Profit Controller Concepts Reference Guide

Honeywell Process Solutions
Advanced Process Control

Profit Controller
Concepts Reference Guide
RM09-400
R310
12/07
Release 310
Notices and Trademarks
Copyright 2007 by Honeywell International Inc.
Release 310 December, 2007
While this information is presented in good faith and believed to be accurate, Honeywell disclaims
the implied warranties of merchantability and fitness for a particular purpose and makes no
express warranties except as may be stated in its written agreement with and for its customers.
In no event is Honeywell liable to anyone for any indirect, special or consequential damages. The
information and specifications in this document are subject to change without notice.
Honeywell, PlantScape, Experion PKS, and TotalPlant are registered trademarks of Honeywell
International Inc.
Other brand or product names are trademarks of their respective owners.
Honeywell International
Process Solutions
2500 West Union Hills
Phoenix, AZ 85027
1-800 343-0228
R310 Advanced Process Control Profit Controller Concepts Reference Guide ii

12/07
About This Publication
Statement of Work
The following table describes the audience, purpose, and scope of this document:
Purpose This book provides a technical though broadly penned

orientation to Profit Controller (RMPCT) concepts and
functionality.
Audience Process and control engineers
Field technicians
For Product Release All Profit Controller (RMPCT) releases
LCN releases 500 and 600 series.
AxM X-side releases 200 and higher
Release Information
This is document version 1.11 for all releases of Profit® Controller (RMPCT) software.
Who Should Use This Book

Engineers This book is intended for process and control engineers responsible for
installing and tuning Profit Controller. The education and practical experience
appropriate for these tasks is assumed.
This book provides a detailed description of RMPCT functionality without going into a
mathematical description of the algorithms. See RMPCT Course Information for
information about the courses Honeywell offers that explain the mathematical
underpinnings of RMPCT.
Operators Operators can probably find useful information here and there, but this book
is not composed with Operator duties in mind.
R310 Advanced Process Control Profit Controller Concepts Reference Guide iii
12/07
What This Book Tells you about RMPCT
What This Book Tells you about RMPCT

Honeywell’s Robust Multivariable Predictive Control Technology (RMPCT) is a multi-
input multi-output control application that controls and optimizes highly interactive
industrial processes.
This RMPCT Concepts Reference examines the design concepts and mathematical
principles behind the application, and gives advice about configuration and tuning
appropriate to any implementation.
RMPCT Course Information

Honeywell offers several classes that explain the math behind RMPCT and how to
implement an RMPCT application.
Engineers who would like a more technical introduction to RMPCT should contact
Honeywell Automation College
2500 W. Union Hills Drive
Phoenix, AZ 85027
How This Book Is Organized

The following table summarizes what each section in this book tells you about this
publication and about Profit Controller (RMPCT):
In This Section You Can Find This Information
About This Publication How to make the best use of this book,
(You are here) and how the information is ordered.
What information you can find in the

different sections.
What writing conventions have been used

throughout this and other publications in
the Profit library.
iv Advanced Process Control Profit Controller Concepts Reference Guide R310

12/07
How This Book Is Organized
In This Section You Can Find This Information
Section 1, “Quick Tour” The different RMPC and Profit Controller

(RMPCT) versions that are supported.
What Profit Controller (RMPCT) utilities

are available to help you collect data,
identify a process, and build and install a
controller.
How to implement a controller on an

automated control system.
Section 2, “Profit Controller (RMPCT) How Profit Controller (RMPCT) uses

Variables” controlled variables (CVs), manipulated
variables (MVs), and disturbance variables
(DVs) to control a process.
Section 3, “Control Interval” How settling time, time constants, CV

constraints, independent variables, and
blocking help you choose the best control
interval.
Section 4, “Economic Optimization” How Profit Controller (RMPCT) uses an

objective function for optimizing control,
and how the optimization horizon and the
CV and MV soft limits influence
optimization calculations.
Section 5, “Robust Control” How Profit Controller (RMPCT) control

solutions get their robustness through
range control, limit funnels, and singular-
value thresholding.
Section 6, “Quick Reference to the Profit How to quickly locate on the Profit
Controller (RMPCT) Displays” Controller (RMPCT) displays the
configuration, tuning, and optimization
parameters described in this book.
R310 Advanced Process Control Profit Controller Concepts Reference Guide v

12/07
Writing Conventions Used in This Book
Writing Conventions Used in This Book

The following writing conventions have been used throughout this book and other books
in the Profit Suite library.
• Words in double quotation marks " " name sections or subsections in this
publication.
• Words in italics name book titles, add grammatical emphasis, introduce words that
are being referenced or defined, or represent mathematical variables. The context
makes the meaning and use clear.
• Words in bold type indicate paragraph topics or bring important phrases to your
attention.
• Shading brings paragraphs and table entries to your attention.
• Windows pull down menus and their options are separated by an angle bracket >.
For example, Under Settings> Communications, set the baud rate.
• Messages and information that you type appear in Courier font.
• Acronyms, Scan parameters, point names, file names, and paths appear in
UPPERCASE. The context makes the meaning and use clear.
• Command keys appear in UPPERCASE within angle brackets. For example, press
<ENTER>.
• TPS user station touch-screen targets appear in rounded boxes. For example, touch
MODIFY NODE
.
• Graphic buttons appear in UPPERCASE within brackets [ ]. For example, touch

[TAG].
• Point-dot-parameter means a point name and one of its parameters. For example,
point-dot-SP means the SP parameter for the point.
• Zero as a value and when there is a chance for confusion with the letter O is given as
Ø. In all other cases, zero as a numerical place holder is given as 0. For example, 1.0,
10, 101, CVØ1, parameter PØ.
• The terms screen and display are used inter changeably in discussing the graphical
interfaces. The verbs display a screen and call a screen are also used inter
changeably.
vi Advanced Process Control Profit Controller Concepts Reference Guide R310
12/07
Selected Bibliography
Selected Bibliography
The following technical publications can be helpful. More recent editions might be
available since compiling this list. Contact your company or university librarian, or local
bookstore for complete bibliographic and ordering information for these and other books
about model identification and multivariable control.
Author Title City/Publisher/Year
Åström and Computer-Controlled Systems Englewood Cliffs: Prentice, 1987

Wittenmark
Choi ARMA Model Identification NY: Springer-Verlog, 1992
Ljung System Identification Theory for Englewood Cliffs: Prentice, 1987

the User
Golub and Loan Matrix Computations Baltimore: John Hopkins UP, 1989
Morari and Zafiriou Robust Process Control Englewood Cliffs: Prentice, 1989
Pankratz Forecasting with Dynamic NY: Wiley, 1991

Regression Models
Seborg, et al Process Dynamics and Control NY: Wiley, 1989
Stephanopoulos Chemical Process Control Englewood Cliffs: Prentice, 1990
Zhu and Backx Identification of Multivariable London: Springer-Verlog, 1993

Industrial Processes for
Simulation, Diagnosis, and
Control
R310 Advanced Process Control Profit Controller Concepts Reference Guide vii
12/07
This page is left intentionally blank
R310 Advanced Process Control Profit Controller Concepts Reference Guide viii
12/07
Contents
1. QUICK TOUR ...............................................................................13

1.1 Overview ........................................................................................................ 13
In This Section .....................................................................................................................13
1.2 Profit Controller (RMPCT) Applications...................................................... 14
Typical Applications .............................................................................................................14
SISO vs Multivariable Control ..............................................................................................14
1.3 Implementation.............................................................................................. 15
The Variables .......................................................................................................................15
The Controller Model............................................................................................................15
Predict-Back and Estimated Disturbance Models ................................................................16
Implementing a Controller ....................................................................................................16
2. PROFIT CONTROLLER (RMPCT) VARIABLES .........................19

2.1 Overview ........................................................................................................ 19
In This Section .....................................................................................................................19
CVs, MVs, and DVs Defined ................................................................................................19
2.2 Controlled Variables ..................................................................................... 20
Characteristics .....................................................................................................................20
Feedback Performance Ratio...............................................................................................20
Correction Horizon ...............................................................................................................20
Tuning for Response............................................................................................................21
Speed vs Accuracy ⎯ The Tradeoffs...................................................................................22
Finding the Best Performance Ratio ....................................................................................22
Degrees of Freedom ............................................................................................................22
Minimizing the Error .............................................................................................................23
Setting the Engineering Unit (EU) give-up factors................................................................23
EU give-ups vs Controller Speed .........................................................................................24
CV Tracking .........................................................................................................................24
Setpoint vs Range................................................................................................................24
Limit Ramping ......................................................................................................................25
Periodic Sampling ................................................................................................................25
Bad Value Treatment ⎯ Critical CVs ...................................................................................26
Bad Value Treatment ⎯ Non Critical CVs............................................................................26
Predicted Values ..................................................................................................................26
State Estimation ...................................................................................................................27
State Estimation for Integrating Processes ..........................................................................27
Ramp Correction Settings ....................................................................................................27
R310 Advanced Process Control Profit Controller Concepts Reference Guide ix

12/07
Contents
A Caution about Ramp Correction....................................................................................... 28

Ramp Correction Trade-Offs ............................................................................................... 28
State Estimation Settings Summary .................................................................................... 29
2.3 Manipulated Variables...................................................................................30
Characteristics..................................................................................................................... 30
Rate-of-Change (Max Move) Limits..................................................................................... 30
Limit Ramping ..................................................................................................................... 30
Movement Weights.............................................................................................................. 31
Degrees of Freedom ........................................................................................................... 31
Setting MV Priorities............................................................................................................ 32
Use Lower Numbers............................................................................................................ 32
MV Tracking ........................................................................................................................ 32
Bad Value Treatment .......................................................................................................... 33
Anti-Windup......................................................................................................................... 33
Predict-Back and Estimated Disturbance Compensation .................................................... 33
Predict-Back ⎯ Implementation Suggestions ..................................................................... 34
Predict-Back ⎯ An Illustration............................................................................................. 34
Predict-Back Illustration ⎯ The Disturbances ..................................................................... 35
Predict-Back Illustration ⎯ The Problem............................................................................. 36
Predict-Back Illustration ⎯ The Solution ............................................................................. 36
Predict-Back Illustration ⎯ The Benefits ............................................................................. 37
On/Off Control ⎯ Configuration .......................................................................................... 37
On/Off Control ⎯ An Example ............................................................................................ 38
Automatic Mode Switching .................................................................................................. 38
Configuring Automatic Mode Switching............................................................................... 39
Taking MVs Off Profit Controller (RMPCT) Control ............................................................. 39
MV Move Accumulation....................................................................................................... 39
2.4 Disturbance Variables ...................................................................................42
Characteristics..................................................................................................................... 42
DV Influence on CV Tuning ................................................................................................. 42
Feedforward Performance Ratio ......................................................................................... 42
3. CONTROL INTERVAL................................................................. 45
3.1 Overview .........................................................................................................45
Read This ............................................................................................................................ 45
In This Section..................................................................................................................... 45
3.2 Control Interval and Settling Time ...............................................................46
The Ideal Interval-to-Settling Time ...................................................................................... 46
Shorter Intervals .................................................................................................................. 46
Longer Intervals................................................................................................................... 46
When Time Constants Vary ................................................................................................ 46
Recommended Intervals ..................................................................................................... 46
Intervals > Settling Time...................................................................................................... 46
x Advanced Process Control Profit Controller Concepts Reference Guide R310

