Sei sulla pagina 1di 57

Control


Control Station
 Analysis,
Software for Process Control Software For Process
Tuning andControl
Training

Station Analysis, Tuning & Training

Software For Process Control


Analysis, Tuning & Training

Hands-on Workshop Series


A Companion to
Practical Process Control using Control Station 3.7

Douglas J. Cooper

Hands-on Workshop Series by Douglas J. Cooper


Copyright  2004 by Control Station LLC
All Rights Reserved
Table Of Contents

Page

Workshop 1: Exploring Dynamics of the Gravity Drained Tanks 2

Workshop 2: P-Only Control of Tank Level 5

Workshop 3: The Hazard of Tuning PI Controllers by Trial and Error 8

Workshop 4: PI Control of Heat Exchanger Temperature 10

Workshop 5: PI Control of Distillation Top Composition 14

Workshop 6: PID Control of Heat Exchanger Temperature 16

Workshop 7: PID with Filter and Control of the Multi-Tank Process 21

Workshop 8: Adaptive PI Control of Nonlinear Processes 24

Workshop 9: Modeling and Simulation of Single Loop Processes 27

Workshop 10: Cascade Control of the Jacketed Reactor 31

Workshop 11: Feed Forward Control of the Jacketed Reactor 34

Workshop 12: Advanced Feed Forward Control of the Heat Exchanger 38

Workshop 13: Multivariable Decouplers and Distillation Control 42

Workshop 14: Modeling and Decoupling of Multi-Loop Processes 47

Workshop 15: Dead Time Compensation Using the Smith Predictor 51

Workshop 16: Dynamic Matrix Control (DMC) of the Gravity Drained Tanks 55

1
Hands-on Workshop Series by Douglas J. Cooper
Copyright  2004 by Control Station LLC
All Rights Reserved
Workshop 1: Exploring Dynamics of the Gravity Drained Tanks
Objective: To generate open loop step test data and learn how to describe the observed dynamic
process behavior with a first order plus dead time (FOPDT) model. Also, to learn about the
nonlinear nature of processes.
Reference: “Practical Process Control” Chapters 1-3

1) We always begin a controller design by analyzing the dynamic behavior of the process, or how the
measured process variable responds to changes in the controller output signal. Here we analyze
the dynamic behavior of the gravity drained tanks process.

Click the Case Studies button on the main Control Station screen. From the pop-up list of
processes, click "Gravity Drained Tanks" to start the simulation. Study the graphic and observe
how a change in the controller output signal will cause the valve position to change, which
changes the flow rate of liquid into the top tank, which ultimately changes the level in the lower
tank. Liquid level in the lower tank is our measured process variable.

2) The formal approach for analyzing the behavior of a process is to fit a dynamic model to process
data. Good data starts with the process at steady state. The controller output signal is then moved
far enough and fast enough so that it forces a clear response in the measured process variable.
For a step test, the controller output signal is stepped from one value to another and the response
of the measured process variable signal is recorded from initial steady state to final steady state.
At the upper right of the draining tanks graphic on your screen, locate the white number box
below the Controller Output label. The number displayed should be 70, corresponding to a 70%
output signal from the controller.
The most convenient way to step the controller output signal is to click once on this white number
box. Alternatively, you may double click (click twice in rapid succession) the box.
If you clicked once, the box will turn blue. For the first step test, type 80 into the box and press Enter.
This will cause the controller output to step from its current value of 70% up to the new value of
80%, increasing the flow of liquid into the top tank and causing liquid level to rise.
If you double clicked, a pop-up menu gives you the option of performing a step, oscillate, ramp or
PRBS experiment. With the Step option active, type 80 as the Controller Output Value and then
click Done at the bottom of the pop-up menu.

3) Watch as the process responds. When the measured process variable (liquid level) reaches its new
steady state, click the “Pause” icon on the tool bar above the strip charts to stop the simulation.

To view a full screen plot, click the printer icon (with floating label "View and Print Plot") on the
tool bar. Alternatively, click File on the menu list and choose View and Print Plot. With the plot
displayed, click the Plot Options icon on the plot tool bar and use the various options to create a
plot best suited for graphical analysis.
(Aside: When reading graphs, your results will be more precise if the plot axes are adjusted so the
data of interest fills the screen. Also, data points can be read most accurately if tick marks are
positioned frequently across the grid.)

2
Hands-on Workshop Series by Douglas J. Cooper
Copyright  2004 by Control Station LLC
All Rights Reserved
Workshop 1: Dynamics of the Gravity Drained Tanks (cont.)
4) Using the methodology described in the Practical Process Control book, fit a first order plus dead
time (FOPDT) dynamic model to the process response plot. That is, compute from the response plot
the FOPDT model parameters: steady state process gain, KP, overall time constant, P, and apparent
dead time, P. The work sheet on page 4 will help you in these calculations. Record your results:

Process gain, KP = Time constant, P = Dead time, P =

5) Enter a controller output of 50% and let the measured level reach steady state. Repeat the above
procedure for a second step in the controller output from 50 up to 60% and record your results:

Process gain, KP = Time constant, P = Dead time, P =

6) The two steps in the controller output (7080% and 5060%) were both of the same size. Are the
model parameters the same for these two steps? How/why are they different?

7) Set the controller output to 20% and let the measured level in the lower tank steady out (you should
notice a low alarm condition). Step the controller output from 20% up to 40%. When the measured
level reaches a steady value, make a next step up to 60%. Continue making steps, next up to 80%
and finally up to 100%. Let the process complete its response between each step.

View a plot of this response over the whole range of operation. The change in the controller output is
constant at each step. If the process is linear, then the measured liquid level response will be the
same for each step. If the process is nonlinear, then the KP, P and P will change as operating level
changes. Is the gravity drained tanks a linear or nonlinear process?

(optional)

8) Return to the main Control Station screen, click the Case Studies button and start the Heat
Exchanger process. Take a moment to study the process graphic. The controller output adjusts the
flow rate of cooling liquid on the shell side. The measured process variable is stream temperature
exiting the exchanger on the tube side.

Step the controller output signal from 39% up to 50%. This causes an increase in the cooling flow
rate and hence a decrease in the measured exit stream temperature. When the process response is
complete, click the “Pause” icon on the tool bar above the strip charts to pause the simulation.

Using the skills you have learned, fit a FOPDT model to the data and record the model parameters:

Process gain, KP = Time constant, P = Dead time, P =

What is a notable difference in the process gain for the heat exchanger compared to that of the
gravity drained tanks?

9) Perform the experiments necessary to determine if the heat exchanger is a linear or nonlinear
process. What experiments did you perform and what do you conclude?

3
Hands-on Workshop Series by Douglas J. Cooper
Copyright  2004 by Control Station LLC
All Rights Reserved
Workshop 1: Dynamics of the Gravity Drained Tanks (cont.)
This work sheet follows the terminology and procedures detailed in the Practical Process Control book.
In this exercise, u(t) = controller output and y(t) = measured process variable (level in the lower tank)

Controller output step Controller output step


from 70%  80% from 50%  60%

u1 = 70%  y1 = u1 = 50%  y1 =

u2 = 80%  y2 = u2 = 60%  y2 =

u = y = u = y =

KP = (y/u) = KP = (y/u) =

tYstart = tYstart =

y63.2 = y1 + 0.632(y) = y63.2 = y1 + 0.632(y) =

t63.2 = t63.2 =

P = t63.2 - tYstart = P = t63.2 - tYstart =

tUstep = tUstep =

P = tYstart - tUstep = P = tYstart - tUstep =

4
Hands-on Workshop Series by Douglas J. Cooper
Copyright  2004 by Control Station LLC
All Rights Reserved
Workshop 2: P-Only Control of Tank Level
Objective: To design and test a P-Only controller for tracking changes in set point and rejecting
disturbances. Also, to explore how controller gain impacts P-Only performance.
Reference: “Practical Process Control” Chapters 4-5

1) All controllers should be designed for a specific level of operation, which includes specifying the
expected or design value for the set point, and typical or baseline values for the important process
disturbances. Here we design a controller for the gravity drained tanks process. Click the Case
Studies button on the main Control Station screen and from the pop-up list of processes, click
"Gravity Drained Tanks" to start the tanks simulation.

For this P-Only level control design, consider the case where the design level of operation is a
measured liquid level in the lower tank of 3.3 m. The flow rate of the disturbance stream out of
the bottom of the lower tank is expected to be about 2.0 L/min during typical operation. When
your controller is put in automatic, the set point should be at the design level of the measured
process variable:
ysetpoint =

A P-Only controller design requires determination of the controller bias, or the value of the
controller output that causes the measured liquid level in the lower tank to steady at the design
value of 3.3 m when the disturbance flow rate is at its design value of 2.0 L/min. Change the
controller output, search for this value and record your result:

ubias =

2) The next step, similar to workshop 1, is to describe the dynamic behavior of this process with a
first order plus dead time (FOPDT) dynamic model. That is, we need to compute the steady state
process gain, KP, overall time constant, P, and apparent dead time, P, for the process at the design
level of operation.

In step 1 above it was (hopefully) found that a controller output of 64% causes the measured level in
the lower tank to steady at the design value of 3.3 m. Because this process is nonlinear, best practice
is to perform an experiment that includes data both a little below and a little above this design level.

Set the controller output to 60% and wait until the process steadies. With this as a starting point,
step the controller output up to 68%. The result is a data set that on average describes the design
level of 3.3 m. When the process response is complete, click the “Pause” icon on the tool bar above
the strip charts to pause the simulation.

View a full screen plot of the response by clicking the printer icon (with floating label "View and
Print Plot") on the tool bar. Alternatively, click File on the menu list and choose View and Print Plot.
With the plot displayed, click the Plot Options icon on the plot tool bar and use the various options
to create a plot best suited for graphical analysis.

Compute and record the FOPDT process model parameters:

Process gain, KP = Time constant, P = Dead time, P =

5
Hands-on Workshop Series by Douglas J. Cooper
Copyright  2004 by Control Station LLC
All Rights Reserved
Workshop 2: P-Only Control of Tank Level (cont.)
3) One important reason for fitting a FOPDT model to dynamic process data is because the resulting
model parameters can be used in correlations to estimate controller tuning values. One popular
controller design correlation is the ITAE (Integral Time Weighted Absolute Error) for Set Point
Tracking, which in the form below yields an initial estimate of the P-Only controller gain, KC,
when set point tracking is the primary objective.

Compute and record P-Only controller gain. Remember it has a sign (plus or minus) and units.

ITAE for Set Point Tracking tuning relation: KC =


0.2021
θ P /τ P 1.2188
KP

KC =

4) Now validate your completed controller design. Click the level controller icon (the LC in the
white circle on the gravity drained tanks graphic). When the controller design menu opens, note
that "Manual Mode" is displayed at the top of the menu indicating that no controller is in service
(the control loop is open). Click "Manual Mode" and from the selection of controllers, click "PID"
to enable a proportional-integral-derivative controller.

The default configuration is a PI controller. Thus, the proportional and integral modes are “On”
while the derivative mode is “Off.” Implement a P-Only controller by turning the integral mode
off. To do this, click “ON: Integral with Anti-Reset Windup” (located half way down the
controller design menu) and choose “off: Integral Mode.” Tuning values entered for the integral
and derivative terms have no impact on controller computations when these modes are off.

With this P-Only configuration, click "Set Point," then "Bias" and finally "Controller Gain" and
enter the values you computed for these selections. When finished, click "Done" at the bottom of
the controller design menu to put your controller in automatic. If you have performed the steps
correctly, the strip chart should show the white measured level signal tracing (with some noise or
random error) over the straight yellow set point line. There should be no sustained offset.

5) Explore the performance of this controller in tracking set point steps. The set point is most
conveniently changed by clicking on the "Set Point" box on the process graphic. Step the set point
from 3.3 m up to 4.3 m and back to 3.3 m. Observe the performance of your controller as it
attempts to track the changes. Do you notice offset when the set point is different from the design
value?

Now explore how the magnitude of KC impacts offset and the oscillatory nature of the controller
response. Click the level controller (the LC in the circle on the graphic) and in the controller
design menu, change your controller gain. If you think your controller is too aggressive because of
large response oscillations, cut your KC in half. If your controller is rather slow or sluggish in its
response, double your KC.

Now make the same set point step as before. Keep changing your controller gain to search for a
performance that balances offset against oscillatory behavior. Call this your "best" KC tuning
value.

