practical. Perhaps surprisingly, the algorithm
has proved to be very good for most single-loop applications and remains the most widely
employed in spite of todays powerful digital control systems.
The continuous PID algorithm is implemented in pneumatic and electronic calculation
systems, and the discrete PID algorithm, which approximates the continuous using standard
numerical methods, is implemented in digital control systems. Both of the algorithms are given
in the following.
(6.1)

E(t) = SP(t) CV(t)


Continuous:

1
MV (t ) K c E (t )
TI

E (t ' )dt 'Td


0

d CV
I
dt

(6.2)
(6.3)

Discrete:

where
CV = controlled variable
MV = manipulated variable
SP = set point
E = error
I = constant of initialization (bias)

t = time
t = execution period
n = current controller execution counter
(integer number)
Kc = controller gain
TI = controller integral time
Td = controller derivative time

6-15

(6.4)

Operability in process design

Chapter 6 Process Control

Many variations to these algorithms are used in practice. Some of the variations have (a)
proportional on measurement, CV, (not error), (b) filtered measurement for the derivative
calculation, and (c) interaction between the proportional and integral modes (Witt and
Waggoner, 1990).
The tuning constants in the PID algorithm (KC, TI, and Td) are adjusted to achieve good
performance for each control loop. Many correlations are available to determine the constants
based on the dynamic model for the loop, i.e., the response of the CV to a change in the MV.
When selecting tuning methods, the engineer should be aware of the following.

There is no one method that will give good performance for all loops because each loop
has unique performance goals, measurement noise and model uncertainty.
All tuning correlations should be considered an opening gambit that helps the engineer
in an iterative procedure.
Proper tuning depends on the form of the PID algorithm, and as previously mentioned,
many forms are in use industrially. Therefore, the engineer must match the appropriate
tuning correlation with the applicable version of the PID algorithm.

Since control performance is the ultimate goal, lets discuss loop performance for a
moment. Some of the key performance factors are discussed in the following.

Controlled variable behavior We want the controlled variable to remain close to


(ideally, exactly equal to) the set point. Performance is often measured by the IAE
(integral of absolute value of error) for single step disturbances or variance for a long
sample of data. Other issues can also be important; for example, large deviations (single
overshoot or oscillations) can be especially deleterious to performance, so that
overshooting a set point by a large amount during a disturbance should be avoided.
Manipulated variable behavior The controller must adjust the manipulated variable to
compensate for disturbances, but very aggressive adjustments are undesirable in some
processes. For example, boiler pressure is typically controlled by adjusting the fuel flow
rate to the combustion flame. If the fuel is adjusted rapidly and frequently, the boiler is
subjected to thermal stresses through rapid expansions and contractions that over time
that can damage the equipment and require expensive shutdown and repairs. In addition,
measurement noise is propagated by the controller to the manipulated variable, so that
tuning constant values need to be selected to moderate the propagation of noise.
Robustness The process dynamics are not known exactly, and the dynamics change,
sometimes substantially, because of changes in the process operating conditions. For
example, the changes in production rate affect the dynamics on all control loops in the
plant. When the controller tuning is constant but the loop dynamics changes, the control
loop performance can degrade and even become unstable. Therefore, the controller
tuning must be robust to process dynamics changes. Tuning with good robustness
performs well (or acceptably) over a range of loop dynamics, while poor robustness
yields rapid degradation as the dynamics change from the base case.

6-16

Operability in process design

Chapter 6 Process Control

We see the key conflict in feedback control. Keeping the controlled variable near its set point
requires aggressive adjustment of the manipulated variable, while robustness and limitations to
MV variability favor more moderate adjustment of the manipulated variable.
Achieving good control performance involves a tradeoff among competing performance
objectives. The importance of each goal and the circumstances in which it is achieved
(e.g., amount of measurement noise and model mismatch) differs. Therefore, the
engineer must evaluate each loop to determine the proper process and control design,
control algorithm, and controller tuning.
Example 6.1 Good Loop Performance - The process considered in this example involves
mixing two streams and has a pure dead time due to transportation delay and a first-order mixing
tank. It is shown in Figure 6.4. A typical dynamic response plot for this single-loop PI control is
shown in Figure 6.5.
The dynamic response is given in Figure 6.5 for a step change in the controller set point at 2.5
minutes. We note the following.
- The immediate response is a step in the manipulated variable due to the proportional mode.
- After the set point change, the controlled variable does not respond for the duration of the
dead time. During this time, the manipulated variable changes due to the integral mode
- For typical loop tuning, the manipulated variable will slightly overshoot its final value and
the controlled variable will slightly overshoot the set point.
- Finally, the controlled and manipulated variables achieve steady state. The controlled
variable is equal to the set point because of the integral mode.

Figure 6.4 Single loop PI control for Example 6.1.


(This is not a good process design; we should minimize dead time and time constants in the feedback process, but it
is used here for demonstration purposes because of the clear relationship between the pipe length and tank volume
and the feedback dynamics.)

6-17

Operability in process design

Chapter 6 Process Control

Figure 6.5 Typical good feedback response to a step set point change.
(Variables plotted as deviation from initial steady state.)
Example 6.2. Control performance diagnostics - Lets apply the insights gained in the
previous example to diagnose control loop behavior. The situation involves a loop that has
typical performance goals, so that a pattern similar to that in Figure 6.5 is desired. The
performance with current tuning is given in Figure 6.6. Decide (a) whether the performance is
acceptable or not and (b) if not acceptable, determine as much possible, the cause of the
unacceptable performance. Can you suggest a corrective modification?
In diagnosing a control loop, we first evaluate the performance of the instrumentation. For example,
valve stiction and hysteresis could seriously degrade control performance. We will assume that this
has been done and that that the instrumentation is functioning well. Our diagnose proceeds as follows.
-

The control loop appears to be stable, which is essential.


The controlled variables behavior is too oscillatory, as is the manipulated variables
behavior. This suggests that the controller is too aggressive for the process dynamics.
Too aggressive would be caused by the controlled gain that is too large, the integral time
that is too small, or both. (Remember that the integral time is in the denominator.)
We would like to evaluate the tuning.
We see that the proportional kick when the set point is initially changed is not too large;
the manipulated variable changes near to but less than its final steady-state value.
Therefore, we conclude that the proportional gain (KC) has a reasonable value.

The manipulated variable has excessive overshoot. Since the controller gain is OK,
we conclude that the integral time is too small. (Remember that the integral time is
in the denominator.
We would increase the integral time (by about 50%) and perform another set point
step test.
Continue until good performance is achieved.

6-18

Operability in process design

Chapter 6 Process Control

Figure 6.6 Initial control loop


performance for Example 6.2
Example 6.3 Tuning for plant with variable dynamics Generally, the PID controller will
operate for months or years with one set of tuning constants. During this time, the process
dynamics will change because of changes to operating conditions, like the production rate.
Therefore, the tuning constant values should be the best performance for a range of operating
conditions. The goal will be to minimize the IAE for a range of plant realizations (samples from
the range of parameters), while observing a limitation on the MV variation.
Lets consider the following example.

Nominal process model:

with the following nominal values


process gain
dead time
time constant

= KP
=
= P

= 1.0
= 5.0 minutes
= 5.0 minutes

Assume for a minute that the process


dynamics were constant and known exactly.
Then, the tuning constants yielding minimum
IAE and the transient responses are given in
Figure 6.7.

LimittoMVvariation

Figure 6.7 Transient response for nominal


model and KC= 1.18 and TI = 8.59 min.
Realistically, the process conditions change. From plant operating experience, we expect that
the parameters change about 35% from their nominal values and in a very highly correlated manner,
as would often be the case when the process change is production rate. The best tuning and transient
responses are given in Figure 6.8.

6-19

Operability in process design

Chapter 6 Process Control

LimittoMVvariation
LimittoMV variation

a. Plant has - 35% all parameters

b. Plant has nominal parameters

LimittoMVvariation

c. Plant has + 35% all parameters

Figure 6.8 The best tuning when 35% dynamic model parameters are possible. Variables expressed as
deviation from initial steady state. Tuning: KC = 0.77 and TI = 7.65 min

We observe the following from the previous three examples.

Perfect control, maintaining the controlled variable at its set point is not possible, because
of feedback dynamic delays.
The controller can be adjusted to be more or less aggressive in its feedback
compensation. In general, more aggressive feedback improves the controlled variable
performance and degrades the manipulated variable performance.
The controller tuning must be selected to match the nominal process dynamics,
variability in dynamics, and control objectives.
Very aggressive feedback can lead to undesirable oscillations and potentially, instability.

Although we will not demonstrate it here, these results are generally true for all feedback
controllers.
These examples demonstrate that the major factor influencing control performance is
the process dynamics and that proper controller tuning is an important factor in
enabling the control loop to perform as well as possible for the process dynamics.

6-20

Operability in process design

Chapter 6 Process Control

6.2.2 Cascade Control


When the feedback dynamics are slow and a single-loop controller cannot achieve the desired
performance, single-loop control can often be enhanced to achieve much better performance with
low-cost and simple technology. The two most common are cascade and feedforward control;
cascade is introduced briefly in this section, with much more detail available in, for example,
Marlin (2000).
In many processes, an intermediate variable exists that provides an early indication of an
important disturbance, and we would like to use the early indication of a disturbance to improve
control performance. Cascade accomplishes this using a hierarchical, feedback approach. Lets
consider the stirred tank heating process in Figure 6.9 where the outlet temperature is controlled
by adjusting the heating temperature. In this process, common disturbances include heating
medium pressure variation and lack of precise valve stem positioning. Is there a measurable
variable that would indicate that these disturbances have occurred? The answer is clearly yes;
the heating medium flow. We are half way to designing a cascade controller.
Since cascade involves a hierarchy of feedback controllers, a causal relationship must
exist between the control valve and the intermediate or secondary measured variable. By
observation, we confirm the causal effect, so that a feedback controller can be implemented
controlling heating flow by adjusting the valve. This controller will quickly correct for
disturbances in the heating medium pressure and valve stem position errors, but it will not
achieve the desired temperature control. Therefore, we direct the output of the temperature
controller to the set point of the heating flow controller. This is a hierarchy and is called a
cascade control system. The design is sketched in Figure 6.10, and a comparison of single-loop
and cascade control performance is given in Figure 6.11; clearly, cascade control performs
better, with an IAE reduction of over 90% from single-loop feedback.

(b)

(a)

Figure 6.9. Single-loop feedback.

Figure 6.10 Cascade control of the stirred


tank heater outlet temperature

6-21

Operability in process design

Chapter 6 Process Control

Figure 6.11 Comparison of (a) single-loop and (b) cascade control for the same process and
disturbance. Cascade performs better for the same disturbance.
A properly designed cascade control system requires that certain criteria be satisfied.
These five criteria are summarized in Table 6.2 that the engineer can apply when deciding
whether cascade is applicable. Standard PID controllers can be used in cascade control.

6.2.3 Feedforward Control


When the feedback dynamics are slow and a single-loop controller cannot achieve the desired
performance, single-loop control can often be enhanced to achieve much better performance with
low-cost and simple technology. Feedforward control is introduced briefly in this section, with
much more detail available in, for example, Marlin (2000).
Feedforward uses another approach for taking advantage of an early indication of an
important disturbance to improve control performance. Lets again consider the stirred tank
heating process in Figure 6.9 where the outlet temperature is controlled by adjusting the heating
temperature. Another common disturbance involves variation in the feed temperature.
Table 6.2. Cascade design criteria
(Affirmative answer required for every entry)

1.
2.
3.
4.
5.

Cascade is desired when


Single-loop performance unacceptable
A measured secondary variable is available
A secondary variable must
Indicate the occurrence of an important disturbance
Have a causal relationship from valve to secondary (cause effect)
Have a much faster response than the primary

6-22

Operability in process design

Chapter 6 Process Control

Is there a measurable variable that would indicate that this has occurred? The answer is
clearly, yes, the feed temperature. We are half way to designing a feedforward controller.
Since feedforward involves a different principle from feedback, a causal relationship must not
exist between the manipulated control valve and the surrogate or feedforward measured
disturbance variable. By observation, we confirm the absence of a causal effect.
Feedforward control uses process models to determine the valve adjustment that will
exactly compensate the measured disturbance. (Naturally, exact compensation is the goal of
the calculation, but imperfect compensation is expected because of plant-model mismatch.)
The feedforward controller compensates for the measured disturbance, but it will not
achieve the desired temperature control because of model mismatch and other unmeasured
disturbances. Therefore, we retain the feedback temperature controller to correct for all other
disturbances and model error in the feedforward scheme. This is called a feedforward/feedback
control system. The design is sketched in Figure 6.12, and a comparison of single-loop and
feedforward/feedback control performances is given in Figure 6.13. Clearly, feedforward
achieves a substantial improvement in control performance.
A properly designed feedforward control system requires that certain criteria be satisfied.
These criteria are summarized in Table 6.3. The engineer can check a proposed design using the
five criteria before completing the design. A standard PID controller can be used for the
feedback control, and a special algorithm must be used for the feedforward controller. For
details, see Marlin (2000).
The engineer should recognize the similarities and differences in cascade and
feedforward. Use of the criteria in Tables 6.2 and 6.3 is highly recommended to ensure that a
proper design is selected.

Figure 6.12 Feedforward/feedback control design

6-23

Operability in process design

Chapter 6 Process Control

Figure 6.13 (a) Feedback and (b) Feedforward /feedback control performance

Table 6.3. Feedforward design criteria


(Affirmative answer required for every entry)

1.
2.
3.
4.
5.

Feedforward is desired when


Single-loop performance unacceptable
A measured disturbance variable is available
A measured disturbance variable must
Indicate the occurrence of an important disturbance
Not have a causal relationship from valve to measured
disturbance sensor
Not have a much faster effect on the CV than the MV

6.2.4 Non-square process systems


Many process systems have an unequal number of controlled and manipulated variables. In this
section, we will see why these situations occur and learn the basics for controlling non-square
systems. We will restrict ourselves to the simplest systems with either (a) several controlled and
one manipulated variables or (b) one controlled and several manipulated variables.
More Controlled variables: We encounter incidences of multiple controlled variables in
everyday life. For example, we might be driving to a city and aim to arrive at exactly 3:00, but
we do not want to drive faster than 65 miles per hour (about 105 km per hour). When more
controlled than manipulated variables exist, it is not possible to maintain all controlled variables
at their set points. Fortunately, a hierarchy of goal importances often exists, as it does in the
driving example. In the driving example, we would set our speed to achieve our arrival time,
unless the speed exceeded the maximum; then, we would select the maximum limit and accept a
late arrival. Lets consider a process application of this concept.

6-24

Operability in process design

Chapter 6 Process Control

Example 6.4. Signal select The chemical reactor in Figure 6.14a has a cooling coil. The
conversion in the reactor should be controlled by adjusting the cooling, but the temperature
should never exceed its maximum limit (to prevent side reactions or equipment damage). We
note that the coolant valve is fail open to provide the safest condition, maximum cooling, if the
signal to the valve should go to zero. Design a control system.
We can achieve these control objectives using PID controllers. We can start by designing the
composition controller, using the outlet analyzer for the measured variable and the coolant flow valve
as the manipulated variable. This should work well, but it can reduce the cooling and allow the reactor
temperature to exceed its maximum. Therefore, we also design a temperature controller to adjust the
coolant valve. Clearly, the valve cannot obey both commands simultaneously, so when should it obey
each? We note that the control objectives call for selecting the controller output that demands the most
cooling and ignoring the other controller output. This is easily achieved by sending both controller
outputs to a signal select device (or algorithm) that reads all inputs and sends an output that is the
highest or lowest of the inputs. Here, we use a low signal select, because the valve is fail open, so that
the smallest signal gives the largest valve opening. The control system is shown in Figure 6.14b. The
dynamic performance of signal select is given in Figure 6.15.

Figure 6.14a Reactor with cooling coil and Figure 6.14b


single-loop control.
control.

Reactor

with

signal-select

Figure 6.15 Dynamic response of the signal select control design to an unmeasured disturbance,
a feed impurity that reduced the rate of reaction.

6-25

Operability in process design

Chapter 6 Process Control

More manipulated variables: We encounter incidences of multiple manipulated variables in


everyday life. For example, we want to regulate our speed at (near) 65 miles per hour by using
the accelerator and the brake pedal. When more manipulated than controlled variables exist,
many combinations of manipulations can regulate the controlled variable to the same values.
Fortunately, an ancillary goal usually enables us to select the best operating conditions by
defining an order of manipulation. In the driving example, the ancillary goal of energy
conservation would indicate that we should never use the accelerator and brakes simultaneously;
therefore, one would use the brake only after the accelerator is not being depressed, and one
would begin to accelerate only after braking has ceased. Lets consider a process application of
this concept.
Example 6.5. Split Range The fuel gas system in Figure 6.16a has two sources of fuel and many
consumers. The goal is to supply the ever-varying consumers by purchasing the appropriate
amount of fuel at the lowest price. We note that Fuel A is less costly than Fuel B.
We decide to control pressure to balance the consumption and purchase of fuel gas. Since all
consumers are independently determined to meet the moment-to-moment fuel needs of the process
units, the purchase must be adjusted to control pressure. We would like to purchase the less expensive
fuel and purchase the more expensive fuel only when required, i.e., when the less expensive is at its
maximum. The pressure PID controller would split its output and send signals to both valves, which is
called split range control. The exact details depend on the hardware implementation; here, we will
assume that the signal splitting occurs in the control computer and that the fuel valves are fail closed.
With the controller output scaled to 0-100 percent, the 0-50 percent values will affect only the lower
cost Fuel A, with the Fuel B valve fully closed. When the controller output is 50-100 percent, the Fuel
A valve will be fully opened. When the controller output is 50-100 percent, the Fuel B valve will be
opened. The control strategy is shown in Figure 6.16b. The dynamic response of the system to two
disturbances is shown in Figure 6.17.

Figure 6.16a Fuel distribution Figure 6.16b Fuel distribution system with split range
system with two sources.
pressure control.

6-26

Operability in process design

Chapter 6 Process Control

Figure 6.17 Dynamic response of the pressure split range control to two disturbances.
Marlin (2000)
A general rule in control design is never control the same variable with two controllers,
which in generally true. Nevertheless, lets consider an example where controlling the same
variable with two PID controllers is correct and serves as an alternative for split range control.
Example 6.6. Two controllers for one CV - The tank in Figure 6.18a accepts a flow that is
controlled elsewhere in the plant. The level is to be controlled using valve v101 that regulates
flow to a profitable downstream process. If the flow through v101 is inadequate to prevent
overflow, flow through v102 may be used, but a severe economic penalty is incurred. Design a
control strategy.
We can start with a straightforward PI (or P-only) level controller, LC101, with a set point in the
middle of the vessel height; this controller will manipulate v101. If the flow through v101 is not
sufficient to control the level because of a very large increase in the flow to the vessel, v102 must be
adjusted. Therefore, we introduce PI level controller LC102 that could use the same or redundant
level sensor. LC102 has a set point of 90% and manipulates v102. The key to this design is the
difference in the level controller set points; the two controllers must have significantly different set
point values. The completed design is shown in Figure 6.18b.

In most introductory process control courses non-square process systems are not
discussed. Is this because they are rare? No, non-square systems are common and are designed
for the following reasons.

Extra controlled variables provide actions to avoid violating important limitations. This
can prevent equipment damage and severe process disturbances.
Extra manipulated variables provide an expanded operating window, so that the
process can function well over a larger range of set point changes and disturbances.
Extra sensors and manipulated variables increase the reliability of the process and control
system.
6-27

Operability in process design

Chapter 6 Process Control

Figure 6.18a Single level controller. Operator Figure 6.18b Two level controllers
action is required when valve v101 is fully open.
(Split range control could be applied.)
When should the process and control system provide the additional flexibility? Designs
provide the additional flexibility when the economics and safety benefits justify the added
investment. These features are not included in most units in a plant, but large plants will typically
have many instances of non-square systems.

6.2.4 Inferential control systems


In all of the control methods considered to this point, the important variables have been measured,
a situation that is desirable and most often possible. However, not every important variable can be
measured in real time, i.e., fast enough so that timely control actions can be based on their
measurements. There are various reasons for the lack of key measurements. First, some sensitive
analyses have not been sufficiently automated to provide accurate, reliable measurements without
human management; thus, these measurements can be obtained only infrequently in a laboratory.
Second, if the real-time measurement is possible the cost of installing a sensor in the plant may not
be justified by the potential benefits derived from the additional sensor, especially considering the
alternative methods in this section. The cost is not typically high for conventional sensors for
measuring temperature, pressure, flow, and level but may be prohibitive for an expensive analyzer
with sample system and on-going maintenance. Third, the sensor may not provide information in a
timely manner if it must be located far downstream or it may have a long delay due to processing
time.
The lack of measurements of key variables in a timely manner certainly makes automated
feedback control more difficult, but not always impossible. In some situations, we can add extra
measured variables that, while not giving a perfect indication of the key unmeasured variable,
provide a valuable inference. For example, a temperature rise along a packed bed reactor might be
very useful in determining the conversion of a single reactant to a single product. However, the
temperature rise might not be useful in a packed bed reactor for determining the distribution of
many reaction products. Lets consider a typical example of inferential control.

6-28

Operability in process design

Chapter 6 Process Control

Example 6.7. Inferential composition. Suppose that we wish to regulate the multicomponent
flash separation process in Figure 6.19. In the process, a liquid has its temperature increased by
heat exchange and subsequently has its pressure reduced across a valve. The resulting two-phase
flow enters a drum, where the liquid and vapor are separated and exit in by the bottoms and
overhead pipes, respectively. The goals for control are given in the following.

Safety, which we will achieve by preventing excess pressure


Equipment protection, which we will achieving by ensuring that liquid flows through the
pump
Production rate by controlling the feed flow
Composition of the light key ethane in the liquid product. The desired value is 10%
ethane, and the allowable range is from 9 to 12 %.
Relevant disturbances are the temperature measurement error bounded by 0.50 C and
changes in feed butane composition of 5 mole% occurring simultaneously with a
change in methane of equal magnitude and opposite sign. (While we consider two
disturbances here, many more might be relevant in an industrial process.)

Figure 6.19 Typical flash separation process. (Marlin, 2000)

6-29

Operability in process design

Chapter 6 Process Control

We investigate the best manner for achieving the final goal. One possibility would be to install
an on-stream analyzer, likely a gas chromatograph that extracts samples from the bottoms stream and
determines the percentage ethane in the sample. Perhaps, the economics do not justify the on-stream
analyzer. Therefore, we will seek an inferential variable. From our knowledge of the flash process, we
know that three factors influence the compositions: temperature, pressure and feed composition. The
feed composition is not measured (and would be as expensive as the product measurement), so it is
eliminated from consideration. The pressure has a strong effect on the compositions; however,
changing the pressure of the vessel requires a more costly vessel to accommodate a wide range of
pressures; therefore, we do not select pressure. Now, we proceed to determine whether the flash
temperature is an appropriate inferential variable for this process equipment, operating conditions,
and control goals.
First, we evaluate the relationship between the candidate inferential variable and the true
controlled variable, which is shown as the base case in Figure 6.20 at the base case pressure and feed
composition. We see that the relationship is strong, especially in comparison to the measurement error
in temperature.
Second, we determine whether the disturbances influence the composition-temperature
relationship. The results in Figure 6.20 display the effects of temperature measurement error and the
feed composition disturbance. As we would expect, the disturbances introduce errors into the assumed
single-variable relationship. The key question is, Is the inferential relationship, while not exact, good
enough to achieve the performance goal? The analysis to answer the key question is shown
graphically in Figure 6.20, where the effects on the estimated composition of the maximum sensor
error and composition disturbances are evaluated. We see that the actual composition will remain
within the stated goal (9 to 12 percent) when maintaining the measured temperature at 25 C.
Third, we evaluate the dynamics of the potential temperature control loop. The temperature
can be measured with minimal delay, and the steam to the preheater can be adjusted rapidly, so that
the closed-loop dynamics would be on the order of minutes. Therefore, the loop dynamics are fast
enough for the application.
Based on this analysis, we conclude that the T6 inferential variable can be used in this process
application. Naturally, other disturbances or a tighter goal for composition deviations might change
this conclusion.

An inferential variable can be employed to improve control performance even when the
true controlled variable can be measured, but with significant delay. As shown in Figure 6.21, an
analyzer has been installed downstream to measure the butane concentration in the propane
product. At essentially no cost, the same analyzer could measure the ethane concentration for
feedback to the flash process! Because of the long delay from the flash drum to the analyzer, the
inferential temperature control would be retained, with its set point being adjusted by the
analyzer feedback in a cascade design.

6-30

Operability in process design

Chapter 6 Process Control

Changes in methane are accompanied by changes in butane of equal magnitude and opposite
sign. Pressure is constant at the base case value.

Figure 6.20. Analysis of the inferential relationship for the flash process at 1000kPa.
From Marlin (2000)

The general criteria for successful inferential control are summarized in Table 6.4. The application
of these criteria requires process insight from the engineer to develop reasonable candidate
inferential measurements.
The use of an inferential variable must be tailored to the specific process application.
Applications of similar concepts on unit operations must be evaluated because of differing
control goals, disturbances, inferential sensor accuracy, materials and operating
conditions.
Much more can be learned about the important topic of inferential control. Further
process examples of single measured inferential variables and building calculated inferential
variables (using multiple measurements) are available in Marlin (2000). An alternative method
for designing inferential variables utilizes the enormous amount of information available in plant
operating data. This data can be used for determining key correlation relationships among
process variables that provides a basis for building inferential control variables. A good
introduction to using plant data for inferential design is given in Kresta et. al. (1994).