12/07
Contents
3.3 Blocking ......................................................................................................... 48

Defined.................................................................................................................................48
Control Quality .....................................................................................................................48
CV Constraints .....................................................................................................................48
MV Moves ............................................................................................................................48
Independent Variables .........................................................................................................48
Reducing the Calculation Time ............................................................................................49
4. ECONOMIC OPTIMIZATION .......................................................51

4.1 Overview ........................................................................................................ 51
In This Section .....................................................................................................................51
Optimization Concepts .........................................................................................................51
4.2 Objective Function........................................................................................ 52
Degrees of Freedom ............................................................................................................52
General Form .......................................................................................................................52
Weighting Coefficients .........................................................................................................53
Maximizing Profit vs Minimizing Cost ...................................................................................53
Setting Linear Objective Coefficients ...................................................................................53
When Limits Do Better Than SetPoints ................................................................................54
Range Limits and Product Quality........................................................................................54
When Setpoints Do Better Than Limits ................................................................................55
4.3 Optimization Horizon and Optimization Speed Factor.............................. 56
Optimization Horizon............................................................................................................56
Optimization Speed Factor...................................................................................................56
4.4 Soft Limits...................................................................................................... 57
How Soft Limits Work ...........................................................................................................57
CV Soft Limits ......................................................................................................................57
MV Soft Limits ......................................................................................................................58
5. ROBUST CONTROL ....................................................................59

5.1 Overview ........................................................................................................ 59
In This Section .....................................................................................................................59
Revisiting the Process Model ...............................................................................................59
Model Error is Unavoidable ..................................................................................................59
Coping with Model Error.......................................................................................................59
5.2 Range Control................................................................................................ 61
Defined.................................................................................................................................61
Setpoint Control ⎯ An Example...........................................................................................61
The Problem.........................................................................................................................61
Range Control ⎯ An Example .............................................................................................62
The Control Matrix................................................................................................................62
Ranges and Optimization .....................................................................................................62
R310 Advanced Process Control Profit Controller Concepts Reference Guide xi

12/07
Contents
5.3 Limit Funnels .................................................................................................63

Defined................................................................................................................................ 63
Three Examples .................................................................................................................. 63
How the Funnels Work ........................................................................................................ 65
Funnel Type and Decouple Ratio ........................................................................................ 65
When to Use Funnel Type 1 and 2...................................................................................... 66
5.4 Singular-Value Thresholding........................................................................68
Defined................................................................................................................................ 68
Matrix Condition .................................................................................................................. 68
Determining a Threshold ..................................................................................................... 68
Maximizing the Matrix Condition.......................................................................................... 68
5.5 Min-Max Design .............................................................................................70
Your Input to the Offline Design .......................................................................................... 70
Overview of the Design Procedure...................................................................................... 70
Benefits ............................................................................................................................... 71
5.6 Optimal Scaling..............................................................................................72
Purpose of Optimal Scaling ................................................................................................. 72
How Values Display on Profit Controller (RMPCT) Screens ............................................... 72
6. QUICK REFERENCE TO THE PROFIT CONTROLLER DISPLAYS

73
6.1 Overview .........................................................................................................73
In This Section..................................................................................................................... 73
Target Access ..................................................................................................................... 73
6.2 Finding Targets on the Profit Controller (RMPCT) Displays .....................74
Tuning and Configuration Parameters................................................................................. 74
Optimization Parameters ..................................................................................................... 75
Control (Operating) Parameters .......................................................................................... 76
xii Advanced Process Control Profit Controller Concepts Reference Guide R310
12/07
1. Quick Tour
1.1 Overview
In This Section
This section introduces you to Profit Controller (RMPCT). Read this section to learn
about:
• What you can expect from an Profit Controller (RMPCT) controller, and
• What it takes to implement a controller on an automated control system.
R310 Advanced Process Control Profit Controller Concepts Reference Guide 13

12/07
1. Quick Tour
1.2. Profit Controller (RMPCT) Applications
1.2 Profit Controller (RMPCT) Applications

Typical Applications
Here are typical applications for Profit Controller (RMPCT).
• An entire process unit, such as a distilling unit or a catalytic cracking unit.
• A unit operation, such as a reactor and associated equipment.
• Complex equipment, such as a paper machine.
• Any system that encompasses variables that are related or interact with each other.
SISO vs Multivariable Control

The variables in the process that must be maintained at some value or within some range
are referred to as controlled variables (the CVs). In order to keep the CV values where
they should be, the controller adjusts the values of the manipulated variables (the MVs).
In a single-input single-output controller, there is one CV (the controller input or process
value) and there is one MV (the controller output). With RMPCT, there are multiple CVs
and multiple MVs. The controller views all the variables taken together as one system
and simultaneously considers the effects of all MVs on the CVs. Quite often the number
of MVs and CVs are not equal; there can be more or fewer MVs than CVs.
14 Advanced Process Control Profit Controller Concepts Reference Guide R310

12/07
1. Quick Tour
1.3. Implementation
1.3 Implementation
The Variables
There are three types of process variables that RMPCT uses as control input and output:
• Controlled Variables (CVs) variables that the controller attempts to keep at
setpoint or within a range that the Operator specifies. The first priority of the
controller is to keep the CVs within their constraints.
• Manipulated Variables (MVs) variables that the controller adjusts in order to keep
the CVs within constraints and to optimize the process, while not moving any of the
MVs outside of their constraints.
• Disturbance Variables (DVs) measured variables not under control of the controller
(they may come from an upstream process, for example) but which affect the values
of the CVs. By predicting the future effects of the DVs on the CVs, the controller
can take action to prevent CV excursions outside constraints before they develop.
DVs provide feedforward information to the controller.
The Controller Model

Sub-Process Models RMPCT uses a model to predict process behavior. The overall
process model is composed of a matrix of dynamic sub-process models, each of which
describes the effect of one of the independent variables (MVs and DVs) on one of the
CVs. A sub-process model describes how the effect of an independent variable on a CV
evolves over time.
Sub-process models are null when a particular independent variable has no effect on a
particular CV.
Dynamic Response of Sub-Processes RMPCT uses a generic form of sub-process
model that provides a reasonably good description of the dynamic behavior of the vast
majority of processes that are encountered in the processing industries. This generic
model contains a number of coefficients whose values determine the dynamic response
of a sub-process.
Identifying the Model To make the generic models into specific models, you have to
determine coefficient values where the predicted process responses agree with the actual
process responses. This procedure is known as identifying the model, or fitting the model
to the process. This model identification is typically done once when the controller is
installed.

12/07
1. Quick Tour
1.3. Implementation
To identify the model you obtain data from the process while any existing control loops
between the MVs and CVs are open. During this open-loop testing, the MVs are moved
independently. The MV input values and the CV response values are recorded.
Model Coefficients This test data is used to determine the model coefficients. With the
appropriate coefficients, the model can predict CV responses quite accurately.
Predict-Back and Estimated Disturbance Models

With its predict-back and estimated disturbance models, RMPCT can predict the effect
unmeasured disturbances have on CVs:
• Predict-back models back-out the known effects of MVs and DVs from a signal that
contains disturbance information, but that is also correlated with one or more of the
MVs or DVs. This result is an estimate of the disturbance.
• Estimated disturbance models describe the effect of the estimated disturbance on the
CVs.
RMPCT incorporates predict-back and estimated disturbance models beginning with
software release 130.0.
Implementing a Controller
Here is a summary of how you implement a controller:
1. Decide what the CVs, MVs, and DVs are.
2. Open existing loops between CVs and MVs. Apply test input signals to the MVs
(the simplest test signal is a series of steps made by the Operator). Apply test signals
to the DVs, if possible; otherwise, a period of time must be found when the DV
value is undergoing significant change.
3. Record the CV, MV, and DV signals during the test. The variables’ values are
sampled at the interval (or a sub interval) at which the controller is expected to
execute. The sampled values are collected into one or more files.
4. Identify the process model, using the collected data from the open-loop testing.
5. Build the controller from the model. Using the Honeywell Controller Builder, the
result is two files that define this particular controller application. These are read by
the controller to define its operation when it is first activated.
6. Run the controller in simulation to verify that the controller works as expected.
7. Install the controller. Use WARM mode to test control before turning control ON.

12/07
1. Quick Tour
1.3. Implementation
The remaining sections in this book provide more detailed information about RMPCT
variables and about the control functions that RMPCT performs.

12/07

12/07
2. Profit Controller (RMPCT) Variables
2.1 Overview
In This Section
This section explains how Profit Controller (RMPCT) uses controlled variables (CVs),
manipulated variables (MVs), and disturbance variables (DVs) to control a process.
CVs, MVs, and DVs Defined

This is what these variables mean to the Profit Controller (RMPCT):
• CVs Controlled variables are the process conditions to be controlled (that is why
they are also called PVs, process variables). The temperature of an outflow stream
can be a CV.
• MVs Manipulated variables are the control handles on the process. These are the
variables whose conditions are manipulated (changed) to control the CVs. The gas
flow to a furnace can be an MV.
• DVs Disturbance variables are measured disturbances (changes) in the process that
influence the CVs, but that are not under RMPCT control. The temperature of an
inflow stream can be a DV.

12/07
2.2. Controlled Variables
2.2 Controlled Variables

Characteristics
In a typical Profit Controller process, some or all of the CVs interact with each other.
This means that any action taken to change the value of one CV (to bring it back within
limits, for example) can change the value of other CVs, perhaps in unwanted directions.
The controller must coordinate changes to a number of MVs to move a particular CV a
desired amount in a desired direction without causing undesired changes to other CVs.
This is why a collection of single-loop controllers fares badly in attempting to control
interacting variables—none of the controllers know what the other controllers are doing.
A CV can have either a setpoint that defines the desired value for the CV, or a high and
low limit that define a range of allowable values. The Operator can switch between
controlling by setpoint or by range at any time.
A setpoint is treated as a range with the high and low limits set to the same value.
Feedback Performance Ratio

Defined The feedback performance ratio is the ratio of the closed-loop to open-loop
settling times for a CV.
The nominal open-loop settling time for a CV is the gain-weighted average of the settling
times for the sub-process models for the CV.
The nominal dead time for a CV is the gain-weighted average of the dead times for the
CV.
Controller Response Each CV has a performance ratio. The performance ratio specifies
how fast the controller returns the CV to setpoint or within limits when a control change
or a disturbance causes the CV to stray.
The performance ratios are the only knobs that you need to tune the controller response.
You can change the performance ratios while on control.
Correction Horizon
The controller uses the performance ratio to determine the correction horizon. The
correction horizon is the time within which the controller must bring the CV to zero
error.
To determine the correction horizon, the controller multiplies the performance ratio times
the nominal open-loop settling time of the CV, then adds the nominal dead time.