6
Hands-on Workshop Series by Douglas J. Cooper
Copyright  2004 by Control Station LLC
All Rights Reserved
Workshop 2: P-Only Control of Tank Level (cont.)
6) Using your "best" KC, put the set point at 3.3 m and let the measured liquid level steady out. Now,
step the set point from 3.3 m up to 4.3 m, then up to 5.3 m, then 6.3 m and finally 7.3 m. Let the
measured level steady out after each set point step. You may need to use the "Change Strip Chart
History" icon on the tool bar to increase the strip chart display to 80 minutes.

How does offset and oscillatory nature of the response change as the set point moves further from
the design value? Does the controller output ever become constrained (hit a maximum or
minimum limit)? If so, how does this impact the set point response?

7) Study disturbance rejection performance of a P-Only controller. Using your "best" KC , put the set
point back at 3.3 m and let the measured level steady out. Use the "Change Strip Chart History"
icon from the tool bar and return the strip chart display to 30 minutes. Wait for previous changes
to scroll off of the moving strip charts and when they show essentially straight lines, expand the
plots using the "Rescale Vertical Axis" icon from the tool bar.

Step the disturbance flow rate from 2.0 L/min up to 5 L/min by clicking on the Pumped Flow
Disturbance box located to the lower left of the process graphic. The set point is at its design value,
so why does the controller now show offset? Step the disturbance back and forth between 2.0 L/min
and 5 L/min using different values of KC. How does controller gain impact offset and oscillatory
behavior for P-Only disturbance rejection? Is your best KC for disturbance rejection the same as that
for set point tracking?

(optional)

8) Design and test a P-Only controller for the Heat Exchanger process. The design level of operation is
a measured exit stream temperature of 144 oC. The warm oil flow disturbance stream is expected
to be about 10 L/min during typical operation. Test the controller for set point tracking from 144
o
C up to 150 oC and back again. Also test the controller in rejecting warm oil flow stream
disturbance changes from 10 L/min up to 15 L/min and back again.

7
Hands-on Workshop Series by Douglas J. Cooper
Copyright  2004 by Control Station LLC
All Rights Reserved
Workshop 3: The Hazard of Tuning PI Controllers by Trial and Error
Objective: To build intuition about PI controller performance and tuning when the control objective is
disturbance rejection. Also, to experience the drawbacks of trial and error tuning.
Reference: “Practical Process Control” Chapters 6-9

1) Here we explore the tuning of a PI controller by trial and error when disturbance rejection is the
control objective. The process will be an ideal linear simulation designed using Custom Process.
Start by clicking the Custom Process button on Control Station’s main screen and then choose
Single Loop Process from the list.

When the simulation starts, notice that the graphic to the right of the scrolling plots is comprised
of a Process button, Disturb(ance) button and Controller button (the C in the white circle).

To create the simulation, click the Process button on the graphic. This opens a “Construct Process
and Disturbance Models” form. First specify the controller output to measured process variable
dynamic behavior. Click the Process Model tab (it should already be active if you have not done
any exploring) and select from the list of models available, the:
- Overdamped Linear Model
- Self Regulating (Stable) Process.
Enter the process gain, three time constants and dead time that define the controller output to
measured process variable dynamic behavior of this overdamped linear third order plus dead time
model:
Process Gain, KP = 1.0
First Time Constant, P1 = 10.0
Second Time Constant, P2 = 1.0
Third Time Constant, P3 = 1.0
Lead Time, PL = 0
Dead Time, P = 1.0

Next specify the behavior of the disturbance to measured process variable dynamics. Click the
Disturbance tab at the top of the form, ensure that Overdamped Linear Model is selected, and
enter:
Disturbance Gain, KD = 1.2
First Time Constant, D1 = 12.0
Second Time Constant, D2 = 1.2
Third Time Constant, D3 = 1.2
Lead Time, DL = 0
Dead Time, D = 1.2

Click Done at the bottom of the form to start the simulation.

2) Controller output and disturbance variable changes differ in their impact on the measured process
variable because the dynamic models defined above are different. Verify this by stepping the
controller output from its default value of 50% up to 60% and when the response is substantially
complete, step it back to 50%. Next, step the disturbance variable from its default value of 50% up
to 60% and when the response is substantially complete, step it back to 50%.

8
Hands-on Workshop Series by Douglas J. Cooper
Copyright  2004 by Control Station LLC
All Rights Reserved
Workshop 3: Tuning PI Controllers by Trial and Error (cont.)
Compare the two responses side by side (you may need to use the Change Strip Chart History icon
on the toolbar to see both responses). Does a step in the disturbance variable show a larger but
slightly slower response associated with a larger gain and longer overall time constant?

3) Click the Controller icon on the graphic and select PID from the controller design menu. Specify a
very conservative (sluggish) PI controller by entering the tuning values:
Set Point = 50
Controller Gain = 1.0
Reset Time = 50

Choose the Integral with Anti-Reset Windup mode and be sure you are using a PI controller by
leaving Derivative Mode off. Click Done to put the controller in automatic.

4) Trial and error tuning can use up significant production time and create expensive off-spec
product. To track how inefficient a trial and error approach can be, zero the simulation clock by
clicking on the Reset Process Clock to Zero on the upper left of the button tool bar. From this
point forward, do not reset the clock or pause process execution until instructed to do so.

5) Start tuning for disturbance rejection by considering the current performance of the controller.
Click the white “Disturbance, D” box on the graphic, step the disturbance variable from 50% up
to 60% and after the response, step it back to 50%.

Now try and improve controller performance by adjusting KC and/or I on the controller design
menu. The PI controller tuning map in Chapter 8 of the Practical Process Control book is for set
point tracking but may assist your thought process as you search for your best tuning values. Test
your new values by again stepping the disturbance from 50% up to 60% and back again. Repeat
this procedure as necessary to arrive at “best” tuning values for disturbance rejection. Use your
own experience and opinion to define “best.”

When you have determined best tuning values for disturbance rejection, pause the simulation and
record them below. Also record the elapsed time displayed on the process clock at the lower left
of the screen.

KC = I = Elapsed Time = minutes

(optional)

6) Repeat the above exercise with the objective of controller tuning for set point tracking. Start with
your best disturbance rejection tuning values above and determine by trial and error whether they
are best when the control objective is tracking set point steps from 50% up to 60% and back
again. Record your results:

KC = I = Elapsed Time = minutes

Is your best tuning for set point tracking the same as your best tuning for disturbance rejection?

9
Hands-on Workshop Series by Douglas J. Cooper
Copyright  2004 by Control Station LLC
All Rights Reserved
Workshop 4: PI Control of Heat Exchanger Temperature
Objective: To learn about the performance of a PI controller and to explore the interaction of the two
tuning parameters on controller performance. Also, to learn how Design Tools can
automate the dynamic modeling and controller design task.
Reference: “Practical Process Control” Chapters 6-9

1) A formal or systematic procedure for designing and tuning a controller is detailed in the “Practical
Process Control” book and will be used in this study. This procedure is important because it is the
one used in the plant, the lab and with Control Station simulations:
(Aside: A summary of the procedure we will follow in this workshop is:
a) Move the process to the design level of operation and when it reaches steady state, generate
and record controller output to measured process variable dynamic data.
b) Fit a FOPDT (first order plus dead time) model to this data. Design Tools is well-suited for this
modeling task, though graphical methods explored in earlier workshops can be used.
c) Use the resulting FOPDT model parameters in a correlation to compute initial PID controller
tuning values.
d) Implement your controller on the actual process and perform final tuning by trial and error until
control objectives are satisfied.)

Begin this study by starting the Heat Exchanger process in Control Station. Consider a case where
the design temperature of the exit stream is 134 oC and where the warm oil disturbance flow rate is
expected to be about 30 L/min during typical operation. Record your design set point here:

ysetpoint =

Your first goal is to move the process to the design level of operation. Click the disturbance box
shown on the graphic and set the warm oil flow rate to its expected value of 30 L/min. Now
search for the value of the controller output (that manipulates the cooling flow rate) that causes
the measured exit temperature to steady at the design value of 134 oC.

When you have moved your process to the design level of operation, wait until all of the dynamics
have scrolled off screen on the data plots (wait until the plots show essentially straight lines) and
expand the plots using the "Rescale Vertical Axis" icon from the tool bar.

2) Just as for the P-Only controller design, we start a PI controller design with experiments to generate
dynamic process data around the design exit stream temperature of 134 oC. Since a controller
output of 42% causes the measured exit temperature to steady at 134 oC, make your changes
starting at this controller output value.
(Aside: Design Tools is not limited to simple step tests. In this study, we make a quick pulse up
followed by a quick pulse down in the controller output. Such a doublet generates data both above
and below our design level of operation, a desirable result when the process has a nonlinear
character. This method also returns the process to the initial design value relatively quickly, which
is a popular notion with operations staff.)

10
Hands-on Workshop Series by Douglas J. Cooper
Copyright  2004 by Control Station LLC
All Rights Reserved
Workshop 4: PI Control of Heat Exchanger Temperature (cont.)
Before starting the experiment, turn file storage on with the "Save Data to File" icon on the tool
bar. You must enter a file name before data storage begins. After entering a file name, a pop-up
menu asks for a data storage rate and starting time. Accept the default values by clicking “OK.”

Storage of process data to file is now in progress. Step the controller output from 42% up to 47%.
When the measured temperature has shown a clear response to this change (you need not wait for
the response to reach steady state), step the controller output from 47% down to 37%. And after a
clear response to this change, step the controller output back to its original value of 42%. As soon
as the final response is complete, use the "Stop Saving Data" icon on the tool bar to turn file
storage off.

3) Now use Design Tools to automatically fit a FOPDT model to this data. First pause execution of
the Heat Exchanger process using the “Pause” icon on the tool bar. Then, use the “Navigate” icon
on the tool bar to open Design Tools.

From the Design Tools main screen, click the "Open Data File" icon on the toolbar. In the Open
File pop up window, click the file name where you saved the process data and then click “OK” to
import your dynamic process data. After the file is read, it will be displayed on your screen with
the Time, Manipulated Variable and Process Variable columns of data already properly labeled. If
you agree that the data labels are correct, click “OK” to accept the column labels.
(Aside: You can use the “Edit” button located at the upper right of this Label Data window and read
the column labels if you are not sure which data each column contains.)

4) You should be back to the Design Tools main screen now. Make sure the "First Order Plus Dead
Time (FOPDT)" model is displayed in the equation box at the bottom of the screen. If not, use the
“Select Model” icon on the tool bar to select it from the library of models.

Click the “Start Fitting” icon on the tool bar to start fitting the FOPDT model to the process data.
If the fit is successful, Design Tools will display a plot. Use your judgment to determine whether
the model reasonably describes the data you collected. Hopefully, the FOPDT model plot line
(shown in yellow) will substantially approximate the response of the measured process variable
data (shown in white). If the model does not describe the data reasonably well, then the tuning
values computed from the model parameters will be suspect.

If the model describes the data to your satisfaction, close the plot screen and note the FOPDT
model parameters displayed to the left on Design Tools’ main screen.

KP = P = P =

5) The main Design Tools screen will also display tuning parameter values for the P-Only, PI and
PID controller.

11
Hands-on Workshop Series by Douglas J. Cooper
Copyright  2004 by Control Station LLC
All Rights Reserved
Workshop 4: PI Control of Heat Exchanger Temperature (cont.)
The PI controller gain, KC, and reset time, I , are computed using the IMC (Internal Model
Control) correlation. The correlation is shown below, but you need not perform the calculation as
Control Station has done it for you. Choose the Standard PID tab on the tuning chart and record
the IMC tuning values for the PI controller:

Standard Tuning:  C is the larger of 0.1  P or 0.8  P

1 p
IMC Tuning: KC = I = P
K p ( p   c )

KC = I =

6) After recording the PI controller tuning values, close Design Tools to return to the Heat
Exchanger process. Click the “Continue” icon on the tool bar to restart the simulation. (Note: if
you had closed the heat exchanger simulation rather than navigating from it, you must again enter
the warm oil disturbance flow rate of 30 L/min and a controller output value of 42% to move the
process back to the design conditions.)

Click the temperature controller (the TC in the circle on the graphic) to open the controller design
menu. Click “Manual Mode” at the top of the menu and from the list of controllers, choose PID.
The default design is for a PI controller with anti-reset windup logic.
(Aside: When the integral mode is on, the controller bias need not be entered. This is because,
when the controller is put in automatic, the bias will be automatically set to the current value of the
controller output. This provides a smooth or bumpless transfer to automatic.)

On the design menu, click "Set Point," then "Controller Gain" and "Reset Time" boxes and enter
the values you recorded above. Because this process has a negative gain, you must select the
“Proportional - Direct Acting” mode and enter KC as a positive number. Leave derivative time as
zero, thus making the PID controller into a PI controller. When you are finished, click "Done" at
the bottom of the menu to put your controller in automatic.