6-31

Operability in process design

Chapter 6 Process Control

Figure 6.21 Inferential control combined with analyzer feedback with substantial delay.
Marlin(2000)

Table 6.4 Criteria for designing inferential control systems

Necessary situation
Measurement of the true controlled variable is not available in a timely manner
Not measured: on-stream sensor not possible or unreliable
Not measured: on-stream sensor too costly
Unfavorable feedback dynamics: sensor has poor dynamics, e.g., long dead time or analysis
time, or is located far downstream
Measured inferential variable(s) is available
Inferential variable features
The inferential variable must have a good relationship to the true controlled variable for changes
in the manipulated variable
Relationship above is insensitive to changes in operating conditions, i.e., unmeasured
disturbances, over their expected ranges
Favorable (fast) dynamics for use in feedback control
Correction of inferential variable
By primary controller in automated cascade design
By plant operator manually based on periodic information
When inferential variable is corrected frequently, the sensor for the inferential variable must
provide good reproducibility, not necessarily high accuracy

6.3 The influence of process design on control


The process being controlled has a profound effect on our ability to apply control and on the
quality of the control performance. For example, we could choose to drive the bus or the ride
bicycle depicted in Figure 6.22. If we need to implement a command for a direction change,

6-32

Operability in process design

Chapter 6 Process Control

Figure 6.22 Control of different vehicles.


(from Marlin (2000))

specifically to execute a 180-degree turn, which vehicle would be superior? Naturally, the
bicycle could turn more rapidly and in a small diameter, so it would be superior. Now, lets
consider another situation; consider both vehicles travelling at the same speed and encountering
a large bump in the road. In this new situation, the bus is superior because it is less sensitive to
the disturbance. Similarly, the design of chemical processes can have profound effects on
dynamic performance.

Chemical engineers design processes that are safe and reliable, satisfy production
requirements (product qualities, production rates, etc.) over a range of conditions, have
good steady-state efficiency, and have dynamic behavior that favors good dynamic
control performance.
In this section, a number of process characteristics will be introduced that affect control
performance. Each characteristic can have favorable or unfavorable effects of control
performance depending on the details of the process design. This does not imply that all
characteristics can be designed for favorable performance because the process chemistry and
physics set requirements that override dynamic behavior is some cases. For example, a reactor
volume is required to achieve a specific conversion and yield, even if a smaller volume would be
easier to control. However, much opportunity exists for the design engineer to improve (or
degrade) the operability and dynamic behavior of the process.
This section will concentrate on characteristics that affect a single controlled variable.
The characteristics are summarized in Table 6.5. In general, the conclusions presented here will
extend to multivariable processes as well. Important aspects of multivariable control are
addressed in Section 6.4.

6-33

Operability in process design

Chapter 6 Process Control

Table 6.5 Process characteristics that influence control performance


Characteristic
Favorable
Unfavorable
1. Process equipment in the feedback loop
a. Dead time
b. Inverse response
c. Time constant(s)
d. Limitation in the manipulation rate
of change
e. Limitation to manipulation range
f. Process non-linearity

Short
None
Small, few
None
None
Minimal

g. Sensitivity to manipulation

Moderate
(~1% CV/1% manipulation)

h. Process stability without control

Stable

Long
Large
Large, many
Slow compared with needed
dynamic response
Small, limits achievable
steady-states
Significant, affects damping
and stability
Too low (small CV range)
Too large (very small
manipulation needed)
Unstable

2. Process elements in (only) the disturbance path


a. Disturbance time constant(s)
b. Disturbance frequency

Many, large
Small or large period
compared with feedback
dynamics
Small

c. Disturbance magnitude

Few, small
Near critical frequency of the
feedback loop
Large

3. Instrumentation elements in the feedback loop


a. Sensor and final element dynamics
b. Measurement noise
c. Non-ideal final element behavior
d. Controller execution period

Fast compared with feedback


process
Small magnitude compared
with allowable variation
Element follows controller
output closely
Short compared with feedback
dynamics

Slow compared with feedback


process
Large magnitude compared
with allowable variation
Large deviations between
element and controller output
Long compared with feedback
dynamics

4. Process structure
a. Feedback from integrated
processes
b. Interaction among control loops*
(Control structure also a factor)

Negative feedback

Positive feedback

Small

Strong

5. Control performance goals


a. Quality specifications
b. Penalty for constraint violation
c. Production rate specifications
d. Profitability sensitivity for
different feasible operations
e. Safety

Large short-term variation


allowed
Small
Specification on average, shortterm fluctuations allowed
Small

Small short-term variation


allowed
Large
Must meet demands
immediately
Large

Safety is paramount. See Chapter 7 for safety hierarchy.

* The entries for interaction are simplified. Some designs with strong interaction can yield good control
performance. See Section 6.4

6-34

Operability in process design

Chapter 6 Process Control

The structure of Table 6.5 indicates that the location of some characteristics, in the
feedback or disturbance path, is critical to the characteristics effect on feedback control
performance. The distinction is shown schematically in Figure 6.23 using both a block diagram
and a simple process drawing. The block diagram shows the distinction between the two paths,
the feedback path is every element in the loop, including the temperature sensor, the control
valve v3, tank 2 liquid volume, and the coiled heat exchanger. The disturbance path is external to
the feedback, e.g., tank 1 liquid volume, and does not affect stability.

6.3.1 Process Equipment in the Feedback Loop


The feedback loop performance is improved by fast and strong effects of the manipulated
variable on the controlled variable. Lets look the characteristics in the feedback loop.

(a)

(b)

Figure 6.23 Single-loop control system for TC-3. a. Block diagram, b. Process schematic

6-35

Operability in process design

Chapter 6 Process Control

6.3.1a Dead time During the dead time, the effect of a change in the manipulated variable
does not influence the controlled variable, as shown in Figure 6.4a. This delay in information
degrades feedback control performance, as shown clearly in Figure 6.4b. The red box shows an
area of controlled variable deviation from set point that no feedback control algorithm can
reduce; the cause is the process dead time. The engineer should make every effort to reduce or
remove dead time from the feedback loop.
Feedback control performance is improved by process designs with short dead times in
the feedback path.

Figure 6.24a Response of a dead time to a Figure 6.24b. Effect of dead time on feedback
control performance
step input
6.3.1b Inverse response An inverse response occurs when the manipulated variable has two
effects on the controlled variable. One effect is faster and has a smaller magnitude positive
(negative) gain, while the other path is slower and has a larger magnitude negative (positive)
gain. A typical inverse response is shown in Figure 6.25a. This wrong-way initial response
delays return to desired operation, and therefore, degrades feedback control performance.
Feedback performance is shown in Figure 6.25b. For further details on modeling inverse
response behavior, see Marlin (2000) Chapter 5 and Appendix I.
Feedback control performance is improved by process designs without inverse response
in the feedback path.

6-36

Operability in process design

Figure 6.25a Typical open-loop


inverse response

Chapter 6 Process Control

Figure 6.25b Typical closed-loop inverse response for


a set point change

6.3.1c Time constants One can think of each time constant being the result of a dynamic
balance, e.g., material or energy balance, for the process. When modeled using fundamental
balances, typical process systems consist of many balances and thus time constants. For
example, a distillation tray has one energy balance and as many material balances as
components; therefore, a distillation tower model has many balances. When we model the same
process empirically, we usually find that one or a few time constants can represent the process
adequately for the purpose to selecting a control strategy and tuning the controller.
Naturally, these time constants introduce delays in the feedback loop. In additional,
numerous series time constants can introduce a behavior very similar to dead time, which we
have seen is particularly deleterious for feedback. This effect is shown in Figure 6.26.
Therefore, design engineers should reduce time constants in the feedback path, when possible.

Figure 6.26 Step response for time constants in series.

6-37

Operability in process design

Chapter 6 Process Control

Feedback control performance is improved by process designs with few and smallvalued time constants in the feedback path.
6.3.1d Manipulated variable rate of change Process equipment is designed for extreme
conditions, e.g., high temperature and pressures, and appears very sturdy. One might wonder
why we need to observe limitations to the maximum rate of change of manipulated variables.
Lets look at an example of a fired heater In Figure 6.27, in which the outlet temperature of the
process stream is controlled by adjusting the fuel flow to a burner. The process is typically
operating near its material limitations, and exceeding the maximum temperatures can reduce the
operating life of brickwork and piping. In addition, changes to the fuel firing can cause thermal
expansion and contractions that reduce equipment life. Therefore, the process controller must
balance the needs of maintaining the controlled temperature near its set point with the need to
limit manipulation velocity and extend equipment life. This balance is achieved through
appropriate controller tuning.

Figure 6.27 Aggressive manipulations can reduce equipment service factor.

Feedback control performance is limited by the allowable rate of change of the


manipulated variable. Design engineers must recognize where these limitation are
required and in these situations, not expect to achieve excellent controlled variable
performance through very aggressive manipulations.

6.3.1e Manipulated variable range Naturally, a limitation to the range of manipulation


reduces the range of achievable set points and compensation for disturbances. This limitation
affects the steady-state operating window, but it can be overcome if the design engineer provides
additional manipulated variable(s) and a control system for this non-square system.

6-38

Operability in process design

Chapter 6 Process Control

The range can also affect the dynamic response of a control system. Lets consider a
scenario in which you have rushed home and are preparing soup for dinner. You are in a rush
because you want to get working on your process control assignment. (Apparently, this scenario
is hypothetical.) How would you adjust the heat to the burner? If you turned the burner
adjustment to the value that just brought the soup to a slight boil, which is the desired end point,
you would have to wait a long time. You could follow an alternative strategy in which you
turned the burner to maximum heat and when the soup began to boil, reduced the burner
adjustment accordingly. This second strategy has a faster response, which is possible because
the burner has spare capacity. This concept can be applied in process control equipment
design. The feedback controller tuning can be selected to provide some manipulated variable
overshoot during the transient response. Naturally, the aggressiveness required for rapid
controlled variable response must be balanced with the potential equipment damage due to
overly aggressive manipulated variable behavior. The control engineer must understand process
and equipment operating goals!
Feedback control is limited by the range of the manipulated variable. For steady-state
behavior, returning the controlled variable to its set point is possible when the range of
the manipulated variable is sufficiently large and is not possible outside of the operating
window. For dynamic behavior, the feedback response can be rapid if the equipment is
capable of manipulated variable overshoot

6.3.1f Process Non-linearity We recognize that the controller algorithm and tuning values are
constant in most control designs. The engineer adjusts tuning once and the control system
functions for a long time. This approach can give acceptable dynamic performance when the
feedback behavior is also unchanging. Since essentially all processes are non-linear to some
extent, we have to evaluate when non-linearity causing changing feedback process behavior
is insignificant and when significant. This analysis is specific to each control loop. However,
for a rough guideline a feedback controller should function reasonably well (but not optimally)
for a change of 25% in all parameters, e.g., dead time, time constant and gain. This is a
minimum change that a process can be expected to experience, and much larger changes can
occur in some processes. We have seen an example of a closed-loop behavior for a single
controller with 35% that can be achieved with PI control in Figure 6.8.

Some chemical processes are very non-linear and are challenging to control. An example
is the process for mixing a strong acid and base shown in Figure 6.28. When the pH set point is
7.0, the process is strongly non-linear, with the gain changing by orders of magnitude! This is a
challenging problem. When tight control is required, a typical design provides two mixing tanks
and sizes them to attenuate high frequency fluctuations (Hoyle, 1976; Moore, 1978). When
larger, short-term fluctuations are acceptable, one tank can be used.
An approach to compensate for process gain non-linearity is to include a correcting nonlinearity in the feedback loop. If this is designed well, the product of the two non-linearities can
be approximately linear input-output behavior. The PID controller gain can be adjusted to
account for process non-linearities; for an example applied to pH control, see Liptak (2003).

6-39

Operability in process design

Chapter 6 Process Control

Figure 6.28. pH neutralization mixing process.

Control performance over a range of operating conditions, including disturbances, is


improved by a linear feedback loop. Non-linearities can degrade performance, and for
significant process non-linearities, the design engineer should evaluate including
compensating non-linearities in the control equipment or calculation.
6.3.1g Sensitivity to manipulation The gain relationship between the manipulated and
controlled variables is clearly important. To have a general discussion, we will consider the
variables expressed as percentage of range. For controlled variables, the range will be the sensor
range, which should span the values expected under normal variation, excluding startup and
extreme disturbances due to, for example, a pump failure. The manipulated variable is expressed
as 0-100% valve opening (closing), or if the output is to a secondary in a cascade, as percent of
the secondary sensor range. With scaled variables, the process gain is expressed in units of
%CV/%MV.
A desired value for the process gain is around 1 %CV/%MV. With this value, the control
variable can be directed to the entire range of the sensor span by adjusting the manipulated
variable through its entire range. If the process gain value is too small, the manipulated variable
has a weak effect and cannot compensate larger disturbances. If the process gain value is too
large, the lack of precision in adjusting a valve will lead to poor control. (It is not possible to
adjust the valve from 65.2% to 65.4% precisely.) The situations with inappropriate process gains
are shown in Figure 6.29.
For good control performance, the process gain should be approximately 1 %CV/%MV.
If the value is too low, the design engineer should provide an additional manipulated
variable with wider range. If the value is too high, the design engineer should include a
manipulated variable with a smaller gain.

6-40

Operability in process design

Chapter 6 Process Control

Figure 6.29 Schematic of effect of process gain (KP = CV/MV).


a. Gain is too large, b. Gain is too small
6.3.1h Process stability without control One final effect of the process on dynamic
performance is the stability of the process without control. Most processes are stable and
therefore, will approach a steady state when all input variables, i.e. disturbances and manipulated
final elements, are constant. However, some important processes are not stable.
A stable process has output variable(s) that are bounded in magnitude when all input
variables (disturbances and manipulated variables) have bounded magnitudes.
We see that the definition of stability considers whether a process variable is unbounded
and tends toward infinity. From an engineering viewpoint, no variable gets near infinity because
something very bad (equipment damage, explosion, and so forth) will occur in the plant before
infinity. However, the concept of stability is useful because an unstable variable will not tend to
a steady state, and operating personnel would have to spend an inordinate amount of time
monitoring and adjusting the process for each unstable variable. Therefore, we anticipate that
unstable variables must be controlled.
Inventories - One important category of unstable process variables is inventories, both liquid
and solid, that serve to store materials without changing the material properties. A simple
material balance model for the inventory shown in Figure 6.30a is given in the following.
(6.5)

6-41

Operability in process design

Chapter 6 Process Control

with
A
F
L
m

=
=
=
=
=

cross sectional area (assumed constant)


volumetric flow rate
level
mass of material in the vessel
density

When the flows in and out do not depend on the inventory, the system is termed non-selfregulatory. Non-self-regulatory inventories have all inputs and outputs independent of the
inventory itself; therefore, as the inventory changes (increases or decreases) no change occurs to
an input or output that might stabilize the inventory. Without intervention by people or
controllers, a non-self-regulatory system is unstable. As an example, consider a situation in
which the flows in and out are initially equal, and the outlet flow increases. Then, the derivative
of level would be a negative constant. The dynamic response for a step change in the flow out
from an initial steady state is given in Figure 6.30b. A mathematician would say that the
inventory would decrease without limit; in contrast, an engineer would say that the inventory
would increase until the occurrence of an incident, such as a zero level. The unstable behavior is
clearly demonstrated.

Figure 6.30a. Liquid inventory with flow


out pumped.

Figure 6.30b. Step response for a non-selfregulating level, clearly showing instability.

Fortunately, control of inventories is generally quite easy, because essentially no dead


time exists between the level and potential manipulated variables, the flows in and out.
Therefore, a feedback controller can maintain the level near its set point or can allow it to vary
about the set point to modulate the manipulated flow variations. Details on level modeling,
control and tuning for various process objectives are available in Marlin (2000).

6-42

Operability in process design

Chapter 6 Process Control

Should unstable levels be controlled? Lets consider some situations.

Large tanks with only one continuous flow rate. Raw material feed tanks have periodic
inflows during deliveries but can have continuous outflows to the process. Finished
product tanks can have continuous inflows from the process but only periodic outflows
for shipment to customers. These tank levels cannot be controlled. The design engineer
must provide adequate storage volumes to allow continuous process operation between
deliveries and shipments, including consideration for disruptions in transportation due to
weather and other factors.
Large tanks with continuous flows in and out. These tanks can store material between a
sequence of process units. This material can be termed intermediate products or work
in progress. The size of the tanks can be large to provide reliability, e.g., time for
maintenance, as discussed in Chapter 4. Usually, the daily production rates of the
individual units are set independently, although they must balance over a long period of
time. Therefore, these tank levels are not usually controlled. When designed with
adequate volumes, operating personnel have sufficient time to monitor and manage the
flows to maintain the levels in acceptable ranges.
Chemical processes have many smaller tanks and drums, such as reflux drums, vaporliquid separators, kettle reboilers, and so forth. The ratio of inventory to flow rate (V/F)
is usually in the range or 5 to 15 minutes; therefore, disturbances can rapidly lead to
overflow or emptying of the vessel. All of these liquid inventories must be controlled.
Many chemical reactions occur in a liquid inventory. These reactors are different from
storage facilities because the reactor volumes influence the process operation (Fogler,
1986). These reactor levels must be controlled because they are unstable and influence
the product quality and process profitability.

In summary, inventories with very large volumes are not controlled automatically; people
provide the feedback action. For smaller inventories, people cannot reliably monitor the process,
so that automatic feedback control is essential. In addition, when the volume of the inventory
affects the process performance (or safety), with chemical reactors being the most common
example, feedback inventory control is essential.
Unstable processes - A second category of unstable processes involves processes that have selfregulation. Most self-regulatory processes are stable; however, some self-regulatory processes
can be unstable for specific design parameters. An example is a CSTR with an exothermic
reaction and cooling, which can experience multiple steady states and unstable steady states; for
a clear exposition of this behavior, see Fogler (1986). The reader might be thinking, This
doesnt happen frequently; so, is it important to evaluate the design for stability? The answer
comes from industrial experience related by Bush (1969). The ICI Chemical Company was
developing a process for chlorinated hydrocarbons using a pilot plant reactor; unfortunately, it
operated in continuous cycling operation, so it never achieved steady state (as desired). A
thorough dynamic analysis confirmed that the reactor had a wide range of conditions where the
operation was unstable without control. We can learn two lessons from this industrial
experience. One, unstable process operation, while not common, does occur. Second, the
engineer is wise to evaluate a potentially unstable system before completing the design.

6-43

Operability in process design

Chapter 6 Process Control

Figure 6.31a Reactor with feed-effluent heat


exchanger. If each process is stable, the system
can be unstable.

Figure 6.31b Reactor with feed-effluent heat


exchanger and one possible stabilizing
feedback control design.

Even when each individual process is stable, the process integration of material and
energy can cause the entire unit to be unstable. A typical design structure involves feed-effluent
heat exchange that increases the energy efficiency of a design by exchanging heat to reduce
utility costs for heating (or cooling). An application of this design structure is shown in Figure
6.31a for a packed bed reactor that has exothermic chemical reactions. Note that an increase in
the reactor outlet temperature causes an increase in the reactor inlet temperature. This
disturbance can continue around the process loop, increasing in magnitude each time. The
result is instability. For this design, when the individual processes (heat exchanger, fired heater,
and reactor) are stable, the positive feedback due to the feed-effluent exchanger can result in the
process system being unstable!
Such a system is analyzed by Silverstein and Shinnar (1982). Process controls shown in
Figure 6.31b stabilize the system by tightly controlling the reactor inlet temperature, preventing a
reactor outlet deviation from affecting the reactor inlet. Silverstein and Shinnar also provide
guidance on the information and models useful in analyzing reactor dynamics at the design stage.
Depending on the strength of the positive feedback in the process system, other design
modifications might be required to moderate the effects of disturbances. Some other possibilities
are summarized in the following.

Control the reactor outlet temperature by adjusting the inlet temperature via a cascade
Reduce the amount of heat exchanged in the feed-effluent heat exchanger and increase
the duty of the fired heater
Split the reactor into multiple beds with inter-bed heat exchangers for cooling
Split the reactor into multiple beds and inject cold feed into the inlet of each bed to
control temperature. The cold feed would by-pass all preheat equipment.
Add inert packing in the reactor to reduce the rate of change of bed temperature

6-44

Operability in process design

Chapter 6 Process Control

Note that the use of multiple reactor beds also corrects uneven flow distribution, reducing the
chance for catalyst bypassing, low fluid flow near dense packing, and local hot spots.
Clearly, process structures with heat integration improve efficiency and introduce
challenges to the dynamic operation. The creative design engineer can realize the energy savings
without sacrificing safety or stable operation with consistent product quality.
Good control performance requires the control of unstable variables, with the exception
of inventories that have a very large V/F (volume/flow) ratio enabling effective manual
operation.

6.3.2 Process elements in (only) the disturbance path

It makes sense that the disturbance characteristics will influence the variability of the controlled
variable. Here, we will consider key factors and their influences. Before we begin, we should
emphasize an overriding perspective in control.
The first approach should be to eliminate or reduce disturbances. The design steps
discussed here are performed after reasonable effort has been expended to prevent
disturbances.
6.3.2a Disturbance time constant(s) Disturbance time constants are between the disturbance
origin (flow rate, composition, temperature, and so forth) and the control loop. These time
constants will slow the disturbance effect, which will be beneficial by reducing the deviations
experienced by the controlled variable.
Example 6.8 Disturbance time constant - Lets look at an example of the effect of a
disturbance time constant on control performance. In Figure 6.32a, the step disturbance affects
the feedback loop directly, and the maximum deviation of the controlled variable from its set
point is greater than 7. For the same feedback process, disturbance and controller in Figure
16.32b, the step disturbance passes through a tank (first order system) before affecting the
feedback loop, and the maximum deviation is much lower, about 3.2. Clearly, the performance
is better with the extra tank.
Control performance is improved by large disturbance time constants (that do not
appear in the feedback loop). The improvement is in reducing the maximum deviation
from set point, or equivalently smaller variance.

6-45

Operability in process design

Chapter 6 Process Control

Figure 6.32 Feedback control performance for (a) a disturbance directly entering the feedback
loop and (b) a disturbance through a time constant before entering the feedback loop.
6.3.2b Disturbance frequency A periodic disturbance can usually be characterized by a
frequency. The question is, Does this frequency affect control performance? The answer is a
resounding Yes! As explained in Appendix 6.A, disturbances near the critical frequency are
not controlled well by a feedback controller. This is because the controller is not fast enough to
compensate for the positive half wave before the negative half wave occurs. The control system
is chasing its tail, and the performance can be worse than if the controller were switched to
manual.
On the other hand, disturbances with much higher frequencies contribute little variation
because the process (time constants) attenuates the disturbances. In addition, disturbances with a
lower frequency are easily attenuated by the feedback controller. Therefore, disturbances near
the critical frequency lead to poor control performance. This situation is shown in Figure 6.33,
and further discussion of this topic is given in Appendix 6.A.
Process modifications are needed for disturbance frequencies near the critical frequency.
One modification would be to add a mixing tank between the disturbance source and the
feedback loop. Another modification introduces a fast control loop to compensate for the
disturbance before affecting the loop; an example would be a heat exchanger to regulate the
temperature of an input stream that affects the feedback loop.

6-46

Operability in process design

Chapter 6 Process Control

We might wonder if disturbances near the critical frequency of typical control loops
occur at all. Sadly, they do, and the cause is often other control systems that are poorly
implemented. Many process units have similar dynamics, and when integrated units have poorly
tuned control loops, the oscillations from one process are near the critical frequency in another
unit. Therefore, one poorly tuned loop can cause difficulties in many other interacting control
loops. This is the reason that some PID tuning rules that yield oscillatory responses, such as decay ratio, are not used in practice.
Control performance is good for disturbances with very low (easily controlled) or very
high (attenuated by process time constants) disturbances. The design engineer should
consider process or control modifications for disturbances near the feedback loop
critical frequency.

Figure 6.33 The effect of a sine disturbance on the process with control at three frequencies.
A is very low frequency where feedback is effective
B is near the critical frequency where control is ineffective
C is high frequency where the process attenuates disturbances

6-47

Operability in process design

Chapter 6 Process Control

6.3.2c Disturbance magnitude Naturally, large magnitude disturbances are worse than smaller
magnitude disturbances. For a truly linear system, the controlled variable deviation from set
point is proportional to the disturbance magnitude. While the strict proportionality is not valid
for non-linear systems, the general conclusion remains valid. The process control system
(controller and process equipment) might not be able to compensate for very large disturbances
because the manipulated variable could be adjusted to its upper or lower limit, where no
compensation will be possible. Engineers must account for very large disturbances by including
additional systems with greater capacity for compensation. These systems are addressed in
Chapter 5 on Safety and here in Section 6.2.4.