12/07
So performance ratio of:

• 1.0 means the controller returns the CV to zero error within the nominal open-loop
settling time.
• 0.5 means the controller returns the CV to zero error in half the nominal open-loop
settling time.
• 2.0 means the controller returns the CV to zero error in twice the nominal open-loop
settling time (with the nominal dead time added to all these).
Tuning for Response

The following figure illustrates how controller aggressiveness changes when the
feedback performance ratio is set to less than and greater than 1.0.
Figure 2-1 Tuning for Response - Speed vs Accuracy
The performance ratio determines the trade-offs that inherently exist between speed of
response, model accuracy, and MV movement.
• A smaller performance ratio gives faster response, results in larger MV movement,
and requires a more accurate model for stable control.
• A larger performance ratio gives slower response, results in smaller MV movement,
and works well with a less accurate model.

12/07
Speed vs Accuracy ⎯ The Tradeoffs

Here are the important trade-offs that you tend to encounter when increasing and
decreasing the feedback performance ratio:
Figure 2-2 Speed vs Accuracy - The Tradeoffs
Finding the Best Performance Ratio

A 1.0 ratio (the default) works well for most processes. If you find that the CV response
is too slow, decrease its performance ratio. In testing ratios with the controller on-line, do
not stray too far in a single change, perhaps .1 at a time.
If control performance becomes oscillatory or unstable, the remedy generally is a larger
performance ratio. If this longer response time is not acceptable, the alternative is a more
accurate model.
Degrees of Freedom
Defined The controller keeps all CVs at setpoint or within range if there are sufficient
degrees of freedom to do so.
Basically, the number of degrees of freedom is the number of MVs not at a limit minus
the number of CVs that either have setpoints or are at or outside a limit. The controller
chooses MV values so as to minimize the number of CVs that are away from setpoint or
outside limits.
As long as the degrees of freedom are zero or positive, all CV constraints can be
satisfied. If the degrees of freedom become negative, it is physically impossible to keep
all CVs at setpoint or within range.
Example Take, for example, a fractionation column with two CVs (top composition and
bottom composition), and two MVs (reflux rate and reboil rate). Both CVs have
setpoints. As long as the reflux and reboil flow control valves are not fully open, the
degrees of freedom are zero and both CVs can be kept at their setpoints.

12/07
However, if feed to the column continues to increase (and nothing bad happens, such as
flooding), there can come a point when the reflux valve, for example, becomes wide
open. Now the degrees of freedom are negative, and it is impossible to maintain both
CVs at setpoint. This limitation is imposed by the physical process, not by any limitation
in the controller.
Minimizing the Error

When there are negative degrees of freedom, the controller maintains the best possible
compromise. The best compromise is defined as minimizing the weighted sum of the
squared CV errors, where a CV error is the amount that the CV is away from its setpoint
or outside a limit:
where i is the CV index.

In this formula, the error is the scaled CV error. An advantage in using the scaled error is
that the scaling automatically results in equal increments of different CVs having equal
importance on the process. You can influence error trade-off by specifying engineering
unit give-ups for each of the CVs. Weights are inversely related to scaling factors and EU
give-ups by:
For more information, see “Optimal Scaling” on page 72.
Setting the Engineering Unit (EU) give-up factors.

You set the value of a CV’s EU give-up based on the importance of keeping the CV
within constraints. The smaller the EU give-up, the more the controller attempts to
minimize the error for that CV. EU give-ups are relative to each other. The controller
uses an average value of the low EU give-up and high EU give-up.
All CVs are assigned an initial EU give-up based on CV scaling factors that give an error
weight of 1.0, which is neutral weighting.
Continuing the fractionation column example from the previous page, the errors in CV1
and CV2 can be traded off equally. If you set the EU give-up of CV1 to 3.0 and the EU
give-up of CV2 to 1.0, CV1 will approximately have 3 units of error for every 1 unit of
error in CV2 (when there are not sufficient degrees of freedom to keep errors on both
CVs zero). When the optimization problem is infeasible the controller calculates a

12/07
minimum effort solution using error weights determined by EU give-ups. CV error offset
ratios are not guaranteed to be equal to those specified by EU give-up factors when
infeasibility occurs. EU give-up factors only specify a relative preference for CV errors.
EU give-ups are analogous to concern factors used in some other controllers.
EU give-ups vs Controller Speed

It is important to know that EU give-ups have no effect when there are sufficient degrees
of freedom to bring all CV errors to zero, which usually is true most of the time. The
only purpose of EU give-ups is to influence steady-state error trade-offs when it is
physically impossible to bring all errors to zero.
ATTENTION
CV EU give-ups do not affect the speed at which the controller corrects CV
errors, and should not be used in an attempt to tune the controller response.
Use performance ratios for this.
EU give-ups can be changed while on control. You do not have to take RMPCT off-
process.
CV Tracking
Configuration You can configure a CV to tracks its limits or its setpoint. Tracking
means that the controller adjusts the Operator-set CV limit or setpoint so that there is no
CV error on initialization.
Objective of CV Tracking CV tracking and limit ramping have the same objective: to
prevent an initial jolt to the process that can occur if CVs are far outside limits or are
away from setpoint when control is initiated. The difference is this:
• CV tracking moves both the external (Operator-set) and internal (controller-honored)
violated limit or setpoint to the current CV value. It is up to the Operator, then, to
return the limit or setpoint gradually to the desired or appropriate value.
• Limit ramping moves only the internal, violated limit to the current CV value. The
controller then automatically returns the internal limit gradually to the external limit
or setpoint.
Setpoint vs Range
The limits that are adjusted depend on whether the CV has a setpoint (high limit equal to
low limit) or a range:

12/07
• Setpoint: Both high and low limits are set equal to the CV value (which therefore
becomes the setpoint value) when the CV is placed on control.
• Range: If the CV is within limits when the CV is placed on control there is no
adjustment. If the CV is outside of a limit, the value of the violated limit is set equal
to the CV value when the CV is initialized, and the other limit is not modified.
A CV is placed on control when the Operator turns RMPCT control ON, or when the
Operator changes the state of an individual CV from dropped to ON.
Limit Ramping
The controller can make excessively large MV movements to bring a CV back within
limits or to setpoint within the error correction horizon when:
• The Operator makes a large change to a CV limit or setpoint, or
• A CV is far outside a limit or away from setpoint when control is turned on.
Limit ramping minimizes the disruption, establishing the rate at which the old limit
ramps to the new limit.
You configure limit ramping by specifying the amount that the controller must move the
old limit towards the new limit at each control interval. Different ramp settings can be
entered for the high and low limits.
• Limit ramping applies (1) When a CV is placed on control, and (2) Anytime an
Operator change violates an internal (active) CV limit or setpoint.
• CV tracking applies only when a CV is placed on control.
Periodic Sampling
Configuration You can configure periodic (asynchronous) sampling for any CV to
indicate that its source value is updated at irregular intervals or at an interval longer than
the control interval.
Example What if you have a control interval of one minute, but one of the CVs receives
its input from an analyzer that has a sampling cycle of 10 minutes? In this case the value
read by the controller changes only every 10 minutes and is constant in between.
If the controller uses the constant value between updates, there is the potential for a
noticeable jolt every 10 minutes if the value changes in one direction at a fast rate. The
jolt happens because the controller sees a large change every 10 minutes instead of a one-
tenth change every one minute. This problem is avoided by configuring the CV for
periodic sampling.

12/07
Bad Value Treatment ⎯ Critical CVs

By configuring a CV critical or non critical, you can specify the action the controller
takes when the source value for a CV goes bad.
For critical CVs, the controller sheds all control if the CV goes bad and remains bad for
longer than the number of controller executions that you specify.
While the CV is bad but before control is shed, the controller uses the predicted value of
the CV. You can set the number of bad reads to zero to make the controller shed control
immediately when a critical CV goes bad.
Bad Value Treatment ⎯ Non Critical CVs

If the CV is non critical, then overall control continues when the CV source value goes
bad.
An additional configuration option lets you set whether or not a non critical CV with a
bad source value continues to be controlled using its predicted value with no feedback, or
whether it is dropped from the controller.
A non critical CV is dropped if its value remains bad longer than the number of bad reads
allowed. You can keep the controller from dropping a non critical CV with persistent bad
values by setting the number of bad reads allowed to -1.
Predicted Values
A non critical CV continues to be controlled if its value goes bad, using its predicted
value rather than measured feedback.
The predicted value of a non critical CV that remains on control will begin to deviate
from the true (but unknown, because it is bad) process value. The prediction generally
continues to worsen as time passes.
ATTENTION
The Operator screens indicate when a CV is being controlled without

feedback and when a CV has been dropped from the controller. It is
important that Operators understand the consequences of this control.
In such cases, Operators need to find other measurements to ensure that

operating problems do not develop.

12/07
State Estimation
State Estimation – A Quick Definition State estimation in RMPCT is offset and ramp
rate correction.
See Table 2-1 for a discussion of state estimation settings and their usage.
Prediction – A Quick Review As explained when we talked about Bad Value
Treatment, RMPCT adjusts its predicted values into the future based on a comparison of
the predicted and actual values at the present and past intervals. This constitutes the
controller's feedback information.
Controller Output The controller can correct either (1) The offset (the prediction error
only), or (2) Both the offset and the rate-of-change of the error (the ramp).
Default Settings RMPCT’s default state estimation settings are determined by the
presence or absence of integrators across a CV (in model terms these are the CV-MV [i j]
pairs, the sub-processes):
When there are no integrating processes across a CV, ramp correction is OFF.
When at least one sub-process across a CV is an integrating process, ramp correction is
ON. Integrators require both offset correction and ramp correction to avoid sustained
offsets.
State Estimation for Integrating Processes

For a CV that has one or more integrating sub-processes, the controller corrects for both
offset and rate-of-change (this is the default setting).
Correcting for both offset and rate-of-change eliminates long-term CV error in the
presence of model errors and unmeasured disturbances. If you turn rate-of-change
correction off for a CV that has integrators, the CV develops an error offset that the
controller cannot account for.
Ramp Correction Settings

Correcting for Offset (Ramp Correction OFF) For a CV that has sub-process models
that are all stable, the default is to correct for the offset only. The assumption is that the
unmeasured disturbance is a step change that occurred at the current interval.
Correcting for Offset and Rate of Change (Ramp Correction ON) If the disturbance
is, in fact, a long-duration ramp, then the controller lags in its correction because it does
not know that the error caused by the disturbance will steadily increase.
In this case, you might improve performance by having the controller correct for both the
offset and its rate-of-change.

