Professional Documents
Culture Documents
APC-4500VE SECCION 2.1. Sintonia de Lazos de Control
APC-4500VE SECCION 2.1. Sintonia de Lazos de Control
Tabla de contenido
Ricardo Caballero 1
Febrero 01, 2003
PROPRIETARY AND CONFIDENTIAL ©HONEYWELL HI-SPEC SOLUTIONS 2001
CONTROL DE PROCESO Hi-Spec Solutions
This section provides insight into the limitations of widely used PID tuning
algorithms by tracking the development path, which lead to profit PID (RPID).
What is Profit PID (RPID)? This section provides an overview of the profit PID
(RPID) software package and provides answers to general question about profit PID
(RPID).
Robust Control Concepts For many engineers, robust control is a new concept.
This section provides a brief technical overview of robust control theory as it applies
to profit PID (RPID).
Many process model based techniques (Ziegler- Nichol, Cohen- Coon, IMC, etc)
have been developed to provide tuning parameters for conventional PID controllers.
Unfortunately, normal change in plant operations such as increasing process
throughput or running of different feedstock’s can widely affect PID controller
performance. None of the previously mentioned techniques attempts to find tuning
parameters that provide consistent control performance over a range of expected
dynamics.
Honeywell addresses this problem with its Profit PID (RPID) software package.
Recent advances in control theory have been incorporated into this PID controller-
tuning package that optimizes PID controller tuning parameters for a specified range
of process dynamics.
2
Ricardo Caballero 2
Febrero 01, 2003
PROPRIETARY AND CONFIDENTIAL ©HONEYWELL HI-SPEC SOLUTIONS 2003
CONTROL DE PROCESO Hi-Spec Solutions
Robust Control
Robust Control theory was developed to allow for uncertainty in dynamic process
models. By considering model uncertainty, PID controller can be tuned to perform
consistently across a range of dynamics.
Profit PID (RPID) design software is based on a proprietary min-max algorithm that
calculates the best tuning parameters for the worst-case process dynamics. By finding
the best worst-case scenario, the design software finds a control solution that is
insensitive to model error and yet still provides good control performance. The net
effect of using this solution is that the average performance across the range of
dynamics is improved.
PID controllers have been around for a long, long time. The basic PID algorithm
includes three user specified tuning parameters (K, T1, T2) whose values
determine how the controller will behave. Historically, these parameters have
been determined either by trial and error or through approaches that require
knowledge of the process. Process based approaches to these parameters Include
the familiar:
Ziegler - Nichols, Cohen-Coon, Gain and Phase Margin Methods, and Optimal
tuning Methods (IAE, ISE, ITAE, ITSE).
Although many of these techniques have provided satisfactory control in the past the
performance of controllers tuned by these methods usually degrades as process
conditions change. Therefore a process engineer is required to periodically monitor
PID controller performance. If control performance has deteriorated the engineer is
forced to retune the controller.
Controller performance can degrade for a variety of reasons. However, the most
common cause of these phenomena is that system dynamics have change (for
example different feedstock and equipment fouling). Recall that the methods
mentioned above (excluding Ziegler-Nichols) require a nominal model of the process
to enable calculation of tuning parameters. Therefore, the performance of controller
tuned by these methods is directly related to accuracy of the process model which was
3
Ricardo Caballero 3
Febrero 01, 2003
PROPRIETARY AND CONFIDENTIAL ©HONEYWELL HI-SPEC SOLUTIONS 2003
CONTROL DE PROCESO Hi-Spec Solutions
This is not to say that these other tuning methods do not allow for some model error.
In fact, all of the above mentioned techniques do account for discrepancies in the
dynamics model. The problem with these tuning methods is that they assume the
model error is effectively fixed and do not attempt to best tune the controller for a
range of dynamics. Therefore, commissioning of these controllers usually involves
detuning the controllers during the implementation phase to offset model uncertainty.
During the 1970s and 1980s, academic and industrial research organizations
developed many methods for dealing with uncertainty in dynamic process models.
Many of the techniques, which include H and m-synthesis, have already been
successfully applied within the aerospace industry. However, these techniques were
not readily applied to the process industries.