Preventing disturbances of large magnitude requires actions involving process elements


outside of the loop being considered. For example, fuel gas systems often mix purchased natural
gas with plant by-products. Occasionally, a unit upset will result in a large flow of by-products
into the fuel gas, introducing major disturbances in the fuel gas pressure and molecular weight
(and heat of combustion). Control systems can be implemented that compensate for these
disturbances and maintain the desired heat release in burners in boilers and fired heaters. From
this brief example, we note that the engineer must know typical disturbance sources to properly
design control systems!
Control performance is good for small disturbance magnitudes.
6.3.3 Instrumentation elements in the feedback loop

Sensors, final elements, signal transmission and the controller calculation all play vital roles in
successful process control. In the introductory process control course, we often assume that the
instrumentation functions perfectly, so that we can concentrate on the process elements.
Excellent instrumentation performance can be approached when engineers select, install, and
maintain instrumentation that is well matched to the plant requirements. Here, we will introduce
some common instrumentation factors that affect control performance.
6.3.3a Sensor and final element dynamics The sensor and final element are in the feedback
control loop, so that any delays introduced by the instrumentation will degrade control
performance. Since process equipment has rather slow dynamics, the instrumentation can often
be assumed instantaneous. Typically, signals from F, T, P and L sensors respond within a few
seconds of a change in the process variable, and control valve stem positions respond in a few
seconds from a change in the signal from the control room. This is much faster than the
dynamics of heat exchangers, distillation towers and most industrial chemical reactors. (In
contrast to process equipment, think about controlling a disk drive or a machine tool cutting
metal.)

Sensors can introduce significant delay when performing a complex chemical analysis on
material sampled from the process. The sampling and analysis can be entirely automated and
performed periodically without manual intervention. Two aspects of this procedure can slow
feedback control. The first is the sampling, involving extracting the material and transporting it
a short distance (tens of meters) to the analyzer. Since the analyzer needs only a small amount of

6-48

Operability in process design

Chapter 6 Process Control

material, the flow from the process to the analyzer could be small, which would result in a very
long transportation dead time, and we know that dead times are bad! An easy solution involves
providing a fast loop with a high flow rate that brings the material close to the analyzer, which
can then extract a small amount. The fast loop returns the material to the process, so that no
product is lost. A schematic of a fast loop is given in Figure 6.34. The second aspect of analyzer
for control is the execution period that is addressed in a subsequent section.
Typical final element response can be too slow for some machinery control applications.
An example is recycle control around a centrifugal compressor. The recycle is needed to ensure
a minimum flow rate through the compressor when the feed rate is too low, because a low flow
through the compressor causes surge and damages the compressor. Surge occurs very quickly,
so that the recycle response must be rapid (Staroselsky and Laudin, 1979; Smith and Kurz, 2005,
Engencyclopedia, 2012). An example is shown in Figure 6.35, with the set point for the FC
defining the minimum flow rate. The control valves for this application must be selected to have
fast response, which can involve volume boosters to increase the air supply flow rate to the
pneumatic actuators.
Control performance is good when the instrumentation dynamics are negligible
compared with the process dynamics in the loop.

Figure 6.34 A typical analyzer sample system Figure 6.35 Compressor anti-surge recycle
with a fast loop to reduce dead time.
control where fast instrumentation is essential.
6.3.3b Measurement noise We use the term noise for contributions to a measurement or
signal that is not repeatable and generally does not represent the behavior of the variable being
measured. In this situation, electrical interference and mechanical vibrations that artificially
modify a signal certainly qualify as noise. Generally, we also categorize high frequency process
variation, like liquid level oscillations, as noise; they represent the real process but are high
frequency and should not be considered for control. The controller acts on the measurement, and
any part of the measurement that does not represent true process changes will lead to incorrect
controller actions. Therefore, we would like to separate the noise from the true signal. The
effects of electrical noise can be reduced by proper cabling. We use filters to remove additional
high frequency components from the signal, as shown in Figure 6.36a.
6-49

Operability in process design

Chapter 6 Process Control

(a)

(b)

Figure 6.36. Filtering measurement noise.


a. Block diagram showing the filter location and decrease in manipulated variable variation
b. Bode plot for CVf(s)/CVm(s); the distinct signal-noise boundary is not actually known
(red dashed line is ideal filter; blue solid line is first order filter)

6-50

Operability in process design

Chapter 6 Process Control

We often think about the frequency of control signals according to bands. Frequencies
much higher than the process dynamics are considered noise. However, overlap usually exists so
that there is no clear boundary separating noise from signal. The ideal filter performance is
shown in Figure 6.36b as a red, dashed line; it does not affect the true signal, reduces the noise
amplitude to zero, and introduces no delay. This ideal filter is not possible. The most common
filter for process control signals is a first order filter, which is similar to a mixing tank. The
performance of a first order filter is compared with the ideal filter in Figure 6.36b. The filter
time constant is chosen to reduce the part of the signal thought to be noise.
A word of caution is appropriate here. The filter appears in the feedback loop, so it
delays control. A common error is to over-filter a signal, i.e., apply a filter with too large a
time constant. With (too) large a filter time constant, trend plots of the filtered signal look
smooth, with little variation. However, the real process variable could have more variation
because a large filter slows feedback, requiring looser PID tuning.
Control performance is improved when measurement noise is small. Engineers can use
filtering, as long as it does not significantly influence the feedback loop dynamics.
6.3.3c Non-ideal final element behavior The most common final element in chemical
processes is a pneumatically actuated control valve. These valves are well manufactured, but
they are not precision instruments. The typical valve has some difference between static and
dynamic friction, so that a minimum force is required to start motion, and it has some gap
when changing direction (e.g., opening to closing) (Choudhurya et. al., 2005; Wallen, 1997).
The result is non-ideal behavior. As shown in Figure 6.37, oscillations and poor performance
occur in the control loop when the final element non-ideality is large. An important step in the
design stage to prevent this behavior is to ensure that control valves are sized properly. The
guideline always oversize equipment to prevent mistakes will lead to a valve that is operating
in a narrow range, e.g. 10-30%, which will exacerbate the effects of valve friction.

(a)

(b)

Figure 6.37 Effects of non-ideal valve performance (Wallen, 1997)


a. Behavior during an experiment without control (u = control signal)
b. Behavior of closed-loop feedback control with poorly performing valve

6-51

Operability in process design

Chapter 6 Process Control

Several solutions to inadequate valve performance are given in the following.

A valve with smaller maximum flow rate can be used. This will only be appropriate if
the initial valve were oversized, requiring adjustment in much less than a normal range
of 50-100% of opening.
A valve positioner can be added to the valve. A valve positioner is essentially a fast
secondary in a cascade structure. It measures the actual valve stem position and controls
it to be close to the command signal from the controller. A schematic of a valve
positioner is given in Figure 6.38a.
The valve can be repaired or replaced.
If small adjustments are required for a large base case flow rate, two parallel valves can
be included in the design. Normal, small adjustments from the controller are made to the
smaller valve. The larger valve is only adjusted infrequently to ensure that the smaller
valve is able to affect the flow, i.e., that it is neither fully opened nor closed. A
schematic of this control is shown in Figure 6.38b

Good control performance relies on final elements responding precisely to controller


commands. Valve performance can be monitored so that poorly performing valves can
be repaired.

(a)

(b)

MV
p/i

Process
controller

Change signal to large


valve only when small
valve is near its upper or
lower range.
Keep smaller valve in
its adjustable range.

p/i

Small valve

Large valve

Figure 6.38. Improving the performance of control valves


a. Schematic of valve positioner.
b. Strategy using small valve for high precision with small adjustments.
6.3.3d Controller execution period Digital controller computations are performed at a fixed
period. Naturally, we would like this period to be short compared with the feedback process
dynamics. Usually, this is not difficult to achieve with modern digital computation. The most
frequent factor that lengthens controller execution is on-stream analyzer time, where a
substantial amount of time is required for analysis. In general, control performance is not
6-52

Operability in process design

Chapter 6 Process Control

affected significantly if the controller execution time is a few percent of the sum of process dead
times and time constants in the feedback loop; a goal is to keep this percentage below five
percent. A comparison of continuous and digital control with an excessively long execution
period is given in Figure 6.39; note that the IAE has nearly doubled for less frequent feedback
control.
When a long execution period cannot be avoided, the design engineer should seek an
inferential variable for the analyzer measurement. The fast inferential variable can be used to
compensate for most disturbances, while the much slower analyzer feedback can reset the
inferential set point in a cascade structure.
Good feedback control performance requires rapid digital execution when compared
with the process dynamics.
(a)

(b)

Figure 6.39 Feedback control performance (a) continuous controller and (b) digital controller
with execution period equal to the sum of the time constants in the feedback path.

6.3.4 Process Structure


Engineers like to concentrate on individual processes and design them to function well. This is a
good start, but we must remember that many units share material and energy, so that there is a
recycle or feedback due to the process structure. These recycles are included in the plant to
improve the profitability by recovering energy and materials that would otherwise be lost for
productive purposes. It is the job of the design engineer to ensure that efficient performance can
be maintained during dynamic operation.
6.3.4a Feedback from integrated processes The streams exiting a process proceed to other
units in the plant. In some designs, these streams return to exchange energy of material with the
6-53

Operability in process design

Chapter 6 Process Control

original process. This feedback is nearly always positive feedback, i.e., the effect is to drive the
process away from the desired operation. Special process and control structures are required to
moderate the unfavorable dynamic effects of integration. Since this topic is addressed in detail
in Section 6.5.4, no further discussion is given to this important topic here.
Generally, recycle systems are more difficult to control, with higher sensitivity to
disturbances and slower dynamics.
6.3.4b Interaction among control loops Most individual processes require many single-loop
controllers to achieve desired dynamic performance. Because each valve adjustment affects
more than one controlled variable, interaction among loops occurs. This interaction has a
profound effect on control design and achievable performance. Since this topic requires
considerable development, interaction is covered fully in Section 6.4.
Feedback control design is simpler when little interaction occurs between loops. When
strong interaction exists, determining the best loop pairing requires analysis by the
engineer, which is addressed later in this chapter.
6.3.5 Control performance goals

Naturally, the performance goals have a substantial impact on the challenges to achieve the
goals. Very restrictive goals are more challenging and usually require more complex and
expensive designs.
6.3.5a Product quality specifications Material quality specifications can allow considerable
variation or they can be very restrictive. Naturally, the more restrictive the quality
specifications, the more challenging it becomes to achieve the control objectives. Restrictive
specifications follow from customer requirements. For example,

Monomer purities must be very high to produce polymers with consistent molecular
weight distributions
Food produces must prevent foreign materials
Pharmaceuticals must adhere to specifications agreed by licensing agencies, such as the
FDA

A key aspect of the quality specification is whether it applies to the time-average property
or to each individual sample of the product. Lets consider the gasoline blending process shown
in Figure 6.40. The product must satisfy many specifications, like octane, reed vapor pressure,
volatility, and many compositions, e.g., percent alcohol. These specifications are imposed on
material at the end of the batch in the product tank, so that the instantaneous properties at the
blending point need not satisfy the specification. This situation makes the control easier because
a short-term deviation (such as a low octane) can be compensated by a subsequent deviation (of
higher octane) during the batch blend. The key process factor here is the mixing of the product
before specifications are imposed.

6-54

Operability in process design

Chapter 6 Process Control

Figure 6.40 Quality control where material is mixed before qualities are determined.
In contrast, a pharmaceutical manufacturer making pills must meet specifications for
every pill. What if the process produced four pills containing no active ingredient and the fifth
containing five times the desired ingredient. It would not be acceptable, and it might be deadly!
A similar situation occurs for a manufacturer of sheet materials, such as paper, steel and plastic,
as shown in Figure 6.41 for sheet steel. The entire sheet must be uniform; the quality would be
unacceptable if part of the sheet were half as thick as desired, while another part of the sheet
were 50% too thick. The key process factor here is the lack of mixing or other compensation for
off-specification material.
In general, material mixing before imposing quality specifications eases achieving these
goals. Instantaneous and spatial specifications are more challenging to achieve.

Figure 6.41 Quality control where every location must be manufactured to specification.
(Substech, 2012)
6-55

Operability in process design

Chapter 6 Process Control

6.3.5b Penalty for constraint violations Essentially, every process involves many limitations
to operating conditions. The boundaries of the operating window are defined by limits on

Stream properties, for example, product qualities.


Operating conditions affecting equipment performance, for example, liquid and vapor
flows on distillation trays.
Capacities of equipment, for example, maximum flow rate for a pipe, pump and control
valve combination.
Safe operation, for example, the rate of reactant additional in a batch reactor.
Conditions leading to equipment damage, for example, maximum pipe temperatures in a
fired heater.

The penalty for constraint violation can be minimal or it can be very high. A high penalty could
be due to very undesirable consequences for the violation such as (i) loss of production due to
spoiling a batch of product, (ii) damaging equipment, or (iii) activating a Safety Instrumented
System (SIS) that shuts down the process.
Example 6.9 Approach to constraint - When the penalties are high, plant personnel tend to
operate equipment far from the limiting conditions. In many situations, operating near such
constraints generates high economic returns. As a result, a conflict exists concerning how much
short-term risk should be accepted to realize high profit (if a violation does not occur) versus the
penalty for the violation. Operating near limitations requires excellent feedback control
performance, i.e., low variance, that is usually too demanding for personnel to achieve manually.
Lets consider an example of a pyrolysis fired heater that has a chemical reaction occurring in the
piping; a process schematic is given in Figure 6.42. There can be a high incentive to operate at a
high temperature that gives a high conversion, but a maximum temperature must not be violated
to protect the equipment. This situation is shown in Figure 6.43. Therefore, we anticipate that a
large economic benefit exists to operate at a high temperature without violating the temperature
constraint.

steam

Figure 6.43. Pyrolysis reactor, relationship


between outlet temperature and conversion

Figure 6.42. Pyrolysis furnace process


schematic.

6-56

Operability in process design

Chapter 6 Process Control

We will consider three different plant performances achieved by feedback control; the
distributions of temperature data for all cases are shown in Figure 6.44 in histograms. In the
base case (a), the temperature control is not good; the temperature experiences a large variation,
and the set point must be located far from the limitation to ensure that (essentially) all operation
will be below the maximum temperature. In Case (b), the control has been improved, and the
variation is much lower. However, the set point has not been modified, so that no advantage has
been realized from the improved control performance. In Case (c), the temperature control
performance is improved and the set point has been increased to take advantage of the reduced
variation. Note that Case (c) achieves higher average conversion without increasing the risk of
violating the maximum limitation. Further discussion on process control benefits and the use of
historical data in histograms is given in Appendix 6.B, and more detailed case studies are
available in Marlin et.al. (1987).
Where process limitations with high penalties exist, the tendency is to operate far from
the limitations. Often, good control performance can reduce the variability and enable
more profitable operation nearer to the limitation.

Figure 6.44 Histogram of reactor temperature variation.


Case a. Base case with poor initial control
Case b. Improved control with same set point as the base case, Case (a)
Case c. Improved control with modified set point closer to the temperature limit
6.3.5c Production rate specifications We must ultimately manufacture the amount that we
need to satisfy sales, or if the process is meeting another need, such as a waste treatment plant,
the process must handle all effluent it receives. Production control involves issues such as the
following.

Periodic delivery of raw materials and dispatch of products


Need to immediately process inputs or provide products on demand
Storage of raw materials, intermediate products (work in progress) and final products
Balancing the production in integrated plant sections
Observing product rate limitations throughout the plant (bottlenecks)
Scheduling maintenance while providing products when required
6-57

Operability in process design

Chapter 6 Process Control

Defining the product rate control structure is one of the first decisions is plant-wide control. This
topic is addressed in Sections 6.4 and 6.5; therefore, it will not be discussion further here.
Production rate specification, specifically how rapidly the process must respond to
demand changes, has a strong impact on control design. This topic is addressed later in
the chapter.
6.3.5d Profit sensitivity The feasible operating window might be very large for a process,
allowing a range of operating conditions to safely manufacture products satisfying quantity and
quality specifications. Usually, the profitability of the plant operation varies within the operating
window, so that the control system should maintain operation near the maximum profitability
within the window. If the location of the highest profit does not change (or changes
infrequently), the engineer can perform a study once to determine the best conditions and define
these in the operating procedures. In many plants the highest-profit operating conditions change
with disturbances like raw material composition, production rate, product grade, equipment
performance (e.g., reactor coking), equipment maintenance schedule (e.g., reactor decoking), and
environmental conditions (e.g., refrigeration system capacity).

There are many approaches to tracking a changing optimal operating condition. These
are presented briefly in the following.

Conditions enforced by process control In some instances, a study determines that the
highest profit can be at least partially achieved using relatively simple strategies. In the
pyrolysis fired heater reactor in Figure 6.42 discussed above, studies show that the steam
flow should be maintained in a ratio to the hydrocarbon feed rate, which can be achieved
with a simple flow ratio control design.
Standard operating procedures In other situations, the best operating conditions
depend upon a few key variables, such as the type of raw materials. The engineer can
perform studies to determine the best conditions and define rules for plant operations.
Considering again the pyrolysis reactor, the proper steam to hydrocarbon feed ratio
depends on the feed type (ethane, propane, naphtha, gas oil, etc.).
Direct search In direct search methods, the operating personnel introduce (small)
changes to key operating variables and determine the profit at each experimental point
from plant measurements. The experiments indicate a direction in which the profit
increases. The process conditions are changed to a new point with higher profit. Then,
the procedure is repeated until experiments do not yield discernible profit improvement.
These methods are also termed response surface approaches.
Real-time optimization In challenging situations, the profit changes frequently and by
a large amount and the best conditions cannot be determined by rules or simple strategies.
In these situations, an advanced control strategy can characterize the plant by a model
that is corrected or updated using current plant data, and the model can be optimized to
determine the optimal operating conditions. Naturally, this is the most complex
approach, and it is justified where the economic benefits are substantial (Marlin and
Hrymak, 1996; Pedersen, 1995; Vermeer, 1996).

6-58

Operability in process design

Chapter 6 Process Control

Process automation at a higher level in the control hierarchy can address profitability
directly. These profit controllers adjust the process slowly to maintain safety, product
quality, and production rates.
6.3.5e Safety- We conclude this discussion of factors affecting control design with the most
important factor, safety. The Basic Process Control System (BPCS) provides the first of many
barriers between a potential disturbance and an accident. The engineer must consider potential
disturbances and provide BPCS designs to compensate for the disturbances and maintain
conditions within safe limits. Naturally, the control systems cannot compensate for the largest
disturbances, which is the reason for many additional barriers. However, the control system
should keep the process in acceptable conditions nearly all of the time, so that activation of
higher barriers (alarms, Safety Instrumented Systems (SIS), pressure relief valves, and
containment) should be very infrequent. Process safety and the safety hierarchy are the topic of
Chapter 5, where much more on safety is presented in an compelling and informative manner.
Good control performance requires tight control of all safety-related process variables
by the Basic Process Control System (BPCS). The process design should provide a
manipulated variable with a fast and strong effect on each safety-related variable.

6.4 Multivariable Process Systems


Most processes require the control of several variables for successful dynamic operation. The
engineer needs to understand some basic concepts in the behavior of multivariable systems to
design their controls. In this section, these concepts are introduced and are applied in example
multiloop control designs.

6.4.1 Which variables can be controlled?


We must ensure that the process design allows all key variables to be controlled. Lets start with
a single controlled variable and determine what is required for feedback control of the variable.
As we saw in Section 6.1, a feedback loop requires a sensor, a control calculation, and a final
element, usually a valve. Naturally, the sensor is selected to measure a key variable requiring
control. What is a fundamental requirement for the valve? The valve must influence the control
variable, i.e., there must be a causal relationship between the valve and the sensor. Therefore,
For a single-variable process, feedback control is possible when KP 0.

The existence of a causal relationship (or its absence) is relatively easy to determine for a
single controlled variable using fundamental principles and/or empirical data, but the analysis
becomes considerably more complex for multiple controlled variables. Next, we consider how

6-59

Operability in process design

Chapter 6 Process Control

many variables can be controlled in a process. Since we require an independent valve


adjustment for each controlled variable, we conclude the following.
Degrees of Freedom: In a multivariable system, the maximum number of controlled
variables is equal to the number of adjustable final elements, i.e., valves.
We must recognize that this statement does not guarantee that every valve can be used to
independently control a meaningful process variable. Without evaluating the process and valve
location, the only thing that can be stated with certainty is that all remote valve positions can be
adjusted. Therefore, we proceed to the critical issue of controllability.
In automatic control literature, the term controllability has many definitions, each of
which can be applied to a specific set of relevant control issues. We will use a limited definition
here that will enable us to analyze most continuous process control designs.
Controllability: A multivariable process is controllable if the controlled variables can
be maintained at their set points by adjusting the selected manipulated variables as
disturbances enter the system.
Example 6.10. Process Controllability: Chemical plants have boilers to generate steam for
power and heat transfer. The simplified process in Figure 6.45 shows a boiler vessel where
boiling water and steam separate. We would like to control the temperature and the pressure in
the vessel, the design shows two valves adjusting the fuel combusted and the steam leaving the
vessel. Is this an appropriate design that enables the two variables (T and P) to be controlled?

Figure 6.45 Schematic of fired boiler

6-60

Operability in process design

Chapter 6 Process Control

Lets first determine whether each valve influences at least one controlled variable, i.e., do
individual input-output causal relationships exist?
Opening valve
steam temperature
vessel pressure
v101
increase
increase
v102
decrease
decrease
Opening the fuel valve will result is a hotter flame and greater heat transfer to the vessel
contents, which will generate more steam, increase the pressure and increase the temperature.
Opening the steam exit valve will decrease the pressure in the vessel and decrease the
temperature.
We might conclude that sufficient causal relationships exist for controlling the temperature
with the fuel valve (v101) and the pressure with the steam valve (v102). However, this would
not be correct!
We have to use our chemical engineering knowledge of the process. The water in the vessel is
boiling. We know that for a single component at its boiling point, the temperature and
pressure are related. (Consult steam tables for saturated steam.) Therefore, the temperature
and pressure cannot be controlled to independent values. Since pressure is important for
safety, we would control pressure by adjusting the fuel flow. In most designs, the steam
control valve (v102) would be removed and saturated steam temperature would not (cannot) be
controlled. To control the steam temperature, an additional heat exchanger, termed a steam
superheater, must be added to the process design (Ganapathy, 2001; ISA, 2007).
We conclude that the system is not controllable as originally designed.
From Example 6.10, we conclude that having the proper number of individual causal
relationships is not sufficient for controllability of a multivariable system. We must determine
whether the multiple input-multiple output system has the sufficient number of independent
relationships. We can evaluate the controllability (as defined above) by evaluating the steadystate relationship between manipulated and controlled variables. The system is controllable if a
solution exists for this set of linear equations for arbitrary values for the disturbances. (A twovariable system is shown and extension to n variables is straightforward).
6.6

where the Kijs are steady-state gains.


A solution exists for a square system of linear equations when the inverse exists for the
steady-state gain matrix. Therefore, a process system modeled by the equation is
controllable (in the steady state) if the determinant of the gain matrix is non-zero.

6-61

Operability in process design

Chapter 6 Process Control

Example 6.11 Controllability: The two streams in Figure A6.46 are being blended to make a
desired quantity with a desired concentration of component A. One stream is pure A, while
the other has no A, and there is no volume change on mixing. Can the desired total flow and
concentration be achieved by adjusting both component flow rates?

Figure 6.46 Blending process


We begin by deriving the following total and component material balances on the mixing
point.

FA FS FM
FA x A FS x AS FM x AM
1

The linearized model is directly determined to be the following.


F 'M F ' A F 'S

' FA '
FS
FS
FA
x' AM
2
2
( Fs FA ) ss
( Fs FA ) ss

The determinant of the gain matrix is given in the following.

Det ( K )

FA
1
FS

0
2
2
( FA FS )
( FA FS )
( FA FS )

Therefore, we conclude that the system is controllable.


What if the number of inputs and outputs are not equal?

If the number of (independent) controlled variables is greater than the number of


manipulated variables, the system is not controllable.
If the number of controlled variables is less than the number of manipulated variables, the
rank of the gain matrix must be equal to the number of controlled variables. This means
that at least one selection of manipulated variables can be used to control all outputs.

The conclusion about controllability depends on the process and not on a specific control
algorithm. If the system is not controllable, the conclusion is valid for any possible feedback
control. In addition, the conclusion depends on the type of controllability desired. The
definition used here for controllability is applicable to stable, continuous processes that are
6-62

Operability in process design

Chapter 6 Process Control

operated at steady state. Batch processes are inherently dynamic, so that a different approach is
required to determine controllability. Since controllability determines only whether control is
possible; it does not evaluate the control performance. Therefore, a controllable system could
provide excellent or unacceptable dynamic performance; consequently, further analysis is
required. However, a system that is not controllable certainly is not acceptable. For a broader
analysis of controllability, see Skogstad and Postlethwaite (2005).

6.4.2 Operating Window


The controllability criterion ensures that feedback control is possible, but it is exactly valid for
only small (differential) changes. Therefore, we also need to determine whether the manipulated
variables are powerful enough to compensate for expected disturbances. Therefore, the next
stage of process analysis addresses the operating window.
The operating window defines the range of variables (set points and disturbances) over
which the process can operate.
This issue is the topic of an operability chapter, so it is not covered here. However, one
example is given to reinforce the relationship between the equipment design (that provides
capacity) and process control (that can utilize the available capacity).
Example 6.12 Operating window: We wish to determine the operating window for the total
flow and composition for the blending process in Figure 6.46. The maximum flow rates are 60
m3/h for FS and 30 m3/h for FA.
The solution is plotted in Figure 6.47, which for this simple system can be presented in two
dimensions. The results represent the achievable set points for the base-case stream
compositions. We see that a total flow of 60 m3/h at a composition of 80% A is not possible.

Figure 6.47 Operating window for the twocomponent blending process.