12/07
A Caution about Ramp Correction

Long-duration cyclic disturbances can be approximated by a series of ramps (a saw-tooth
signal).
If a CV experiences this type of disturbance, the rate-of-change correction might provide
significant improvement during the times when the ramp rate is approximately constant.
This correction, then, can more than compensate for the poorer performance when the
ramp rate is changing.
STATE EST = ON means both offset correction and ramp correction are ON:
• Always use STATE EST = ON for CVs with integrators.
• Use with caution STATE EST = ON for CVs with sub-process models that are all
stable. Use state estimation to better reject the slow drift disturbance. Do not use
state estimation when fast disturbance is more significant than the slow drift
disturbance.
Ramp Correction Trade-Offs

For a stable process, here is what happens to prediction on the transition slope of a long
duration ramp. See how the prediction is good, until ramping stops or changes direction?
Figure 2-3 Ramp Correction

12/07
However, when the ramp stops or changes sign, which it eventually must, prediction is
worse during transition when both offset and ramp correction are used than prediction
otherwise would be using offset correction only.
State Estimation Settings Summary

The following table summarizes what state estimation settings work best for integrating
and non integrating processes:
Table 2-1 State Estimation Settings and Usage
Display Corrects Corrects

Setting Integrator? Shows Offset Ramp Usage
RMPCT- No D-OFF Yes No Use for non

Set Default integrating
CVs
(Always
Works) Yes D-ON Yes Yes Use for
integrating
CVs
User-Set No OFF Yes No Use for non

Off integrating
CVs
Yes OFF Yes No Do not use

for
integrating
CVs
User-Set No ON Yes Yes Use for non

On integrating
CVs with
slow drift
disturbance
s
Yes ON Yes Yes Use for

integrating
CVs

12/07
2.3. Manipulated Variables
2.3 Manipulated Variables

Characteristics
RMPCT adjusts MV values to achieve control and optimization objectives. Each MV has
a high and low limit. The controller never moves an MV outside its limits, and returns an
MV within limits when:
The controller is started up with the MV outside its limits (unless tracking is on).
The Operator changes an MV limit such that the MV value is outside the limit.
An MV can be either a direct output to some actuator, or the setpoint of a secondary

controller. The usual case is that an MV is the setpoint of a PID controller, which can
output directly to an actuator or can be cascaded to other, downstream PID control loops.
Rate-of-Change (Max Move) Limits

In addition to positional limits, rate-of-change limits can be set on each MV.
These limits are the maximum move that the controller can impose on an MV in a single
control interval. Separate rate-of-change limits can be set for positive and negative
changes.
Rate-of-change limits prevent the controller from making excessively large changes
when an abnormal event occurs. This gives the Operator a chance to intervene.
If a rate-of-change limit is set so small that the controller is hitting it much of the time,
this takes away some freedom for the controller to determine the most stable and robust
trajectory to use in correcting CV errors.
ATTENTION
If MV movement is excessive, it is generally better to increase some of the
CV performance ratios to reduce MV movement instead of reducing rate-of-
change limits.
Limit Ramping
Although the controller never moves an MV outside a limit, an MV can be outside a limit
when the MV initializes if MV tracking is not configured. Also, the Operator can change
a limit to a value such that the MV is outside the new limit.

12/07
Rather than force the controller to move an MV to a violated limit in one interval, you
can specify the minimum amount that the controller must move the MV toward the
violated limit at each control interval. This in effect determines the minimum rate at
which the MV ramps to its limit. The controller, however, is free to move the MV a
larger amount, subject to the rate-of-change limits.
If the MV is not violating the old limit but is violating the new limit, the limit ramps to
the new limit at ten times the configured ramp rate until the limit reaches the MV value.
The limit continues to ramp at the configured ramp rate until the new limit is reached. To
avoid unnecessary interference with possible future MV values that the controller may be
planning, the limit is not immediately moved to the MV value.
Different limits for high and low ramping can be set.
Movement Weights
Defined Conceptually, MV weighting is analogous to CV weighting in that weights are
applied to encourage, or discourage, controller action on particular variables.
• CV engineering give-ups encourage the resolution of particular CV errors at the
expense of other CV errors.
• MV movement weights discourage the movement of particular MVs in resolving CV
error, which results in larger movement of other MVs.
When you want an MV to move less than others, or not to move at all unless necessary,
you apply a movement weight. The movement weight penalizes movement of the MV,
and influences the controller’s choice of alternate MV moves.
All MVs are assigned an initial movement weight of 1.0, which is a neutral weighting.
Example If you set the movement weight of MV1 to twice that of MV2, the controller
moves MV1 half as much as MV2, all other conditions being equal.
Degrees of Freedom
The controller minimizes MV movement whenever possible while still meeting both the
operating and the economic objectives.
When there are more MVs than are required to meet the objectives, the controller spreads
the total MV movement across the MVs to minimize the sum of the squared changes.
The controller minimizes the sum of the squared changes of the MVs, with each change
multiplied by the movement weight for the MV.

12/07
Setting MV Priorities
You can force the controller to move particular MVs by assigning much larger weights to
the MVs you prefer to move very little or not at all.
If you give MV1 a weight of 5, for example, and MV2 a weight of 1, the controller tends
to leave MV1 alone, moving MV2 instead so long as MV2 is not constrained and can
affect the CVs that need to be changed. If MV2 hits a constraint, only then does the
controller move MV1, and only so much as to achieve the control and economic
objectives.
Use Lower Numbers

Use lower numbers to establish ranges (ratios). The controller is happier when the
average MV weighting is around one.
This does not mean, however, that you have to strive to average the weighting across
MVs. It means that in setting weights, it is better to establish a range with the low end of
the range nearer to one.
For example, a range between 10 and 1.0 is a better range than a range between 100 and
10. The ratio is 10-1 for both ranges. However, when the lower end of the range is closer
to one the average weight is closer to one, and the controller is happier.
ATTENTION
An important distinction between RMPCT and some other controllers is that
in RMPCT movement weights do not affect the speed of response or the
stability of the controller. The feedback performance ratio is used to tune the
dynamic response.
Movement weights are used only to set priorities, which MVs you prefer to
move when more than one MV can do the job. If there are redundancies in
the MVs, the MV movement weights have no affect.
MV Tracking
MV tracking is analogous to CV tracking. When you configure an MV for tracking, the
controller adjusts the appropriate MV limit so that the MV is not violating a limit when
the MV initializes. An MV is initialized when RMPCT control is turned on or when an
individual MV is returned to RMPCT control.
If the MV is within limits when it initializes, there is no adjustment. If the MV is outside
a limit, the value of the violated limit is set equal to the current MV value; the other limit
is left alone.

12/07
Bad Value Treatment

The action the controller takes when an MV goes bad depends on whether you configure
the MV as critical or non critical:
• If a critical MV goes bad, control sheds.
• If a non critical MV goes bad, control continues with the MV value frozen at its last
good value.
Anti-Windup
The controller checks the windup status of all control loops cascaded from the MV to the
ultimate output.
If MV movement up or down worsens a windup condition anywhere in the cascade, the
controller does not move the MV in that direction.
Predict-Back and Estimated Disturbance Compensation

Capturing Information from Disturbances Usually, the requirement that MVs and
DVs be independent of each other does not restrict the controller design. Sometimes,
however, important disturbance information can be contained in a variable that is
affected by one or more MVs or DVs, but the variable cannot be used directly as a DV
because the variable is not independent.
Valuable disturbance information in such processes, then, is normally lost. But, with the
predict-back and estimated disturbance compensation features of RMPCT, this
information can be gleaned, giving the controller an intelligence about disturbances it
otherwise could not have.
Tactics Predict-back control is an open loop, feedforward control that uses two models:
1. A predict-back model estimates the unrejected disturbance (the leak-through).
2. An estimated disturbance model then rejects the leak-through effects of the
disturbance.
Finding the Models Predict-back requires models of the effects of the MVs and DVs on
the variable that contains the disturbance information, which in the following illustration
is the process unit’s inlet temperature.
Estimated disturbance compensation requires models of the effects of the estimated
disturbance on the CVs. Control quality is equally influenced by the estimated
disturbance models as by the predict-back models.

12/07
Tuning The predict-back ratio tuning parameter filters the portion of the estimated
disturbance that is used for feedforward compensation. Its value is zero to 1.0. The
default is 1.0, which uses the full value of the estimated disturbance for compensation.
Predict-Back ⎯ Implementation Suggestions

The decision to use predict-back is typically made at the design phase. Predict-back is
usually used when:
1. The MV is a setpoint of a downstream controller, and
2. The disturbance rejection is slower than the effects of the unrejected (leak-through)
disturbances on the RMPCT controlled variables.
Predict-back on MVs that have relatively fast PID control yields negligible results.
Before building predict-back models, try first to improve PID performance.
Predict-Back ⎯ An Illustration
To see how predict-back and estimated disturbance compensation can be used to
advantage, examine the variables and the controls on this process:

12/07
RM09-400
Figure 2-4 A Predict-Back Problem
Predict-Back Illustration ⎯ The Disturbances

In this illustration, one of the MVs is the setpoint to the temperature controller of the feed
heater. The measured temperature of the feed into the process unit is controlled by the
temperature controller, but it is also affected by disturbances such as changes in the fuel
composition, the feed flow rate, or the feed heater inlet temperature.

12/07
Predict-Back Illustration ⎯ The Problem

True, the temperature controller eventually compensates for these disturbances. But, if
the dynamics of the temperature controller and the furnace are slow, relative to the
dynamics of the process unit, some of the CVs of the Profit Controller are disturbed.
These CVs stay disturbed until the temperature controller can return the temperature to
setpoint.
The Profit Controller takes action independently of the temperature controller to correct
the CV errors that develop before the effects of the temperature controller's actions are
felt. The Profit Controller then has to undo these actions as the temperature controller's
actions take effect.
Predict-Back Illustration ⎯ The Solution

It is better to use the inlet temperature as a DV so that RMPCT can use the disturbance
information to do a better job. However, this is not possible because the inlet temperature
is dependent on the temperature controller's setpoint, which is an MV. Predict-back and
estimated disturbance compensation get around this, though.
Here’s how:
• Predict-back uses a model of the effects of the MVs and DVs on the variable that
contains the disturbance information. In the preceding illustration, the predict-back
model includes the dynamic effects of the:
− Temperature controller,
− Fuel-flow controller,
− Furnace on the inlet temperature to the process unit, and
− Feed flow rate, if the feed flow rate is included as a regular DV.
• The effects predicted by the predict-back model are backed out of the variable that
contains the disturbance information, in this case the inlet temperature to the process
unit. This gives an estimate of the disturbance.
• The estimated disturbance compensation models are then used to obtain the effects
of the estimated disturbance on the CVs.
In this way, the Profit Controller obtains the disturbance information and can compensate
for the disturbance.

12/07
Predict-Back Illustration ⎯ The Benefits

The predict-back model includes the temperature controller's behavior of eventually
returning the feed inlet temperature to its setpoint. This way the Profit Controller sees
that the CV errors due to a disturbance are eventually corrected even if Profit Controller
does nothing.
Profit Controller may decide to do something to help out in the interim, but its actions are
smoothly coordinated with the temperature controller. The over-compensation that
occurs when Profit Controller operates independently of the temperature controller is
eliminated.
On/Off Control ⎯ Configuration

Toggling Control The Operator can switch non critical MVs on or off Profit Controller
control at any time. If a non critical MV is off control, Profit Controller does not move
that MV but continues controlling the process using the MVs that remain on control.
Critical MVs cannot be taken off control. Profit Controller sheds control when a critical
MV mode is changed.
Configuring Feedforward A feedforward option determines how the controller treats a
non critical MV when the MV is off Profit Controller control.
An MV that is off Profit Controller control but that has the feedforward option set is still
monitored so that its effects on the CVs are taken into account in case the MV is moved
by something else (typically the Operator).
In other words, when a non critical MV with the feedforward option is taken off control,
the effect is to convert it to a feedforward variable.
Dropped MVs When a non critical MV without the feedforward option is taken off
Profit Controller control, the controller stops using the MV value to update the CV
predictions. In effect, the MV is entirely dropped from the controller.