7) We first validate the performance of the controller in tracking set point steps. Start with a set point
temperature of 134 oC and step the set point from 134 oC up to 144 oC, and after the response is
complete, back again to 134 oC (It is most convenient to make step changes in set point by
clicking on the set point box on the heat exchanger graphic).

Ideally, set point tracking performance should balance a quick response with a modest overshoot and
rapid decay. Use the PI controller tuning map toward the end of Chapter 8 of the Practical Process
Control book as a guide to evaluate the set point tracking performance you observe. Adjust controller
KC and I as needed to obtain a “best” tuning.

12
Hands-on Workshop Series by Douglas J. Cooper
Copyright  2004 by Control Station LLC
All Rights Reserved
Workshop 4: PI Control of Heat Exchanger Temperature (cont.)
8) Study how nonlinear behavior impacts controller performance. Using your best KC and I from step 7,
make a set point step from 134 down to 124 oC and back again. How does the nonlinear behavior of
this process impact the set point response performance when compared to step set point increases?

9) Explore how each tuning parameter impacts the performance of a PI controller when tracking set
point steps. Double your KC while leaving I at its best value and step the set point from 134 oC up
to 144 oC and back. Now return KC to its best value, double your I and again step the set point
from 134 oC up to 144 oC and back. Try other tuning combinations to build your intuition on how
these parameters impact controller performance.

10) Investigate the disturbance rejection capability of the PI controller. Return to your best KC and I
values from step 7. Click the warm oil flow rate box and step this disturbance flow rate from 30
L/min up to 40 L/min. After the response is complete, step it back to 30 L/min. Determine through
trial and error the best KC and I for disturbance rejection. Is it the same as your best tuning for set
point tracking?

Now step the disturbance from 30 down to 20 liters/min and back. Does the nonlinear character of
this process impact disturbance rejection performance?

11) Briefly explore the hazards of reset windup (windup occurs from improper programming of the
integral term of the PI algorithm. The result is decreased performance whenever the final control
element reaches a maximum or minimum constraint).

With the set point at 134 oC, the warm oil disturbance flow rate at 30 L/min and the process steady,
step the set point up to 180 oC. This set point is higher than the process can reach. In fact, the valve
hits its low limit as the temperature approaches 170 oC. After the process has steadied at its extreme
value, step the set point back to 134 oC.

On the controller design menu, switch the integral term to Integral with Windup, repeat the set point
step up to 180 oC, and after the response is complete, back to 134 oC. Do you see a difference
between the two algorithms when the set point is stepped back down? Specifically, note the delay
that occurs before the process starts responding to the return set point step for the reset windup
algorithm. The integral term must first "unwind" before the response can begin.

Now compare Integral with Windup against Integral with Anti-Reset Windup logic for set point
steps from 134 oC up to 144 oC and back again. Do you notice that as long as the controller output
signal does not reach a maximum or minimum constraint, controller performance is unaffected by
choice of algorithm?

(optional)

12) Now that you have learned a formal or systematic method of tuning PI controllers, return to
exercise 3 and repeat the workshop using your new skills. Compare the systematic method to the
trial and error approach and make some decisions about the strengths and weaknesses of each
method.

13
Hands-on Workshop Series by Douglas J. Cooper
Copyright  2004 by Control Station LLC
All Rights Reserved
Workshop 5: PI Control of Distillation Top Composition
Objective: To learn about the PI controller and to explore the interaction of the two tuning
parameters on controller performance when disturbance rejection is the control objective.
Also, to learn how Design Tools can automate dynamic modeling and controller design.
Reference: “Practical Process Control” Chapters 6-9

1) A formal or systematic procedure for designing and tuning a controller is detailed in the “Practical
Process Control” book and will be used in this study. This procedure is important because it is the
one used in the plant, the lab and with Control Station simulations:
(Aside: A summary of the procedure we will follow in this workshop is:
a) Move the process to the design level of operation and when it reaches steady state, generate
and record controller output to measured process variable dynamic data.
b) Fit a FOPDT (first order plus dead time) model to this data. Design Tools is well-suited for this
modeling task, though graphical methods explored in earlier workshops can be used.
c) Use the resulting FOPDT model parameters in a correlation to compute initial PID controller
tuning values.
d) Implement your controller on the actual process and perform final tuning by trial and error until
control objectives are satisfied.)

Begin this study by starting the Distillation Column process in Control Station. Here we focus on
the control of the top composition loop in this study. The bottom composition loop will remain in
manual mode at the default startup values throughout this workshop.

Consider a case where the top composition is 90% when the feed flow disturbance is at its typical or
expected value of 600 Kg/min. Record your design set point:

ysetpoint =

Your first goal is to move the process to the design level of operation. Click the disturbance box
shown on the graphic and set the column feed flow rate to its expected value of 600 Kg/min. Now
search for the value of the controller output (that manipulates the reflux flow rate) that causes the
measured top composition to steady at the design value of 90%.

When you have moved your process to the design level of operation, wait until all of the dynamics
have scrolled off screen on the data plots (wait until the plots show essentially straight lines) and
expand the plots using the "Rescale Vertical Axis" icon from the tool bar.

2) Just as in the previous controller designs, we start a PI controller design with experiments to
generate dynamic process data around the design top composition of 90%. Since a controller
output of 45.4% causes the top composition to steady at 90%, make your changes starting at this
controller output value.
(Aside: Design Tools is not limited to simple step tests. In this study, we make a quick pulse up
followed by a quick pulse down in the controller output. Such a doublet generates data both above
and below our design level of operation, a desirable result when the process has a nonlinear
character. This method also returns the process to the initial design value relatively quickly, which
is a popular notion with operations staff.)

14
Hands-on Workshop Series by Douglas J. Cooper
Copyright  2004 by Control Station LLC
All Rights Reserved
Workshop 5: PI Control of Distillation Top Composition (cont.)
Before starting the experiment, turn file storage on with the "Save Data to File" icon on the tool
bar. You must enter a file name before data storage begins. After entering a file name, a pop-up
menu asks for a data storage rate and starting time. Accept the default values by clicking “OK.”

Storage of process data to file is now in progress. Step the controller output from 45.4% up to
47.4%. When the top composition has shown a clear response to this change (you need not wait
for the response to reach steady state), step the controller output from 47.4% down to 43.4%. And
after a clear response to this change, step the controller output back to its original value of 45.4%.
As soon as the final response is complete, use the "Stop Saving Data" icon on the tool bar to turn
file storage off.

While we focus only on the top composition in this workshop, note that every change in the reflux
rate has a significant impact on the bottom composition. Exploring this multivariable interaction
will be the subject of a later workshop.

3) Now use Design Tools to automatically fit a FOPDT model to this data. First pause execution of
the Distillation Column process using the “Pause” icon on the tool bar. Then, use the “Navigate”
icon on the tool bar to open Design Tools.

From the Design Tools main screen, click the "Open Data File" icon on the toolbar. In the Open
File pop up window, click the file name where you saved the process data and click “OK” to
import your dynamic process data. After the file is read, it will be displayed on your screen with
the Time, Manipulated Variable and Process Variable columns of data already properly labeled. If
you agree that the data labels are correct, click “OK” to accept the column labels.
(Aside: You can use the “Edit” button located at the upper right of this Label Data window and read
the column labels if you are not sure which data each column contains.)

4) You should be back to the Design Tools main screen now. Make sure the "First Order Plus Dead
Time (FOPDT)" model is displayed in the equation box at the bottom of the screen. If not, use the
“Select Model” icon on the tool bar to select it from the library of models.

Click the “Start Fitting” icon on the tool bar to start fitting the FOPDT model to the process data.
If the fit is successful, Design Tools will display a plot. Use your judgment to determine whether
the model reasonably describes the data you collected. Hopefully, the FOPDT model plot line
(shown in yellow) will substantially approximate the response of the measured process variable
data (shown in white). If the model does not describe the data reasonably well, then the tuning
values computed from the model parameters will be questionable.

If the model describes the data to your satisfaction, close the plot screen and note the FOPDT
model parameters displayed to the left on Design Tools’ main screen.

KP = P = P =

15
Hands-on Workshop Series by Douglas J. Cooper
Copyright  2004 by Control Station LLC
All Rights Reserved
Workshop 5: PI Control of Distillation Top Composition (cont.)
5) The main Design Tools screen will also display tuning parameter values for the P-Only, PI and
PID controller.

The PI controller gain, KC, and reset time, I , are computed using the IMC (Internal Model
Control) correlation. The correlation is shown below, but you need not perform the calculation as
Control Station has done it for you. Choose the Standard PID tab on the tuning chart and record
the IMC tuning values for the PI controller:

Standard Tuning:  C is the larger of 0.1  P or 0.8  P

1 p
IMC Tuning: KC = I = P
K p ( p   c )

KC = I =

6) After recording the PI controller tuning values, close Design Tools to return to the Distillation
Column process. Click the “Continue” icon on the tool bar to restart the simulation. (Note: if you
had closed the distillation column simulation rather than navigating from it, you must again set the
disturbance feed flow rate to 600 Kg/min and the controller output to 45.4% to move the process
back to the design level of operation.)

Click the top composition controller (the CC in the circle at the top of the column graphic) to open
the controller design menu. Click “Manual Mode” at the top of the menu and from the list of
controllers, choose PID. The default design is for a PI controller with anti-reset windup logic.
(Aside: When the integral mode is on, the controller bias need not be entered. This is because
when the controller is put in automatic, the bias will be automatically set to the current value of the
controller output. This provides a smooth or bumpless transfer to automatic.)

On the design menu, click "Set Point," then "Controller Gain" and "Reset Time" boxes and enter
the values you recorded above. Because this process has a positive gain, you must select the
“Proportional - Reverse Acting” mode. Leave derivative time as zero, thus making the PID
controller into a PI controller. When you are finished, click "Done" at the bottom of the menu to
put your controller in automatic.

7) Test the performance of your controller in rejecting disturbances when the feed rate to the column
suddenly changes. Starting with the design feed rate to the column of 600 Kg/min, step this
disturbance feed rate up to 650 Kg/min and observe the response (it is most convenient to make
step changes in the disturbance flow rate by clicking on the disturbance box on the distillation
column graphic).

After the response is complete, return the feed rate to 600 Kg/min to complete the disturbance
rejection test.

16
Hands-on Workshop Series by Douglas J. Cooper
Copyright  2004 by Control Station LLC
All Rights Reserved
Workshop 5: PI Control of Distillation Top Composition (cont.)
8) Review the PI controller tuning map at the end of Chapter 8 of the Practical Process Control book.
Here we will develop a corresponding map that shows disturbance rejection performance as a
function of PI controller tuning.

In the center "base case" cell, write the KC and I you computed in Step 5. Document the
disturbance rejection performance you observed in Step 7 as the base case performance.

2.0KC = 0.5I = 2.0KC = I = 2.0KC = 2.0I =

Base Case
KC = 0.5I = KC = I = KC = 2.0I =

0.5KC = 0.5I = 0.5KC = I = 0.5KC = 2.0I =

9) Explore the disturbance rejection performance of the PI controller for the eight other cases in the
grid. That is, for each set of tuning parameters, step the column feed rate from 600 Kg/min up to
650 Kg/min, and after the response is complete, back again. Document the controller performance
for each case.

Can you see parallels in relative performance between tuning for disturbance rejection and tuning
for set point tracking?

17
Hands-on Workshop Series by Douglas J. Cooper
Copyright  2004 by Control Station LLC
All Rights Reserved
Workshop 6: PID Control of Heat Exchanger Temperature
Objective: To learn about the tuning and performance capabilities of a PID controller and to explore
the interaction of the three tuning parameters on controller performance.
Reference: “Practical Process Control” Chapter 10

1) The systematic controller design procedure presented in workshop 4 can be applied to three mode
PID control. For the Heat Exchanger process, consider a case where the design or expected operating
temperature of the exit stream is 137 oC when the warm oil disturbance flow rate is at its expected
value of about 10 L/min.

Begin by performing a test to generate dynamic process data around the design operating conditions.
Use this same dynamic test data when computing the various PID tuning parameters in the steps that
follow. If you are uncertain how to proceed with a dynamic test, review Steps 1-4 in Workshop 4.