6-63

Operability in process design

Chapter 6 Process Control

6.4.3 Multi-loop systems: Interaction


Lets consider processes that are controllable and have an acceptable range for compensation,
i.e., have a sufficiently large operating window. We now venture into the large and complex
topic of designing multi-loop control systems; by multiloop, we mean a control design consisting
of multiple single-loop controllers using PID algorithms. This can be the topic of an entire book,
so the discussions will be limited to understanding some basic concepts and key guidelines.
Fortunately, a good understanding of process dynamics and the basics from control engineering
can often provide insight for good designs, but considerable expertise can be required for
complex processes.
First, we consider why multi-loop control is different from single-loop control, or stated
differently, why cant each individual loop be designed without consideration for the other
loops? The main reason is interaction.

Interaction exists when one manipulated variable can influence more than one
controlled variable.

The importance of interaction is shown schematically in Figure 6.48. Manipulated variable MV1
influences controlled variable CV1, but MV1 also influences other parts of the process, which
have other controllers. We see that the total effect of MV1 is the sum of

the direct effect CVA, which is independent of potential effects from other control
loops (in manual) and
the interaction effect CVB, which is the result of (only) other control loops responding
to the change in MV1.

It is important to note that the measured variable CV1 is the sum of CVA and CVB and that CVA
and CVB do not exist as separate process variables they cannot be independently measured. To
determine the importance of interaction on the original control loop, we could evaluate the
steady-state gain between MV1 and CV1 (CV1/MV1) for two scenarios, one with all other
controllers in manual and another with all other controllers in automatic. This ratio is given in
the following, with all changes based on the same change in MV1 (MV1).
Measure of interaction = [CVA/MV1]/[(CVA + CVB)/ MV1]
= CVA/(CVA + CVB)
If this ratio is 1.0, interaction does not affect the steady state of MV1-CV1; if it differs from 1.0,
interaction affects the steady state of MV1-CV1.

6-64

Operability in process design

Chapter 6 Process Control

Figure 6.48 Interaction effects on MV1(s)-CV1(s) due to associated control loops


In fact, this ratio was invented by Bristol (1966), who termed it the relative gain and
represented it by the symbol ij, with i = controlled variable and j = manipulated variable.
CVi

MV
j

other loops open


ij
CVi

MV
j other loops closed

(6.7)

Details on calculating the relative gain values for all possible control loop pairings are given in
Sidebar I on Calculating the Relative Gain.

6-65

Operability in process design

Chapter 6 Process Control

Sidebar I: Calculating the Relative Gain*


The relative gain is defined in the following equation.

CVi
CVi

MV

MV
j MV constant
j other loops open

ij
CVi
CVi

MV
MV
j CV constant
j other loops closed

k
The relative gain array, which is a matrix, can be evaluated directly from the open-loop process
gain matrix, K. (We use the Hadamard product, which is element-by-element multiplication.)

K K1

ij kij kI ji

Thus, knowledge of the steady-state process gains enables us to evaluate the relative gains.

k11 k12
k
21 k 22
K ..

..
kn1 ..

k13
k23
..
..

11 12

21 22
..

..
n1 ..

k1n

..

.. knn

13
23
..
..

1n

..

.. nn

The relative gain ij gives insight into behaviour for multiloop systems when CVi is paired with
MVj. For example, the following proposed design controls CV1 By adjusting MV2, controls CV2
by adjusting MV1, and so forth.

CV
1
CV2

..
CVn

MV1

MV2

11
21

12
22

n1

..

MVn

..

.. nn

CVi

MVj

We can determine some important properties about the control design from the values of the
relative gains; see Table 6.6.
* The relative gain calculation and interpretation presented here is limited to stable processes

6-66

Operability in process design

Chapter 6 Process Control

Possible values for the relative gain and how each affects control design are summarized
in Table 6.6. We observe that interaction can have a profound effect on the behavior of
multiloop systems.
While the relative gain only addresses the steady-state behavior of the system in
Figure 6.48, it provides considerable insight into the effects of interaction. Even with only
steady-state information, the contents of Table 6.6 show that many potential designs can be
eliminated from further consideration, e.g., those with one or more negative relative gain values.
Therefore, we will use relative gains as one tool for deciding the proper control loop pairing.
We conclude that the relative gain provides useful insights and quantitative analysis.
Some potential designs can be eliminated, but loop-pairing designs cannot be based
solely on relative gain values.

Many additional performance issues are important in designing the feedback controller
loop pairing.

Pair manipulated and controlled variables with fast feedback dynamics, emphasizing the
reduction in loop dead times.
Pair manipulated and controlled variables to give large process gains (Kij). This provides
the controller with a large operating window and tends to reduce interactions. Here,
large does not necessarily mean a large absolute value; remember that the gain has
units. When evaluating the process gain, it is best to express a dimensionless gain, as
given in the following.

The range of the manipulated variable would be (i) typically 0-100% when adjusting a
valve directly or (ii) the secondary sensor range when adjusting a set point in a cascade.
The range of the controlled variable is the expected range of the measured variable would
experience, which would normally be the sensor range. When expressed as a
dimensionless value, the process gain (Kij*) should be near 1.0.

Associated with the above point, a design should favor loop pairings that have relative
gains around 1.0. Loops with positive relative gains but with a significantly different
from 1.0 (e.g., 5.0 or 0.20) could require controller tuning to be adjusted depending on
the manual-automatic statuses of interacting loops. While such retuning can be
implemented automatically, the resulting control design is substantially more complex.

Designs should avoid nested loops, where one control loop will function properly only
when another interacting control loop is in automatic. An example of a design with
nested loops is shown in Figure 6.49a. The flow controller depends on the proper
functioning of the level controller. An acceptable alternative design is shown in
Figure 6.49b.

6-67

Operability in process design

Chapter 6 Process Control

Table 6.6. Possible relative gain values and implications for control design.*
Case

CVA

CVB

Relative
gain,
0

II

CVA

1.0

III

+
|CVA|>|CVB|

> 1.0

IV

0<<1

|CVA|<|CVB|

0<

VI

0
|CVA|=|CVB|

(CVA+CVB)

Control Design
Feedback control is not possible when other loops are in manual. In
nearly all situations, this loop pairing would be eliminated from
consideration. (However, a special case where this design is used is
presented later in this chapter.)
Feedback control is possible, and the steady-state effect of interaction
is zero. Interaction does not affect the steady-state relationship
between CVi and MVj.
This loop remains a candidate for
implementation.
Interaction reduces the steady-state gain between MVj and CVi. This
results in smaller feedback gain, but the loop would remain a candidate
for implementation.
Interaction increases the steady-state gain between MVj and CVi. This
could result is overly aggressive feedback unless the controller gain
(KC) is reduced for the multi-loop situation. The loop would remain a
candidate for implementation.
Interaction changes the sign of the steady-state gain! For stable
feedback, the controller gain (KC) would have to be modified to
maintain negative feedback as the status (manual-automatic) of the
interacting loops changes! In essentially all situations, this loop
pairing would be eliminated from consideration.
Feedback control is not possible when other loops are in automatic.
This loop pairing would be eliminated from consideration.

* Table includes only CA 0. Student exercise is to extend table for CA 0.


* These interpretations are applicable to processes that are stable (without control)
* All feedback controllers have an integral mode

Figure 6.49a Example of nested control


loops (relative gain = 0)

Figure 6.49b. Example with nested loops


reconfigured.

6-68

Operability in process design

Chapter 6 Process Control

Example 6.13 Interaction measure in distillation You desire to control both product
compositions in the distillation tower shown in Figure 6.50. The pressure and both level
controllers have already been designed. You have performed experiments to determine
approximate linear models for the composition control, with the results given in the following
expression; note that the matrix contains individual input-output models.
0.0747 e 3 s
XD ( s ) 12 s 1
XB ( s ) 0.1173e 3.3 s


11.75s 1

0.70e 5 s
0.0667 e 2 s

15s 1 FR ( s ) 14.4 s 1 X ( s )
F
2 s
3 s

0.1253e FV ( s ) 1.3e

10.2 s 1
12 s 1

with
XD
XB
FR
FV
XF

=
=
=
=
=

Light key in distillation


Light key in bottoms
Reflux flow rate
Vaporized flow from reboiler
Light key in the feed

Evaluate the interaction measure and discuss the influence of these results on the control loop
pairing composition control.
The steady-state gains in the dynamic model can be used to evaluate the relative gain array, which is
given in the following.

XD
XB

FR
6.09
-5.09

FV
-5.09
6.09

We observe that only one of the possible two loop parings has positive relative gains. Therefore, the
designer must select the positive pairing, following the recommendations in Table 6.6.

Figure 6.50 Distillation tower for


Example 6.13.

6-69

Operability in process design

Chapter 6 Process Control

6.4.4 Multi-loop systems: Control loop pairing


With the process design complete with sensors and final elements included, the multi-loop
design decision becomes the pairing of measured controlled variables with manipulated
variables, which can be final elements of set points in cascade designs. The approach usually
taken is to eliminate possibilities that cannot provide adequate control performance. Then,
evaluate the remaining for dynamic performance. The criteria used here are summarized in
Table 6.7 and are discussed and applied in the following.
6.4.4a. Ensuring process controllability This topic has been addressed in Section 6.4.1. The
process must be controllable before the engineer proceeds to designing a control strategy.

Table 6.7. Criteria to determine loop pairing


Ability to control selected CVs with selected MVs
Controllability

Integrity

Ability for control system to function properly when one or


more control loops is not functioning

Range

Sufficiently large operating window when control system is


compensating for disturbances

Important variables

Expected scenarios

Performance of the most important variable(s) is the best


possible, perhaps at the expensive of poorer performance of
less important variable(s)
Control performance is best possible for the disturbances
that are expected to occur most frequently, perhaps at the
expensive of poorer performance for other disturbances

6.4.4b. Ensuring design integrity Control integrity is defined in the following.


A control system has integrity if, when one or more of the controllers is not functioning,
the remaining feedback control system is stable without changing the sign of any
controller gains.

When a feedback controller is not functioning, it is not adjusting the manipulated variable. This
can occur when the controller is placed in manual (off) or when the controller remains in
automatic (on) but the manipulated variable reaches an upper or lower bound.
Control designs with pairings on loops with positive relative gains satisfy this integrity
criterion. Therefore, positive relative gain elements are required for good control design. A
rigorous check requires that all sub-systems also have positive relative gains. For example, a
6-70

Operability in process design

Chapter 6 Process Control

3x3 system must be paired on positive relative gains. Then, each 2x2 system resulting in one
controller being off must have positive relative gain pairings. This checking procedure can
involve many instances for a large number of controlled and manipulated variables.
Note that the system is not guaranteed to be stable with one or more controllers off.
However, the system will be stable for retuned feedback controllers, with the retuning not
changing the sign of the feedback controllers.
Example 6.14 applies the integrity requirement for distillation control.
Example 6.14. Ensuring integrity in Blending- We must control the composition and total
blended flow rate from the blending process in Figure 6.46. We have already concluded in
Example 6.11 that the process is controllable. The product specification is 5% component A.
Select a design with good integrity.
We evaluate the relative gain matrix, which can be done using the steady-state gain matrix.

xAM
FM

FA
0.95
0.05

FS
0.05
0.95

By applying the criteria in Table 6.6, we conclude that we cannot eliminate either loop pairing, because
the relative gains for each are finite and positive. Therefore, both have acceptable integrity. Dynamics
are important, so we consider the feedback dynamics and observe that the responses will be fast for all
valves to all sensors (as long as the composition analyzer is close to the mixing point).
When the relative gain is near 1.0, the effective process gain and therefore, the controller
gain remains relatively unchanged, whether the interacting loop in on or off. However, if the
relative gain is far from 1.0, the effective process gain changes significantly depending on whether
the interacting controller is on or off. Therefore, the controller gain yielding stable feedback with
good performance is very different whether the interacting control is on or off. For the case with
relative gains far from 1.0, the controller gain must be adapted in real-time based on the status of the
interacting controller. The adaptation can be achieved, but it is to be avoided unless absolutely
necessary. Therefore, we select the pairing xAM FA and FM FS., which has a relative gain much
nearer 1.0.
Note that these results depend strongly on the nominal steady-state operating conditions.
Would the results change if the nominal operation were changed from 5% to 95% A in the mixture?

6.4.4c. Achieving control range The range addresses an acceptably large operating window,
so that the control system can compensate for expected disturbances and achieve the desired
range of set points. This topic was addressed in Section 6.4.2; the extension of the design
guideline presented here is that the control system should be able to achieve the required region
of the operating window based on the actions of the process controllers.

If the original design does not provide adequate coverage of the operating window, the
control design can be enhanced using split-range technology. Example 6.5 shows a control

6-71

Operability in process design

Chapter 6 Process Control

design with split range technology to ensure that the automatic controller can achieve the
appropriate range of fuel gas manipulation.
6.4.4d. Emphasizing the most important variables Often, one or more of the controlled
variables are much more important than the remaining controlled variables. In this situation, the
control design should be designed for fast, large range feedback for the more important
controlled variables. This might result is somewhat poorer performance for controlled variables
of lesser importance. A proper balance must be achieved by the design engineer.

Another approach for achieving better performance of selected controlled variables is to


tune their feedback controllers more aggressively. This can be achieved if other feedback
controller are tuned loosely, giving slower and poorer feedback control to the less important
controlled variables. Again, a proper balance must be achieved by the design.
Example 6.15 Gaining fast control for important variable In the fuel gas distribution
system in Figure 6.51, two sources of fuel (fuel gas and vaporized fuel) are provided to supply
the total demand for numerous consumers. The consumers demand flow rates independent of
each other and of the fuel gas system. The control system must provide the quantity of fuel, and
in this design, it must also regulate the heating value of the fuel to the consumers.
The control system balances the demand and supply by controlling the pressure in the distribution
system (header). The heating value is controlled using an on-stream analyzer that adjusts one of the
fuel sources. How should the two loops be paired? We base this decision on the need for fast pressure
control, while the heating value can experience short-term deviations from its set point. Therefore, we
control pressure by adjusting the faster manipulated variable, the fuel gas, and we design the
controllers to control heating value by adjusting the remaining manipulated variable, steam to the
liquid fuel vaporizer.

Steam

PC

Fuel
gas

Liquid fuel
AC

Heating
value
analyzer

Consumers,
not regulated
by the fuel
distribution
system

6-72

Figure 6.51 Fuel distribution


system in Example 6.15

Operability in process design

Chapter 6 Process Control

Example 6.16 Tuning matched to control goals Lets consider a hypothetical two-variable
process with the following dynamic model.
1 . 0 e 1 .0 s
CV1 ( s ) 1 2 s
CV ( s ) 0.75e 1.0 s
2
1 2 s

0.75e 1.0 s

1 2 s MV1 ( s )
1.0 s

1 .0 e
MV2 ( s )
1 2 s

We desire to control both output variables, and we will pair the variables on their largest gains, CV1MV1 and CV2-MV2. We encounter three common situations (i) CV1 is more important, (ii) CV1 and
CV2 are of equal importance, and (iii) CV2 is more important. How can we adjust the control system to
achieve one of these goals, while noticing that the dynamics are equal for both manipulated variables?
The answer is controller tuning. The controller for the more important variable can be tuned tightly,
with the interacting controller detuned to maintain stability. Note that simply increasing the controller
gain of the more important while leaving the interacting controller(s) unchanged could lead to overly
aggressive and even unstable behavior.
If the controlled variables are of equal importance, the two controllers are tuned to give approximately
the same aggressiveness and performance. Typical dynamic performances for three situations are
given in Figure 6.52. More details on this approach and case study, including the tuning, is available
in Marlin (2000, Chapter 20).
(b)

(a)

(c)

(Note the differences in CV2 scales.)


Figure 6.52 Multiloop PID tuning to match the
controlled variables goals. (CVs plotted as
deviation from initial steady state.)
(Marlin, 2000)
a. CV1 and CV2 are equally important
b. CV1 is more important
c. CV2 is more important

6-73

Operability in process design

Chapter 6 Process Control

Figure 6.53 Directionality for feedback and two disturbances depicted for an automobile.
6.4.4e. Achieving good performance for expected disturbance scenarios Many possible
disturbances can affect a process, but often, one or a few disturbances occur most frequently at
significant magnitudes to affect process performance. In single-loop control, the type of
disturbance was not considered because the guidelines in Table 6.5 apply to all disturbances,
e.g., feed rate, feed composition, cooling water, etc. A major difference of multivariable systems
is that they have key characteristic not appearing in single-loop systems. While single-loop and
multiloop systems are influenced by disturbance size and frequency, only multiloop systems are
also affected by direction.
To understand the concept, lets consider the automobile shown in Figure 6.53. If the
automobile were pushed forward (or backward), the driver could easily compensate and return
the automobile to its the original position. However, if the automobile were pushed sideways,
the driver would have to iterate forward and backward, slowly making progress returning to the
original position. Disturbances of the same magnitude are easy or difficult to control based on
the directions of the disturbance and feedback process. If they are aligned, control
performance is good. If they are orthogonal, feedback control is more difficult and performance
degrades. Directionality analysis applied to the automobile also applies to process control.
Example 6.16 Distillation directionality We will investigate the performance of the two
distillation quality control strategies shown in Figure 6.54. The designs are two common
approaches to achieve level and composition control in two-product distillation. We note that
the material balance design has a relative gain for the two composition loops of 0.39, while the
energy balance has 6.09. Clearly, the energy balance control has greater interaction, as measured
by the relative gain metric. But, how do the designs compare for disturbances?
The most common unmeasured disturbance is feed composition. The dynamic responses for
each control system responding to a step disturbance in feed composition are given in Figure 6.55. It
is clear that the energy balance design has better performance, with lower total IAE (and lower ISE)
and a shorter settling time.
The result contradicts s design rule sometimes proposed that the best control design relative
gain values nearer to 1.0. In this case, and in others of importance in industry, the dynamic
performance is best for the design when the feedback direction is aligned with the disturbance
direction, whether or not the relative gain is closer to 1.0.

6-74

Operability in process design

Chapter 6 Process Control

A full coverage of the directionality concepts is not possible in this process-design oriented
chapter. The interested reader is referred to material on quantitative analysis of directionality in
McAvoy et. al. (1985), Skogestad and Morari (1987) and Marlin (2000, Chapter 21).

Figure 6.54 Candidate distillation control strategies for Example 6.16.

(a) Energy balance control

(b) Material balance control

Figure 6.55. Dynamic behavior for two distillation tower control designs. (Marlin, 2000)

6-75

Operability in process design

Chapter 6 Process Control

There are situations where engineers can encounter conflicts in applying the five guidelines for
loop pairing in Table 6.7. Lets consider an example where a conflict among the guidelines
occurs.
Example 6.17Conflicts in the control of a pyrolysis heater Two basic control objectives for
a fired heater are (i) process fluid throughput flow control and (ii) process fluid outlet
temperature control. The manipulated variables are the feed flow valve and the fuel flow valve.
(We will assume that adequate air is provided to the burner.) What is the proper loop pairing?
The conventional design is shown in Figure 6.56a. There is a causal relationship between the
controlled and manipulated variables, the pairing is on positive relative gain elements, and the
dynamics of the temperature control are usually adequate, although having minutes of delay and
time constant. However, the conventional control does not provide extremely tight temperature
control and cannot provide separate control for multiple passes (pipes). Furnaces with multiple
passes operating at high temperatures require independent temperature control for each pass.
The strategy in Figure 6.56b achieves the tight, multi-pass temperature control. However, it
suffers a disadvantage. The causal relationship between the total feed flow controlled variable
and the manipulated fuel requires the temperature controllers to be in automatic (on). Thus, the
design in Figure 6.56b has nested loops and has a pairing on a zero relative gain. These
characteristics are to be avoided if possible, but they can be tolerated if there is an overwhelming
advantage. When applied, the design in Figure 6.56b should have a monitor that does not allow
the flow controller to be in automatic unless all temperature controllers are in automatic.
The conventional design in Figure 6.56a is applied in essentially all heaters whose
purpose is to raise a stream temperature. The unconventional design in Figure 6.56b is used by
some practitioners for heaters with multiple passes where chemical reactions occur in the passes.

(a)

(b)

Figure 6.56 Two fired heater control strategies.


(a) Conventional with positive relative gain
(b) Unconventional with zero relative gain but fast, independent temperature loop dynamics for
every pass

6-76

Operability in process design

Chapter 6 Process Control

This example serves as a warning about guidelines. Good guidelines are often correct
and give useful assistance to the engineer; however, guidelines are not fundamental principles.
The engineer must understand the basis for the guidelines, know when they should be observed,
and know when they should be violated, with care.
The presentation of multiloop control in this section has introduced many key issues and
provided guidance on analysis methods and design technology. The reader should recognize that
this is an enormous topic and full coverage would require a separate book. The reader can access
helpful materials given in Additional Learning Resources. In should also becoming clear that
control design requires the understanding of both process and control technology, and each
design is based on a thorough definition of the pertinent issues, including performance goals,
process behavior, relevant constraints, key disturbances, and instrumentation available. The next
section provides guidance on a structured design procedure to integrate all pertinent issues.

6.5 Control System Design Procedure


As is often stated by professors, If you dont know where you are going, any path will do.
Certainly, this old adage applies to control system design. A great deal of important information
is contained in the process material and energy balances and the process drawings. However, the
complete definition of the required control performance (safety, economics, production rate,
product quality, etc.) and the situations expected in the plant (disturbances, constraints, etc.) are
not. Here, a method for defining the problem and steps for completing the design are presented
and applied to a simple example, the flash separation process in Figure 6.57.

Figure 6.57 Flash process with preheat. (Marlin, 2000)

6-77

Operability in process design

Chapter 6 Process Control

6.5.1 Defining the control design problem


The design definition is summarized in the control design form. The definition is a
summary of business data, performance targets, and preliminary process flowsheet and
equipment information. Business data includes feed, product and fuel values, the ranges for raw
material properties, production rates, and product qualities, including multiple product
specifications for similar materials. Performance targets involve safety and reliability targets
(covered in more detail elsewhere in this monograph). Preliminary process design includes
equipment type (e.g., centrifugal or positive displacement pump), equipment sizing, pressure
rating and instrumentation selection. When designing controls, the engineer must re-evaluate all
of these decisions and modify them as required.
A sample control design form is presented in Table 6.8 for the example process. The first
entry in the table after the heading presents the control objectives using the seven categories
proposed by Marlin (2000). The definitions should be as detailed as possible, giving, for
example, sizes of disturbances, set point changes, and allowable deviations in key variables. The
control objectives should not contain or imply design solutions; they should be based on desired
operation for safety, product quality, profit, and so forth. Note that issues that do not fit the
general categories can be documented in the tables last entry, additional considerations.
The next two entries define the instrumentation, measurements and final elements.
Clearly, these preliminary decisions will be evaluated and possibly improved during the control
design.
The next category defines the constraints or limitations that influence dynamic operation.
Some of the constraints result from equipment performance limitations. Other constraints result
from bounds on operating conditions outside of which the process could be unsafe or produce
useless products. Violations of hard constraints lead to unsafe operation or extreme economic
loss, such as shutting down the process for repairs. Violations of soft constraints can be
tolerated, at least for limited duration, but with economic loss; an example could be a high
temperature in a reactor that would degrade the catalyst activity.
The following category defines the key disturbances, giving approximate ranges and
frequencies of occurrence. Engineers accustomed to operability analysis will be able to identify
these disturbances from knowledge of the integrated processes and the equipment being used.
Certainly, the process dynamics should be understood for control design. Dynamic
models for all inputs and outputs would be ideal; however, empirical models cannot be
determined because the process does not exist, and fundamental models often require excessive
time and cost to develop. As a minimum, all key aspects of the dynamics should be noted,
including open-loop unstable variables, variables affecting safety, very slowly responding
variables, etc. For new processes, where experience does not exist, a fundamental analysis of
dynamics using simulation might be warranted.

6-78

Operability in process design

Chapter 6 Process Control

Table 6.8. Preliminary Control Design Form for the flash process in Figure 6.57
(Marlin, 2000)
TITLE:
Flash drum
PROCESS UNIT: Hamilton chemical plant
DRAWING :
Figure 6.57

|ORGANIZATION:
|DESIGNER:
|ORIGINAL DATE:

Profit, Inc.
I. M. Learning
January 1, 2011

CONTROL OBJECTIVES:
1) Safety of personnel
a) the maximum pressure of 1200 kPa must not be exceeded under any (conceivable) circumstances
2) Environmental protection
a) material must not be vented to the atmosphere under any (conceivable) circumstances
3) Equipment protection
a) the flow through the pump should always be greater than or equal to a minimum
b) cavitation in the pump should be prevented
4) Smooth, easy operation
a) the feed flow should have small variability
b) product flows should have modest variability, not greater in percentage than the feed flow variability,
which includes production rate changes
5) Product quality
a) the steady-state target of value 10 mole% of ethane in the liquid product should achieved for operating
condition changes of +20 to -25% feed flow, 5 mole% changes in the ethane and propane in the feed,
and -10 to +50 C in the feed temperature.
b) the ethane in the liquid product should not deviate more than 1 mole % from its set point during
transient responses for the following disturbances
i) feed temperature experiences a step from 0 to 30 C
ii) feed composition experiences a step of +5 mole% ethane and -5 mole% of propane
iii) feed flow set point changes 5% in a step
6) Efficiency and optimization
a) the heat transfer should be maximized from the process integration exchanger before using the more
expensive steam exchanger
7) Monitoring and diagnosis
a) sensors and displays needed to monitor the normal and upset conditions of the unit must be provided.
Typical faults for this process would involve sensor or valve failures that would affect the flash
temperature and/or pressure.
b) sensors and calculated variables required to monitor the product quality and thermal efficiency of the
unit should be provided for longer term monitoring. One typical slowly changing aspect of this process
would be heat exchanger fouling.
MEASUREMENTS:
Variable
Principle
A1
chromatograph
F1
orifice
F2
orifice
F3
orifice
F4
orifice
F5
orifice
L1
delta pressure
P1
piezo electric
T1
thermocouple
T2
thermocouple
T3
thermocouple
T4
thermocouple
T5
thermocouple
T6
thermocouple

Value, Sensor Range


10, 0-15 mole%
100, 0-200
120, 0-200
100, 0-200
45, 0-90
55, 0-110

range is lower half of drum


1000, 500-1500 kPa
0, (-)50-100 C
25, 0-100 C
90, 0-200 C
45, 0-200 C
25, 0-100 C
25, 0-50 C
6-79

Special Information
update every 2 minutes

Operability in process design

Chapter 6 Process Control

MANIPULATED VARIABLES:

I.D.