The major obstacle blocking application of the techniques within the process industry
was the lack of an easy-to-use controller design package, which incorporated
uncertainty-based controller design techniques. Without an easy-to-use controller
design package, most of these methods require an extensive understanding of
frequency domain control theory, a theory that is not familiar to most process
engineers.
To introduce these new control techniques into the process industries, Honeywell
developed Profit PID (RPID). Profit PID (RPID) is an easy-to-use controller-tuning
package that incorporates modern uncertainty-based control techniques to enhance
the performance of conventional PID control loops.
Profit PID (RPID) is a controller design package which is available as part of the
APC development Environment. The design software is used to calculate optimal PID
tuning parameters based on engineering inputs. The calculated tuning parameters are
then implemented on Honeywell distributed control systems to create Profit PID
(RPID) controllers.
When properly used, Profit PID (RPID) controllers are able to provide consistent
control performance over the range of identified process dynamics. The net result is
improved regulatory control.
4
Ricardo Caballero 4
Febrero 01, 2003
PROPRIETARY AND CONFIDENTIAL ©HONEYWELL HI-SPEC SOLUTIONS 2003
CONTROL DE PROCESO Hi-Spec Solutions
Profit PID is not a solution to all control problems. (Remember that a Profit PID
controller is simply a PID controller that has been tuned to operate over a range of
dynamics.) If tuning parameters cannot be found for simple PID controller under
current dynamics. Then the Profit PID design software will not have any better luck
finding controllers tuning parameters for the associated range of dynamics. In lieu of
valve or other process related problem, a multivariable control system that account
for disturbances and controller interactions may be more appropriate.
Profit PID provides improved control over a range of dynamics; it does not provide
improved control for all dynamics. Systems where process dynamics change
significantly may be inappropriate for Profit PID implementations. Examples include
systems where the process gain changes in sign or the value of the process gain varies
by several orders of magnitude. In situations like this, engineering judgment must be
used prior to implementation of Profit PID.
Although Profit PID can predictive controller, it is not a multivariable predictive
controller. When control response to disturbance variables, manipulation of multiple
variables, or constraint control is required, the use of multivariable predictive
controller such as Honeywell’s RMPCT should be considered.
Profit PID should be used wherever improved PID control has the potential to
increase process yield, improve product quality, reduce equipment maintenance, or
minimize safety hazards. Profit PID generates a PID controller that is best suited to
control the process over a range of identified dynamics. Additionally, Profit PID is
ideal for tuning loops underneath a multi-variable predictive controller, since Profit
PID controllers maximize closed loop performance consistency.
Process dynamics move within a range for a variety of reasons. For example, the
following list provides common plant occurrences that can change the way a process
will respond to PID control:
settling time, or dead time, the calculated solutions is simply detuned to compensate
for the uncertainty. Unlike Profit PID, these packages do not find the most responsive
controller parameters for all models within the uncertainty range.
Profit PID determines the optimal controller tuning for the specified range of process
dynamics.
Profit PID has been tested in process plant applications with exceptional result. In one
study, various PID tuning methods were applied to a stripper temperature controller.
149
148
147
146
145
140
0 1000 2000 3000 4000 5000 6000 7000 8000 9000 10000
Minutes
Stripper Bottom PCT Control
Honeywell
RMPC Concepts Course 2/10/2003 Page 3-28
6
Ricardo Caballero 6
Febrero 01, 2003
PROPRIETARY AND CONFIDENTIAL ©HONEYWELL HI-SPEC SOLUTIONS 2003
CONTROL DE PROCESO Hi-Spec Solutions
As this table shows, Profit PID, performs significantly better than the other tuning
methods. Since process economic performance is often closely linked to process
variability, this illustrates that the use Profit PID can lead to superior control
performance and higher process profitability.
In more interesting study, a detailed comparison between IMC and Profit PID tuning
methods was performance on a fluidized bed reactor system.
Fluidized-Bed Reactor
PC
Gas
Solids
Cooling
Feed
LC Water
TC
Gas
Feed
Product
Honeywell
RMPC Concepts Course 2/11/2003 Page 3-30
7
Ricardo Caballero 7
Febrero 01, 2003
PROPRIETARY AND CONFIDENTIAL ©HONEYWELL HI-SPEC SOLUTIONS 2003
CONTROL DE PROCESO Hi-Spec Solutions
In this case, both the pressure and temperature controller had been previously tuned
using Internal Model Control (IMC) techniques. Then, Profit PID was implemented.