2) Using Design Tools, fit a FOPDT model to the dynamic test data collected in Step 1. Use the
resulting KP, P and P in the IMC (Internal Model Control) tuning correlations to compute PID
tuning values. Choose the Standard PID tab on the tuning chart and record the Ideal (Non-
interacting) PID controller values:
Standard, Ideal PID KC = I = D =

Now record Conservative PID tuning values for the Ideal (Non-interacting) PID controller:

Conservative, Ideal PID KC = I = D =

3) Return to the heat exchanger and implement your PID controller using the Standard PID tuning
values. On the Controller Design menu, select:
- Integral with Anti-Reset Windup mode
- Ideal (Non-interacting) Derivative mode
- Derivative computed on Measurement option.

Test the performance of this controller by stepping the set point from 137 oC up to 142 oC and back
again, and then from 137 oC down to 132 oC and back. How does the nonlinear behavior of this
process impact the set point tracking performance for the step up vs. the step down?

Input the Conservative Tuning values and repeat the set point steps from 137 oC up to 142 oC and
back, and then 137 oC down to 132 oC and back.

When you compare Standard versus Conservative tuning, which performance is "best." That is,
which do you believe as control designer provides the most desirable set point tracking performance?

4) We will study how noise in the measured process variable impacts derivative action later. For now,
set the measurement noise to zero (a convenient way to do this is by clicking the right mouse key).
Build your intuition on the interaction that KC , I and D have with each other. Starting with your
Standard PID parameters of Step 2, double and then halve each of the tuning parameters in turn
while leaving the other two fixed at their best value. Note how each change affects controller
tracking performance for set point steps from 137 oC up to 142 oC. Do your experiments show that
derivative action works to dampen oscillations?

18
Hands-on Workshop Series by Douglas J. Cooper
Copyright  2004 by Control Station LLC
All Rights Reserved
Workshop 6: PID Control of Heat Exchanger Temperature (cont.)
5) Starting with your Standard PID parameters of Step 2, explore the impact of measurement noise on
derivative action.
Set the signal noise in the measured process variable to 0.5 (click the right mouse key) and step the
set point from 137 oC up to 142 oC and back again. Now double the noise to 1.0 and repeat the set
point steps. Double the noise to 2.0 and repeat the set point steps. Does derivative action lead to
degraded controller performance when the measurement process variable signal is corrupted with
significant noise? How does the controller output signal change as the level of noise changes?

6) Compare PID with PI control to verify that it is derivative action that is causing performance
problems in the presence of noise.

Return to the controller design menu, turn derivative mode “Off” and click Done. Does the set point
tracking improve under what is now PI control? Do the oscillations in the measure process variable
increase because derivative action is no longer damping them?
Compare the controller output signal when derivative action was present to the way it looks now
under PI control. To do this, you may need to click the View and Print Plot icon on the tool bar and
use the fixed plot options to view back to where this data is located. Can you see that derivative
action amplifies process variable noise in the controller output signal?

7) Study the disturbance rejection capability of the PID controller. Leave the measurement noise at
zero, return the derivative mode to Derivative on Measurement, and set the controller parameters
to their best tuning values. Test the ability of the controller to reject steps in the warm oil
disturbance flow from 10 L/min up to 20 L/min and back again. Determine the “best” tunings for
disturbance rejection. Are there different "best" tunings for disturbance rejection as compared to set
point tracking?

Return the warm oil disturbance flow to 10 L/min and let the process steady. Step the disturbance
flow from 10 L/min up to 20 L/min. When the process steadies, step the flow up to 30 L/min, and
then 40 L/min. Is the nonlinear character of the process evident through changes in disturbance
rejection performance?

(optional)

8) Show that with proper tuning, the performance of the Ideal and Interacting PID algorithms behave
the same.

In Design Tools, choose Standard PID tuning for the Interacting PID controller (you will have to
refit the dynamic test data if you have closed Design Tools). Record your results:

Standard, Interacting PID KC = I = D =

Return to the heat exchanger and restore the design operating conditions. That is, enter a set point
of 137 oC and a disturbance flow rate of 10 L/min. Leave measurement noise at zero.

Click the Advanced option at the top of the Control Design Menu to get access to the Interacting
derivative mode and implement you PID interacting controller.

19
Hands-on Workshop Series by Douglas J. Cooper
Copyright  2004 by Control Station LLC
All Rights Reserved
Workshop 6: PID Control of Heat Exchanger Temperature (cont.)
Test the performance of this controller by stepping the set point from 137 oC up to 142 oC and back,
and then from 137 oC down to 132 oC and back. Compare this performance to that of the Ideal PID
controller tuned using the Standard, Ideal tuning values. Is the performance the same?

9) Test the performance of the PID controller when using the tuning parameters from the wrong
derivative form.

Choose the Interacting derivative mode and input the Standard, Interacting tuning values (these are
the correct values to establish a base case). Make the set point steps up and down as before. Now
keep the tuning values fixed, but on the controller design menu, change to the Ideal derivative mode
and repeat the set point steps. Does the performance of the controller change? How?

20
Hands-on Workshop Series by Douglas J. Cooper
Copyright  2004 by Control Station LLC
All Rights Reserved
Workshop 7: PID with Filter and Control of the Multi-Tanks Process
Objective: To learn about the tuning and performance capabilities of a PID controller with derivative
filter when measurement noise is large, and to explore the interaction of the four tuning
parameters on controller performance.
Reference: “Practical Process Control” Chapter 10

1) The three mode PID control algorithm can be modified with the addition of a derivative filter to
create a four mode PID controller. The PID controller with derivative filter offers improved
performance when there is significant noise or random error in the measured process variable. It
achieves this by limiting the size of (or more precisely, the rate of change of) each controller
output move. The consequence is that there are now four tuning parameters that must be specified.

The systematic controller design procedure presented in workshop 4 can be applied to the four mode
PID controller with derivative filter. We study the performance of this controller and the interaction
of the four tuning parameters using the multi-tank process. Start the multi-tank process from Control
Station's main screen and observe the process graphic.
(Aside: The multi-tank process is a multivariable version of the single loop gravity drained tanks
process. There are two sets of freely draining tanks positioned side by side. The two measured
process variables are the liquid levels in the lower tanks. To maintain level, the two level controllers
manipulate the flow rate of liquid entering their respective top tanks. Similar to the single loop case,
the gravity driven flows are proportional to the square root of the height of liquid in the tank
(hydrostatic head), so this process displays moderately nonlinear behavior.

An important feature of this process is that each of the upper tanks drain into both lower tanks.
This creates a multivariable interaction because actions by one controller affect both measured
process variables. Note that the characteristics of each drain stream are all different so there is no
symmetry between feed flow rate and steady state tank level.)

For this study, we focus only on the left two tanks (upper and lower tank 1). We will explore
multivariable interaction in a later workshop. For now, simply observe that as we make changes to
tank 1 on the left, there is a corresponding reaction in tank 2 on the right.

The design level of operation for this study is a liquid level of 3.0 m in lower tank 1 when the
disturbance drain flow rate out of that tank, D1, is at its expected value of about 1 m3/min.

Begin by performing a test to generate dynamic process data around the design operating conditions.
You should only move controller output 1, CO1, to generate this data. Use this same dynamic test
data when computing the various PID tuning parameters in the steps that follow. If you are uncertain
how to proceed with a dynamic test, review Steps 1-4 in Workshop 4. Controller 2 should remain in
manual mode throughout this workshop and controller output 2, CO2, should remain at the default
value of 61.5%.

2) Using Design Tools, fit a FOPDT model to the dynamic test data collected in Step 1. When you read
in the data file, you will notice the Label Data Columns screen displays nine columns of data. This
is because the file contains data for the control loops for both tanks 1 and 2. It is essential in this
(and any) study that the columns be properly labeled for model fitting and controller tuning.

21
Hands-on Workshop Series by Douglas J. Cooper
Copyright  2004 by Control Station LLC
All Rights Reserved
Workshop 7: PID with Filter & Control of Multi-Tanks Process (cont.)
If you are unsure which columns of data in the file are for tank 1, click the Edit Data button in the
upper right corner of the Label Data Columns screen and read the headings contained in the file.
For this workshop, the second column will hold the tank 1 controller output (manipulated
variable) data and the third column will hold the tank 1 measured process variable data. Be sure
these columns are labeled accordingly.

After fitting the FOPDT model, use the resulting KP, P and P in the IMC (Internal Model Control)
tuning correlations to compute PID tuning values. Choose Standard PID tuning for the three mode
Ideal PID controller and record your results:
Standard, Ideal PID KC = I = D =

Now record Standard PID tuning for the four mode Ideal PID controller with derivative filter:

Standard, Ideal PID with Filter KC = I = D = =

3) Return to the multi-tanks process and set the measurement noise for tank 1 to 0.05 m (use a right
mouse click to access the Change Measurement Noise feature, or click Tasks on the menu list and
then select Change Measurement Noise for tank 1). Remember that tank 2 will remain in manual
mode throughout this workshop.

We will now run three set point step tests right in a row so you can compare them side by side. For
this workshop, you will need to select the Advanced option at the top of the Controller Design Menu
to have access to all of the options you will need.

a) First implement your three mode PID controller tuning values on lower tank 1. Be sure you have
changed measurement noise as described above. On the Controller Design menu, select the Ideal
(Non-interacting) Derivative mode, the Derivative computed on Measurement option, and make
sure the Derivative Filter is Off.

Put the controller in automatic and step the set point from 3 m up to 4 m and back again. Pause the
process after the set point steps are complete. Does the measured process variable "wander" on the
way up rather than moving crisply to the new set point? View a fixed plot (click on the View and
Print Plot icon) and see if you can figure out why (note that the controller output signal becomes
constrained at the upper limit).

b) Next implement your four mode PID controller with derivative filter tuning values on lower
tank 1. Again select the Ideal (Non-interacting) Derivative mode, the Derivative computed on
Measurement option, and here, make sure the Derivative Filter is On.

Put the controller in automatic and step the set point from 3 m up to 4 m and back again. Pause the
process after the set point steps are complete. How does the filter impact the behavior of the
controller output signal? Does the measured process variable now move crisply to the new set
point? Does there appear to be any disadvantages to using the derivative filter?

22
Hands-on Workshop Series by Douglas J. Cooper
Copyright  2004 by Control Station LLC
All Rights Reserved
Workshop 7: PID with Filter & Control of Multi-Tanks Process (cont.)
c) Set the measurement noise for tank 1 to 0.0 m (i.e. no measurement noise) and again implement
your three mode PID controller tuning values on lower tank 1. On the Controller Design menu,
select the Ideal (Non-interacting) Derivative mode, the Derivative computed on Measurement
option, and make sure the Derivative Filter is Off.

Once again step the set point from 3 m up to 4 m and back again. Pause the process after the set
point steps are complete. How does the three mode PID controller, when no measurement noise
present, compare to the four mode PID controller with derivative filter in the presence of
measurement noise? How do the controller output signals compare? Does the filter appear to
eliminate the negative effects of noise in the measured process variable?

4) Investigate how the size of the derivative filter, , impacts controller performance. Set the
measurement noise for tank 1 back to 0.05 m and again implement your four mode PID controller
with derivative filter tuning values on lower tank 1.

Step the set point from 3 m up to 4 m and back again. Double the derivative filter, , and repeat the
set point steps. Continue doubling  and stepping the set point until the trend becomes clear. What
do you conclude about the impact of derivative filter on set point tracking performance?

(optional)

5) Compare the Ideal vs. Interacting PID Controller with derivative filter. Use Design Tools to compute
the Interacting PID with Filter tuning parameters using the dynamic test data collected in Step 1 and
record your results:

Standard, Interacting PID with Filter KC = I = D = =

Implement your Ideal PID controller (as you did in Step 3b) and step the set point from 3 m up to 4
m and back again. Be sure you select the Ideal (Non-interacting) Derivative mode on the Controller
Design menu for this test.

Now implement your Interacting PID controller and again set the set point. Be sure you have
selected the Interacting Derivative mode on the Controller Design.

When assigned their proper tuning values, is the performance of the two controller forms the same?

23
Hands-on Workshop Series by Douglas J. Cooper
Copyright  2004 by Control Station LLC
All Rights Reserved
Workshop 8: Adaptive PI Control of Nonlinear Processes
Objective: To explore how a parameter scheduled adaptive controller can improve controller
performance over a range of nonlinear operation.
Reference: “Practical Process Control” Chapters 6-10

1) Here we explore the design and implementation of a parameter scheduled adaptive PI controller.
The process will be a nonlinear simulation created using Custom Process. Start by clicking on the
Custom Process button on Control Station’s main screen; then choose Single Loop Process. When
the simulation starts, notice that the graphic to the right of the scrolling plots is comprised of a
Process button, Disturb(ance) button and Controller button (the C in the white circle).