Maximum capacity (at design pressures)


(%open at design, maximum flow)
100%, 100
53%, 189
50%, 200
14%, 340
52%, 106

v1
v2
v3
v4
v5
CONSTRAINTS:

Variable

Limit values

drum pressure
drum level
Ethane in F5
product

1200 kPa, high


15%, low
10 1 mole%,
(max deviation)

Measured/
inferred
P1, measured
L1, measured
A1, measured &
T6, inferred

Hard/
soft
hard
hard
soft

Penalty for violation


________________
personnel injury
pump damage
reduced selectivity in
downstream reactor

DISTURBANCES:

Source
feed temperature (T1)
feed rate (F1)
feed composition

Magnitude
-10 to 55C
70 to 180
5 mole% ethane

DYNAMIC RESPONSES:

(input = all manipulated variables and disturbances)


(output = all controlled and constraint variables)
Gain
Dynamic model

Input

Output

Dynamics
infrequent step changes of 20C magnitude
set point changes of 5% at one time
approximately periodic changes (every 2-3 hr)

(Determined empirically after the process has been started up)


Bottoms liquid level
Drum Pressure
Flash temperature
Feed flow rate
Bottoms % ethane

- This variable is unstable and must be controlled.


- This is a critical safety and process environment variable that changes rapidly and can
exceed limits quickly. Feedback control is required.
- The temperature is not high enough to damage equipment. Good temperature control
is essential to achieve the desired separation
- The feed flow rate can vary but should do so slowly so that the flash pressure and
temperature do not deviate significantly from their set points
- This is the purpose of the equipment! In this process, the specifications allow little
variation, so an analyser is provided for on-stream analysis

ADDITIONAL CONSIDERATIONS:
a)
Liquid should not exit the drum via the vapor line
b)
The composition analyzer is much less reliable than other sensors

6-80

Operability in process design

Chapter 6 Process Control

The final entry addresses the important topic of monitoring and diagnosis, which should
be addressed at the design stage to ensure that adequate sensors and sample locations are
provided. Knowledge of equipment enables engineers to define the likely faults that result from
equipment malfunction and human failure. Then, important sensors for identifying each fault
could be determined using a cause-effect diagram.
This form may seem a bit pedantic, requiring excessive documentation for every
decision, and in fact, most control designs are performed in practice without such extensive
documentation. The form is used here because it provides an excellent structure for beginning
engineers who after gaining proficiency, may perform the analysis without the form. However,
even the most experienced engineers benefit from this type of documentation for complex
designs.
It is important to recognize that experienced engineers can sometimes by-pass the
Control Design Form (CDF) documentation, but they always perform a thorough
analysis involving information and issues included in the CDF.

6.5.2 Designing the control system


Describing a design procedure is always difficult, because individuals solve problems using
many good (and some poor) procedures that are tailored to the specific problem. A typical
sequence of analysis steps are proposed here that should help the reader and inexperienced
engineer address complex control design issues. As the individual gains experience, he/she can
personalize the sequence.
The control design procedure shown in Table 6.9 begins by collecting basic information
on the preliminary process design. Then, the control design form is completed to document the
information required for the design. Next, the feasibility of achieving the goals is evaluated by
ensuring that the design has the adequate number of degrees of freedom and adequate capacity,
i.e., the operating window is large enough. The following step involves understanding the
process and operational goals, approximate dynamics, constraints, disturbances, and so forth that
influence the design. More quantitative analysis, such as relative gain evaluation, could be
performed at this step. Then, the design is begun by defining controls in the following sequence.

The overall flow and inventory control


Each unit control
- Flow and inventory
- Process environment (pressure, temperature, flow ratios, etc.)
- Product or intermediate stream qualities
- Safety as required
Optimization
Monitoring and diagnosis
- Short-term monitoring of fast-occurring situations
- Long-term monitoring of process behavior
6-81

Operability in process design

Chapter 6 Process Control

Table 6.9. A Control Design Procedure. Marlin (2000)

6-82

Operability in process design

Chapter 6 Process Control

The design sequence does not necessarily involve a linear sequence of steps. As some
design decisions are evaluated, the engineer could decide that either the process equipment
design or previous control decisions should be modified. If so, iteration is required to modify
previous decisions in a manner that enables the engineer develop designs that meet all goals. To
perform the design with minimal iterations, the design engineer must be able to look ahead and
anticipate later aspects of the design. For example, when designing the production rate and
inventory controls, the engineer should not use valves for production rate control that are more
appropriately used for important safety and product quality control.

6.5.3 Control design for the flash process


The control design procedure can be applied to the flash process in Figure 6.57. The entire
procedure is presented in Chapter 24 of Marlin (2000). Here, a few key results are presented
along with the final design. We begin by provisionally defining five controlled variables as the
vessel pressure, liquid level, feed flow rate, flash temperature, and ethane composition in the
liquid product. Next, we determine if a sufficient number of valves exist. We see that five
valves are present, which is sufficient to control the five controlled variables. However, we must
investigate further by evaluating the controllability using the steady-state gain matrix. (Note that
the level is unstable without control and has no steady-state gain, but the analysis can proceed
using the derivative of the level (McAvoy, 1983).)

0
2.0
0
0 v1
F1 0

T6
.0708
.85
.44
0
.19 v 2

0
.043 v3
A1 .00917 .11 .44

P1 .567
6.80 1.39
0
5.36 v 4


dL1 / dt .0113 .136 .31 .179 .0265 v5
We evaluate the determinant of the gain matrix.
x 12

6-83

Operability in process design

Chapter 6 Process Control

First, we observe that the determinant is essentially zero, from which we conclude that
the five measured variables cannot be controlled using the five valves. We also observe that the
first and second columns can be made identical my multiplying by a constant value of
approximately 12. This indicates that valves v1 and v2 have the same effects on the measured
variables. We will remove one manipulated variable from this analysis and as a result, must
reduce the number of controlled variables by one. We note that valves v1 and v2 heat the feed,
so that they have the same effect of flash temperature (T6), pressure, level and composition. The
least important controlled variable is the flash temperature; it does not affect safety and is a
process environment variable. The resulting gain matrix is non-singular, indicating that the
process can be controlled (at least in a small region around the design point).

F1


A1
P1


dL1 / dt


v 2

0
.043 v3
.11 .44

6.80 1.39
0
5.36 v 4
.136 .31 .179 .0265 v5
0

2.0

Next, we determine if the process has the capacity to respond to defined changes in
disturbance variables while maintaining safe operation and achieving the desired liquid product
composition. For this small process, we can sketch the operating window in Figure 6.58. We see
that the process has the capacity required for large changes in feed temperature and flow rate.
Since the process equipment is satisfactory, we move to the loop pairing. We want a
design that functions properly when one controller is placed in manual or equivalently, when one
valve saturates. Therefore, we evaluate the relative gain array below.

The relative gain analysis indicates that there is only one loop pairing design for this
process with positive relative gains for all control loops! Next, we quickly evaluate the
dynamics for the loops indicated by the relative gain analysis. We see that the dynamics are
favorable, from which we conclude that the valve that has the strongest effect on the variable
also has fast dynamics. Therefore, we accept the design.
We still have some important variables to integrate into the design. We recognize that the flash
temperature is important. In a previous section, we determined that the flash

6-84

Operability in process design

Chapter 6 Process Control

Figure 6.58 Operating window for the flash


process. Marlin (2000)
temperature is a good inferential variable for the liquid composition. Therefore, we include T6
as a secondary variable in the analyzer feedback control loop. (This does not violate the
controllability analysis because there is no specific value for this temperature; its set point is
adjusted by the analyzer controller.)
In addition, we have an additional valve, v1. The valves v1 and v2 both heat the feed, but
the costs of the two are different. The process heat integration is much less costly than the steam
heat that requires fuel combustion. However, we need both exchangers to provide the capacity
for changes in operating conditions. Therefore, we introduce split range control for the flash
temperature that maximizes the use of preheat before opening the steam valve.
The completed design is shown in Figure 6.59, which includes additional equipment for
safety. An example of a dynamic response to a feed composition disturbance is shown in
Figure 6.60 that shows the successful regulation of all variables.

Figure 6.59 Control design for the flash process. Marlin (2000)
6-85

Operability in process design

Chapter 6 Process Control

Figure 6.60 Dynamic response of the flash control system to a step disturbance in the feed
composition. (Time in minutes, temperatures in degrees Celsius and compositions
in mole percent.) Marlin (2000)

6-86

Operability in process design

Chapter 6 Process Control

6.5.4 Unique features of plant-wide control


The definitions of plant-wide control problems often contain common features. Here, we discuss
a few of the common features and control designs to achieve good performance. We do not
introduce new basic control methods; we tailor the technology to achieve the demands of the
process dynamics, safety, product quality, production rate, and profitability.

6.5.4.1 Production rate and inventory control


One important goal of any process operation is to ensure that material entering equals the
material leaving, corrected as needed for some accumulation or depletion consistent with the size
of inventory storage. The goal of production rate control is to ensure that the major flow
variables match the production rate without human intervention. Several general issues that have
to be considered for production rate control are discussed in the following.

Inventory control Most process plants contain inventories, so that materials flow from
one intermediate inventory to the next. Therefore, production rate control cannot be
achieved by a single flow controller; a series of flow and inventory controllers are
required because setting one flow into or from an inventory does not ensure that the other
(from or into) flow will match the desired production rate. For example, most liquid
levels are pumped as shown in Figure 6.61. Changing the inlet flow rate to the vessel
does not affect the flow rate exiting the vessel. The system is unstable, and feedback
control of the level is required to ensure that the flows in and out are equal at steady state.
In the other common example, the inventory of gas in a vessel is measured by the
pressure as shown in Figure 6.62. A change in the flow in affects the vessel pressure that
subsequently affects the flow out, so that the pressure is a stable variable without control.
While this system would theoretically attain a steady state with the flows in and out
matching, the large changes in vessel pressure are usually not acceptable due to the high
cost for a high-pressure vessel. Therefore, pressure is controlled to maintain a safe
condition in the vessel as well as to match flows in and out.

Figure 6.61 Liquid level control with


pumped out flow and control

Figure
control

6-87

6.62 Gas

inventory

pressure

Operability in process design

Chapter 6 Process Control

Figure 6.63 Production rate and level control.

Location of production rate controller In a series of flows and inventories, only one
flow can be set independently by operating personnel; all other flows are determined by
the inventory controllers so that material balances are achieved. Therefore, the engineer
has to select the best location for the production flow controller. The most obvious
choice is the feed of raw materials into the first unit as shown in Figure 6.63 referred to
as a feed push design. (In the figure, each liquid level represents what might be a more
complex process with inventory.) While this design results in little variability in the first
unit, it allows more variability in later units and cannot rapidly achieve a specified
product flow rate. A second choice could be the product flow rate as shown in
Figure 6.64; this design provides tight product rate control but allows variability in
upstream flows. This is often referred to as a product pull design and is used where the
process must rapidly satisfy a variable demand. Other choices are possible within the
process as shown in Figure 6.64. The unit with feed flow control is usually selected to be
the most sensitive to flow variation, so that very low variability flow rate yields the best
performance, such as high reactor yields, highest production rate or profit. This is often
referred to as bottleneck control.

Figure 6.64 Bottleneck production rate and level control

6-88

Operability in process design

Chapter 6 Process Control

Multiple flows Often, a plant processes several materials that must be supplied
concurrently. Usually, these materials are provided in a specific ratio, and the ratio is
adjusted based on feedback from process behavior, such as product yields or purity. A
flow ratio design can achieve the desired flow rates with only one flow being independent
and all others being supplied according to specific ratios.

A caution on ratio control is noteworthy. Ratio control should not be applied to


the output flows from reaction or separation processes. An example is shown in
Figure 6.65. Controlling both product flows from a distillation tower is generally not
acceptable, because the ratio of flows depends on the feed composition (and reflux ratio);
this design will likely result in inventories completely draining or filling. At least one of
the product flows for a distillation tower should be determined by an inventory controller.

Discontinuous units Many process plants consist of some continuous processes and
some discontinuous processes. For example, the discontinuous process could be a
chemical reactor, and the continuous process could be a separation sequence after the
reactor. Plants are often designed as shown in Figure 6.66, with multiple discontinuous
processes in parallel and some inventory between the discontinuous and continuous
sections of the plant. The plant is operated so that one discontinuous process completes a
batch and transfers its material to the inventory, which must have capacity to store the
complete batch quantity. The flow rate to the continuous process is set to process, on
average, the entire batch by the time that the next batch will be completed and transferred
to storage. In this case, the inventory is not controlled automatically.

Figure 6.65 Distillation with both products determined by flow control (Not recommended)
6-89

Operability in process design

Chapter 6 Process Control

Figure 6.66 Batch reactors feeding continuous processes with intermediate inventory.

Raw materials and finished product inventories - Whenever inventories are expected
to vary for good plant operation, the inventories are not controlled automatically, and
people monitor the inventories and redirect flows as required. Naturally, sensors and
alarms are provided to aid the operations personnel. A typical situation involves raw
material and finished product inventory. These inventories are very large, containing
material for many days of plant operation when transfers to and from the plant occur
infrequently.

6.5.4.2 Utilities control


An important, if often overlooked, aspect of plant-wide operations are the utilities generated and
consumed in the plant. These utilities include, but are not limited to, steam, cooling water,
compressed air, liquid and gas fuels, hydrogen, oxygen, heating or cooling fluids, and electricity.
A utility is an entity that is required to produce the products; we exclude raw materials and
consider them separately. The plant operation is simpler when the plant purchases the utilities
from an outside company; however, external purchasing is often more costly because of the
profit gained by the outside company. In addition, some by-products can be most economically
used as utility streams. Therefore, a process plant often generates some or all of the utilities it
consumes.
The operation of a utility process is different from a conventional process because it is
required to satisfy the second-to-second demand of the plant. Some of the requirements include
matching the production with consumption, responding very rapidly to changes in demand,
operating over a wide range of generation, providing the utility reliably, and generating the
utility at low cost. The importance of utilities becomes clear when we recognize that the entire
plant depends on immediate supply of each utility to match every individual process unit
requirement.
6-90

Operability in process design

Chapter 6 Process Control

Some key issues in utility design, operation and control are discussed in the following.

Balancing production and consumption The utility system must match its generation
(perhaps, including external purchases) to the total process consumption. Since many
process consumers exist and they function independently, this can be challenging. Lets
take steam as an example utility. One method for determining the consumption of steam
would be to measure the flow rate of steam to every process consumer, heat exchanger,
reboiler, steam turbine, and so forth, as shown in Figure 6.67. Then, we could ensure that
the generation equaled the calculated consumption. However, this would be a terrible
approach because errors in flow measurement would always result in an erroneous
calculated value of total steam consumption. The pressure would increase or decrease
until a major incident occurred in the plant.
A better method is to rely on the steam system to perform the material balance, as shown
in Figure 6.68. Whenever the flows into and out of the steam pipe are unequal, the
pressure changes. Therefore, we use the pressure as an indication of how well the
generation and consumption match. To automate this concept, we install a pressure
controller that adjusts the steam generation. This ensures that the total (integrated) steam
consumption and generation are equal and that deviations are corrected quickly, due to
the fast response of the pressure in the small pipe volume.

Wide range of operation Each of the many units in the plant consumes the utility and
might be operating at low or high production rates (and utility consumption rates), while
some might be shut down. Therefore, the total utility rate can experience wide variations
from day-to-day and very wide variations from month-to-month. In additional, process
disturbances, including emergency shutdowns, can cause large variation over minutes.
The challenge is to design a generating system that can operate over such a wide range
efficiently. Again, we can consider steam as an example utility that is generated in one or

Figure 6.67 Steam balance by equating


manipulated to measured flows (not
recommended).

6-91

Figure 6.68 Steam balance by pressure


control (recommended).

Operability in process design

Chapter 6 Process Control

steam to consumers

F
stm
PC

FC

FC

fuel
FY

fuel

fuel

fuel

Y=dX+e
automatic-

A/M

FC

FC

X
same control to other boilers

manual
switch

Figure 6.69 Typical mutliboiler steam header control.


The ratio (d) and bias (e)
parameters can be adjusted
independently for each boiler to
achieve the highest efficiency.
Marlin (2000)

more boilers heated by fuel combustion. A boiler can operate over a relatively wide
range, perhaps 25-100% of its maximum capacity. However, a boilers efficiency (heat
to boil steam/total heat of combustion) is low at lower generation rates. Therefore, a
typical design has several boilers that can be placed in operation when needed. As a
result, a wide range of operation and a reasonable efficiency can be achieved
concurrently. A typical control design is given in Figure 6.69.

Reliable utility generation The entire plant depends on utilities, so that reliability is
essential. For steam generation, a boiler is a complex unit operating near material and
combustion safety limits. Therefore, it is not unusual for a boiler safety and protection
system to detect an incipient fault (e.g., potential loss of water circulation or loss of
sufficient airflow) and shutdown the boiler one or more times per year. This situation
would be disastrous if only one boiler were included in the design. In addition, most
utility generators require considerable time for startup; for example, a boiler start from
ambient temperature might require many hours. Therefore, more boilers are typically in
operation than needed to satisfy the total demand, so that upon a single boiler failure the
remaining boilers can satisfy the total plant steam demand. In the event of two boilers
failing, the plant will have a load shedding plan that will stop some less important
consumers and switch others to an alternative power source such as electricity. The load
shedding is devised to maintain units in operation whose shutdown costs the most and
requires the greatest time for restarting.

In some instances, alternative sources of the utility increases the reliability of the utility
system. For example, alternative sources of pipeline natural gas provide a backup for a
utility that is designed to consume by-product methane and ethane in combustion
systems. The backup sources are more expensive (if they were not, they would be the
primary sources). Therefore, the control systems are designed to call on them only when
needed. A typical fuel gas distribution for a process plant is shown in Figure 6.70.
6-92

Operability in process design

Chapter 6 Process Control

Figure 6.70. Typical fuel gas distribution system in a process plant. Marlin (2000)
Another method for providing reliability in a utility in the plant is to include an inventory
that can be used to satisfy demand, at least for a short time until a fault is corrected and
utility operation resumes. For many situations, this is not possible, but it is an option for
some utilities such as air separation, where storage of liquid oxygen and nitrogen is
possible. Naturally, liquefying and subsequently vaporizing the material increases the
cost.

Rapid response to disturbances The feedback control systems for utilities using fastresponding variables can maintain most systems in balance. However, situations arise in
which very large step disturbances occur that result in large deviations and potential
utility system collapse, where the generation is so much lower (or greater) than demand
that the plant cannot remain in operation. For example, a large steam demand occurs
where a unit is started up, and a steam excess occurs when a unit is shutdown. Naturally,
the first approach to improve the dynamic behavior will be to slow the disturbance;
however, this is not always possible or cost effective. In these situations, the control
system must be enhanced with feedforward control.

Profitability Utilities involve fuel and power consumptions that have a major impact
on overall plant profitability. Therefore, these systems need to be operated efficiently. As
discussed, these systems typically have numerous parallel methods of generation and
sometimes, external sources as well. The control system should be designed to utilize the
least costly generation while satisfying the total demand.
In some instances, substantial improvements can be made by scheduling process
operations to lower utility costs. For example, periodic operations such a material

6-93

Operability in process design

Chapter 6 Process Control

movements among storage locations that consume electrical power can be performed
when the electricity prices are low, usually during the night.

6.5.4.3 Process recycle


Process integration through recycle of material and energy can provide substantial economic
advantages. However, it is well recognized that integrated units can be more difficult to operate
because a change in one unit propagates a disturbance to other units that ultimately returns to the
originating unit. Thus, a poorly designed and operated plant with recycle can experience many
disturbances that cycle through integrated units. Fortunately, proper design and control can
substantially reduce the negative operability issues without losing the economic advantages of
recycle.
The key complicating effect of recycle systems is the feedback occurring as part of the
process. A process sketch and a block diagram of a simplified reactor process with recycle are
shown in Figures 6.71a and 6.71b. The transfer function between a disturbance in the inlet
temperature and the reactor outlet temperature is given below.

T 4( s )
GR ( s )GH 1 ( s )

T 0( s ) 1 GR ( s )GH 2 ( s )
Now, we will give specific transfer functions, which could be for an exothermic chemical
reaction (KR = 3 > 1).
Process without recycle

Figure 6.71a. Typical process with


recycle.

Process with recycle

Figure 6.71b. Block diagram of the typical recycle


process in Figure 6.71a

6-94

Operability in process design

Chapter 6 Process Control

The question we first investigate is, How does the recycle change the effect of a disturbance in
T0? The responses of two designs are given Figure 6.72 for a step change in the inlet
temperature, T0. In the top figure, the response for a process without recycle is plotted, and in
the bottom figure, the response for the recycle system is plotted. We note that the effect of the
disturbance is much larger and slower in the recycle system. To understand why, we rearrange
the overall transfer function to obtain the following.
4
0

We see that the effective gain and time constant are influenced by recycle, and as the recycle
gain becomes larger (and KH2KR approaches 1.0), the disturbance effect becomes larger. This is
process feedback, but unlike the feedback provided by process control, this is bad feedback.
When we look at the definition of feedback, we recognize the process feedback in this example
as positive feedback, which tends to drive the output variable away from the desired value.
Process recycle systems almost always introduce positive feedback that magnifies the
effects of disturbances and lengthens time to steady state.
Recycle is introduced into process designs to realize desirable efficiency effects, but it also
introduces positive feedback in the process. The engineer should include control designs that
provide the good efficiency while ensuring that the process has good dynamic performance A
few approaches for the design and control of recycle systems are discussed in the following.

Figure 6.72 Dynamic response of two designs, with and without recycle. Marlin (2000)

6-95

Operability in process design

Chapter 6 Process Control

Break the positive feedback A direct method for reducing the positive feedback from
recycle is to adjust a compensating source or sink of the recycled entity (material or
energy). As an example, lets consider the reactor with feed-effluent heat exchange in
Figure 6.71a. The positive feedback results from the reactor effluent temperature
affecting the reactor inlet temperature. This can be prevented (perfectly in steady state
and with some deviation dynamically) by introducing an additional heat exchanger as
shown in Figure 6.73, which would increase the capital cost. The heat exchanger heats
the reactor feed to a fixed temperature. In this case, the heat recovered by the feedeffluent exchanger must be reduced and a heating fluid must be provided, which could
result in higher operating costs.
Applying this principle requires that some recycled entity not be recovered; however,
most of the recycle is returned to the process. This type of design is very effective as
long as the process and control dynamic response for the unit breaking the positive
feedback is fast with respect to the other elements in the recycle system.

Smooth dynamics through inventory control In some processes, there is a long time
between a change in the input to the process and the response of the recycle. For the
example in Figure 6.74, a solvent is added to a reactor and is recovered, purified, and
returned to the reactor. The recycle loop is lengthy, with delay between an increase in the
solvent flow to the reactor and the solvent return. Therefore, the design in Figure 6.74
includes an inventory for storing solvent, which ensures that the reactor can be provided
the appropriate amount without delay. The direction of level control involves
manipulating the outflow (push system) for flows heading towards the intermediate
storage tank and manipulating the inflow (pull system) for flows away from the
intermediate storage tank. We note that the storage inventory is not controlled; it rises
and falls to account for temporary changes in solvent processing rates. Level control in
recycle has been discussed by Buckley (1974).

Figure 6.73 Recycle process with reactor


inlet temperature control.

6-96

Operability in process design

Chapter 6 Process Control

FEED

PRODUCT

PC

LC
FLASH DRUM

X
LC

FY
REACTOR

SOLVENT

SOLVENT

FC
LC
LC

REACTOR

(PURIFYING
SOLVENT)

FC

HEATER

PURGE

FC

MAKE-UP

Figure 6.74 Solvent recycle process with uncontrolled recycle inventory. Marlin (2000)

Prevent component accumulation Material recycle systems provide the opportunity


for a component to accumulate in the process. In an extreme example, a component
might not exit the process in any product stream. In such a case, the component would
accumulate in an unlimited manner, and upset the process operation. The classical
solution is to introduce a purge stream from the process in a location where the
accumulating material is at a high concentration.

The purge will contain some of the accumulating material and will enable steady state to
be achieved. Note that many other components exit in the purge as well, and if these are
valuable, the purge stream should be processed in a separation unit and the valuable
components returned.
Component buildup is also possible in recycle systems with a chemical reactor in the
loop. This topic is addressed in detail in Luyben et. al. (1999).