The Following figures illustrate the result of this study.
8
Ricardo Caballero 8
Febrero 01, 2003
PROPRIETARY AND CONFIDENTIAL ©HONEYWELL HI-SPEC SOLUTIONS 2003
CONTROL DE PROCESO Hi-Spec Solutions
Sample Result
In this case, the implementation of Profit PID decreased the standard deviations of
both the fluidized bed pressure and temperature by a factor of about six.
9
Ricardo Caballero 9
Febrero 01, 2003
PROPRIETARY AND CONFIDENTIAL ©HONEYWELL HI-SPEC SOLUTIONS 2003
CONTROL DE PROCESO Hi-Spec Solutions
In the field of process control, robustness may be loosely defined as the ability of a
control system to operate consistently under a variety of operating conditions.
Therefore, a Profit PID controller should to able to provide reliable control
performance even when process condition may vary or errors may be present in the
original process model.
Most PID controllers in use today have been tuned by process model based
techniques. When using these techniques, an engineer must obtain knowledge about
how a given process responds to specified variable change. This knowledge is then
used to formulate empirical models, which can be used to design the control system.
Unfortunately, there is always a limit to the amount of knowledge one can acquire
about a process. Therefore, there is always an inherent uncertainty with regard to the
final controller design. The following example will illustrate this point.
Robustness an Example
From this information, the engineer concludes that seasonal temperature changes are
affecting the process dynamics of the tower. To design a PID controller that will
control work well during both seasons, he decides to uses average or nominal process
model.
10
Ricardo Caballero 10
Febrero 01, 2003
PROPRIETARY AND CONFIDENTIAL ©HONEYWELL HI-SPEC SOLUTIONS 2003
CONTROL DE PROCESO Hi-Spec Solutions
SISO Example
PV
Reflux
Vapor
t
Drums MV
Coke
Coking Heater t
Honeywell
RMPC Concepts Course 2/11/2003 Page 3-5
The use of a nominal model is not bad choice since this is the best guess of the actual
process model under the given circumstances. Furthermore, the nominal PID
controller design, which is based on the nominal model, can always be adjusted to
speed up or slow down control performance. In this case (ol) represent the open-loop
time constant and (cl) represent the closed-loop time constant.
As the following figure illustrates, the nominal controller design can tuned to be as
fast or as slow as required.
11
Ricardo Caballero 11
Febrero 01, 2003
PROPRIETARY AND CONFIDENTIAL ©HONEYWELL HI-SPEC SOLUTIONS 2003
CONTROL DE PROCESO Hi-Spec Solutions
1. 2
1
A fast controller:
2.5 times faster ( c= )
0. 8
0. 6
A sluggish controller:
4 times slower (c = 48)
0. 4
0. 2
0
0 20 40 60 80 100 120 140 160 180 200
So far, the engineer has done nothing that seems intuitively wrong. However, there is
a subtle problem. The engineer assumed that the process models were accurate. This
is usually not the case. In fact, most identified process models are actually off by 20-
40%. Model uncertainties of the magnitude can often cause problem with nominal
controller designs. When model mismatch occurs, nominal PID controller designs can
yield undesirable results.
For example, notice the increase in oscillatory behavior as model mismatch (20-50%
uncertainty) is introduced into the control system. (See Figures)
12
Ricardo Caballero 12
Febrero 01, 2003
PROPRIETARY AND CONFIDENTIAL ©HONEYWELL HI-SPEC SOLUTIONS 2003
CONTROL DE PROCESO Hi-Spec Solutions
1.2
0.8
0.6
4 times slower (c = 48)
0.4
0.2
0
0 20 40 60 80 100 120 140 160 180 200
Honeywell
RMPC Concepts Course 2/11/2003 Page 3-11
What is actually desired to control this system is a nominal PID controller design that
is insensitive to model mismatch. This is found using robust design techniques. In
order words, robust design minimizes the bandwidth of the nominal PID controller
design to yield a continuum of good performance across the expected dynamics
range.