To create the nonlinear simulation, click the Process button on the graphic. This opens a
“Construct Process and Disturbance Models” menu. Click the Process Model tab. To specify a
nonlinear controller output to measured process variable dynamic behavior, select Overdamped
Nonlinear Schedule Model from the list.

The model input form provides for the entry for three linear process models, one each at a low,
middle and high basis value of the measured process variable.
(Aside: The nonlinear simulation method used in Custom Process exploits the fact that a linear
model can accurately describe the behavior of a nonlinear process for a narrow range of operation,
and that nonlinear behavior can be simulated by averaging the dynamics of multiple linear models.

During simulation, Custom Process identifies the two linear models whose basis values bracket the
current value of the measured process variable. The dynamic behavior of these two linear models
are then interpolated to simulate the nonlinear behavior of the process. The closer the measured
process variable is to a particular basis value, the more weight that linear model is given in the
interpolation. If the current measured process variable is outside the bounds of the basis values,
the dynamics of the two closest models are extrapolated.)

Enter the process gain, time constants and dead time listed below. Each set of parameters defines
the dynamic process behavior at a specific value of the measured process variable. These linear
models are then weighted and combined to provide a nonlinear behavior over the complete range
of process operation:
Low PV Middle PV High PV
Process Variable Basis 20 50 80

Process Gain, KP 3 2 1
First Time Constant, P1 30 20 10
Second Time Constant, P2 3 2 1
Third Time Constant, P3 0 0 0
Lead Time, PL 0 0 0
Dead Time, P 3 2 1

We will not enter a disturbance model because we will not be exploring disturbance rejection
performance in this workshop.

When finished, click Done at the bottom of the form to start the simulation.

24
Hands-on Workshop Series by Douglas J. Cooper
Copyright  2004 by Control Station LLC
All Rights Reserved
Workshop 8: Adaptive PI Control of Nonlinear Processes (cont.)
2) Verify that the behavior of your Custom Process simulation matches the model values you entered
in Step 1. To do this, generate dynamic test data around each of the process variable basis values
and then use Design Tools to model the data. Our goal is to have each Design Tools fit match the
parameter values we entered into Custom Process.

Adjust the controller output signal until the measured process variable is very near the low basis
value of 20 (hint: try a CO of 37%). When the process is steady, start saving data in a file, say
PV20.txt, and step the controller output a small amount (hint: try stepping the CO from 37% up to
39%). To get the best data set and thus the most accurate model fit, you will want to save every
sample of data to file. To do this, after you click on Save Data, a Save Simulation History form
opens. Change the storage rate in that form to Every 1 Sample.

Read PV20.txt into Design Tools. From the Select Model list, choose the Second Order Plus Dead
Time (SOPDT) model (after all, we entered a SOPDT model into our Custom Process). Fit the
data and verify that the SOPDT fit yields values that are close to those entered in the Low PV
column in Step 1.

Repeat this procedure very near the measured process variable basis values of 50 and then 80.
Verify that Design Tools yields SOPDT model values that reasonably match those entered in Step
1 for each of these levels of operation.

3) Design and implement a fixed-tuning PI controller (not adaptive) around the design level for the
measured process variable of 20 (hint: your file PV20.txt already has the process data you need for
this task). As always, you will need to use a FOPDT model to obtain controller tuning values.
Select the Standard PID tab in this workshop and record your PI tuning values:

KC = I =

4) With this fixed-tuning PI controller in automatic, start with a set point of 20 and let the process
steady. Step the set point from 20 up to 35, and then 50, and then 65. Let the process steady after
each step.

How does the set point tracking performance change as the level of operation moves higher? Does
the controller output signal become constrained and contribute to the change in performance? If
not, why does the controller become increasingly sluggish?

5) Design and implement an adaptive PI controller. To use the adaptive feature, select Advanced at
the top of the controller design menu. Then, half way down the menu, turn Adaptive PID to ON.
Next to that, choose Edit Schedule. Much like the Custom Process nonlinear process model input
form, a Construct Adaptive PID Parameter Schedule form will open.

This form lets you enter tuning parameters at three basis values of the measured process variable.
(Aside: The adaptive mechanism continually adjusts the tuning of the PID controller. At each
sample time, the controller computes a weighted average of the tuning parameters whose basis
values bracket the current measured process variable. The closer the measured process variable
is to a particular basis value, the more weight those tuning parameters receive in the interpolation.
If the measured process variable is outside the bounds of the basis values, the tuning parameters
are extrapolated.)

25
Hands-on Workshop Series by Douglas J. Cooper
Copyright  2004 by Control Station LLC
All Rights Reserved
Workshop 8: Adaptive PI Control of Nonlinear Processes (cont.)
Since we are testing our controller in a set point range of 20 to 65, then we should select basis
values for tuning that bracket this range. Step 2 gave us dynamic test data at measured process
variable basis values of 20, 50 and 80. These bracket our range of operation and therefore are
reasonable choices for basis values.

You might also choose to collect new dynamic data at measured process variable values of, say,
20, 43 and 65 and use these more targeted values in our Adaptive PID Parameter Schedule.
Neither choice is wrong and the decision is left to you.

Whatever your choice, record your three sets of tuning values and their associated basis values.
Use Standard IMC tuning:

Low PV Basis = KC = I =

Middle PV Basis = KC = I =

High PV Basis = KC = I =

6) Enter your parameter schedule and put the adaptive PI controller in automatic. Set the set point to
20 and let the process steady. As with the fixed PI tuning, step the set point from 20 up to 35, and
then 50, and then 65. Let the process steady after each step.

How does the set point tracking performance change as the level of operation moves higher. Does
the adaptive controller maintain performance across the range of operation? Can you see in the
controller output signal trace that the controller changes its effort as operating level changes?

(Optional)

7) Repeat this workshop with a three mode PID controller.

8) Repeat this workshop with a four mode PID controller with derivative filter.

26
Hands-on Workshop Series by Douglas J. Cooper
Copyright  2004 by Control Station LLC
All Rights Reserved
Workshop 9: Modeling and Simulation of Single Loop Processes
Objective: To learn how Design Tools and Custom Process can be used to model, simulate and
explore controller design of single loop processes.
Reference: “Practical Process Control” Chapters 6-7

1) In this workshop, we learn how to model and simulate a single loop process. We pretend here that
the Jacketed Reactor is our "real" process, and we explore how to create a simulation of it using
Custom Process. The procedure detailed here is important to understand because it is the same one
you would follow on a real process in the plant or lab.
(Aside: A summary of the procedure we will follow in this workshop is:
a) At the design level of operation, collect controller output to measured process variable dynamic
data. Also, for the dominant disturbance, collect disturbance to measured process variable
dynamic data.
b) Use Design Tools to fit a first order plus dead time (FOPDT) model to the controller output data
set. Use the resulting model parameters to compute controller tuning values.
c) Use Design Tools to fit second order models to both controller output and disturbance data
sets. The resulting models are implemented in Custom Process to simulate the process.
d) Implement a controller in Custom Process and test it in tracking set points and/or rejecting
disturbances. Refine the design and tuning as needed.
e) When satisfied with the controller performance in Custom Process, return to the actual process
for final implementation.)

2) To begin, start the Jacketed Reactor (not the cascade) process. This will serve as our “real” process
for the study. The first step outlined above is to collect data at the design level of operation, which in
this study is a reactor exit stream temperature (the measured process variable) of 92 oC. The cooling
jacket inlet temperature (the disturbance) is expected to be about 50 oC during normal operation but
may occasionally spike as high as 60 oC. Rejecting the influence of this disturbance spike so the
reactor exit stream temperature remains as close as possible to 92 oC will be the focus of our design.
(Aside: The temperature of the reactor exit stream is directly related to percent conversion of
reactants to products and thus can be used to infer product purity. Though product purity is our true
control objective, reactor temperature is an inexpensive and reliable measured variable substitute.)

With the reactor at steady state at the design level of operation, note the values of the controller
output, measured process variable and disturbance. We will need these values for the simulation:

uDesign = yDesign = dDesign =

Begin saving data to a file (called, say, PROCESS.TXT). Step the controller output from the design
level of 42% up to 47%, then down to 37%, and back to 42%. Let the process show a clear response
between steps. Stop saving data after the process variable returns to steady state.

Again begin saving data to a file (called, say, DISTURB.TXT) and step the cooling jacket inlet
disturbance temperature from 50 oC up to 60 oC, and after the process responds, back to 50 oC. Stop
saving data after the measured process variable steadies.

27
Hands-on Workshop Series by Douglas J. Cooper
Copyright  2004 by Control Station LLC
All Rights Reserved
Workshop 9: Modeling and Simulation of Single Loop Processes (cont.)
3) Exit the reactor process and return to the main Control Station screen. At this point we pretend that
we have just returned from the plant with dynamic process data on disk.

Start Custom Process and choose Single Loop Process from the list. Once the simulation has
started, click Pause to suspend activity and then navigate to Design Tools. Read in file
PROCESS.TXT, fit a FOPDT model to the data and use the model parameters to compute a PI
controller gain, KC, and reset time, I , using the Standard tuning correlation. Record the results.
Also, record the FOPDT SSE for this fit in the space provided in step 4 below.

KC = I =

(Aside: The sum of squared errors (SSE) shown at the bottom of the Model Parameters table
indicates how well the linear model describes the process data contained in the file. A smaller SSE
indicates a better goodness of fit. Thus, SSE is useful for comparing how accurately different
models describe the data.)

4) Next determine a second order model that best describes the open loop dynamic process behavior.
The open loop behavior of the jacketed reactor, like a great many chemical processes, is well
described with a second order plus dead time (SOPDT) model.
(Aside: The SOPDT with Lead model available in Design Tools is also useful in describing the
open loop dynamic behavior of certain processes and we will investigate this model form in later
workshops. We do not consider the second order underdamped model in any workshops because
very few chemical processes require its inherent oscillating nature to describe their open loop
dynamic behavior.)

Click on the Select Model button on the Design Tools tool bar and choose the SOPDT model
form. Fit the data with the SOPDT model and record the SSE below:

FOPDT SSE = SOPDT SSE =

Hopefully, the SOPDT model yields a smaller SSE value, indicating that it best describes the
process data. Record the SOPDT model values for use in the Custom Process simulation.

KP = P,1 = P,2 = P =

5) Repeat step 4 above for the data in file DISTURB.TXT. Note: after reading in a data file, Design
Tools requires that you label the time, manipulated variable and process variable data columns. In
previous exercises, the default labels have been correct. Recall that for this file, it was the cooling
jacket inlet temperature that was manipulated, not the controller output signal. Thus, you must
click on the Manipulated Variable label currently above the second column of data, and then drag
and drop the label over to the top of the last column of data in the file. This properly labels the
disturbance changes as the manipulated data. If you are unsure about this, scroll down the data to
verify that this column contains the step disturbance manipulation between 50 oC and 60 oC.

28
Hands-on Workshop Series by Douglas J. Cooper
Copyright  2004 by Control Station LLC
All Rights Reserved
Workshop 9: Modeling and Simulation of Single Loop Processes (cont.)
Fit the disturbance data with the FOPDT and SOPDT model and record the SSE for each fit:

FOPDT SSE = SOPDT SSE =

Again the SOPDT model (hopefully) shows improved ability to describe the process data. Note
the values for the disturbance driven SOPDT model:

KD = D,1 = D,2 = D =

6) The next step in the procedure is to implement the process and disturbance model parameters in
Custom Process to create a jacketed reactor control simulation. Exit Design Tools and return to
the single loop Custom Process. The graphic to the right of the scrolling plots contains a Process
button, Disturb(ance) button and Controller button (the C in the white circle).

To create the simulation, click on the Process button to open a Construct Process and Disturbance
Models menu. The Process Model tab will be active. The SOPDT model is an overdamped linear
model so be sure that model is selected on the menu. Enter the process gain, two time constants
and dead time that define the controller output driven data as determined in step 4. Take care to
enter each parameter in its proper box. Parameters not used in the model should read zero.

Now click on the Disturbance Model tab and enter the process gain, two time constants and dead
time that Design Tools computed to describe the disturbance driven data in step 5. This also is a
overdamped linear model form. Parameters not used in the model should read zero.

Custom Process assumes as a default that the controller output, process variable and disturbance
variable have a minimum value of 0, a maximum value of 100, and a design level startup value of
50. As in this study, these values are not always appropriate for a simulation. In step 2 we noted
the design level start up values. The minimum and maximum values along with these start up
values are listed in the table below.
(Aside: The minimum and maximum values for the reactor process actually have a larger span
than that listed below. Recall that we are approximating a “real” nonlinear process with simplifying
linear models. The farther we move from our design conditions, the more suspect the linear
models become in describing actual process behavior. Hence, we deliberately narrow this span to
a range where we believe the linear models will be reasonably accurate.)