A.5.4.4 Partial control

Partial control is a concept closely akin to inferential control. Inferential control is generally
confined to product quality in a single unit, so that all measured and manipulated variables are
located in the unit. In contrast, partial control addresses a wider range of objectives, such as
profitability. In addition, partial control may require adjustment of variables in integrated units.
(As an aside, this general approach has also been termed self-optimizing control.)

6-97

Operability in process design

Chapter 6 Process Control

Typically, the goal of partial control is profitability of the entire plant, such as the
expression given below.
Profit = product sales revenues raw material costs fuel costs power purchase
changes in inventory (raw material, products, work in progress)
- catalyst and chemicals cost of utilities
We note that important costs that are not affected by the control system, like capital, personnel,
laboratory, and so forth are not included in the measure of operating profit. In special
circumstances, profit can be more concisely expressed as maximum selectivity or minimum
energy or another simple goal, but the simplification should be investigated thoroughly before
being accepted.
When designing partial control strategies, the engineer must determine surrogate
variable(s) that, when controlled in feedback to a constant set point(s), result in process
performance that closely approximates the desired goal, i.e., optimum profit. The surrogate
variables are often termed dominant variables.
Clearly, the advantage of controlling the surrogate dominant variables is much faster and
simpler feedback control. An inherent disadvantage is the approximate relationship of dominant
variables to the true profit. This disadvantage could be overcome by a higher-level profit
controller that uses models and real-time data to reset the values of the dominant variables in a
cascade design.
The design of the dominant variable(s) is a task for the engineer. To begin with, the
dominant variable(s) must be controllable, as discussed previously in this appendix. In addition,
they should provide fast feedback information about process performance, to ensure fast
feedback control response. Finally, the variable(s) should be highly correlated with the ultimate
control goal, such as product quality, profit, etc.
Controlling dominant variables at their set points will maintain the process in
conditions yielding profit close to the maximum as disturbances occur.

Engineers can use two methods to determine an appropriate selection of dominant


variables for partial control design. One method uses fundamental models, while the other uses
historical data. The reader is referred to Luyben et.al. (1999) and Skogestad (2004) for details on
partial control design. Some typical dominant control variables include the following.

Ratio control of components to blending and reactors


Ratio of utilities to feed rate to separation units
Concentration of a limiting component in a chemical reactor

6-98

Operability in process design

Chapter 6 Process Control

6.5.5 Final step - safety review

The final stage of the control design is a safety review. Remember that an improperly designed
control system can lead to hazards and that even a properly designed control system introduces
new faults that can lead to hazards. Therefore, a process safety analysis is performed after the
process control system has been designed and shown on the P&I Drawings. Generally, a hazard
and operability study (HAZOP) would be performed followed by layer of protection analysis
(LOPA), if necessary. Also, changes to the process control design in an operating plant might
have unforeseen consequences that compromise safety. Therefore, a management of change
program must be performed for changes proposed after the initial plant-wide process safety
analysis.

6.6. Unit operation control


Most process plants are composed of a complex interconnection of standard unit operations.
Here, control of several of the more common unit operations is discussed. The emphasis is on
the integration of process and control principles to select the best manner for control. In
addition, many options exist in the design of the unit operation, and the impact of changes in
process design on the control performance is addressed.
We observe that many factors influence control performance. In designing a control system,
we desire a process design and control system that has

Range: A wide range of operation


Gain: A large steady-state gain, i.e., a strong effect
Dynamics: A fast response
Linearity: A nearly linear relationship between manipulated and controlled variable, so
that a controller functions well with constant values for tuning parameters

6.6.1 Flow control


The simplest control loops with one measured, controlled variable and one manipulated variable.
In this section, we will introduce issues and designs that are often misunderstood and will see
correct designs.
Examples 6.18 to 6.22 address one of the simplest topics in design, regulation of flow in a closed
conduit. Even this simple control system requires knowledge of the equipment performance.
From many designs that could provide good flow control, one design often provides the highest
energy efficiency. While learning these designs is useful, the more important lesson is that
control design requires in-depth understanding of the process goals and equipment performance.

6-99

Operability in process design

Chapter 6 Process Control

Figure 6.75 Flow process with


centrifugal pump and variable
resistance by valve.

Example 6.18. Loop linearity using valve characteristic: The most common method of flow
control involves placing a variable resistance in a closed conduit. A typical system is shown in
Figure 6.75 that includes a centrifugal pump and an automatic control valve. A proportionalintegral controller can be used for feedback control.

Range: A wide range of operation can be achieved, from zero flow to the maximum with the
valve fully opened.
Gain: If a substantial percentage of the pressure drop occurs across the valve, the gain will be
large.
Dynamics: The dynamics of flow measurement and valve actuation are fast, and the flow
process is very fast. (How long after you open a faucet does it take the flow to increase?)
Linearity: A nearly linear relationship between manipulated and controlled variable is desired.
The following analysis must be performed for each flow system, and a compensating nonlinearity introduced into the control loop where needed.

Next, we determine the relationship between the valve position and the flow rate, which can be derived
by applying the Bernoulli equation to the flow system and ignoring small friction losses.

F Fmax

Cv (v )
100

Pv

ave

with
Cv(v)
F
Fmax
Pv
v

ave

the inherent valve characteristic (v in percent open)


the volumetric flow rate
the maximum flow rate at 100% valve opening
the pressure drop across the valve
is the valve opening in %
is the average density of the fluid

Note that inherent characteristic is the relationship between the valve stem position (v), which is
adjusted by the controller, and the flow rate through the valve at constant pressure drop (Pv) across
the valve. Control valves are manufactured with various inherent characteristics, and a few common
examples are given in Figure 6.76. The manufacturers achieve various functions for inherent
characteristic by modifying the relationship between the stem position and the opening between the
valve plug and seat. Since the manufacturer cannot know whether or how much the pressure drop

6-100

Operability in process design

Chapter 6 Process Control

changes with flow, it is the task of the design engineer to select the proper characteristic for a
particular flow system.
The key pressures in the typical flow system in Figure 6.75 are shown in Figure 6.77 as a
function of the flow. The top curve is a typical head curve for a centrifugal pump; the design engineer
can acquire the curve matching the installed pump from the pump manufacturer. The bottom curve is
the system curve and can be determined by calculating the pressure drops as a function of flow using
friction factor correlations.
We note that for the example process in Figure 6.75 the pressure drop across the valve
decreases with increasing flow rate; see Figure 6.77. For good control performance, we desire a
constant process gain between the adjusted control valve stem position (v) and the controlled variable
(F). We can achieve a (approximately) constant gain by selecting a function for the valve
characteristic (Cv) that results in the same change in flow rate for each one percent change in stem
position, independent of the starting value of the stem position.

F Fmax

Cv (v) Pv
v
100 ave

with a constant

The desired function is selected from standard characteristics provided by valve manufacturers. For
this example, the proper characteristic is equal percentage, which yields a nearly linear process gain
between the valve position and the flow rate.

The characteristic for every control valve must be determined considering the
relationship between the adjusted valve stem and the measured controlled variable for
the specific process and operating conditions.

quick
opening

linear

equal
percentage

Figure 6.76 Typical (inherent) valve


characteristics

valve stem position (%)

6-101

Operability in process design

Chapter 6 Process Control

Figure 6.77 Valve pressure drop


for the process in Figure 6.75
Marlin (2000)
The relationship between stem position and controlled variable will differ from loop to
loop, so the engineer needs to evaluate each process to determine the appropriate linearizing
characteristic. Also, note that the controlled variable does not have to be flow rate; it could be
temperature, pressure or other measured process variable.
Example 6.19. Flow control. Different flow control designs are shown in Figure 6.78. The
sensor is an orifice meter; the material is a liquid at its bubble point; and the goal is to control the
fluid flow rate. Determine which, if any, of the designs is acceptable
a. This design has a serious flaw. A valve is located in the pump suction, which introduces a
significant pressure drop before the pump. This design will likely lead to cavitation in the
pump. Valves should be placed in pump outlets, not in suction lines.
b. This design is likely OK. But, it would be better to locate the orifice sensor before the valve,
where the pressure is higher and the likelihood of cavitation in the sensor is lower.
c. This design does not have pump, which is proper if the pressure in the vessel is sufficient
for the flow to the downstream unit; we will assume that this is the case. The liquid at the
bottom of the drum is at its bubble point, which could vaporize in an orifice meter, due to the
pressure decrease where the liquid velocity increases. Therefore, the orifice meter must be
located sufficiently below the vessel to provide a head that can prevent cavitation.
d. This design has a serious flaw. The orifice meter is located after the valve, which
introduces pressure drop. There is a high likelihood of flashing in the sensor.

6-102

Operability in process design

Chapter 6 Process Control

Figure 6.78 Possible flow control designs for the liquid from a flash drum.
(Orifice meters used in all designs)

Example 6.20 Two possible flow control designs are given in Figure 6.79. The fluid is far from
its bubble point. Which would you recommend?
The key question for feedback control is, Does a causal relationship exist between the
manipulated and controlled variables? For both designs, a change in valve opening will
affect the measured flow rate. Therefore, either of the designs will function. As stated in the
previous example, the flow meter is typically placed at the location of higher pressure to
prevent cavitation.
Example 6.21 Energy consumption. The generic flow loop in Figure 6.80 would involve some
pressure drop across the sensor and valve. In some processes, especially high pressure gas, the
cost of these pressure drops can be substantial. What design steps can be taken to reduce the
pressure drops and reduce the cost of energy?
Sensor: The pressure drop is affected by the choice of sensor technology. For example, the
following list ranks some common sensors based on their non-recoverable pressure drop.

Orifice
Venturi
Pitot tube

(highest pressure drop)


(lowest pressure drop)
6-103

Operability in process design

Chapter 6 Process Control

Figure 6.79 Two possible flow control designs.

Figure 6.80 Flow control loop.

Valve: The choice of valve body affects the pressure drop across the valve. For example, the following
list ranks some common valve bodies based on their non-recoverable pressure drops.

Globe
Butterfly

(higher pressure drop)

Valve sizing: The size of the valve refers to the orifice opening with the stem in the 100% open
position. Valve manufacturers document the options for each valve type, and naturally, only a discrete
number of options are available. The valve size affects the system (lower) curve in Figure 6.77, with a
larger valve size or capacity having a lower pressure drop. With a lower pressure drop, the engineer
can select a pump (upper) curve that requires less energy. However, the equipment must be able to
process the desired flow rate, which requires some minimum capacity, investment and energy
consumption. A general guideline is that the pump-valve-pipe combination should process the design
rate at ~70% valve opening. One should use such a guideline with caution it certainly would not
apply in the case of reactor cooling water where perhaps triple the normal flow rate is required when
the temperature of the reactor contents is too high and a runaway is possible. Engineers need to
understand the range of variation to be achieved by the equipment before designing the equipment!
Pump driver: The pumps speed of rotation of the centrifugal pump affects the top curve in
Figure 6.77. A variable speed pump driver can achieve the desired flow rate by changing the speed
without a control valve. The key advantage of this design is lower energy consumption. Variable speed
drivers can be steam turbines or variable speed electric motors. The choice is between (a) the simpler,
lower capital cost, and higher operating cost constant speed pump or (b) the more complex, higher
capital cost, and lower operating cost variable speed driver. Economics provides the basis for the
choice.

Example 6.22. Two valves in a pipe: The flow system is Figure 6.81a has two valves in one
pipe. The pump is a centrifugal pump with a constant speed driver. What can be controlled in
this process?
Lets consider the flow rate first; how many flow rates can be controlled independently? By material
balance, we note that the flow rates at all sensors are identical. Therefore, we conclude that no more
than one flow rate can be controlled. Since there is a causal relationship between valve opening and
the flow for each valve, either valve may be used as a manipulated variable to control flow.

6-104

Operability in process design

Chapter 6 Process Control

Figure 6.81a. Pipe with two valves.

Figure 6.81b. One solution for control using two


valves in one pipe.

Now, we will consider the pressures? Lets assume that flow is being controlled and consider three
locations, numbered 1 to 3 in Figure 6.81a.
1.
The pressure at the outlet of the centrifugal pump depends upon (i) the suction pressure and
(ii) the pump head, which in turn depends on the flow rate (as defined by the pump
performance curve). When the flow is determined, the pressure is dependent. Therefore, it
cannot be controlled when the flow is controlled.
2.
The pressure between the two valves depends upon the relative amount of pressure drop
occurring at each of the valves. Note that the total resistance to flow depends on the flow rate
but that the same flow rate can be achieved by different combinations of the two valve
openings. Therefore, the flow rate and intermediate pressure can be controlled independently
using the two control valves.
3.
The pressure downstream of the two valves depends on the downstream process and cannot be
controlled by adjusting the two valves available in this example.
One solution for this problem is shown in Figure 6.81b. This design achieves a pressure reduction or
letdown along with flow control. The pressure letdown could be desired when a high-pressure
source is used in a process with lower pressure equipment. One common example is natural gas
distribution in a city. The distribution network would be at high pressure because of the long
distances, but local gas users would not like to invest in expensive equipment required by high
pressures. Therefore, pressure letdown allows local users to have equipment appropriate for lowerpressure operation.

6.6.2 Heat exchanger control


Nearly every process plant involves heating and cooling, and many unit operations require heat
transfer, e.g., distillation reboilers and condensers. Therefore, we begin the coverage of unit
operations with some common heat exchangers. Let us begin with a counter current shell and
tube heat exchanger shown in Figure 6.82. A simplified model of the exchanger is given in the
following.

6-105

Operability in process design

Chapter 6 Process Control

Figure 6.82. Schematic of typical shell and


tube countercurrent heat exchanger.
(Padleckas, 2012)
Q FHot C p H (THout THin )
Q FCold C pC (TCout TCin )
Q UAY (T )lm
U

(6.8)

1
1 x 1 1
hi
k
hF
h0

hi i Fi

ho o Fo o

where
Q
= heat transferred
F
= flow rate
Cp
= heat capacity
U
= overall heat transfer coefficient
h
= film heat transfer coefficient
Y
= correction factor for non-exact counter current flow patterns
T
= temperature
k
= thermal conductivity
x
= wall thickness
, = empirically determined coefficients
Subscripts F, i and o represent fouling, inner and outer, respectively
Example 6.23. Basic heat exchanger control - Lets see how some designs stack up with the
desired behavior. We begin with a heat exchanger without phase change. The first design we
will consider is the most obvious, in which one stream is adjusted to control the temperature of
the other stream, as shown in Figure 6.83.

6-106

Operability in process design

Chapter 6 Process Control

Figure 6.83. Heat exchanger with one flow


rate manipulated to control the other
streams outlet temperature.

The design is evaluated in the following.

Range: The design has a wide range, as the manipulated flow can be adjusted from
zero (no heat transfer) to its maximum (maximum heat transfer)
Gain: The gain changes dramatically over the range of manipulated flow. At low
flows, the gain (T/F at steady state) is very large, but at high flows, the gain becomes
very small.
Dynamics: The dynamics are reasonably fast, but the time to steady state involves
heating or cooling all liquid and metal of the heat exchanger.
Linearity: As discussed under the gain, the process gain will be highly non-linear.
This will require some compensation in other elements to achieve a nearly linear
control loop. Options exist for non-linear compensation in the final element (control
valve characteristic) or in the controller gain (KC), which can be modified through a
more complex algorithm.
Other considerations: The fouling rate is usually higher at low flow rates leading to
high exit temperatures, especially when the adjusted stream is cooling water.
Therefore, low flows through a heat exchanger should be avoided.

Generally, the design in Figure 6.83 is used when tight control is not required and fouling is
not an issue because of the materials and operating conditions.
Example 6.24. Faster Heat exchanger control - Now, lets consider a slightly more complex
process in Figure 6.84 with an adjustable by-pass around the exchanger. The control system
adjusts the ratio of fluid through the exchanger and through the by-pass to influence the heat
transfer.
This design, which is used extensively in practice, is evaluated in the following.

Range: The design has a wide range, as the manipulated flow can be adjusted from
zero by-pass (maximum heat transfer) to all by-pass (no heat transfer)
Gain: The process is basically a mixing process. Therefore, the gain effect on the
temperature is strong.

6-107

Operability in process design

Chapter 6 Process Control

Figure 6.84. Heat exchanger bypass


manipulated to control the streams
outlet temperature.

Dynamics: The dynamics of the mixing process is very fast, much faster than the heat
exchanger itself.
Linearity: The linearity depends upon the relationship between the valve adjustment
and the flow. This will not be exactly linear, but it will be closer to linear than the
previously considered design.
Other considerations: Limitations can be imposed to prevent a low flow through the
exchanger. The by-pass adjustment can be effected by either (i) one three-way valve or
(ii) two valves with opposite failure positions. The three-way valve is less expensive
and does not ensure tight closure. The two-valve design is used for large flows (pipes)
and where (nearly) complete shut off is desired.

Example 6.25 Manipulate the heat exchange area - Now, lets consider a heat exchanger with
phase change where a liquid boils on the shell side. This example will enable us to evaluate
novel methods for influencing heat transfer. In the first design, we will consider changing the
area! Naturally, it is not possible to change the area of the tubes, but we can influence the area
available for heat exchange. The heat transfer coefficient on the shell side is very large for the
area where boiling occurs, and it is much smaller for the area where the vapor contacts the tubes.
As a first approximation, all heat transfer takes place across the area where boiling occurs.
Therefore, the heat transfer is strongly affected by the liquid level in the shell.
One control system is shown in Figure 6.85, and it is evaluated in the following.

Range: The design has a wide range, since the level can vary from zero (no boiling; no
heat transfer) to full (maximum boiling and maximum heat transfer)
Gain: Changing the effective value of UA has a strong influence on the heat transfer.
Dynamics: The dynamics are moderately fast. Increasing the level can be achieved by
increasing the liquid flow to the shell, but decreasing the level requires a longer time to
boil more liquid than is entering the shell.
Linearity: If the shell had straight sides, the area would be linear with the level, but
the shell is typically a horizontal cylinder. The relationship between level and heat
transfer is close to linear near the middle of the shell and deviates strongly at the
extreme ranges of very low and high level.
6-108

Operability in process design

Figure 6.85. Process and control design that


influences exchanger duty by influencing the
liquid level in the exchanger.

Chapter 6 Process Control

Figure 6.86. Process and control design that


influences exchanger duty by influencing the
boiling pressure.

Other considerations: Often, this design is used with refrigerants, where the vapor
returns to a compressor. In such situations, liquid carryover in the vapor could
damage the compressor.

The design in Figure 6.85 is very commonly used.


Example 6.26 Manipulate the exchange temperature difference - Lets evaluate an
alternative in which the temperature is used to affect the heat transfer! The boiling temperature
of the bubble-point liquid in the shell depends on the shell side pressure of the heat exchanger.
A design is given in Figure 6.86, and it is evaluated in the following.

Range: The design has a moderate range, because operating at high pressures requires
a more expensive shell design.
Gain: Changing the boiling temperature has a strong influence on the heat transfer.
Dynamics: The dynamics are fast, which is the major advantage for this design.
Linearity: The relationship between pressure and heat transfer are not linear, but not
strongly non-linear.
Other considerations: Often, this design is used with refrigerants where the vapor
returns to a compressor. In such situations, the extra pressure drop across the vapor
line valve increases the work by the compressor; this can be a decisive negative factor
in selection of a control system.

In general, the design in Figure 6.85 is preferred for heat exchangers with a boiling refrigerant
because of its lower work requirement; the design in Figure 6.86 is used when fast dynamics are
essential.

6-109

Operability in process design

Chapter 6 Process Control

Example 6.27 Steam condensation in an exchanger - Now, lets consider the very common
heat exchanger using steam as the heating source. Steam near saturation is provided to the
exchanger to heat the other fluid. The steam condenses at nearly constant temperature to provide
the energy being transferred.
We could use the principle discussed above by influencing the area for steam condensation, as
shown schematically in Figure 6.87a.

Range: The range is large because the liquid can cover from none (maximum
exchange) to all (no exchange) of the surface.
Gain: The gain would be large.
Dynamics: The dynamics are not symmetric. To increase the area for condensation,
the valve can be opened to drain liquid, which would be fast. However, to decrease
exchange the valve is closed and the water must accumulate from the condensing
steam, which could be slow.
Linearity: From the discussion on dynamics, it is clear that the response is non-linear.
Other considerations: A minor advantage for this design is a smaller valve in the
liquid line.

An alternative steam exchanger design shown in Figure 6.87 drains all condensate so that no
area is covered by water. The condensate in collected in a steam trap that periodically releases
the condensate to return to the boiler, without allowing steam to escape. There are many
steam traps using various principles; one float-type design is shown schematically in
Figure 6.88.

Range: The range is large because the steam valve can be adjusted from fully opened
to closed.
Gain: The gain would be large.
Dynamics: The dynamics are symmetric. All of the area is available for condensation,
and the transfer depends upon the steam available. The dynamics are moderately fast,
although the liquid and metal of the exchanger influence the dynamic response.
Linearity: The linearity of the system depends on the relationship between the valve
position and the steam flow, which can be made nearly linear by the selection of the
proper valve characteristic.
Other considerations: Minor disadvantages for this design are a larger valve in the
steam line and the maintenance required for the steam trap.

6-110

Operability in process design

Chapter 6 Process Control

steam

a)

steam

b)

LC
condensate
condensate

Figure 6.87 Control for a steam heat exchanger. (a) manipulating the condensate flow and (b)
manipulating the steam flow. The condensate collection in (b) is typically achieved using a
steam trap. Marlin (2000)

Figure 6.88 Typical float-type steam trap. (Sugartechnology, 2012)

6-111

Operability in process design

Chapter 6 Process Control

6.7. Conclusions
We will consider our progress in learning process control by considering the control hierarchy
shown in Sidebar II. This hierarchy provides an overview of plant operations decision making
with several categories based on a temporal decomposition. The fastest responses are required at
the lowest level, with each subsequent level involving lower frequency decisions. The location
of the decisions in the hierarchy is based on the importance of the decision and the process
dynamics of the part of the process being considered. Lets briefly consider each level,
discussing topics covered and topics for future study.

Level
A problem-solving approach
(all levels)

Topics covered
In Section 6.5, a systematic
control design problem definition
form is provided.

Topics for future study


Plant-wide control design
remains a fertile topic for
investigation, and likely will for
a very long time.

A flowchart for design is given.

The process

Some plant-wide issues are


presented.
The influence of the process
design on control performance is
presented in Section 6.3.

Process equipment and structure,


along with the control equipment,
Instrumentation is addressed in
strongly influences dynamic
Appendix A.
behavior
Chapter 5 on Safety addressed
Protection
the safety hierarchy, including
the basic process control system.
This level addresses personnel
safety, equipment protection and
In this chapter, further examples
appropriate processing of
of basic process control are
effluents.
covered.
Smooth Operation and
Stability
This level provides a consistent
process environment and is
critical for successful production
rate and product quality control.

Product Quality
Direct measurement and control
of product quality is essential.

Inherently safe process design

Quantitative safety analysis


Fault tree analysis
Reliability analysis
Computing structures for
highly reliable process
control systems
Multiloop control remains the
In Section 6.2, classical control
methods beyond single-loop were basis for much control at this
level.
introduced, including cascade,
feedforward, signal select, split
Further analysis of
range, and inferential,.
multiloop control
More examples of
Some examples of unit-operation
control for unit
control are given in Section 6.6.
operations
Greater depth in partial
control concepts
Multiloop control techniques
A major advance in process
introduced in Section 6.4 is often control has been achieved
required because of strong
through Model Predictive
interactions among quality
Control (MPC) that provides a
variables.
centralized or coordinated
6-112

This topic provides opportunity


for innovation, especially with
highly non-linear processes and
newly developing process
technology.

Operability in process design

Chapter 6 Process Control

method for controlling numerous


variables simultaneously.
Inferential control is introduced
in Section 6.2.5.

Profit
Plant operation has a strong
influence on profitability.

Methods for determining the best


operating conditions have not
been addressed here.
Appendix B gives a good insight
into the economic benefits for
process control and gives
methods for estimating benefits
for project development.

Monitoring and Diagnosis


People play a crucial role in plant
automation. They must monitor,
diagnosis and intervene to correct
faults.

Plant planning and scheduling


(Not shown in Sidebar)

Chapter 9 on Troubleshooting
addresses this issue directly for
short-term faults that need
correction.

This issue is not addressed in this


chapter.

Important decisions include raw


material contracts and purchases,
daily and weekly production
schedules, plant inventory
management, and product
dispatch

In some instances, quality cannot


be measured in real-time.
Inferential estimates play an
increasingly important role, and
multivariate statistical methods
have proven successful in
industrial applications.
This is a major area of innovation
in process control. Controls can
be designed to approximately
follow a high-profit strategy, see
Partial control
Self-optimizing control
Models can be used in real-time
to predict the best operating
conditions, see
Real-time optimization
Optimal blending
Every process must be monitored
for performance to measure the
following.
Performance (efficiency,
yield, energy
consumption, etc.)
Equipment status and
likelihood of future fault
Need for maintenance
(e.g., heat exchanger
cleaning)
Longer-term operations decisions
require detailed economic
calculations involving process
performance predictions and
analysis of dynamics. The
dynamics include inventories,
shipping, and responses of
multiple companies in a supply
chain.
These are challenging and
interesting problems for
engineers.