12/07
On/Off Control ⎯ An Example

Consider this example:
SP PID
RMPCT
MV FC
SP PID
Operator
TC
RM09-400
Figure 2-5 Dropping Non Critical MVs
When the MV is taken off Profit Controller control, the backup regulatory scheme
controls the valve. The setpoint to the flow controller no longer moves the process.
Instead, the Operator-controlled setpoint to the temperature controller moves the process.
When non critical MVs are dropped and backup control assumes the process, such as it
does in the illustration, the MV value cannot represent the results on the CVs that are
presumed in the model. To avoid this, configure the MVs for DROP instead of
FEEDFORWARD. This prevents Profit Controller from assuming that the MV is moving
the process congruently with the models, which it is not, when the backup PIDs have
control.
Automatic Mode Switching

In many applications there is a conventional control scheme in place prior to Profit
Controller. The existing control scheme may not be completely compatible with the
choice of variables that you make for the Profit Controller.
The Profit Controller may output to the setpoint of a flow controller that is the secondary
of a temperature controller in the conventional control scheme. When Profit Controller
control is turned on, the temperature controller cascade must open so that RMPCT has
control of the flow setpoint.
If the Operator turns Profit Controller (RMPCT) control off or if Profit Controller
(RMPCT) control is automatically shed, perhaps due to a bad value of a critical variable,
control should shed to the conventional control scheme. This means that the temperature
cascade must close. There could be a number of control mode changes of this type that
have to be made in order to switch between conventional control and Profit Controller

12/07
(RMPCT) control. These changes are difficult for an Operator to make, so a mechanism
is provided to automate the mode switching.
Configuring Automatic Mode Switching

To configure automatic mode switching, you enter the number of loops in the cascade
from each MV to the ultimate output, and fill in an array that specifies the mode of each
loop that should be in effect for conventional control when the MV is not on Profit
Controller (RMPCT) control.
If one or more MVs are taken off Profit Controller (RMPCT) control, the controller sets
up the modes that you configure so that a backup conventional control scheme takes
over.
Taking MVs Off Profit Controller (RMPCT) Control

MVs can be taken off Profit Controller (RMPCT) control in the following ways:
• Operator turns off Profit Controller (RMPCT) control. Here, all MVs are shed to
backup control.
• The value of a critical CV or critical MV goes bad and stays bad longer than the
allowed number of bad reads. Here, all MVs are shed to backup control.
• The Operator opens the cascade from Profit Controller (RMPCT) to an MV’s
ultimate output, and the MV is critical. Here, all MVs are shed to backup control.
• The Operator opens the cascade from Profit Controller (RMPCT) to an MV’s
ultimate output, and the MV is non critical. Here, the MV is shed, subject to these
contingencies:
− If the cascade is opened between Profit Controller (RMPCT) and the first loop,
the MV switches to feedforward while off control if it is so configured.
− Otherwise, the MV value is not used.
The converse of these conditions puts Profit Controller (RMPCT) back on control. When
control starts or resumes, the controller automatically establishes the cascades for each
affected MV based on the configured loops.
MV Move Accumulation
MV moves are calculated as single-precision floating-point values in IEEE format. When
the output to a DCS such as the LCN accepts values in this same format, there is no loss
of information. However, some DCSs have a coarser resolution. For example, a DCS
might be able to store only integer values from 0 to 1023 unscaled. If the scaled range
was 1000 to 3000, an MV output value of 1000 would be represented exactly on the DCS

12/07
(as unscaled 0) but the next greatest value that could be represented on the DCS would be
1001.953 (as unscaled 1). MV output values between 1000 and 1001.953 would have to
be rounded to one value or the other.
Coarse DCS resolution causes a problem when small MV moves are being calculated,
because the change at any one interval may be too small to “kick” the DCS value to its
next increment. The controller reads back the value from the DCS at the next interval to
establish where the MV actually is, and in this case will read the same value as at the
previous interval, and again add a move that will not be big enough to actually change
the DCS value.
MV move accumulation solves the problem of ignoring MV moves that are smaller than
DCS resolution. To implement MV move accumulation you either enter the DCS
resolution if it is known or tell the controller to estimate the resolution. The controller
rounds the MV output value to the nearest value that the DCS can actually represent and
outputs this value. The controller remembers the difference between the desired full
resolution output and the rounded value that was actually output, and algebraically adds
this to the move at the next interval. In this way, small moves accumulate until the
accumulation is eventually large enough to change the DCS value to its next increment.
The following parameters (per MV) are used with move resolution:
• Resolution – This should be set as follows:
-1 Move accumulation calculations are not performed. The MV output value is sent
directly to the external system. Use this value for output to a DCS such as the LCN that
supports IEEE single-precision resolution.
0 Automatic resolution calculation is performed. The controller attempts to
determine the resolution of the DCS by accumulating the maximum difference between
the read back of the MV output and the value that was output at the previous interval.
Use this value if the DCS has coarse resolution and you do not know what the resolution
is.
>0 Resolution calculation is performed using Resolution as the DCS resolution.
The value of Resolution should be the smallest increment to the MV output that can be
represented at the DCS, expressed in the units of MV output. Use this setting if the DCS
has coarse resolution and you know what the resolution is. This may be somewhat more
reliable than the automatic resolution calculation.
• Calculated Resolution – If Resolution = -1 or Resolution = 0, Calculated Resolution
is the estimated resolution of the DCS value (i.e., the smallest increment of MV
output that can be represented at the DCS). This should be zero if the DCS supports
IEEE single-precision. If Resolution > 0, Calculated Resolution is the maximum
mismatch between the MV value output at one interval and read back at the next,

12/07
which should be zero or very small. If Resolution > 0 and Calculated Resolution is
significant, the value entered for Resolution is probably incorrect.
• Resolution Mismatch – The difference between the value read back from the DCS
and the value output to the DCS at the previous interval.
• Resolution Residual – The amount of unrealized MV output (i.e., the difference
between the full-resolution value and the nearest value that can be represented on the
DCS).

12/07
2.4. Disturbance Variables
2.4 Disturbance Variables

Characteristics
DVs affect the CVs in the same way that MVs affect the CVs. The DVs, however, are
not under Profit Controller (RMPCT) control, typically because DV behavior is
determined outside the scope of the process unit being controlled.
But, by including the DVs in the controller model, their future effects on the CVs can be
anticipated by the controller. Profit Controller (RMPCT), then, can take corrective action
to prevent disturbances from driving CVs outside their limits or away from setpoint.
The usefulness of a DV depends on the dynamics between the DV’s measured value and
its effect on the CVs. If a DV instantaneously affects a CV, the DV is less useful because
a CV error already exists by the time the controller can do anything about it.
To be useful, DVs need time constants or a dead time comparable to, or larger than, at
least one of the MVs.
• If a DV’s dead time is not larger than at least one of the MVs, it is impossible for the
controller to cancel all of the effects of the DV.
• If dead times are equal, it is theoretically possible for the controller to cancel the
effects of the DV. However, large MV changes are required if the DV’s time
constants are small relative to the MV’s.
DV Influence on CV Tuning
You can configure the desired feedforward response time of a CV to a DV differently
than the feedback response time of a CV to the effects of unmeasured disturbances and
model error.
Feedback tuning is set by the CV Feedback Performance Ratio. Inevitable errors in the
model limit how fast you can set the feedback response. At some point, attempts to
achieve faster response result in oscillations and eventually unstable control. These
instabilities are caused by feedback, which is necessary to correct unmeasured
disturbances and model error.
If feedforward information is treated outside the feedback loop, as it is with RMPCT,
then feedforward does not contribute to unstable control and the feedforward response
can be tuned much faster.
Feedforward Performance Ratio

Configuration You configure the feedforward response time by setting the feedforward-
to-feedback performance ratio.
12/07
2.4. Disturbance Variables
This ratio multiplies the feedback performance ratio, which in turn multiplies the nominal
open-loop settling time to determine the response time.
Example The feedback performance ratio is set to 0.8 and the feedforward-to-feedback
performance ratio is set to 0.5 for some CV.
Here, the controller corrects:
• Feedback errors (the unmeasured disturbances and the contribution of model error)
in 0.8 of the nominal open-loop settling time.
• Feedforward errors (caused by DVs) in 0.4 of the nominal open-loop settling time
(0.5 x 0.8 = 0.4).
Plus the nominal dead time for both.

12/07

12/07
3. Control Interval
3.1 Overview
Read This
The control interval (also called the execution interval or execution period) is established
during the controller design and build process. This interval cannot be changed on-line.
Once a controller is installed on an automated control system, the only way to change the
control interval is to rebuild and reinstall the controller.
In This Section
This section explains the relationship between the control interval and the settling time,
time constants, CV constraints, independent variables, and blocking. Understanding these
relationships and their interplay can help you pick the best control interval for your
application.

12/07
3. Control Interval
3.2. Control Interval and Settling Time
3.2 Control Interval and Settling Time

The Ideal Interval-to-Settling Time
If all the sub-process models had approximately the same time constants, you would
typically choose the control interval to be approximately 1/40 of the settling time. The
settling time is approximately four times the major time constant, so the control interval
would be about 1/10 of the major time constant.
Shorter Intervals
A smaller interval than 1/10 of the major time constant does not noticeably improve
control because the process does not appreciably respond in less than this interval.
Smaller control intervals entail more processing overhead and more memory to hold the
step response coefficients, so the ideal interval size is a reasonable compromise between
computational efficiency and control performance.
Longer Intervals
The only deleterious effect of a long control interval is that it delays recognition of
disturbances and subsequent corrective action.
When Time Constants Vary

Many processes exhibit a wide range of time constants in the various sub-processes. This
complicates the choice of a control interval.
It can be necessary to choose an interval that is longer than ideal for the sub-processes
with short time constants in order to keep the computational and memory overhead
reasonable for the sub-processes with long time constants.
Recommended Intervals
There are no hard rules, but it is generally desirable to have the longest settling times less
than 200 - 300 intervals, subject to having the shortest settling times for important CVs
more than 10 intervals.
If the time required to detect an error is not a problem, the control interval can be set as
large as the settling time of a CV, or even larger.
Intervals > Settling Time

If the performance ratio for a CV is set to 1.0 and the control interval is set equal to the
open-loop settling time, the controller corrects limit violations within the open-loop
settling time, which in this case is one control interval.