Click on the Zeros and Spans tab and enter the maximum, minimum and start up values for each
variable of the process as listed in the table:

Controller Output Process Variable Disturbance


Minimum Value 0 60 40
Maximum Value 100 110 60
Start up Value 42 92 50

29
Hands-on Workshop Series by Douglas J. Cooper
Copyright  2004 by Control Station LLC
All Rights Reserved
Workshop 9: Modeling and Simulation of Single Loop Processes (cont.)
When finished, click on Done to start the simulation. You now are simulating the Jacketed Reactor
process. Test the open loop dynamic behavior by stepping the controller output and disturbance
variable and observing the measured process variable response. Return the process to the design
conditions when you accept the simulation as appropriate.

7) Click on the controller icon on the graphic, select a PID controller, and enter the controller gain and
reset time determined in step 3 (leave the derivative time as zero). Since the controller gain is
negative, be sure you specify that the sense of the controller is direct acting. The integral mode
should have anti-reset windup logic.

Test the controller performance in tracking a set point step from 92 oC up to 95 oC and back to 92 0C.
Then test the controller in rejecting disturbance steps from 50 oC up to 60 oC and back to 50 oC.

Do your responses look similar to that of the actual Jacketed Reactor process shown below? If so,
you have successfully used Control Station to simulate this “real” process. Why do you think there
are differences between your simulation and the behavior of the actual reactor process as shown?

S in g le Lo o p R eacto r U n d er P I C o n tro l
P r oces s : S ingle L oop J acketed R eactor C ont.: P I D ( P = DA, I = AR W , D= of f , F = of f )
96
D i s t u r b a n c e C o n t r o l l e r O u t p uP t V / S e t p o i n t

94

92

90

88
80

60

40

20

60

55

50

30 35 40 45 50 55 60 65 70 75 80 85 90
Time (mins)

T uning: Gain = - 4.40, R es et T im e = 2.10, S am ple T im e = 1.00

Set point tracking and disturbance rejection performance of the Jacketed Reactor process

8) From the main Control Station screen, start the Jacketed Reactor (not the cascade) process. Click on
the controller icon on the graphic and enter your PI controller tuning values just as in step 7 above.
Repeat the set point tracking and disturbance rejection test and verify the above plot is accurate.

Does your disturbance rejection performance look like that shown in the plot? It is especially
important to verify this because Workshops 10 and 11 focus on using cascade control and feed
forward control for improved disturbance rejection in the jacketed reactor. The above plot will be
used as a baseline to establish benefits for these advanced architectures.

30
Hands-on Workshop Series by Douglas J. Cooper
Copyright  2004 by Control Station LLC
All Rights Reserved
Workshop 10: Cascade Control of the Jacketed Reactor
Objective: To explore cascade controller design and implementation. Also, to compare disturbance
rejection performance of a cascade architecture with that of a lone PI controller for the
jacketed reactor process.
Reference: “Practical Process Control” Chapter 18

1) Select the cascade jacketed reactor in Case Studies and start the simulation. Like all cascade
architectures, the process graphic shows two measured variables, two controllers, but only one
manipulated variable (one valve). The upper strip chart on the screen shows the primary process
variable (the reactor exit temperature) while the lower strip chart shows the secondary process
variable (the cooling jacket outlet temperature). To see controller output signals, use the “View
and Print Plot” icon on the tool bar to generate a fixed plot, then use the “Plot Options” icon on
the plot tool bar to display controller output traces.

The design level of operation is the same as in Workshop 6. The control objective is to maintain
reactor exit temperature at 92 oC. Reactor exit temperature indicates percent conversion of reactants
to products so this exit temperature is the primary or outer loop process variable. The cooling jacket
inlet temperature (the disturbance) is expected to be 50 oC during normal operation but occasionally
may spike as high as 60 oC. Our controller design will focus on rejecting this disturbance. Begin this
study with the process steady at the design level of operation.

Offset is unacceptable for reactor temperature control because off-spec product means lost profit.
Hence, a PI controller will be used on the primary loop. The secondary or inner controller maintains
jacket exit temperature at the design value of 75.6 oC. A fast and simple controller is desired here.
Since we do not sell cooling jacket water, offset does not present a problem and a P-Only
controller will serve our purpose.

Consider the discussion above and the cascade jacketed reactor graphic on your screen, then
record this design information for later:

Secondary (inner) P-Only Controller: Tsetpoint = ubias =

Primary (outer) PI Controller: Tsetpoint =

2) Cascade controller design begins with the secondary controller (the cooling jacket exit
temperature control loop). This will be P-Only controller designed to track set point updates sent
from the primary controller.

Controller tuning follows the same procedure presented in earlier workshops. Begin saving data to
file (say, SECOND.TXT) and generate dynamic data by stepping the controller output from 42%
down to 37%, then up to 47%, and finally back to 42%. Let the response become clearly
established before making each change. Turn file storage off when the response is complete.

3) Pause the reactor simulation, navigate to Design Tools and read in SECOND.TXT. You will notice
the Label Data Columns screen displays eight columns of data. This is because the file contains
data for both the secondary and primary loop. It is essential in this (and any) study that the
columns be properly labeled for model fitting and controller tuning.

31
Hands-on Workshop Series by Douglas J. Cooper
Copyright  2004 by Control Station LLC
All Rights Reserved
Workshop 10: Cascade Control of the Jacketed Reactor (cont.)
If you are unsure which columns of data in the file are for the secondary control loop, click on the
Edit Data button in the upper right corner of the Label Data Columns screen and read the headings
contained in the file. For the cascade reactor, the second column will hold the secondary
controller output data and the third column will hold the secondary measured process variable
data. Be sure these columns are labeled accordingly.

Fit a FOPDT model to the data and if the fit seems reasonable, record the P-Only controller gain
from the ITAE for Set Point Tracking correlation:

Secondary controller P-Only controller gain, KC =

Remember that if the model does not reasonably approximate the data, the controller tuning values
computed from the model parameters will be suspect.

4) Close Design Tools and return to the cascade reactor. Click on the secondary controller icon on the
reactor graphic (the controller for the cooling jacket exit temperature) and implement a P-Only
controller (select a PID controller and turn the integral and derivative action off). Enter your set
point, bias and controller gain as recorded above and click Done. Make sure you have the control
sense (whether it is direct or reverse acting) properly selected.

Test the controller in tracking set point steps near the design conditions. When you are satisfied
that the set point tracking performance balances a rapid response with a short settling time, leave
the secondary controller in automatic and proceed to designing the primary controller.

5) The primary controller output is the set point of the secondary controller. Thus, stepping the set
point of the secondary controller is the same as stepping the controller output of the primary
controller. Note that because of this cascade architecture, the output of the primary controller has
units of oC rather than the more traditional units of %.

Begin saving data to file (say, PRIMARY.TXT) and generate dynamic data by stepping the secondary
set point from 75.6 oC down to 72.6 oC, then up to 78.6 oC, and finally back to 75.6 oC. Allow the
process to respond before making each change. When the data has been collected, turn data file
storage off.

6) Again use Design Tools to tune the controller. Take care to label the proper manipulated and
measured process variable data columns (The default labels are not correct. Click on the Edit Data
button in the upper right corner of the Label Data Columns screen and read the headings
contained in the file).

If the model fit is reasonable, compute PI controller tuning parameters using the IMC (Internal
Model Control) tuning correlation:

Primary controller PI tuning parameters: KC = I =

32
Hands-on Workshop Series by Douglas J. Cooper
Copyright  2004 by Control Station LLC
All Rights Reserved
Workshop 10: Cascade Control of the Jacketed Reactor (cont.)
7) Return to the jacketed reactor simulation, click on the outer primary controller and implement
your PI controller. With both controllers in automatic, test the disturbance rejection capabilities of
the cascade architecture by making step changes in the cooling jacket inlet temperature from 50
o
C up to 60 oC and back to 50 oC again. Compare the performance of this cascade architecture in
rejecting the disturbance to that of the single loop controller shown below.

As shown below, the reactor exit temperature for the single loop (not cascade) case ranges during
the disturbance from about 88 oC to 95 oC.

S i n g l e L o o p R e a c t o r U n d e r P I C o n t ro l
Pro c e ss: Si n g l e L o o p Ja c k e t e d R e a c t o r C o n t . : PID ( P= D A , I= A R W , D = o ff, F = o ff)
D is t u r b a n c e C o n t r o lle r O u t p Pu Vt / S e t p o in t

96

94

92

90

88
80

60

40

20

60

55

50

30 35 40 45 50 55 60 65 70 75 80 85 90
Tim e (m in s )

T u n i n g : G a i n = -4 . 4 0 , R e se t T i m e = 2 . 1 0 , Sa m p l e T i m e = 1 . 0 0

8) Step the set point from 92 oC up to 95 oC and back again and compare the results to the above
single loop case. Does the cascade architecture offer any benefit in set point tracking
performance?

(optional)

9) Explore how P-Only controller tuning affects disturbance rejection performance. What happens
when you increase and decrease the P-Only controller gain of the secondary controller? How does
disturbance rejection performance change if you put a PI controller on this inner loop?

Restore the P-Only controller gain to its design value and explore how tuning adjustments to the
outer primary PI controller affect disturbance rejection performance. Does derivative action show
benefit?

33
Hands-on Workshop Series by Douglas J. Cooper
Copyright  2004 by Control Station LLC
All Rights Reserved
Workshop 11: Feed Forward Control of the Jacketed Reactor
Objective: To explore feed forward controller design and implementation. Also, to compare
disturbance rejection performance of a feed forward controller with that of a lone PI
controller for the jacketed reactor process.
Reference: “Practical Process Control” Chapter 19

1) Start the Jacketed Reactor (not the cascade) process. As in the previous workshops, the design level
of operation is a reactor exit temperature of 92 oC. Reactor exit temperature indicates percent
conversion of reactants to products so maintaining this exit temperature is our control objective. The
cooling jacket inlet temperature (the disturbance) is expected to be 50 oC during normal operation
but occasionally may pulse as high as 60 oC. Our controller design will focus on rejecting the
influence of this disturbance. Begin this study with the process steady at the design level of operation
and record the design set point.

Tsetpoint =

2) A feed forward control element will be used in combination with a traditional feedback controller
in this study (called feed forward with feedback trim). The feed forward element is comprised of a
process model that describes controller output to process variable dynamic behavior and a
disturbance model that describes the disturbance to process variable dynamic behavior.
(Aside: Feed forward, like cascade control, is used for improved disturbance rejection. Feed
forward control is appropriate when one particular disturbance causes a significant disruption to
process operation. For feed forward implementation, a sensor must be installed to measure this
disturbance. A feed forward model must be programmed to compute a sequence of control actions
that will just counter the impact of each disturbance event on the measured process variable.)

As always, tuning requires a controller output to process variable dynamic data set. For this study,
this same data will be used to fit the process model required in the design of the feed forward
element. Start saving data to file (say, PROCESS.TXT) and perform an open loop test by stepping
the controller output from 42% down to 37% then up to 47% and finally back to 42%. Let the
response become clearly established before making each change. Turn file storage off when the
response is complete.

3) Disturbance to process variable dynamic data is needed for the disturbance model in the feed
forward element. With the process at steady state, start saving data to another file (say,
o o
DISTURB.TXT). Step the disturbance (the cooling jacket inlet temperature) from 50 C up to 60 C
o
and after the process has responded, back to 50 C. When the response is complete, stop saving data.

4) Pause the simulation and navigate to Design Tools. Open the PROCESS.TXT file and verify that
the first, second and third data columns are labeled as time, manipulated variable and process
variable, respectively. Fit a first order plus dead time (FOPDT) model to the data and record the
PI controller tuning values. Choose the Standard tuning correlation to compute a PI controller
gain, KC, and reset time, I.

KC = I =

34
Hands-on Workshop Series by Douglas J. Cooper
Copyright  2004 by Control Station LLC
All Rights Reserved
Workshop 11: Feed Forward Control of the Jacketed Reactor (cont.)
Also record the FOPDT model parameters for use in constructing the process model portion of the
feed forward element. Include the SSE to gauge of goodness of model fit.

KP = P = P = SSE =

Later we will compare the benefit of different models on feed forward performance. Select a
second order plus dead time model (SOPDT) and fit it to the process data. Record the results:

KP = P,1 = P,2 = P = SSE =

Based on the SSE, does the SOPDT model show improved capability in describing process
behavior over the FOPDT model?