6-113

Operability in process design

Chapter 6 Process Control

Sidebar II Overview of Control Hierarchy


Variation in
product quality

MONITORING
&
DIAGNOSIS

Feed rate

Time

Operating
Profit

PROFIT
Flow

AC

fb

PRODUCT
ff
FY

FY

QUALITY

FI
LAL

LC

FC

TC

TAH

SMOOTH
OPERATION &
STABILITY
Control F, T, P,& L

PROTECTION
Safety

feed selection

Environment

maintenance

Equipment

etc.
SIS
T > MAX?

fc
PROCESS

6-114

Operability in process design

Chapter 6 Process Control

This brief summary of the control hierarchy and topics covered in operability
demonstrates that (i) you have made considerable progress in building skills and knowledge and
(ii) you have the opportunity to learn and to innovate. Process control is not a solved problem,
requiring mere application of known solutions and calculations. In fact, process control is a
toolbox of technologies and methods that can be tailored to achieve good dynamic performance
for a seemingly unlimited number of process structures.
You are now ready to apply process control to challenging new problems not discussed
here or solved before by anyone!

Additional Learning Topics and Resources


For an outstanding public-domain reference for instrumentation, see the following reference.
Kuphalt, T (2012) Lessons in Industrial Instrumentation,
http://www.openbookproject.net/books/socratic/sinst/
The basic control topics in Sections 6.1 and 6.2 are covered more thoroughly in many process
control textbooks. At the risk of appearing immodest, I would recommend the following.
Marlin, T. (2000) Process Control, Designing Processes and Control Systems for
Dynamic Performance, McGraw-Hill, New York
Multiloop control is the basic approach for regulatory process control. While the principles are
straightforward, developing industrial strength designs requires expertise. The following
references give the insight into the importance of disturbance and process direction.
Skogstad, S. and M. Morari (1987) Effect of Disturbance Directions on Closed-loop
Performance, IEC Res., 26, 20292035.
Stanley, G., M. Marino-Galarraga, and T. McAvoy (1985) Short-cut Operability
Analysis: 1. The Relative Disturbance Gain, IEC Proc. Des.Devel., 24, 1181-1188
A control valve depends on an actuator to provide the force for movement and can have the
additional component of a positioner to improve the precision of movement. The following Web
sites have additional information.

http://www.spiraxsarco.com/resources/steam-engineering-tutorials/control-hardware-elpn-actuation/control-valve-actuators-and-positioners.asp
http://iamechatronics.com/notes/general-engineering/436-final-control-elements-controlvalves?start=10

Engineers build understanding and expertise by learning best practices. There is much to learn
from reviewing designs of common unit operations, such as boilers, firer heaters, distribution
6-115

Operability in process design

Chapter 6 Process Control

networks, refrigeration cycles, compressors, turbines, pumps, and so forth.


references give an introduction to many practical control designs.

The following

Drieger provides designs for several industrial unit operationshttp://www.driedger.ca/


Liptak, Bela (1999) Instrument Engineers Handbook, 3rd Edition, Process Control, CRC
Press, Boca Raton
Luyben, William, Bojrn Tyreus, and Michael Luyben, (1999) Plantwide Process Control,
McGraw-Hill, New York.
Dukelow, S. (1986) The Control of Boilers, ISA, Research Triangle Park, N.C.

Whenever we perform an engineering task, we should ask, How can we monitor the
performance? Process control systems are functioning essentially continuously for years, and
plant personnel are overwhelmed by the data generated every 200 ms for each of the thousands
of control loops in a plant. Fortunately, technology is available to monitor and diagnose
operating control systems without interfering with their normal functioning.
Jelali, M. (2006). An overview of control performance assessment technology and
industrial applications, Journal Process Control, 14, 5, 441466

References
Agreda, V., L. Partin, and W. Heise (1990) High-Purity Methyl Acetate via Reactive Distillation,
CEP, February 1990, 40-46
Barton, G., R. Harris, P. Moxham, and M. Thatcher (1987) ICI Vinyl Chloride Plant Control
Study, in Marlin et. al. (1987)
Bristol, Edgar (1966) On a New Measurement of Interaction for Multivariable Process
Control, IEEE Trans. Auto. Control, AC-11, 133-134
Buckley, P. (1974) Material Balance Control of Recycle Systems, Instr. Tech., May 1974, 29-34
Bush, S. (1969) The Measurement and Prediction of Sustained Temperature Oscillations in a
Chemical Reactor, Proc. R. Soc. Lond. A, 309, 1-26
Choudhurya, Shoukat, N.F. Thornhill, S.L. Shah (2005) Modelling valve stiction, Control
Engineering Practice 13, 641658
Denenberg, L. (2012) http://www.larry.denenberg.com/predictions.html
Engencyclopedia (2012) http://www.enggcyclopedia.com/2012/02/typical-pid-arrangementcentrifugal-compressor-systems/
Fogler, F. S. (1986) Elements of Chemical Reaction Engineering, Prentice-Hall, Englewood
Cliffs
Ganapathy, V. (2001) Superheaters, Design and Performance, Hydrocarbon Processing
http://v_ganapathy.tripod.com/superhtr.pdf
Harismiadis, V., D. Stavrakas, A. Fantolini, L. Pedone, L. DOrazi, R. Prodan, A. Sood and G.
Bhattad (2012)Use dynamic simulation for advanced LNG plant design, Hydro Proc.,
August 2012 http://www.hydrocarbonprocessing.com/Article/3068284/Use-dynamicsimulation-for-advanced-LNG-plant-design.html
Harris, T. (1989) Assessment of Control Loop Performance, Can. J. Chemical Engineering, 67,
856-861.
6-116

Operability in process design

Chapter 6 Process Control

Hoyle, D. L. (1976) Designing for pH Control, Chem. Eng., 471-476.


ISA (2007) Fossil Fuel Power Plant -Steam Temperature Controls, ISA-77.44.01-2007ISA,
Research Triangle Park, N.C.
Jelali, M. (2006). An overview of control performance assessment technology and industrial
applications, Journal Process Control, Volume 14, Issue 5, 441466
Johnman, P., K.Hitz, E. Fyvie, H. Morris, D. Gosden, and B. Taber, Eraring Power Station
Control Study, in Marlin et. el. (1987).
Kresta, James, Thomas Marlin, and John MacGregor, (1994) Development of Inferential
Process Models by Partial Least Squares, Comp. Chem. Eng., 8, 597-612.
Liptak, Bela (1999) Instrument Engineers Handbook, 3rd Edition, Process Control, CRC Press,
Boca Raton
Liptak, Bela (2003) Instrument Engineers Handbook, 4th Edition, Process Measurement and
Analysis, Volume I, CRC Press, Boca Raton
Luyben, William, Bojrn Tyreus, and Michael Luyben, (1999) Plantwide Process Control,
McGraw-Hill, New York.
Marlin, Thomas (2000) Process Control: Designing Processes and Control Systems for Dynamic
Performance, McGraw-Hill, New York.
Marlin, T. and A. Hrymak, (1996) Real-Time Operations Optimization of Continuous Processes,
Proceed. Chemical Process Control, Proceedings of CPC-V, Lake Tahoe, Nevada
Marlin, Thomas, John Perkins, Geoff Barton, and Mike Brisk (1987) Advanced Process Control
Applications, Warren Centre Industrial Case Studies of Opportunities and Benefits, ISA,
Research Triangle Park. (Copyright transferred to editors in 1995)
Marlin, Thomas, John Perkins, Geoff Barton, and Mike Brisk (1991) Benefits from process
control: results of a joint industry-university study, J. Proc. Cont., 1, 68-83
McAvoy, Thomas (1983) Interaction Analysis, Instrument Society of America, Research
Triangle Park
Moore, R. L., (1978) Neutralization of Wastewater by pH Control, ISA, Research Triangle Park.
Newell, R., R. Hayward, D. Bogle, S. Jacobsson, and S. Frayne (1987) CALTEX Kurnell
Petroleum Refinery Control Study in Marlin et. al. (1987)
Padleckas,H. (2012) distributed under creative commons license, (Thanks!)
http://en.wikipedia.org/wiki/File:Straight-tube_heat_exchanger_1-pass.PNG
Patel, V., J. Feng, S. Dasgupta,, and J. Kramer, (2012) Use of Dynamic Simulation in the Design
Of Ethylene Plants, http://www.kemco.or.kr/up_load/blog/Dynamice%20Simualtion.pdf
Pedersen, C., D. Mudt, J.K. Bailey, and J. Ayala, (1995) Closed-loop Real-time Optimization of
a Hydrocracker, Paper CC-95-121, NRPA Computer Conference, November 6-8, 1995
Perlmutter, D. (1972) Stability of Chemical Reactors, Prentice Hall, Englewood Cliffs
Silverstein, J. and R. Shinnar (1982) Effect of Design on the Stability and Control of Fixed Bed
Catalytic Reactors with Heat Feedback. 1. Concepts, IEC Process Des. Dev., Vol. 21, No.
2, 241-256
Skogestad, Sigurd and Ian Postlethwaite (2005) Multivariable Feedback Control Analysis and
Design (2nd Edition), Wiley, Chichester
Skogestad, Sigurd (2004)``Control structure design for complete chemical plants'', Computers
and Chemical Engineering, 28 (1-2), 219-234
Skogstad, S. and M. Morari(1987) Effect of Disturbance Directions on Closed-loop
Performance, IEC Res., 26, 20292035.

6-117

Operability in process design

Chapter 6 Process Control

Stanley, G., M. Marino-Galarraga, and T. McAvoy (1985) Short-cut Operability Analysis: 1. The
Relative Disturbance Gain, IEC Proc. Des. Devel., 24, 1181-1188
Staroselsky, N. and L. Laudin Improved Surge Control for Centrifugal Compressors, CEP, May
12, 1979, 175-184
Smith, R. and P. Varbanov (2005) Whats the Price of Steam?, CEP, July 2005, 29-33.
Substech (2012) distributed under creative commons license, (Thanks!)
http://www.substech.com/dokuwiki/doku.php?id=steel_strip_processing
Sugartechnology (2012) distributed under creative commons license, (Thanks!)
http://sugartechnology.com/steamtraps/mechanical.php
Turton, R., R. Bailie, W. Whiting, J. Shaewitz, and D. Bhattacharyya (2012) Analysis, Synthesis,
and Design of Chemical Processes 4th Edition, Prentice-Hall, Upper Saddle River
Vermeer, P. C. Pedersen, W. Canney, J. Ayala (1996) Design and Integration Issues for Dynamic
Blend Optimization, Paper CC-96-130, NPRA Computer Conference, November 11-13,
1996
Wallen, Anders, (1997) Reproduced with permission from the American Automatic Control
Council from the 1997 American Control Conference paper, Wallen, A., Valve Dynamics
and Automatic Tuning, Proceedings of the American Control Conference Vol. 5,
Albuquerque, New Mexico, June 1997, Pg. 2930-2934
White, R. and R. Kurz (2005) Surge Avoidance for Compressor Systems, Proceedings of the 35th
Turbomachinery Symposium, Texas A&M University,
http://turbolab.tamu.edu/proc/turboproc/T35/16-WHITE.pdf
Witt, S. and R. Waggoner (1990) Tuning Parameters for non-PID Three Mode Controllers,
Hydro. Process, 69, 74-78
Zhang, Y. and M. Dudzic (2005) Industrial application of multivariate SPC to continuous caster
start-up operations for breakout prevention, Control Engineering Practice, 14, 1357
1375

6-118

Operability in process design

Chapter 6 Process Control

Appendix 6.A Understanding Frequency Response


An important factor in dynamic performance is the frequency at which disturbances occur
in comparison to the process dynamics. This topic is more exactly analyzed in using frequency
response, but chemical engineers have difficulty with frequency response calculations. (Setting
s=j in the transfer function and finding the amplitude and phase angle of the resulting complex
number; do you remember that?) Here, we will consider the issue qualitatively in the time
domain. We will build insights that are used in analyzing process systems and complement
detailed frequency-response calculations.
Disturbances occur because of changes to surrounding equipment and varying sources of
utilities and raw materials. These are more or less random and generally do not follow patterns;
however, they generally occur over a narrow range of frequencies. For example, the fluctuation
of feed composition to a distillation tower can be caused by imprecise control of an upstream
unit, and the fluctuation might involve oscillations with periods of 10 to 30 minutes. To simplify
the analysis here, we will consider a sine disturbance, while recognizing that it is only an
approximation for disturbances that change roughly periodically with time.
We will consider the performance of the output variable to sine inputs. To gain some
basic understanding, lets consider the behavior of the concentration at the outlet of a mixing
tank to a sine variation in the inlet flow concentration. The explanation follows from the
drawing in Figure 6.A.1, which shows two disturbances, each with a different sine frequency in
the input concentration but with the same amplitudes. The maximum effect of the input change
on the tank concentration depends on the amount of "difference in" the component A that is
introduced during a half period. Naturally, the + and - deviations from the mean cancel, so that
the net effect of the average output is zero. The size of the input integral highlighted in
crosshatched red determines magnitude of the transient deviation in the output. For the periodic
disturbance, we note that the larger the integral, the larger the output magnitude. Clearly, as
the frequency increases, the integral (crosshatched red) decreases, so that the output magnitude
decreases.
We can now consider the behavior of a typical process without and with control in response to a
sine disturbance.

Without control How would a process respond to a sign disturbance? The response of
the process without control to a sign disturbance is shown graphically in Figure 6.A.2a.
We observe that at very low frequencies (relative to the time constant, so that << 1)
the output magnitude is equal to the product of the system gain and input magnitude, i.e.,
the dynamics have little effect on the disturbance. In addition, we observe that at very
high frequencies (relative to the time constant, so that >> 1) the output magnitude
becomes very small with respect to the product of the system gain and input magnitude,
i.e., the process dynamics attenuate the disturbance without control! The results are
displayed in a Bode plot in Figure 6.A.2b.

6-119

Operability in process design

Chapter 6 Process Control

Figure 6.A.1. Effect of input frequency on the tank concentration


With control - How would a process under feedback control respond to a sign
disturbance? The response of the process with feedback control to a sign disturbance is shown
graphically in Figure 6.A.3a. At very high frequencies, we know that the process itself, without
feedback, decreases the effects of disturbance. What happens at low frequencies? We observe
that at low frequencies (relative to the time constant, so that << 1) the output magnitude is
very low. This behavior results for the effectiveness of feedback, which has lots of time to
respond to the slowly changing disturbance. We also observe that neither the process itself nor
the feedback controller are effective in reducing the effect of the disturbance near the frequency
where c 1, i.e., the resonant frequency. The results are displayed in a Bode plot in Figure
6.A.3b.
The observant reader might be wondering, Since perfect sine waves dont occur in
chemical plants, is any of this useful? The answer is, Yes! Even though perfect sine waves
do not occur, many disturbances can be characterized by a range of frequencies. Combined with
knowledge of the feedback dynamics, an understanding of frequency response can be applied to
predict the location of the dominant disturbance frequency in the major regions of system
behavior shown in Figure 6.A.3b.

6-120

Operability in process design

Chapter 6 Process Control

Figure 6.A.2a The effect of a sine disturbance on the process with control at three frequencies.

Figure 6.A.2b Bode plot of the amplitude ratio for a disturbance response with control.
(CV = controlled variables, D = disturbance)

6-121

Operability in process design

Chapter 6 Process Control

Figure 6.A.3a. The effect of a sine disturbance on the process with control at three frequencies.

Figure 6.A.3b. Bode plot of the amplitude ratio for a disturbance response with control.
(CV = controlled variables, D = disturbance)

6-122

Operability in process design

Chapter 6 Process Control

Lets consider the example in Figure 6.A.4, which shows a process, gives its open-loop
dynamics and shows some characteristic data for a variable without control. The major design
questions are, Can feedback control reduce the variability of the output variables, and if so, by
how much? We note that the data has a very long period (very low frequency) for the dominant
variability compared with the feedback dynamics. In addition, we note that the data contains
some high frequency components that are of lower amplitude. We would predict the following.

For the lower frequency variation, a large fraction of the uncontrolled amplitude can be
reduced by feedback control
For the higher frequency variation, very little (if any) of the uncontrolled amplitude can
be reduced by feedback control

This prediction is confirmed by the data in Figure 6.A.5, which shows the performance with
feedback control.
The preceding analysis was qualitative. However, a more reliable quantitative method
for predicting future control system behavior is available. For example, see Harris (1987) and
Jelali (2006). These quantitative methods are much preferred for engineering practice, and
software products are available to perform the calculations. In addition, they have the advantage
of being able to use original data with feedback control to predict possible performance
improvements with modified controls.

Figure 6.A.4. Typical process operating data without control.

6-123

Operability in process design

Chapter 6 Process Control

Figure 6.A.5. The reduction in variability achieved through feedback control. The process and
disturbances are the same as shown without control in Figure 6.A.4. Marlin (2000)

6-124

Operability in process design

Chapter 6 Process Control

Appendix 6.B Benefits from Process Control


6.B.1 Introduction
Process control is implemented to achieve many goals. The primary goal is safety. The design
of the safety hierarchy is addressed in Chapter 6. Other control strategies are required by various
regulations, e.g., by equipment and process technology vendors, or to satisfy government
regulations, such as effluents to the environment. However, many control strategies are not
essential and therefore can be considered optional methods for achieving product quality,
production rate, and equipment protection. They must be justified based on economic criteria.
The economic analysis of a control system is addressed in this appendix.
Process control achieves benefits by reducing variability. Perhaps, it is better to state that
in most cases, the variability is transferred from very important variables to lesser important
variables. For example, the original data without product quality control in Figure 6.B.1a shows
high variability in the analyzer measurement and no variability in the coolant flow. When
feedback control is implemented as shown in Figure 6.B.1b, there is much less variability in the
product quality but much more variability in the coolant flow. This is why we say that the
variability has been transferred. Naturally, we design feedback control to transfer the
variability from variables where it is costly to variables where the cost is extremely low.
One example of where variability is removed without transfer is the reduction of
variability due to time constants in a process. For example, variability in a liquid feed property
(composition or temperature) can be reduced significantly through mixing in a vessel. In
addition, the flow rate variability can be reduced by averaging level controller tuning. (For
averaging level control, see Marlin, 2000).

Figure 6.B.1. Variability and feedback control.


(a) without feedback control, (b) with feedback control

6-125

Operability in process design

Chapter 6 Process Control

6.B.2 Benefits Calculation


Now, lets introduce the basic equation for estimating control benefits.

100

(6.B.1)

where
B
D0
D1
V(*,*)

=
=
=
=

SF

benefits for the change in dynamic performance ($/y)


original (base case) distribution in controlled variable (histogram)
distribution of the controlled variable under new control strategy (histogram)
Improved economic process performance at the base case operating conditions
($/h)
correction factor for other operating conditions, e.g., production rate,
(dimensionless)
service factor, the fraction of time that the control strategy is improving the
process performance (dimensionless)
Time when the control strategy should be in service (h/y)

A few important issues are worth noting. First, the calculation does not use average values of
variables; it uses the distributions of controlled variable values. Since process control effects an
improvement by reducing variability, this use of distributions before and after control seems
reasonable. Unfortunately, this correct policy is not always followed in practice, which can lead
to erroneous benefit predictions. Second, the benefit can be influenced by other variables in the
process. For example, many benefits are proportional to production rate; an example would be
energy saved in a distillation tower. Third and finally, the benefits depend on the amount of time
in the year when the control strategy is functioning.
The distribution of the key controlled variable is used in the benefit calculation. The
distribution is easily determined from plant data, as shown in Figure 6.B.1. With the data
represented by the distribution, we can readily evaluate the average process performance using
the following equation, as shown schematically in Figure 6.B.2.

with
PRave = average process performance
PRj = process performance in interval j
(evaluated at mid-point of interval)
Fj = fraction of data in interval j
m = number of intervals

6-126

(6.B.2)

Operability in process design

Chapter 6 Process Control

Figure 6.B.2 Evaluating variable distribution from trend data. Marlin (2000)
Note that the process performance is typically expressed as a physical variable, such as
energy/throughput, percent yield, and so forth. Then, a second relationship is employed to
translate the change in process performance to a change in economic performance, i.e., the
variable V in equation (6.B.1). This is shown in the following equation.

V = PR* Iv = [(PRave)base case (PRave)improved ] Iv

(6.B.3)

with
(PRave)base case
(PRave)improved
Iv
PR

=
=
=
=

the valuable of process performance with base case operation


the valuable of process performance with improved process control
the incremental value of the change in average process performance
process performance

Note that this expression includes a constant incremental value, independent of the magnitude of
the change in process operation. In some situations, the incremental value depends on the
magnitude of change, and equation (6.B.3) can be easily modified to reflect the variable
incremental value when calculating the change in economic performance.
Much more detail is important in this calculation procedure, but before addressing further
detail, two examples are presented.

6-127

Operability in process design

Chapter 6 Process Control

steam

Figure 6.B.3 Ethane pyrolysis fired heater/reactor.


Example 6.B.1 Improved temperature control of a pyrolysis fired heater A pyrolysis
reactor cracks hydrocarbon feed to lighter products, with the primary goal of producing olefin
products for subsequent use in polymerization reactors. Here, we will consider a single heater
that processes ethane feed shown in Figure 6.B.3.
To determine the economic performance, the engineer requires the variable
distribution and the relationship between the variable and the process performance. Data with
the initial temperature control is shown in Figure 6.B.4a. (This data was previous considered
briefly in Figure 6.44.) The process performance correlation is also plotted in Figure 6.B.4.
Applying equation (6.B.2) to the data in Figure 6.b.3a, the average ethane conversion can be
evaluated to be 54.6%.
Several methods are available to predict the performance with improved control, and
these will be discussed subsequently in this appendix. For the purposes of demonstrating the
calculation, we will assume that the prediction in Figure 6.B.3b can be achieved. Applying
equation (6.B.2) to the data in Figure (6.B.3b), the average ethane conversion can be
evaluated to be 59%. We conclude that the change in process performance is the following.

P = (Pave)base case (Pave)improved = 59-54.6 = 4.4%

6-128

Operability in process design

Chapter 6 Process Control

(a)

(b)

Figure 6.B.4 Distribution data and process performance correlation for Example 6.B.1.
a. Base case distribution, b. Distribution with improved control Marlin (2000)

To determine the change in economic performance, the incremental value of the


change in feed conversion is needed. This value requires a simulation of the entire plant,
because the unreacted ethane flows through the entire plant, requiring considerable energy,
and is recovered and recycled to the pyrolysis reactor, where it can react again. We will take
the commodity prices of ethane and ethylene in $/kg to be 0.286 and 0.925, respectively. If we
assume that the benefit is 30% of the difference in feed and product prices (based on yields
including recycle), the value of the change in conversion is evaluated in the following.
Value of increased conversion/kg = (PR/100)*Iv = ( 0.044)(0.639)(.30) = $0.00843 $/kg

PR
Iv

=
=

change in yield due to improved control


incremental value

6-129

Operability in process design

Chapter 6 Process Control

The total value depends on the production rate. A typical industrial pyrolysis heater
can produce about 548 m-Ton/day of ethylene, requiring about 913,000 kg/day of ethane.
Therefore, the economic improvement can be calculated to be the following.

V = M (kg/h) * PR (% feed)/100 * Iv ($/kg) = (0.00843 $/kg)* (0.913x106 kg/day)


V = 7.7 x103 $/day
Normally, we evaluate process economics using annual values. This reactor will be
operating continuously when the plant is in operation. However, it requires periodic
maintenance to remove by-product coke, and during this maintenance, the productive
reactions do not occur. The maintenance will require one day, and eight of these maintenance
decokes will occur every year. In addition, the plant will shut down once per year for five days
for maintenance. Summing these periods when the control system cannot contribute
improvements, one obtains the following.
SF/100*T = 365 -8 5 = 352 days/y
Note that this analysis assumes no unplanned shutdowns. The plant history could be reviewed
to determine whether unplanned shutdowns contribute a significant a significant loss in
operating time.
All of the terms can be combined to determine the profit realized through improved
temperature control of a pyrolysis fired heater. The result is given in the following.
B = V * SF/100 * T = (7.7 x103 $/day) * (352day)
B = 2.7 x 106 $/y
This is an enormous economic benefit! It would certainly justify considerable engineering
effort to achieve the reduced variability in temperature control. Note that the base case is
poorer than would likely be observed in practice today. But, we can conclude that the benefit
for operating closer to the high temperature limit is greater than four hundred thousand
dollars per year per degree Celsius. Naturally, these results depend strongly on the costs of
materials and the market conditions, i.e., whether it is possible to sell the extra production.

We note that the reduction in variability in Example 6.B.1 was accompanied by a change
the controller set point that increased the average temperature. In this example, a simple
reduction in variability without the change in set point would not have improved performance.
In general, the engineer must decide whether a set point adjustment is appropriate. Lets
consider an example where the set point is not adjusted.
Example 6.B.2 Flue gas excess oxygen Naturally, air is mixed with fuel in the burner of a
combustion process. Sufficient air for complete combustion is essential so that no combustible
material exists in the flue gas, because the flue gas with combustibles could mix with leaking air
and explode. However, any air not required for complete combustion leads to inefficiency; it is
heated and exhausted to the environment via a smoke stack. As a result, excess air leads to a
waste of fuel. Therefore, the goal is to ensure a slight excess of air at all times but to keep the
6-130

Operability in process design

Chapter 6 Process Control

excess small. This situation is shown in Figure 6.B.5, which depicts the efficiency curve and the
constraint of minimum excess oxygen in the flue gas.
The distributions are given in Figure 6.B.6 for the excess oxygen data for the base
case, with higher variability, and the improved case, with lower variability. We can apply
equation (6.B.2) to determine the efficiency for each distribution, and then, we can determine
the increase in efficiency as given in the following.