12/07
3. Control Interval
3.2. Control Interval and Settling Time
This is no more difficult for the controller than if the settling time consisted of 40 control
intervals.
However, if a large disturbance occurs just after a control calculation, no corrective
action is taken until one settling time elapses. Then it takes an additional settling time to
achieve the correction. For many CVs, this is not a problem.

12/07
3. Control Interval
3.3. Blocking
3.3 Blocking
Defined
In principle, the controller should consider a constraint on each CV at each control
interval in the future out to the correction horizon (where steady state is achieved). In
practice, however, this can result in too large a problem to be solved efficiently. So, the
controller places constraints only at selected intervals. This is blocking.
Control Quality
Blocking doesn’t appreciably reduce the quality of control because there is little
incentive for the controller to generate moves that cause CVs to wiggle outside of
constraints in the intervals between the blocking intervals when constraints are not
actually imposed. Such behavior requires high frequency changes to the MVs, which the
controller tries to avoid because MV movement results in less robust control.
CV Constraints
The default number of constraints for a CV is 10 (10 is also the maximum number of
constraints). You can specify a smaller value for each CV.
The specified number of constraints are distributed approximately uniformly over the
time range from the current interval out to the correction horizon. The last CV blocking
will always be placed around the longest setting time interval of all the associated
submodels of a particular CV. The controller can deviate from a uniform distribution to
place constraints where they do the most good, as determined from the distribution of
dead times and correction horizons.
MV Moves
Ideally, if the computing power were available, the controller would consider each MV
an independent variable at each interval from the present out to the longest correction
horizon of any CV affected by the MV (minus the dead time).
Unfortunately, just as with CVs, this can result in too large a problem to be solved
efficiently. Therefore, the controller places independent variables only at selected
intervals. When the performance ratio of a CV is tuned greater than 1, the associated MV
control horizon will also be extended.
Independent Variables
The default number of independent variables for an MV is 10 (10 is also the maximum
number of independent variables). You can specify a smaller value for each MV.

12/07
3. Control Interval
3.3. Blocking
The specified number of independent variables are distributed from the present time
interval out to the longest correction horizon of any CV affected by the MV, minus the
dead time. The controller places the independent variables closer together at first and
then farther apart the further they are into the future.
Reducing the Calculation Time

If the controller takes too long to execute its algorithms (more than half the length of the
control interval), you can specify a smaller number of blocking intervals to speed the
execution.
For most processes, there is little change in the quality of control when the number of
blocking intervals is reduced to five. Even fewer blocking intervals can often be
satisfactory.
Controller calculation time is reduced more by decreasing the number MV blocking
intervals than by decreasing the number CV blocking intervals.

12/07

12/07
4. Economic Optimization
4.1 Overview
In This Section
This section explains how Profit Controller (RMPCT) uses an objective function for
optimizing control, and how the optimization horizon and the CV and MV soft limits
influence optimization calculations.
Optimization Concepts
To understand how Profit Controller (RMPCT) optimizes a process, here are a few
fundamental terms you should know:
Table 4-1 Important Optimization Terms in RMPCT
Term Meaning
Objective Function What you want the controller to accomplish after the
control objectives are met (improve product spec,
increase product throughput, lower utility costs).
Degrees of Freedom The number of MVs not at a limit minus the number of
CVs that either have setpoints or are at or outside
limits. The controller chooses MV values so as to
minimize the number of CVs that are away from
setpoint or that are outside limits.
Optimization Horizon How fast the controller must bring the objective function
to an optimal value.
Soft Limits Offset from high and low limits to protect future freedom
for moving MVs, and for leaving room in CVs for
unexpected disturbances.

12/07
4.2. Objective Function
4.2 Objective Function

Degrees of Freedom
In many applications, the control requirement to keep all variables inside constraints does
not use up all the degrees of freedom available to the controller.
Even when there are more CVs than MVs, there can be extra degrees of freedom if a
number of the CVs have ranges rather than setpoints, which is often the case.
You can put the extra degrees of freedom to good use by defining an objective function
so that the controller optimizes some aspect of the process in addition to controlling it.
General Form
The objective function is a linear or quadratic function of any or all of the CVs and MVs.
The general form of the objective function is:
minimize J = ∑ bi x C Vi + ∑a i 2 ( CVi − CV0 i )

2
+
i i
∑ bj x M Vj + ∑a j 2 ( MVj − MV 0 j )
2
j j
where
bi are the linear coefficients on the CVs
bj are the linear coefficients on the MVs
ai are the quadratic coefficients on the CVs
aj are the quadratic coefficients on the MVs
CV0i are the desired resting values of the CVs
MV0j are the desired resting values of the MVs.
To maximize rather than minimize the objective function, multiply each term by -1
(minimizing the negative of something is the same as maximizing it).
The controller minimizes the objective function (or maximizes the negative of it) subject
to keeping all CVs within limits or at setpoint, and all MVs within limits.

12/07
Weighting Coefficients
Emphasizing What Is Valuable Weighting of the linear objective coefficients is
relative. A larger absolute value relative to other CV and MV coefficients emphasizes
that variable. A “-100”, for example, emphasizes ten times the desirability that “-10”
attributes, so far as the optimizer is concerned. And 100 attributes ten times the cost that
10 attributes.
Example The objective function can be used for many purposes. For a simple example,
assume that there is one valuable product from a process controlled by RMPCT and you
want to maximize this product.
MV3 is the setpoint of a PID controller that controls the flow of this product. To
maximize its flow rate, set b3 = -1 (or any other negative number with a reasonable
magnitude), and set all the other objective function coefficients to zero.
Then the objective function is
minimize J = -MV3
which is the same as
maximize J = MV3
which causes the controller to maximize the valuable product.
MV3 typically would have a high limit that is the maximum flow rate achievable. Under
some operating conditions the controller can increase MV3 until it hits its high limit.
Under other operating conditions, however, other MV and CV limits are hit first. When
this happens, the controller may have to reduce MV3 to keep other variables within their
limits.
In any case, the controller always determines which constraints should be active in order
for MV3 to be as large as possible.
Maximizing Profit vs Minimizing Cost

In many applications the major product and feed streams and the utility usage are
included in the set of MVs and CVs.
For such applications a typical use of the objective function is to maximize operating
profit, defined as the value of products minus the cost of feeds and utilities.
Setting Linear Objective Coefficients

The linear objective coefficients are the knobs for maximizing operating profit. Here is
how you set the coefficients:

12/07
• Set the linear coefficients (bi or bj) of all CVs and MVs that are products to the
negative of the product value. ⎯ For example, if CV2 is a product stream with units
of kg/hr and the product has a value of $1.2/kg, set b2 = -1.2. This assumes that you
want the units of the objective function to be $/hr.
• Set the linear coefficients of all CVs and MVs that are feeds to the feed cost.
• Set the linear coefficients of all CVs and MVs that are utility streams to the utility
cost.
• Set all other objective function coefficients to zero.
When Limits Do Better Than SetPoints

Define Boundaries Sometimes the objective function is influenced by most or all of the
variables in the controller, which can happen when the objective function is used to
maximize operating profit. In such cases, define the allowable operating boundaries by
setting limits rather than setpoints on most or all CVs.
Setting limits rather than setpoints lets the optimizer determine where to drive the
process.
Example For example, rather than putting a setpoint on a feed rate, you can set an upper
limit based on the pumping limit or on feed availability, and let the optimizer determine
the optimal amount of feed.
In this case, the optimization typically increases feed until one or more constraints are hit
and then rides those constraints. You, therefore, have the responsibility to include in the
controller all the constraints that might be hit under varying operating circumstances.
Range Limits and Product Quality

Controller Freedom With the optimizer configured to maximize operating profit, even
product qualities may be given limits rather than setpoints.
Optimization, by definition, ensures that there is no unnecessary giveaway of valuable
product due to a quality being better than spec. The optimization pushes qualities to their
limits unless some (unusual) constraints prevent this.
The advantage of configuring a limit rather than a setpoint is that the controller has more
freedom to correct control errors.
Example Consider, for example, when two side draws on a crude column both have
end-point specifications. When a disturbance causes one end-point to be outside its high
limit, the controller is temporarily allowed to make the other end-point better than spec.

12/07
When the controller is able to work within a range rather than being constrained to
holding setpoint, the controller corrects the error more easily (and you can configure a
much shorter correction horizon). The optimizer can then more leisurely push both
qualities against their constraints.
When Setpoints Do Better Than Limits

Of course, there are cases where ranges are not appropriate and setpoints need to be used
instead.
For example, sometimes it is as important to minimize the variability of a quality
parameter as it is to keep the quality within some product quality limit. In such a case,
you have to use your knowledge of the process and the product to decide whether a limit
or a setpoint should be set.
Remember, limits allow the controller more freedom to optimize operations and provide
robust control.

12/07
4.3. Optimization Horizon and Optimization Speed Factor
4.3 Optimization Horizon and Optimization Speed Factor

Optimization Horizon
The optimization horizon specifies how fast the controller must bring the objective
function to its optimal value. This is set independently of the error correction horizons on
the CVs.
You typically want to set the optimization horizon considerably longer than the error
correction horizons. This avoids the occasional exposure of letting optimization
compromise the robustness of the controller.
Optimization Speed Factor

You set the optimization horizon indirectly by setting the optimization speed factor.
The default setting for the optimization speed factor is 1.0, which results in an
optimization horizon approximately six times the CV overall response time.
The CV overall response time is defined as the average of the longest CV response time
and the average CV response time.
ATTENTION
Setting the optimization speed factor to zero turns the optimizer off, which
turns off the objective function. CV and MV objective coefficients, then, have
no influence on the direction of the process.
To increase the optimization speed (to decrease the optimization horizon), set a larger
value for the optimization speed factor, and vice versa.

12/07
4.4. Soft Limits
4.4 Soft Limits

How Soft Limits Work
In addition to the hard-set control limits, you can use soft limits on the CVs and MVs.
Soft limits are inside (are more restrictive than) the hard limits.
Hard Limits
Soft Limits
Control Optimization Range

Range
Soft Limits
Hard Limits
RM09-400
Figure 4-1 How Soft Limits Work
You do not set soft limits directly. Rather, you set the amount that the soft limit is inside
the control limit. Soft limits are respected by the optimization of the economic objective
function. In other words, the optimum value is subject to all variables being inside their
soft limits. Soft limits are ignored for the control purposes of keeping the CVs inside
their control limits or at setpoint.
CV Soft Limits
You can set a soft limit on a CV slightly inside the control limit to provide a buffer so
that disturbances of normal magnitude do not bump the CV outside the control limit.
If the optimization limit and the control limit are instead both set equal to the actual spec
on the CV, optimization then tends to keep the CV at the spec value. Consequently, even
a small disturbance can push the CV out of spec until the controller reacts.
If you configure the soft limit some small amount inside of the actual spec, optimization
cannot push the CV all the way to the spec. Optimization, in fact, tends to bring the CV
back to the soft limit if the CV is outside the soft limit but inside the control limit. This
action proceeds at the speed determined by the optimization horizon, and provides a
buffer. If a large disturbance pushes the CV outside the control limit, the much more
aggressive error correction horizon comes into play, and the CV is quickly brought back
within the control limit.