5) Now open DISTURB.TXT in Design Tools and as before, verify that the first and third data columns
are labeled as Time and Process Variable. For this data set, recall that it was the disturbance
variable that was manipulated during the dynamic experiment. Scroll down the data to verify that
the last column in the file contains this disturbance step between 50 and 60 oC. You can also edit
the file to confirm this. After you have verified the disturbance data, click on the Manipulated
Variable label currently above the second column of data, and then drag and drop the label over to
the top of the last column of data in the file. This properly labels the disturbance changes as the
manipulated data for model fitting.

Fit a first order plus dead time (FOPDT) model to the data and record the FOPDT model
parameters for use in constructing the disturbance model portion of the feed forward element:

KD = D = D = SSE =

Also select a second order plus dead time model (SOPDT) and fit it to the disturbance data.
Record the model parameters and SSE:

KD = D,1 = D,2 = D = SSE =

Based on the SSE, does the SOPDT model show improved capability in describing the
disturbance behavior over the FOPDT model?

6) Close Design Tools and return to the jacketed reactor. Click on the controller icon on the reactor
graphic. At the top of the controller design menu, select the PID with Feed Forward controller.

On the left side of the menu, enter the set point from step 1 and PI controller KC and I from step
4. On the right side of the menu, click on the Process tab and enter your FOPDT process model
parameters from step 4. Then click on the Disturb(ance) tab and enter your FOPDT disturbance
model parameters from step 5. Leave parameters not included in the model as zero. Click Done to
put the controller in automatic.

35
Hands-on Workshop Series by Douglas J. Cooper
Copyright  2004 by Control Station LLC
All Rights Reserved
Workshop 11: Feed Forward Control of the Jacketed Reactor (cont.)
(Aside: For the feed forward model to be physically realizable (mathematically solvable), the dead
time of the disturbance model must be greater than or equal to the dead time of the process
model. Otherwise, the math would say that the feed forward corrective actions must start before
the disturbance even occurs! This is not a limitation of Control Station but a physical and
mathematical reality of all feed forward elements. If your model fits from steps 4 and 5 do not meet
this criterion, you must arbitrarily change the disturbance or process model dead time to make
them equal. Unfortunately, such a change will reduce model accuracy and degrade feed forward
performance.)

7) Test the disturbance rejection performance of this feed forward with feedback trim
implementation. Make step changes in the disturbance from 50 up to 60 oC and back to 50 oC
again. When the response is complete, pause the simulation.

To determine the benefits of the feed forward element, click on the controller icon and from the
controller menu select the regular PID controller (without feed forward). Start the simulation and
repeat the step changes in the disturbance from 50 up to 60 oC and back to 50 oC.

How do the two controllers compare? Does the feed forward element show clear benefit? Is the PI
tuning you used for the non-feed forward study correct or to make the comparison fair, do you
need to adjust the tuning values?

8) Return the controller to PID with Feed Forward and on the right side of the menu, enter your
SOPDT model parameters for the process and disturbance. Leave parameters not included in the
model as zero. Click Done to put the controller in automatic. Remember that for the feed forward
model to be physically realizable (mathematically solvable), the dead time of the disturbance
model must be greater than or equal to the dead time of the process model. If need be, shorten the
process dead time to equal the disturbance dead time and test the feed forward disturbance
rejection performance. Make step changes in the disturbance from 50 oC up to 60 oC and back to
50 oC again.

The SOPDT models (hopefully) had lower SSE values than the FOPDT models, implying better
model fits of the data. Does the SOPDT models show clearly improved feed forward disturbance
rejection performance? Why or why not?

9) Compare the disturbance rejection performance of static feed forward to dynamic feed forward.
Open the controller design menu and on the model entry portion of the form, click the box to
switch to static feed forward.
(Aside: Static feed forward only uses the process and disturbance gains in the feed forward
calculation. Thus, the dead time model parameters that account for the delay that occurs before a
disturbance reaches the measured process variable, and the time constant model parameters that
account for the speed at which each disturbance disrupts the measured process variable, are not
considered in the static feed forward calculation. With static feed forward, each feed forward action
is implemented immediately and completely upon detection of a disturbance.

The benefit of static feed forward control is that it requires no programming. Often, static feed
forward can be implemented with conventional control loop instrumentation.)

36
Hands-on Workshop Series by Douglas J. Cooper
Copyright  2004 by Control Station LLC
All Rights Reserved
Workshop 11: Feed Forward Control of the Jacketed Reactor (cont.)
Test the static feed forward disturbance rejection performance by making step changes in the
disturbance from 50 oC up to 60 oC and back to 50 oC again. Is there any clear change in
disturbance rejection performance compared to the fully dynamic model form?

10) Return to the dynamic FOPDT Feed Forward model parameters (Step 6) and perform a
sensitivity study to determine how each feed forward model parameter impacts the disturbance
rejection performance. To do this, first double and then halve the process model gain and perform
a disturbance rejection experiment after each change. Keep all the other model parameters at their
original design values. Repeat this procedure for the process model time constant and finally the
process model dead time. Do this study a second time for the disturbance model parameters.

Which of the model parameters most impacts disturbance rejection performance? For each model
parameter, determine if it is more conservative (less aggressive) to have it be too large or too
small.

11) Explore the benefit of feed forward control on set point tracking performance. Step the set point
from 92 oC up to 95 oC and back again. Now use the controller design menu to change the
controller to traditional PID (no feed forward). Keep the PI controller tunings unchanged and
repeat the set point step. Does the feed forward element offer any benefit in set point tracking
performance?

37
Hands-on Workshop Series by Douglas J. Cooper
Copyright  2004 by Control Station LLC
All Rights Reserved
Workshop 12: Advanced Feed Forward Control of Heat Exchanger
Objective: To explore advanced modeling issues and the impact of plant-model mismatch on feed
forward controller performance.
Reference: “Practical Process Control” Chapter 19

1) The design objective for study is to maintain a constant heat exchanger exit stream temperature of
135 oC. During normal operation, the warm liquid disturbance flow rate is steady at about 10
L/min, but up-stream operation occasionally causes the disturbance stream flow rate to jump as
high as 20 L/min. Rejecting the impact of this disturbance is our design objective.

Start the Heat Exchanger process. Design a PI controller. That is,


- move the process to the design operating level,
- generate controller output to measured process variable dynamic data around this design level
and save the data to file called, say, PROCESS.TXT,
- fit a first order plus dead time (FOPDT) model to the dynamic test data,
- use the resulting parameters in a PI tuning correlation (IMC with Standard tuning) to obtain
initial values for PI controller gain, KC, and reset time, I

If the FOPDT fit describes the dynamic behavior of the data to your satisfaction, record the model
parameters:

KP = P = P = SSE =

Record your PI tuning parameters using the Standard tuning correlation:

KC = I =

Implement your PI controller and verify that it performs reasonably well in rejecting step changes
in the warm liquid disturbance flow from 10 L/min up to 20 L/min and back again.

2) A feed forward control element is comprised of a process model that describes controller output to
process variable dynamic behavior and a disturbance model that describes the disturbance to
process variable dynamic behavior. In step 1, you already computed the required process model.

Make sure the controller is in Manual Mode and generate the dynamic data needed to determine
the disturbance model. With the process at steady state, start saving data to another file (say,
DISTURB.TXT). Step the warm oil disturbance flow rate from 10 L/min up to 20 L/min. Because
of the complicated models we will use to fit this data, let the measured exit temperature pattern
evolve all the way to a clear steady state. When the response is complete, step the disturbance
flow rate back to 10 L/min and again wait for the response to complete. Stop saving data.

Why does the heat exchanger show this inverse response to a disturbance variable step?

38
Hands-on Workshop Series by Douglas J. Cooper
Copyright  2004 by Control Station LLC
All Rights Reserved
Workshop 12: Advanced Feed Forward Control of Heat Exchanger (cont.)
3) Pause the simulation and navigate to Design Tools. Open DISTURB.TXT and label the first and third
data columns as time and process variable. It was the disturbance variable that was manipulated
during the experiment. Verify that the last data column in the file contains this disturbance step
between 10 and 20 L/min. You can edit the file if need be to confirm this. Thus, you must click on
the Manipulated Variable label currently above the second column of data, and then drag and drop
the label over to the top of the last column of data in the file. This properly labels the disturbance
changes as the manipulated data for model fitting.

Fit a first order plus dead time (FOPDT) model to the disturbance data and record the model
parameters for use in constructing the disturbance model portion of the feed forward element:

KD = D = D = SSE =

Fit a second order plus dead time model (SOPDT) to the disturbance data. Record the model
parameters and SSE:

KD = D,1 = D,2 = D = SSE =

Fit a second order plus dead time with lead time model (SOPDT w/ Lead) to the disturbance data.
Record the model parameters and SSE:

KD = D,1 = D,2 = D,L = D = SSE =

Based on the SSE and your observations of the goodness of model fit, does the SOPDT model
show improved capability in describing the disturbance behavior over the FOPDT model? How
about the SOPDT w/ Lead model?
(Aside: Design Tools must search for five model parameters that minimize the SSE when fitting a
SOPDT w/ Lead model. A model with five adjustable parameters can fit a wide range of data. To
learn about this, go to the single loop Custom Process and construct a second order plus dead
time with lead time model. First put in a large positive lead term, say twice the size of the largest
time constant, and perform an open loop test by stepping the controller output. Repeat with a large
negative lead term. Repeat a third time with a zero lead term. The lead term influence should be
clear. If a process has a zero lead term influence, Design Tools may converge on a fit that looks
reasonable but beware of nonsense results. For example, the lowest SSE can sometimes be
achieved mathematically for such data with a fantastically large L and a KP near zero. It is your job
not only to visually accept model fits, but also to ask if the model parameters make sense.)

4) Return to the Heat Exchanger process. Make sure your PI tuning values from Step 1 are still
entered on the Controller Design menu. Keep these values constant for all tests hereafter.

Demonstrate the performance of a feedback controller with no feed forward element in rejecting a
disturbance step from 10 L/min to 20 L/min and back again. Pause the simulation when the
response is complete. This will be our baseline performance.

39
Hands-on Workshop Series by Douglas J. Cooper
Copyright  2004 by Control Station LLC
All Rights Reserved
Workshop 12: Advanced Feed Forward Control of Heat Exchanger (cont.)
Select the PID with feed forward controller. Click on the Process tab of the feed forward model
and enter the FOPDT process model parameters from step 1.

Click on the Disturb(ance) tab of the feed forward model and enter the FOPDT disturbance model
parameters from step 3. Click Done to put the controller in automatic. Show the performance of a
feed forward controller with feedback trim in rejecting a disturbance step from 10 L/min to 20
L/min and back again. Pause the simulation when the response is complete.
(Aside: Recall that for the feed forward model to be physically realizable, the dead time of the
disturbance model must be greater than or equal to that of the process model. If your models do
not meet this criterion, you must change one of the model dead times to make them equal.)

5) Return to the controller design menu, click on the Disturb(ance) tab of the feed forward model,
enter the SOPDT disturbance model parameters and click Done to put the controller in automatic.
Show the performance of the controller in rejecting a disturbance step from 10 L/min to 20 L/min
and back again. Pause the simulation when the response is complete.

Repeat one more time for the SOPDT w/ Lead disturbance model. Pause the simulation when the
response is complete and click on View and Print Plot to view a fixed plot of the investigation.
Use the Plot Options icon to create a plot similar to that shown below.

C o m p arin g Feed Fo rw ard M o d els


P V /S e tp o in t

P ro c e s s : H e a t E x c h a n g e r C o n tro lle r: P ID w ith F e e d F o rw a rd


138
137
136
135
134
133
132 SOPDT w/ Lead
No Feed Forward FOPDT SOPDT
D i s tu rb a n c e

20

15

10

40 48 56 64 72 80 88 96 104 112 120 128 136 144


T im e (m ins )
T u n in g : G a in = -1 . 8 0 , R e s e t T im e = 1 . 1 0 , D e riv T im e = 0 . 0 , S a m p le T im e = 1 . 0 0
P ro c e s s M o d e l: G a in (K p ) = -0 . 3 9 , T 1 = 0 . 9 6 , T 2 = 0 . 0 , T D = 0 . 8 0 , T L = 0 . 0
D is tu rb a n c e M o d e l: G a in (K d ) = -0 . 1 4 6 , T 1 = 0 . 7 3 5 , T 2 = 0 . 7 3 5 , T D = 0 . 8 2 2 , T L = -3 . 4 1

40
Hands-on Workshop Series by Douglas J. Cooper
Copyright  2004 by Control Station LLC
All Rights Reserved
Workshop 12: Advanced Feed Forward Control of Heat Exchanger (cont.)
6) Why does feed forward with a FOPDT disturbance model only show modest benefit over the
disturbance rejection performance of a feedback controller with no feed forward element at all?
Why does a SOPDT disturbance model show no benefit in disturbance rejection performance
compared to that of the FOPDT model?
Why does the SOPDT w/ Lead disturbance model not show perfect disturbance rejection
performance?