When this exercise is carried out, a robust PID controller design emerges. The
following figure illustrates how the robust PID controller performs when model
uncertainty is introduced. (See Figures)
13
Ricardo Caballero 13
Febrero 01, 2003
PROPRIETARY AND CONFIDENTIAL ©HONEYWELL HI-SPEC SOLUTIONS 2003
CONTROL DE PROCESO Hi-Spec Solutions
1.2
2.4 times faster ( c=
1
0.6
0.4
0.2
0
0 20 40 60 80 100 120 140 160 180 200
Honeywell
RMPC Concepts Course 2/11/2003 Page 3-14
As you can see, the robust PID controller design performs well despite the
uncertainties contained within the process models. It also maintains good control as
performance of the controller is pushed from a sluggish controller (cl = 48) to an
aggressive controller (cl=5).
A robust PID controller design is inherently stable and its performance can always be
adjusted to yield smooth control responses. However, there are some trade-offs that
must be considered in the design of robust PID controller.
14
Ricardo Caballero 14
Febrero 01, 2003
PROPRIETARY AND CONFIDENTIAL ©HONEYWELL HI-SPEC SOLUTIONS 2003
CONTROL DE PROCESO Hi-Spec Solutions
Profit PID uses ISE (Integral Squared Error) and weighted two-norm movement
penalty as its control performance criteria to find an optimal PID controller design.
The design process begins with the standard PID controller loop and optimal first
order lag:
Setpoint e u K
y
PID Controller
Kc I D 2 s2 + 2 s +1
Controller Process
Honeywell
RMPC Concepts Course 2/11/2003 Page 3-15
Note: It is not required for Profit PID to be effective. However, for complicated
dynamics (Dead time and inverse response), it can significantly improve controller
performance.
The first step in the robust design is to specify the identify process model in the form
of a laplace model. This is accomplished by using the APC identifier, which is
provided as part of the Profit PID software. The Laplace model can be either based on
the data or manually entered. Uncertainty factors are then used to represent either
process dynamics range or modeling errors.
The robust design software calculates optimal tuning coefficients (Kc, I, d, f,)
through the use of a proprietary min-max algorithm. The algorithm operates on a
number of different PID equation forms and essentially calculates tuning coefficients
that yield the best control performance for the worst-case process dynamics.
This solution is more than just doing the best on the worst case. This solution has the
net affect of improving the overall controller performance within the specified
15
Ricardo Caballero 15
Febrero 01, 2003
PROPRIETARY AND CONFIDENTIAL ©HONEYWELL HI-SPEC SOLUTIONS 2003
CONTROL DE PROCESO Hi-Spec Solutions
2.1 Overview
This section explains how to build a profit PID controller from the multiple-input
single-output model file.
Several files are associated with a Profit PID application. A model file must be
available to start the design. A profit PID model file will have a .pid extension. In
addition to the model file, four other files are associated with Profit PID.
A .soc file contains the tuning constants, operating limits, set point and configuration
options. This file is automatically generated by the build>RPID function. It also
provides the user’s view to the controller(s) when running a simulation.
A .soc file contains the non-model information required to simulate a process: starting
value, biases, load disturbance, noise characteristics, etc.
An .xp file contains the model information from the .pid file in a form that can be
used by the windows based simulation.
An .eb file contains the model information from the .pid file in a form that can be
used by an LCN simulation.
16
Ricardo Caballero 16
Febrero 01, 2003
PROPRIETARY AND CONFIDENTIAL ©HONEYWELL HI-SPEC SOLUTIONS 2003
CONTROL DE PROCESO Hi-Spec Solutions
A robust PID controller can be designed with a faster or slower performance ratio by
adjusting the optional Ratio parameter. For example, decreasing the performance ratio
will increase the speed of response while increasing the performance will decrease the
speed of response. The default value of the performance ratio is 1. Currently, the
design is limited to values >0 and <= 10. Engineering judgment must be used when
adjusting this parameter because an unstable controller may result. The following
table lists practical values for both uncertainty and performance ratios for use in most
Profit PID controller designs.