PR = (PRave)improved - (PRave)base case = 87.7-86.8 = 0.90%


To determine the change in economic performance, the incremental value of the
change in efficiency is needed. The incremental value depends on the fuel cost and the base
case consumption of fuel. We will use a 5 $/GJ cost of fuel and a base case fuel to the boiler
of 100 GJ/h. Therefore, the economic improvement can be calculated to be the following.

V * M= PR (% efficiency)/100 * M (GJ/h) * Iv ($/GJ) =


(0.9/100)*(5)*100 = 4.5 $/h
Normally, we evaluate process economics using annual values. We will use 340 days
operation to allow time for maintenance shutdowns. Therefore, the annual benefit would be
the following.
B = ( 4.5 $/h) * (24 h/day)*(340 day/y) = 37 k$/y
This is a substantial savings by simply improving and existing control system.

Figure 6.B.5 Boiler and efficiency curve versus excess oxygen in the flue gas Marlin (2000)

6-131

Operability in process design

Chapter 6 Process Control

(a)

(b)

Figure 6.B.6 Distribution data and process performance correlation for Example 6.B.2.
a. Base case distribution, b. Distribution with improved control. Marlin (2000)
Note that the base case and improved distributions have essentially the same set points.
In this case study, the entire benefit was realized by reducing the distribution around the same
average value. If the engineer incorrectly evaluated the base case data using the average value of
excess oxygen, the incorrect conclusion would have been that no improvement in efficiency was
possible. This example highlights the importance of using the distribution of dynamic plant data
when evaluating process performance.
The method just demonstrated evaluates the performance and profitability of a process
based on the distribution of values for a key process variable. Like all engineering calculations,
it has specific advantages and is based on assumptions that must be recognized by the engineer.
Therefore, some key aspects of the method will be presented in the following.

6-132

Operability in process design

Chapter 6 Process Control

6.B.3 Issues in the Benefits Calculation


Many potential pitfalls exist when applying the benefits calculation approach. Eight of the more
common pitfalls are discussed here to help the reader avoid committing these errors. The most
challenging issue, predicting the future performance, is addressed in the next section.
6.B.3a. Base case data The calculations for the base case use historical data from the process.
The historical data must be representative. Representative data should characterize the
behavior of the process in the future if no changes were made to the control system. Also, the
variability in the data should be caused by factors that can be influenced by the feedback control
being proposed.
Lets consider the sample data in Figure 6.B.7. The data includes typical variability and
two major deviations from normal operation, one due to a process shutdown and one due to an
equipment limitation. The equipment limitation could have been maintenance on a pump, heat
exchanger, etc. The variability over the period is strongly affected by these major deviations.
But, will the proposed control system prevent the shutdown or the maintenance event? If not, the
base case data must exclude these events.
Obtaining data is very easy with digital control systems that store historical data for all
variables. However, the example just discussed points out the dangers of extracting data from
history without carefully reviewing it. If all of the data were used to characterize the base case,
the improvement possible with improved control could have been grossly overestimated.
6.B.3b. Variable distribution The distribution is not assumed to follow any typical
distribution, normal, etc. The distribution does not have to be symmetric or continuous. It must
only be representative. Representative data should characterize the behavior of the process
prior to any proposed changes by the project.

Figure 6.B.7. Historical data with events not affected by control. Marlin et. al. (1987)

6-133

Operability in process design

Chapter 6 Process Control

Non-symmetric frequency distributions are common when a high penalty is paid for
violations of a constraint. When operators are responsible for adjustments, they will make very
strong corrections when nearing a constraint. When a control system is making adjustments, the
design engineer can introduce a non-linear effect in the controller to take strong corrective
actions when the variable approaches the constraint. The method described here can properly
represent process performance with non-symmetric distributions.
6.B.3c. Process performance curve The process performance curve does not have to conform
to any specific criteria. It need not be linear, convex, or continuous.
Although we do not want process variables to violate constraints, they will occasionally
exceed their limits due to large disturbances. The process performance correlation can represent
the penalty, as shown in Figure 6.B.8. As expected, the performance becomes worse during
violations, and the performance can suffer a step decrease immediately upon the violation.
Naturally, performance curves with other shapes are possible.
6.B.3d. Frequency of variability The condensation of trend data into a histogram results in
the loss of frequency information. Because frequency data is not included in the histogram, the
same histogram could summarize a process with either high frequency or low frequency
variation. While all data are inherently dynamic, the process performance correlation nearly
always applies to steady state. Therefore, the typical assumption made is that the process data
variability is of low frequency and the process is operating at quasi-steady state, so that the
process correlation can be applied.

Figure 6.B.8. Typical process performance curve with discontinuity at constraint violation.

6-134

Operability in process design

Chapter 6 Process Control

6.B.3e. Economic value of change The engineer must determine the economic value for the
change in process operations. This is often not as straightforward as one might initially think,
because the economic change must represent all changes to the operation of the process resulting
from the control system. Lets consider some issues.

- The net change in profit includes positive and negative effects. If improving
product quality requires additional energy, the net improvement is the sum of
the positive quality value and the negative value of additional fuel.
- It is not correct to use average values for energy, feeds, products, utilities and so
forth. Incremental or marginal values must be used, since they represent the
change in profit from a base case, and it is this change that the control system
will effect. These incremental values should not include the fixed costs for
items not influenced by a small change in the control system. These fixed costs
include major capital equipment recovery (equipment not changed by the
control design), personnel, laboratories, offices, waste treatment, etc.
- The process change must be related to costs, i.e., to purchases and/or sales. This
principle is demonstrated in the distillation example in Example 6.B.3.
- The benefit for control improvement is an element in a project profitability
analysis using the standard time-value of money approach, such as net present
value or discounted cash flow. Costs include additional installed equipment
(sensors, valves, computing equipment, etc.), engineering time for design and
technician time for programming.
- The incremental value for a change in operation is not necessary constant. As
the change increases in magnitude (or changes sign), the process responses can
change. The changes can be dramatic, as when constraints are encountered; for
example, Smith and Varbanov (2005) show substantial changes in the cost for
steam as letdowns (direct connections between headers) are opened.
Example 6.B.3 Reduced steam for distillation reboiling A control system has been proposed
to reduce the steam use in the reboiler of a distillation tower in Figure 6.B.9. How is the value of
the steam reduction determined?
The steam is generated in the plant, so that the effect of a reduction in steam must be traced
through the boiler and steam system to determine the reduction in fuel achieved by the
reduction of steam in the distillation tower. (If the steam were being purchased at a constant
value in $/ton, determining the value would be easy.) As shown in the figure, a reduction of
reboiler steam (i) reduces one outflow from the medium pressure steam header, (ii) reduces
the steam through the extraction valve from the turbine, (ii) since the turbine is required to
provide a fixed amount of power, the steam flow through the condensation path must increase
(less that the extraction decreases), (iv) less flow leaves the high pressure header, (v) to
balance the flows in the high pressure header, the pressure controller reduces the fuel flow.

6-135

Operability in process design

Chapter 6 Process Control

Figure 6.B.9. Tracking the steam change to fuel purchase.


Why not use the cost of high-pressure steam, which is easily determined from the cost
of fuel, heat of vaporization and the boiler efficiency? For example, a typical cost of highpressure steam is 5.32 $/ton, while the cost of medium-pressure steam is 2.61 $/ton. (These
costs are for a typical steam network and fuel cost of 4.22 $/GJ or 0.0152 $/kWh (Smith and
Varbanov, 2005).) Using the simplified high-pressure value would have yielded a benefit
estimate about double the correct value!
6.B.3f. Operating conditions on benefits Benefits for a base case operation can be influenced
by changes to the operating conditions. Perhaps the strongest effect is the production rate
through the unit. Lets consider two examples.
Example 6.B.4 Distillation energy savings With poor control, distillation towers are overrefluxed to ensure achieving maximum impurity concentrations. When the distillate product
quality is controlled in real time, the reflux and energy consumption are reduced. How does the
tower feed flow rate affect the energy benefits for control?
With constant tray efficiency, the optimal energy would be proportional to the feed rate.
However, this relationship is limited to the region in which the energy input can maintained
proportional to feed rate. One limitation results from the minimum reflux ratio that is
imposed to maintain the liquid and vapor rates in the region that provides adequate liquidvapor contact without weeping. When the feed flow rate is very low, the compositions cannot
be controlled to set points; the tower must be over-refluxed, which results in products purer
than necessary, but within specifications. This situation occurs when the feed is substantially
below the design value, which can occur due to market fluctuations.
6-136

Operability in process design

Chapter 6 Process Control

Figure 6.B.10. Distillation energy reduction occurs in a limited range of feed flow rate.
When evaluating the benefits for control, the feed rate has to be predicted over the
project length (many years), and the benefits adjusted to account for any time when the feed
rate is expected to be in the no savings region. Any reduction would appear in the T
factor in equation (6.B.1).
Example 6.B.5 Compressor energy savings Rotatory compressors require a minimum flow
rate to prevent damage, which is required to prevent surge that can reverse flow direction in
the high-speed machine. All compressors have at least basic anti-surge control that provides
recycle to provide sufficient flow through the compressor when the feed flow is less than
required. A basic control is shown in Figure 6.B.11. How do benefits for anti-surge control
depend on throughput flow?
The minimum surge flow depends upon operating conditions such as compressor speed and
gas molecular weight. Thus, the simple control design in Figure 6.B.11 would have to a very
conservative, i.e., high, minimum flow because the recycle flow controller does not account for
the other operating conditions. More advanced control designs use additional real-time
measurements to provide updated estimates for the minimum flow required (Staroselsky and
Laudin, 1979; Smith and Kurz, 2005, Engencyclopedia, 2012); lower surge flow limits result
in smaller recycle and less energy consumption by the variable-speed machine powering the
compressor. The comparison of base case and advanced anti-surge control is shown in Figure
6.B.11. It is apparent that benefits occur at low feed flow rates and that no benefits are
realized at high flow rates. Again, the engineer must estimate the distribution of flow rates
that will occur in the future and adjust the energy savings accordingly.

6-137

Operability in process design

Chapter 6 Process Control

Figure 6.B.11. Compressor anti-surge energy reduction occurs in a limited range of feed flow
rate.
6.B.3g. Mixing effects on process performance Often, mixing can be represented as a linear
process. However, complex reactions and other chemical interactions can lead to non-linearities
that strongly affect the behavior of material properties that have been stored in inventory. Often,
engineers place inventories for streams with key composition specifications. The purpose is to
provide mixing so that upstream quality control does not have to be excellent, because the tank
outlet composition is easily maintained at the specification. The following example
demonstrates a potential negative impact of inventories.
Example 6.B.6 Effects of Non-linear mixing A crude distillation unit separates crude oil into
many streams, each of which is processed in subsequent processes. Often, storage tanks are
located between crude distillation and subsequent units so that the production rates in various
units can be determined independently, at least over short durations. The crude distillation unit
and diesel storage tank are shown in Figure 6.B.12. The engineers operating the unit suggested
that good quality control of the stream leaving the distillation tower was not important because
mixing in the storage tank attenuated variability. Is this a good strategy?
The contention that the storage tank attenuated composition variability is certainly correct.
But, at what cost is the variability reduced? Many petroleum product qualities do not blend
linearly. For example, the diesel cloud point specification is non-linear, so that a small
amount of heavier material in the blend requires a larger amount of lighter material;
requiring the lighter materials substantially reduces the yield of valuable product. The
benefits for improved control of streams into storage have been studied by Marlin et. al.
(1987), and typical results are shown in Figure 6.B.12. Improved distillation control could
yield an increase in diesel yield of 1.5% of feed (with a concomitant reduction in heavy gas oil)
that represents an enormous economic benefit.
6-138

Operability in process design

Chapter 6 Process Control

Figure 6.B.12. Crude distillation showing importance of stream composition control before
mixing subject to non-linear blending relationship. Newell et. al. (1987)
6.B.3h. Incidence reduction Reducing the variability of key process variables can prevent
infrequent, large deviations that can damage equipment, activate safety instrumented system
(SIS) shutdowns, and if not properly moderated, lead to accidents. Typical process examples
include compressor anti-surge control, chemical reactor temperature control preventing runaway
temperature excursions, and positive displacement pump recycle to prevent excessive pressure.
This topic is addressed in Chapter 4 on Reliability. Naturally, the reduction in such incidents can
lead to considerable savings and should be included in the economic analysis of process control
designs.
In this section, the clear message was delivered, Understand your data, process
principles, equipment performance, operating window limitations, product quality, production
rate goals, and economics. With mastery of the situation, the engineer can adapt the benefits
calculation to address special features of the problem. However, a major issue remains, i.e., the
prediction of future control strategy performance, which is addressed in the next section.

6-139

Operability in process design

Chapter 6 Process Control

6.B.4 Predicting Future Performance


A pithy quotation is appropriate at this point.
It is difficult to make predictions, especially about the future!

Controversy swirls around the originator of this quote, with attributions to Confucius, Niels Bohr
(Danish Nobel Laureate in Physics), Yogi Berra (American baseball player) and many more
(Denenberg, 2012). Regardless of who said it first, the quotation is valid and serves as a warning
about being overly confident in our predictions. However, we must prepare for the future using
predictions, and in engineering, these predictions can be based on solid principles and practices
that while not guarantying perfection, often yield results adequate for technical and business
decisions.
Here, we are dealing with the challenge of predicting the ability of a control system to
reduce the future variability in an operating process. We will begin with some approaches that
should not be used. Then, we will consider some approaches that can be used.
Do not use these approaches

Do not ignore variability - Consider a situation in which operating closer to a constraint


yields an economic benefit. Sometimes, a person not well versed in plant operation will
suggest simply changing the set point without first reducing the process variability with
better control. An example of this poor approach is shown for temperature control of
steam to a power-generating turbine. High steam temperature improves efficiency of the
turbine; however, higher temperatures lead to much lower turbine blade life. High
temperature alarms are avoided because they indicate operation that will cause shorter
turbine blade life. As demonstrated clearly by data in Figure 6.B.13, raising the
temperature controller set point resulted in excessive alarm activation and if continued,
would have damaged the turbine (Johnman et. al., 1987). This plant experiment was
(appropriately) terminated quickly.

Do not assume the answer - Never simply assume that the base case variability can be
reduced by a fixed percentage without analysis of process dynamics. Often, this
percentage is 50% of the variance. In reality, the improvement can be much smaller or
greater than this arbitrary choice. Assuming the solution is not engineering!

Do not base the prediction on un-reviewed historical data - Do not assume that the
best performance in the historical database can be achieved. The best performance can
be caused by a combination of conditions that might not occur again. For example, the
disturbances might have been temporarily small, the equipment could have been recently
refurbished (heat exchanger cleaned, burner adjusted, catalyst regenerated, etc.), and/or
the feed material might have the most favorable composition. However, we will
reconsider this recommendation with an approach that can be used.

6-140

Operability in process design

Chapter 6 Process Control

Figure 6.B.13. Changing the set point toward the constraint without improved control leads to
excessive constraint violations. Johnman et. al. (1987)

Do not assume set points are correct - When improving the process performance, do
not assume that the original (current) controller current set point value is optimal. For
example, when investigating the best manner for operating a vinyl chloride process in
Figure 6.B.14, engineers recognized that the EDC cracking heater was not being operated
at the best temperature (Barton et. al, 1987). The project modified its direction from
process control to process optimization, with a real-time optimizer using a process model
to predict the best temperature target as operating conditions changed. (For an
introduction to real-time process operation, see Marlin and Hrymak, 1986.)

Figure 6.B.14 Greatest benefits realized by optimizing the reactor conversion.


Barton et. al. (1987)

6-141

Operability in process design

Chapter 6 Process Control

Do not base business decisions on a literature review - Finally, do not assume that
benefits reported in the literature for similar processes can be achieved in your plant.
Control performance depends on factors that could be very different in your plant, such
as disturbances (magnitudes and frequencies), equipment capacities, flexibility, energy
costs, and product markets. Published reports can be useful in helping an engineer select
likely candidates for improvement, but reports on other plants do not replace thorough
analysis.

Use these approaches

Do use zero variance for a quick, limiting estimate - A simple approach can be used
for a quick estimate of whether a significant improvement is possible. The engineer can
assume that the future variance of the key variable is zero around its optimal value. We
know that this cannot be achieved, but we can estimate the economic benefit with this
assumption. If the project is not financially attractive with zero variance, we are sure that
it will not be financially attractive with the actual variance achieved by improved control,
and the project can be dropped from further consideration. If the project might be
attractive, one of the other recommended methods discussed in the following items can
be applied to obtain an improved estimate of potential improvement.

Figure 6.B.17. Example of quick improvement estimate using (a) historical data for the base
case and (b) zero variance for the future performance.
6-142

Operability in process design

Chapter 6 Process Control

Figure 6.B.15 Historical base case data from a reactor where maximizing temperature had a
high economic benefit.

Do use best performance after thorough review of scenario - Lets revisit the
approach using past operating performance. The good aspect of this approach is that the
data represent actual disturbances, equipment capacity and flexibility. Difficulties with
this approach have already been discussed. We can overcome these difficulties by
monitoring the process when the data is collected or reviewing detailed operations
logbooks, if they exist. The data in Figure 6.B.15 shows base case operation of a reactor
where incentive exists to maximize the temperature without exceeding the maximum
constraint. The data shows that the temperature was maintained close to the maximum
for about seven days. Does this indicate that this good performance can be repeated by
improved control for the entire period? The engineer must review the operation to
determine if the process was experiencing typical disturbances such as feed tank changes,
production rate changes and others that will be typical in the future; if the result is yes,
it can be concluded that this period of good operation can be repeated. Second, there are
periods with poor temperature performance. Could control have improved the
performance? The engineer must evaluate the equipment performance, capacity and
flexibility. If adjustments were possible but not made, improvements could be made. If
both answers are yes, it is reasonable to predict that the entire thirty-day period could
have been controlled at close approach to the maximum temperature.

Do use plant tests - We could perform plant tests that emulate the proposed control
strategy. This approach requires that sensors and final elements exist in the plant and that
the process dynamics are slow enough for the control calculation to be performed and
implemented periodically by a person. An example is given in Figure 6.B.16, in which
crude oil entering the desalter is preheated by vacuum distillation (VDU) pumparound
(mid-tower condenser). The goal is to produce a sidestream product in the vacuum
tower. In the base case, the heat duty of in the exchanger is too large, resulting in too
much condensation below the VDU sidestream, so that no product can be withdrawn in
6-143

Operability in process design

Chapter 6 Process Control

the VDU. Why is the duty high? The operators must ensure that the temperature of the
crude oil entering the desalter is above a minimum limit; to be on the safe side, the heat
exchanger duty is maintained very high. Plant tests demonstrated that the exchanger duty
could be reduced, the side stream product could be withdrawn, and the desalter
temperature could be maintained in its desired range. Improvement was possible.
Why was the base case operation poor? The desalter temperature was locally displayed,
and the exchanger by-pass valve was manually adjusted. Therefore, adjustments were
very time-consuming, so that poor operation was not due to lack of effort by the plant
operating personnel. It was a poorly designed plant system; it could not be operated at
peak profitability.

Do use simulation combined with historical disturbance data - The previously


discussed methods might not be possible, because there is no relevant best performance
data and emulating the control system is not possible (a sensor or final element is not
available). However, if the plant is in operation, data is available for the key controlled
variable. Lets look at the situation where (i) trend data is available for the variable
without control and (ii) the engineer can obtain an estimate of the feedback dynamics.
The dynamics can be estimated using standard experimental methods used in controller
design and tuning, like the graphical process reaction curve or statistical methods. The
plant data represents the effects of disturbances on the key variable. This plant data can
be combined with a linear dynamic simulation of the process and controller to predict the
closed-loop behavior of the process. This approach is depicted in Figure 6.B.17;
naturally, the process and controller models can be modified to suit the specific system
under study.

Figure 6.B.16. Process system where crude entering desalter is preheated by vacuum distillation
(VDU) pumparound (mid-tower condenser). The goal is to produce a
sidestream product in the vacuum tower.

6-144

Operability in process design

Chapter 6 Process Control

Figure 6.B.17. Schematic of a simulation approach for estimating the control performance
under feedback control. Historical data must be without control.
The programming and calculations are simple to prepare and perform. The important
limiting assumptions are that (i) the plant data includes no control (including no operator
actions for control) and (ii) the process dynamics can be approximated by a linear model
over the range of variation experienced in the data and (iii) the historical data are
representative of the variability in the plant.

Do use advanced statistical method extending the simulation method above - The
requirement that the historical data be free of influence by control is a significant
limitation because important variables are typically controlled, even if only periodically
by manual adjustments. Fortunately, an advanced approach overcomes this limitation. It
performs statistical analysis of the data and predicts the best achievable feedback control
performance. The approach can be extended to cascade, feedforward and multivariable
control. The topic is more involved than can be presented here; the interested reader can
refer to Harris (1987) and Jelali (2006) for further details, and Jelali provides references
to software products to perform the calculations and provide helpful visual displays.

Do use fundamental models when deviations are large - In a few cases, the process
and control system should be simulated using fundamental, non-linear models. The
engineering effort is justified when developing new processes and when analyzing
process control for large disturbances and complex physiochemical systems where
linearized models do not provide adequate accuracy. An example of dynamic simulation
supporting new process development is the famous Tennessee Eastman reactive
distillation process (Agreda et. al., 1990). An example of dynamic simulation to evaluate
6-145

Operability in process design

Chapter 6 Process Control

common unit operations connected in a complex system is reported by Harismiadis


(2012) for a liquefied natural gas plant design
Developing fundamental dynamic models and simulations could be very time-consuming.
Fortunately, software packages are available to assist the engineer; for example, most
flowsheeting packages have dynamic modeling capability. These packages provide
model libraries, graphical model building, physical property databases and advanced
numerical methods.

Do consider infrequent, large and costly incidents - All of the previous methods
provide estimates for variance reduction. Process control can also provide substantial
benefits by responding properly to very infrequent, large magnitude disturbances that can
cause plant shutdowns if not compensated quickly. The report by Zhang and Dudzic
(2006) describes a good example of an inferential variable monitoring system that can
identify incipient constraint violation and alarm the plant personnel. This application
realized good economic benefits and improved the safety of the process by preventing a
release of molten steel in areas where people worked. Fundamental simulation studies
can also be used to determine the benefits for process equipment and control
modifications. Patel et. al. (2012) give a summary of many applications of simulation to
predict the dynamic behavior of a complex chemical plant.

6.B.5 Conclusion
Estimating the benefits for process control is an essential skill for engineers. Not only are good
estimates developed thorough dynamic analysis, but also information is developed to ensure that
the process equipment (capacity and flexibility), instrumentation (sensors and final elements) and
the control strategy (feedback, cascade, feedforward, loop pairing, etc.) is designed and
implemented to achieve the predictions. In addition, the proposed design must conform to the
safety, reliability, and economic goals of the plant.
The topic might more properly be termed analysis for benefits estimation and
preliminary control design.

The benefits calculation expression is shown schematically in Figure 6.B.18. The reader
must keep in mind that the figure is not rigorous, since it implies linear relationships, while the
benefits calculation in equation (6.B.1) allows non-linear relationships. However, it serves as a
memory aid for the terms in the calculation.
The approaches described in this appendix have been used by practitioners for decades
and are well accepted in industry. However, they do not provide a cookbook. They provide a
suite of concepts, approaches, and calculations that can be tailored by the engineer to solve many
problems in applied process control.

6-146

Operability in process design

Figure 6.B.18.

Chapter 6 Process Control

Schematic of the benefits calculation. (Caution that items might not be


separable as shown because of non-linearities.)

The reader should feel comfortable in creatively applying these approaches and where
necessary developing innovative new approaches for problems not addressed here.

6.B.6 Additional Learning Resources


This appendix has addressed many important issues in control benefits estimation, and it has
provided some practical approaches for engineering practice. Perhaps surprisingly, this topic is
not widely addressed in the open literature, although it is of crucial importance to every
investment in process control. The references in Table 6.B.1 provide additional details on
methods, case studies, and citations for other publications worth reading.

6-147

Operability in process design

Chapter 6 Process Control

Table 6.B.1 Resources for Further Learning in Process Control Benefits


Citation
Marlin, Thomas, John Perkins, Geoff Barton,
and Mike Brisk (1987) Advanced Process
Control Applications, ISA, Research
Triangle Park. (Book out of print)

Comment on contents
This book describes a set of seven industrial
studies applying the approach presented in this
appendix.

Marlin, Thomas, John Perkins, Geoff Barton,


and Mike Brisk (1991) , Benefits from
process control: results of a joint industryuniversity study, J. Proc. Cont., 1, 68-83

This journal article summarizes the results of


the above studies.

Bauer, M. and I. Craig (2008) Economic


assessment of advanced process control A
survey and framework, Journal of Process
Control, 18, 218
White, D. (2004) Determining the true
economic value of improved plant
information, Hydro. Process., December
2004, pg. 5358
Wagner, S., A. Al-Ghazzawi, R. McLeod
(2004) Using Process Models to Justify realtime Optimization, Aspen World, Orlando,

This article provides a survey of industrial


attitudes and practices regarding process
control benefits, and it contains a thorough
literature review.
This article discusses uncertainty in benefits
predictions.
This presentation addresses benefits for RealTime Optimization.

6-148

You might also like