12/07
4.4. Soft Limits
MV Soft Limits
You can set a soft limit on an MV slightly inside the control limit to ensure that the MV
is available to correct disturbances in either direction most of the time. If the controller
moves the MV outside a soft limit to correct CV errors (it cannot move the MV outside
its control limit), then the optimizer adjusts other MVs to gradually bring this MV back
within its soft limit.
Keeping MVs slightly inside their control limits can improve controller robustness by
giving the controller more freedom to correct CV errors.

12/07
5. Robust Control
5.1 Overview
In This Section
This section tells you how RMPCT control solutions get their robustness through these
functions:
• Range control
• Limit funnels
• Singular-value thresholding.
Revisiting the Process Model

Control of interacting variables requires coordinated moves of a number of MVs. The
controller determines how to make these moves by simultaneously considering the
effects of all MVs on all CVs. The controller’s knowledge of these effects is based on the
process model.
Model Error is Unavoidable

If the model contains significant error, the effects of the control actions on the process
differ from the effects predicted by the controller, and the quality of control degrades to
some extent.
The problems caused by model error are more pronounced for processes that have large
interactions between CVs that are assigned setpoints or are at constraints most of the
time.
Unavoidably, all models contain some error. One of the most important aspects of a
multivariable controller is its ability to cope with this error.
Coping with Model Error

RMPCT excels in its ability to provide good control even with large errors in the process
model and large interactions between CVs.
Here is a broad penned, first-look at how RMPCT manages this:
• Range Control You have the option of controlling CVs within a range rather than
to setpoint. This gives the controller greater freedom to reject disturbances.
• Limit Funnels The controller imposes its own constraints on CVs to avoid
introducing transient errors.

12/07
5. Robust Control
5.1. Overview
• Singular-Value Thresholding Determining a threshold of singular values and

dropping values smaller than this threshold reduces the chance of magnifying model
error because of a poorly conditioned matrix.
• Min–Max Design This kind of controller design ensures the best controller
performance under the worst combination of estimated model error.
The rest of this section tells you how to use these functions to maximum advantage.

12/07
5. Robust Control
5.2. Range Control
5.2 Range Control

Defined
In many applications many of the CVs have range limits rather than setpoints. RMPCT
takes advantage of range limits to improve the quality of control when there is significant
model error or when a number of CVs interact strongly.
Basically, range control uses range limits directly in the dynamic control solution, rather
than using a steady state optimization prior to the dynamic solution to generate internal
setpoints for the dynamic solution.
This gives the controller the freedom to allow a CV to be anywhere within a prescribed
range rather than being forced to keep each CV at a particular value.
Setpoint Control ⎯ An Example

To see the advantages of range control, consider a fractionation column with top and
bottom products each with quality specifications. Given:
• The top draw has a CV that is the percent of a heavy key in the top product, and the
bottom draw has a CV that is the percent of a light key in the bottom product.
• The MVs are reflux and reboil.
• The CVs interact strongly. Any changes made to the MVs tend to change both CVs
in the same direction and by approximately the same amount.
Of course, there must be some difference in the influence of the MVs on the CVs;
otherwise, it would be impossible to control the two CVs independently. If the difference
is small, though, the controller must make large changes to the MVs to move the CVs
independently a small amount.
The Problem
Assume that a disturbance pushes the top quality above its high limit, but the bottom
quality remains at its high limit. If setpoints are used, the controller must move the top
CV back to its limit while not moving the bottom CV.
Because these CVs want to move together, the independent movement of one of them
requires large changes to the MVs to move the top CV the desired amount, but with
canceling effects on the bottom CV.
The model separately predicts the effects of the reflux change and the reboil change on
bottom quality, and the controller chooses large changes in reflux and reboil whose
effects on bottom quality cancel out. The problem here is that the controller relies on a

12/07
5. Robust Control
5.2. Range Control
small difference between two large calculated values, both of which have some error.
These errors become greatly magnified in the difference, which is the CV value. The
result can be unstable control.
Range Control ⎯ An Example

Now consider what happens with range control.
As with setpoint control, the controller must move the top CV back to its limit. With
range control, however, the bottom CV can be allowed to move below its high limit (as
long as it does not go below its low limit). In effect, a highly interacting control problem
has been converted into control of just a single variable.
The controller can accomplish this with much smaller MV moves and does not have to
rely on the effects of MV moves canceling each other. Model error, then, is not
magnified. The economic optimization slowly returns the bottom CV to its high limit.
This does not cause instability because the optimization speed is slow relative to the
process dynamics.
The Control Matrix

From a mathematical point of view, the controller inverts a matrix of the dynamic
process model in order to calculate the MV moves. If the matrix contains constraints on
highly interacting variables, it is poorly conditioned, and the solution is sensitive to
model error.
With range control, a constraint is in the matrix only if the CV is at or outside its limits in
the solution. In many cases, the correction of a CV that is outside a limit often moves
interacting CVs inside or within limits. Only one of the interacting CVs is constrained in
the solution matrix. The matrix is, then, well conditioned and is not sensitive to error.
Ranges and Optimization

Unless minimization of the variation of a CV is very important in its own right, use
ranges rather that setpoints to define the operating region.
Use optimization to ensure that the controller operates the process at the best place within
the range (which usually means pushing against a number of CV limits).

12/07
5. Robust Control
5.3. Limit Funnels
5.3 Limit Funnels

Defined
The correction horizon concept is that CV errors are reduced to zero at the correction
horizon in the future. Prior to the correction horizon, the controller is free to determine
any trajectory for the CV as long as the CV is brought within limits or to setpoint at the
correction horizon.
Because no trajectory is imposed on the controller, the controller has the freedom to
determine a trajectory that requires minimum MV movement and is least sensitive to
model error.
However, the correction horizon by itself does not say anything about what happens to
the CV prior to the horizon. It is important that the controller does not transiently move a
CV farther outside a limit while correcting other CV errors, even though all CVs are
brought to zero error by their correction horizons.
Limit funnels are used to prevent the controller from introducing transient errors prior to
the correction horizons, by defining constraints on the CVs that are imposed at intervals
from the current interval out to the horizon.
Three Examples
The following figures illustrate limit funnels for different situations. Constraints are
indicated by the converging lines of the funnel.
Examine these figures to see how Profit Controller (RMPCT) moves the CV within the
funnel when control is made within a range, to a setpoint, and by a setpoint change.

12/07
5. Robust Control
5.3. Limit Funnels
Example 1: Range Limits

CV
Funnel Opens With No Further With Control
Automatically Control Action Action
to Current
Value Plus
Tolerance
High Limit
CURRENT TIME
Low Limit
Minimum
Funnel
Height CORRECTION
HORIZON
MEASURED PREDICTED
PREDICTED
t
RM09-400 Funnel
Figure 5-1 Funnel with Range Limits
Example 2: Constant Setpoint

CV
Funnel Opens With No Further
Automatically Control Action With Control
to Current Action
Value Plus
Tolerance
Setpoint
Minimum
Funnel
Height CURRENT T IME
CORRECTION
HORIZON
MEASURED PREDICTED
PREDICTED
t
RM09-400 Funnel
Figure 5-2 Funnel with Constant Setpoint

12/07
5. Robust Control
5.3. Limit Funnels
Example 3: Setpoint Change

CV
With Control
Action
Setpoint
Minimum
Funnel With No
Height Further
CURRENT TIME
Control
Action
CORRECTION
HORIZON
MEASURED PREDICTED
PREDICTED
t
RM09-400 Funnel
Figure 5-3 Funnel with Setpoint Change
How the Funnels Work

The shape of the high and low constraint lines resembles a funnel turned on its side. The
mouth is opened beyond the Operator-entered limits or the CV value at the current time,
and narrows to the limits at the horizon.
The controller is obliged to keep the CV within the constraints defined by the funnel, but
is allowed to follow any trajectory within these constraints.
The controller automatically determines the amount that the funnel opens beyond the
limit or CV value at the current time. This gives flexibility to the control response (such
as when handling inverse response dynamics), and still prevents an excessive amount of
transient coupling of CV values.
Funnel Type and Decouple Ratio

There are three funnel types that the user can set for each CV. Funnel type 0 is the
default. We recommend the user to use funnel type 0 to achieve the most robust control.
Under certain situations, it may be necessary to tighten the CV funnel opening to prevent
excursions. When the funnel type is set to 1, the opening of the funnel is pinched down to
about 1/3 of the funnel type 0 opening. The funnel opening will be further decreased
when the performance ratio is decreased. When funnel type is set to 2, the opening of the

12/07
5. Robust Control
5.3. Limit Funnels
funnel is decided by a new tuning factor called the decouple ratio (value range from 0 to
1). The smaller the decouple ratio is, the tighter the funnel opening will be. The funnel
opening of type 2 does not change with the performance ratio. Whichever funnel type
and tuning is used, the funnel will always be relaxed to include the current CV value. The
following figure gives an illustration of the new funnel type for a range controlled CV.
Figure 5-4 New funnel types with Range Limits
When to Use Funnel Type 1 and 2

The funnel type 1 and 2 could be used to reduce the amount of the excursion of the
corresponding CV. Usually, a tighter funnel will result in a smaller excursion. Different
funnel types will not affect the setpoint response of the particular CV unless there is a big
lead or inverse response. It is not recommended to use funnel type 1 and 2 unless
necessary.
Funnel type 1 and 2 will only reduce the particular CV error when the total error of all
the CVs could be efficiently reduced. Generally, the CV excursion is not a linear function
of the funnel opening. At the beginning of decreasing the funnel opening, the excursion
may not be reduced if the excursion is still smaller than the funnel opening. Also, when
the funnel opening is small enough, further reducing the funnel opening will bring no
more benefits.
12/07
5. Robust Control
5.3. Limit Funnels
If the CV excursion cannot be prevented by different funnel types and decouple ratio,
small performance ratio or ff2fb ratio may be considered. However, the robustness of the
controller could be compromised.

12/07
5. Robust Control
5.4. Singular-Value Thresholding
5.4 Singular-Value Thresholding

Defined
In Section 5.2, “Range Control,” we found that the problems caused by model error can
be severe when the dynamic process matrix is poorly conditioned.
Range control goes a long way to keep the matrix as well-conditioned as possible.
However, if CVs are violating limits and have to be moved in the direction in which they
interact, or if setpoints are imposed, the matrix can still become poorly conditioned.
Singular-value thresholding prevents unstable control even in these situations.
Matrix Condition
Singular-value thresholding is based on the numerical properties of the singular values of
the dynamic process matrix. The elements in this matrix are the model step response
coefficients, which invariably have some error.
In general, if some of the singular values of a matrix are much smaller than the largest
singular values, the matrix is poorly conditioned (the condition number is the ratio of the
largest to smallest singular values).
For a poorly conditioned matrix, small changes in the matrix coefficients within their
error range result in large changes in the CV values that are predicted by multiplying the
matrix by a proposed set of MV changes.
This magnification of the model error can cause unstable control.
Determining a Threshold
Profit Controller (RMPCT) determines a threshold such that singular values smaller than
the threshold can cause unacceptably large errors in the CV predictions. To get around
this, singular values smaller than this threshold are dropped from the solution. This has
the effect of slowing down the control response slightly.
Typically, stable control is maintained and large reductions in MV movement are
achieved with only small reductions in CV error correction when the control matrix
becomes poorly conditioned.
Maximizing the Matrix Condition

Singular-value thresholding does not drop one or more CVs from control. Rather, the
reduction in CV error correction is spread across the CVs that are causing the poor
conditioning in such a way as to maximally improve the matrix condition.