17
Ricardo Caballero 17
Febrero 01, 2003
PROPRIETARY AND CONFIDENTIAL ©HONEYWELL HI-SPEC SOLUTIONS 2003
CONTROL DE PROCESO Hi-Spec Solutions
Generate Defaults
Choose [Generate default] to use information residing this in the active .pid file and
to ensure equilibrium initial conditions.
Choose [Read from existing file (.soc)] to use any case where biases or controller
constants have been modified. In the latter case it is necessary to make sure that the
.soc file containing the modified parameters be saved. By reading from an existing
.soc file, the build procedure will use existing control settings as the starting point for
the min-max calculations. It is advised to use this approach for large uncertainty
especially for the interactive equation types
Range and control limit are initialized based on the data contained in the .pid
document. These values are taken as the max and min values of the observed data.
The parameters are:
The PVEUHI and PVEULO values are used to scale the input to the PID controller,
and the CVEUHI and CVEULO values are used to scale the out put from the PID
controller. While these values are initialized based on observed data, they would be
manually entered based on those values residing in the target ECS. The input values
18
Ricardo Caballero 18
Febrero 01, 2003
PROPRIETARY AND CONFIDENTIAL ©HONEYWELL HI-SPEC SOLUTIONS 2003
CONTROL DE PROCESO Hi-Spec Solutions
should match the high and lo range of the PV on the target controller. The output
values should match the high and low range of the PV on the downstream controller
if the data test were conducted by stepping the downstream controller set point. The
output values should be 100 and 0 for high and low respectively, if the data tests were
conducted by stepping the PID controller output.
Uncertainty
The uncertainty parameter accounts for modeling error in the specified dynamic
parameters. If no uncertainty parameter is specified, the software assumes perfect
process models with an uncertainty parameter of 0 (%). However, if the overall
uncertainty parameter was set to x (%), each of the individual term (gain, delay, zeros
and poles) would have a dynamic range with high and low bounds defined as in the
1 ni
<= <= 1 +0.01x
1 +0.01x
robust controller design.
Experience has shown that in many cases the model uncertainty is in the range of 20-
60%. Currently the uncertainty is a quantity that must be specified by the user. In
future releases, this quantity will be calculated automatically based on the data if it
exists.
Design Type
The design type option is used to define the manner in which the local-loop controller
is to be used.
Ramp This disturbance type can be use effectively when the process
experiences ramp-like conditions in typical operation and is a common
choice with integrating process model.
Step&Ramp Design for a combination of both the step and ramp disturbance
types.
Grid Point Number of grid point (for each uncertain parameter) three is typical
sufficient five causes solution to increase.
Move Suppression Penalty parameter used in the objective function to trade off
controller movement vs. error reduction.
Noise Penalty parameter used to attenuate controller movement for fast processes
with significant noise (normalized value).
Use Minmax Calculation When unchecked, the solution returns the standard IMC
design for a PID controller.
Use Output Display During Search When checked, the message window displays
critical information as the search proceeds.
Error Type This Parameter defines the type of error contribution of the process
output use in the objective function of the min-max calculations. The objective
function uses a combination of the process error and a penalty due to actuator
movement. The two types of process error are
Mini-max Criteria
Termination Criterion for x. Change in each element in the solution vector must be
less that this value.
20
Ricardo Caballero 20
Febrero 01, 2003
PROPRIETARY AND CONFIDENTIAL ©HONEYWELL HI-SPEC SOLUTIONS 2003
CONTROL DE PROCESO Hi-Spec Solutions
Termination Criterion for f. Change in the objective function must be less than this
value
Filter Type The filter is the only controller parameter that is searched in the discrete
domain. Values are theoretically constrained to be between –1 and +1. Since all
control parameters are finally presented in the continuous domain, the filter must be
mapped to the continuous domain. There is no mapping for values less than zero.
Selecting the continuo option constrain the search on this variable to values between
0 and 1. Final values are display in minutes. Selecting the discrete option allows the
search to permit a full range of value but prevent the discrete transformation. In this
instance, final values are be discrete time filter coefficient appropriate only for the
select design execution interval.
21
Ricardo Caballero 21
Febrero 01, 2003
PROPRIETARY AND CONFIDENTIAL ©HONEYWELL HI-SPEC SOLUTIONS 2003