12/07
5. Robust Control
5.4. Singular-Value Thresholding
CVs that are not contributing to the poor conditioning are not affected. When the matrix
is not poorly conditioned, singular-value thresholding has no effect on the control.

12/07
5. Robust Control
5.5. Min-Max Design
5.5 Min-Max Design

Your Input to the Offline Design
When you build the controller (done offline on a PC), you can optionally specify the
min-max design procedure.
The min-max design requires the following information:
• An estimate of the errors in the model parameters. This can be a rough guess, such as
50% error in all parameters. If you have enough data to identify the process model
twice from two different sets of plant data, the differences between the parameters in
the two models is a good indication of model error.
Model error can be especially conspicuous if the two sets of data are from two
periods of different operating conditions.
• A list of the CVs that have setpoints and CVs that you expect will be against a limit
most of the time.
• The error correction horizons that you expect to use for these CVs.
From your input, the min-max design then determines the controller parameters that
result in the best control, given the combination of model error that cause the worst
problems.
Overview of the Design Procedure

The design uses min-max optimization. You provide the nominal model. With that
model, the min-max algorithm conceptually does this:
1. Determines the controller parameters that give the best control performance. The
best control performance is defined as the minimum integral of the squared CV
errors (Integral Squared Error criterion) in response to CV setpoint changes.
2. Adjusts the model parameters anywhere in their error ranges in any combination to
find the model parameter values that result in the worst control performance.
3. Determines from the worst-performance model parameters a new set of controller
parameters that give the best control performance, given the adjusted model
determined in the previous step.
4. Repeats the previous two steps until the procedure converges. When the min-max
algorithm finds this convergence, the controller parameters provide the best control
performance.

12/07
5. Robust Control
5.5. Min-Max Design
Benefits
Sweat the details. Careful attention to the off-line design, time consuming as it might be,
can result in dramatically better control for processes with highly-interacting variables.
The results are similar to those achieved with H∞ or µ-synthesis design techniques.

12/07
5. Robust Control
5.6. Optimal Scaling
5.6 Optimal Scaling

Purpose of Optimal Scaling
Scaling improves the accuracy of the control solution calculations.
As part of the controller build procedure, Profit Controller (RMPCT) determines the
scaling for CVs and MVs that gives the best condition of the dynamic process matrix.
Scaling normalizes the control problem so we can compare variables in different
engineering units and ranges with each other. This balance of units and ranges is
achieved using scaling factors obtained from the process gain matrix. You can influence
default scaling factors in cases when you have different preferences for trade-off between
different CVs. Specifying Engineering Unit (EU) give-up factors for each of the CVs will
give desired trade-offs. CV EU give-ups are inversely related to CV scaling factors.
As part of the controller build procedure you will be presented with a dialog box for
specifying EU give-ups and MV weights. You are also allowed to change these factors
on-line. However, when changes to EU give-ups or MV weights are large users are
encouraged to include them in the off-line design procedure. This will ensure a proper
singular value threshold for robust control (Section 5.4).
How Values Display on Profit Controller (RMPCT) Screens

CV and MV values and their limits are automatically scaled on input to the RMPCT
calculations. Outputs are automatically unscaled.
Values display on the Operator screens in unscaled units.

12/07
6. Quick Reference to the Profit Controller
Displays
6.1 Overview
In This Section
The references included in this section are to help you quickly locate the configuration,
tuning, and optimization parameters described in this book.
References in this section are not inclusive – there are many more parameter settings
available in RMPCT than are listed here.
Target Access
Configuration, tuning, and optimization parameters require ENG key access.
See Control (Operating) Parameters later in this section for the targets open to Operators.

12/07
6. Quick Reference to the Profit Controller Displays
6.2. Finding Targets on the Profit Controller (RMPCT) Displays
6.2 Finding Targets on the Profit Controller (RMPCT)

Displays
Tuning and Configuration Parameters
The following table tells you where to set tuning and configuration parameters on the
Profit Controller (RMPCT) displays:
Table 6-1 Display Reference to Tuning and Configuration Parameters
Set This Parameter On These Displays
Blocking Use the CV, MV Detail screens, or the Control Tuning

screens.
Control Interval Strictly speaking the control interval is a controller

design parameter.
The control interval is established See “Blocking” on page 48 for a discussion of how
off-line in the Controller Builder. blocking influences the execution speed (cycle time).
Once a controller has been put on
the system, the control interval
cannot be changed. If you want a
controller to execute at a different
frequency, rebuild and reinstall the
controller.
Critical, Non Critical Control Use the [CRITICAL] targets on the CV, MV, DV Detail
screens, or the Process Tuning screens.
Drop, Feedforward Control Toggle the drop/feedforward settings with the [WHEN
MV IN MAN] targets on the MV Detail screen.
EU Give-Ups Use the [EU GIVE UP] targets on the CV Detail

screens.
Weights Use the [WEIGHT] targets on the MV Detail screens, or

the Control Tuning screens.
Feedback Performance Ratio Set the feedback performance ratio with the
[PERFORMANCE RATIO] target on the CV Detail
screen.
Feedforward Response Ratio Set the feedforward response ratio with the [FF TO FB
PERF RATIO] target on the CV Detail screen.
PV Tracking Toggle PV tracking ON/OFF with the [LIM TRACKS]

12/07

targets on the CV, MV Detail screens, or on the CV, MV
Process Tuning screens.
Ramp Rates Use the [HI/LO RAMP RATE] targets on the CV, MV
Detail screens.
Rate of Change Set the MV rate of change with the [MAX MOVE]
targets on the MV Detail screen.
Sampling Frequency Toggle synchronous/asynchronous sampling with the

[Update Frequency] targets on the CV Detail screen.
State Estimation Toggle state estimation ON/OFF with the [STATE EST]
target on the CV Control Tuning screen.
Funnel Type Set the funnel type with the [FUNNEL TYPE} target on
the CV Advanced Tuning screen. If Funnel Type 2 is
selected, use the corresponding [DECOUPLE RATIO]
target to set the opening of the funnel.
Move Accumulation Use the [RESOLUTION] target on the MV Detail screen.
Optimization Parameters
The following table tells you where to set optimization parameters on the RMPCT
displays:
Table 6-2 Display Reference to Optimization Parameters
Objective Function Use the linear and quadratic coefficient

targets on the CV, MV Detail screens, or
on the Optimizer Tuning screens.
Optimization Speed (Optimization Horizon) Use the [optimizer speed factor] target on
the Controller Detail screen.
Soft Limits Use the [HI/LO ∆ Soft LIMIT] targets on

the CV, MV Optimizer Tuning screens.

12/07
Control (Operating) Parameters

• Use the Summary screens for entering setpoints.
• Use the Summary screens or the Detail screens for entering high and low limits and
for setting MV modes.
• The RMPCT controller can be toggled ON and OFF from the Controller Bar on any
display.
Honeywell International
Process Solutions
2500 W. Union Hills Drive
Phoenix, AZ 85027 USA
1-800 343-0228

Profit Controller Concepts Reference Guide

Uploaded by

Copyright:

Available Formats

Profit Controller Concepts Reference Guide

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Profit Controller Concepts Reference Guide

Uploaded by

Copyright:

Available Formats

Honeywell Process Solutions

Advanced Process Control

Other brand or product names are trademarks of their respective owners.

R310 Advanced Process Control Profit Controller Concepts Reference Guide ii

Purpose This book provides a technical though broadly penned

Audience Process and control engineers

For Product Release All Profit Controller (RMPCT) releases

LCN releases 500 and 600 series.

AxM X-side releases 200 and higher

Who Should Use This Book

What This Book Tells you about RMPCT

RMPCT Course Information

How This Book Is Organized

In This Section You Can Find This Information

What information you can find in the

What writing conventions have been used

iv Advanced Process Control Profit Controller Concepts Reference Guide R310

In This Section You Can Find This Information

Section 1, “Quick Tour” The different RMPC and Profit Controller

What Profit Controller (RMPCT) utilities

How to implement a controller on an

Section 2, “Profit Controller (RMPCT) How Profit Controller (RMPCT) uses

Section 3, “Control Interval” How settling time, time constants, CV

Section 4, “Economic Optimization” How Profit Controller (RMPCT) uses an

Section 5, “Robust Control” How Profit Controller (RMPCT) control

R310 Advanced Process Control Profit Controller Concepts Reference Guide v

Writing Conventions Used in This Book

• Graphic buttons appear in UPPERCASE within brackets [ ]. For example, touch

Author Title City/Publisher/Year

Åström and Computer-Controlled Systems Englewood Cliffs: Prentice, 1987

Choi ARMA Model Identification NY: Springer-Verlog, 1992

Ljung System Identification Theory for Englewood Cliffs: Prentice, 1987

Pankratz Forecasting with Dynamic NY: Wiley, 1991

Seborg, et al Process Dynamics and Control NY: Wiley, 1989

Stephanopoulos Chemical Process Control Englewood Cliffs: Prentice, 1990

Zhu and Backx Identification of Multivariable London: Springer-Verlog, 1993

1. QUICK TOUR ...............................................................................13

2. PROFIT CONTROLLER (RMPCT) VARIABLES .........................19

R310 Advanced Process Control Profit Controller Concepts Reference Guide ix

A Caution about Ramp Correction....................................................................................... 28

x Advanced Process Control Profit Controller Concepts Reference Guide R310

3.3 Blocking ......................................................................................................... 48

4. ECONOMIC OPTIMIZATION .......................................................51

5. ROBUST CONTROL ....................................................................59

R310 Advanced Process Control Profit Controller Concepts Reference Guide xi

5.3 Limit Funnels .................................................................................................63

6. QUICK REFERENCE TO THE PROFIT CONTROLLER DISPLAYS

R310 Advanced Process Control Profit Controller Concepts Reference Guide 13

1.2 Profit Controller (RMPCT) Applications

SISO vs Multivariable Control

14 Advanced Process Control Profit Controller Concepts Reference Guide R310

The Controller Model

R310 Advanced Process Control Profit Controller Concepts Reference Guide 15

Predict-Back and Estimated Disturbance Models

16 Advanced Process Control Profit Controller Concepts Reference Guide R310

R310 Advanced Process Control Profit Controller Concepts Reference Guide 17

R310 Advanced Process Control Profit Controller Concepts Reference Guide 18

CVs, MVs, and DVs Defined

R310 Advanced Process Control Profit Controller Concepts Reference Guide 19

2.2 Controlled Variables