7) Compare the disturbance rejection performance of static feed forward element to the performance
of dynamic feed forward element with a SOPDT w/ Lead model.

(Aside: Static feed forward only uses the process and disturbance gains in the feed forward
calculation. Thus, the dead time model parameters that account for the delay that occurs before a
disturbance reaches the measured process variable, and the time constant model parameters that
account for the speed at which each disturbance disrupts the measured process variable, are not
considered in the static feed forward calculation. With static feed forward, each feed forward action
is implemented immediately and completely upon detection of a disturbance.

The benefit of static feed forward control is that it requires no programming. Often, static feed
forward can be implemented with conventional control loop instrumentation.)

Using the SOPDT w/ Lead disturbance model, step the disturbance from 10 L/min to 20 L/min
and back again. Pause the simulation after the response has completed.

Open the controller design menu and on the model entry portion of the form, click the box to
switch to static feed forward. Again step the disturbance from 10 L/min to 20 L/min and back. Is
there any clear benefit in disturbance rejection performance for the static vs. dynamic model
form?

8) Compare the set point tracking performance of a feedback controller (no feed forward element) to
that of your best feed forward with feedback trim design. Use the same PI tuning parameters for
both set point tracking experiments. Does the feed forward element offer any benefit in set point
tracking performance?

41
Hands-on Workshop Series by Douglas J. Cooper
Copyright  2004 by Control Station LLC
All Rights Reserved
Workshop 13: Multivariable Decouplers and Distillation Control
Objective: To observe control loop interaction on a multi-loop process and to design and explore
model-based decouplers that minimize such interaction.
Reference: “Practical Process Control” Chapter 20-21

1) Start the distillation column simulation and observe the process graphic. This column separates a
mixture of benzene and toluene into a top and bottom stream. As shown on the graphic, there is a
controller at the top of the column that manipulates the reflux flow rate to control the top
composition. A second controller at the bottom of the column that manipulates the steam rate to
control the bottom composition. The design level of operation for this study is:
- top composition of 94.5% benzene
- bottom composition of 2.6% benzene (97.4% toluene)
- total feed flow rate to column of 596 Kg/min
Record the design set point values for the top controller and bottom controllers:
top
ysetpoint  % benzene bottom
ysetpoint  % benzene

2) Before designing the controllers, explore the dynamics of this process. Step the top controller
output (CO) from 50% up to 55%, down to 45% and back to 50%. Repeat this experiment with
the bottom controller, stepping it from 47.9% up to 52.9%, down to 42.9% and back to 47.9%.
Compare your results to the plot below and verify that it represents the dynamic behavior of the
distillation column.

top CO step up
gives large KP

bottom CO step down


gives large KP

This plot shows that the top composition loop is only mildly nonlinear.

The bottom composition loop, however, is extremely nonlinear. In particular, the process gain of
the bottom loop is three times larger when the bottom composition is increasing from the design
level of operation compared to when it is decreasing, regardless of whether it is the top or bottom
controller forcing the response.

42
Hands-on Workshop Series by Douglas J. Cooper
Copyright  2004 by Control Station LLC
All Rights Reserved
Workshop 13: Multivariable Decouplers and Distillation Control (cont.)
(Aside: The extreme nonlinear behavior makes this process quite challenging to control. To meet
this challenge, we take advantage of the information in the above plot to design the dynamic tests.
Rather than a doublet, we will pulse each controller output in one direction only, the direction that
yields parameters leading to the most stable design.
Recall that tuning correlations compute controller gain, KC, as proportional to the inverse of
process gain, KP. So for a conservative design, we want to use large values of KP in the
correlations to obtain small values of KC. Exploiting the information contained in the above figure,
we achieve the largest values of KP by pulsing the top controller up and bottom controller down
when generating our dynamic test data.)

3) Remember that both loops should remain in manual mode throughout the dynamic tests.

For the top loop, start saving data to file (say, TOP.TXT) and pulse the top controller output from
50% up to 55% and back to 50%. Let the response become well established before making the
second step. Turn file storage off when the response is complete.

Let the column settle out to steady state. For the bottom loop, start saving data to file (say,
BOTTOM.TXT). Step the bottom controller output from 47.9% down to 42.9% and back to 47.9%.
Again wait for the response to complete before making the return step. Turn file storage off when
the test is complete.

4) Use Design Tools to compute PI tuning values for the top controller. Pause the distillation
simulation, navigate to Design Tools and read in TOP.TXT. The Label Data Columns screen will
display eight columns of data because the file contains data for both the top and bottom loop. It is
essential in this (and any) study that the columns be properly labeled before starting a model fit. If
you are unsure which columns of data are for the top loop, click the Edit Data button in the upper
right corner of the Label Data Columns screen and read the headings contained in the file.

For this study, the second column holds the top controller output data and the third column holds
the top measured process variable data. Be sure to drag and drop the labels to reflect this before
starting the model fit. Fit a FOPDT model to the data and if the fit seems reasonable, record the PI
controller tuning values from the IMC tuning correlation:

Top controller PI tuning parameters: KC = I =

Remember that if the model does not reasonably approximate the data, the controller tuning values
computed from the model parameters will be suspect.

Repeat the process to tune a PI controller for the bottom loop. Read BOTTOM.TXT into Design
Tools. The fifth and sixth columns hold the controller output and measured process variable data
for the bottom loop. Label these columns accordingly. Fit a FOPDT model to the data and check
that the fit seems reasonable. Record the PI controller tuning values from the IMC tuning
correlation:

Bottom controller PI tuning parameters: KC = I =

43
Hands-on Workshop Series by Douglas J. Cooper
Copyright  2004 by Control Station LLC
All Rights Reserved
Workshop 13: Multivariable Decouplers and Distillation Control (cont.)
5) Exit Design Tools and return to the distillation simulation. Implement your PI controller design
for the top loop while leaving the bottom loop in manual. Test the top controller by stepping the
top composition set point from 94.5% benzene up to 96% and back again.

Put the top loop back in manual and implement your PI controller design for the bottom loop. Test
this controller by stepping the bottom composition set point from 2.6% benzene up to 3.6% and
back again.

Verify that each of your controllers performs similarly to that shown below.

In d i v i d u al Perfo rm an ce o f T o p an d B o t t o m L o o p
%

P roc e s s : D is t i l l a t i on C ol um n T op: M a nua l M ode / B ot : P ID ( P = D A , I= A R W , D = off )


C O D istilla te

96
Top %

95
94

55
top loop performance with bottom loop in manual
50
B o tto m s % T o p

45

4
3
2

bottom loop performance with top loop in manual


B o t C O

50
48
46
44

500 1000 1500


T im e (m ins )

T uni ng: G a i n = -3. 70, R e s e t T i m e = 63. 6, D e ri v T i m e = 0. 0, S a m pl e T i m e = 1. 00


6) Now put both PI controllers in automatic, step the top composition set point from 94.5% benzene
up to 96% and watch the loop interaction.
(Aside: As the top controller sends more cold reflux down the column to increase the purity of the
top composition, it has the undesirable consequence of causing the bottom of the column to cool.
This moves the bottom composition from set point. The bottom controller responds by increasing
the steam rate, which heats up the bottom of the column as desired. It results, however, in hot
vapors traveling up the vessel and heating the top of the column at the same time the top controller
is trying to cool it. Over time, you will see both the reflux rate and the steam rate increase more and
more as each controller “fights” for its own measured process variable. The result is a dismal
overall performance.)

When you have established the negative impact of loop interaction, return the set point to 94.5%
and let the process reach steady state.

44
Hands-on Workshop Series by Douglas J. Cooper
Copyright  2004 by Control Station LLC
All Rights Reserved
Workshop 13: Multivariable Decouplers and Distillation Control (cont.)
7) (Aside: A multivariable decoupler is a feed forward element where the measured disturbance is the
output signal of the cross-loop controller. To construct a decoupler, we need to identify a process
and cross-loop disturbance model for both the top and bottom loop (four models total). No
additional process testing is necessary to create these models because the tests just completed
contain the information we need.

When the top controller output signal changes (thus changing the reflux rate), it not only causes a
response in the top composition so we can fit a process model, but it also acts as a disturbance to
the bottom composition. Thus, the cross-loop disturbance data we need for the bottom decoupler
is already contained in file TOP.TXT.

And when the bottom controller output signal changes (thus changing steam rate), it not only
causes a response in the bottom composition so we can fit a process model, but it also acts as a
disturbance to the top composition. Hence, the cross-loop disturbance data we need for the top
decoupler is already contained in file BOTTOM.TXT.)

To design the top decoupler, the process and disturbance models needed are:
- Process model: GTT(s) = top composition response to top controller output model
- Disturbance model: GTB(s) = top composition response to bottom controller output model

For the bottom decoupler, the process and disturbance model parameters needed are:
- Process model: GBB(s) = bottom composition response to bottom controller output
- Disturbance model: GBT(s) = bottom composition response to top controller output

To obtain the process model for the top decoupler, GTT(s), read TOP.TXT into Design Tools. The
second column holds the top controller output data. The third column holds the top process
variable data (you can verify this with the Edit Data button). Label these columns as such and fit a
FOPDT model. If the fit is acceptable, record the model below as the top decoupler process
model.

To obtain the cross-loop disturbance model for the top decoupler, GDB(s), read BOTTOM.TXT into
Design Tools. The fifth column holds the bottom controller output data. The third column holds
the top process variable data. Label these columns as such and fit a FOPDT model. If the fit is
acceptable, record the model below as the top decoupler cross-loop disturbance model.

Top Decoupler:
Process Model (top composition response to top controller output)

KP = P = P =

Cross-Loop Disturbance Model (top composition response to bottom controller output)

KD = D = D =
In absolute value, if |KD| > |KP| , then the decoupler assigns more influence over the process
variable to the cross-loop disturbance than it does to the controller. This can lead to unstable
control. If your fits result in this situation, lower your KD above so |KD| = |KP|.

45
Hands-on Workshop Series by Douglas J. Cooper
Copyright  2004 by Control Station LLC
All Rights Reserved
Workshop 13: Multivariable Decouplers and Distillation Control (cont.)
8) To obtain the process model for the bottom decoupler, GBB(s), read BOTTOM.TXT into Design
Tools. The fifth column holds the bottom controller output data. The sixth column holds the
bottom process variable data. Label these columns as such and fit a FOPDT model. If the fit is
acceptable, record the model below as the bottom decoupler process model.
To obtain the cross-loop disturbance model for the bottom decoupler, GBT(s), read TOP.TXT into
Design Tools. The second column holds the top controller output data. The sixth column holds the
bottom process variable data. Label these columns as such and fit a FOPDT model. If the fit is
acceptable, record the model below as the top decoupler disturbance model.

Bottom Decoupler:

Process Model (bottom composition response to bottom controller output)

KP = P = P =

Cross-Loop Disturbance Model (bottom composition response to top controller output)

KD = D = D =

In absolute value, if |KD| > |KP| , then the decoupler assigns more influence over the process
variable to the cross-loop disturbance than it does to the controller. This can lead to unstable
control. If your fits result in this situation, lower your KD above so |KD| = |KP|.

9) Return to the distillation simulation, select the top controller, choose PID with Decoupler from the
menu list, enter your process and disturbance models for the top decoupler, and click Done.
Repeat this for the bottom controller. For both loops, leave your original PI tuning values
unchanged.

With both the top and bottom loop under automatic PI Control with Decoupler, test the
process by stepping the top composition set point from 94.5% benzene up to 96% and back again.

Is your controller stable? Do the decouplers decrease loop interaction? Is the set point tracking
performance similar to when only the top loop was in automatic (step 5 above)?

10) Now test the benefit of decouplers on disturbance rejection. With the process at steady state, step
the column feed rate from 596 Kg/min up to 650 Kg/min and after the response is complete, step it
back again.

Do this both under PI with Decouplers control, and again under just PI control. Does decoupling help
with disturbance rejection?

46
Hands-on Workshop Series by Douglas J. Cooper
Copyright  2004 by Control Station LLC
All Rights Reserved
Workshop 14: Modeling and Decoupling of Multi-Loop Processes
Objective: To explore multivariable modeling issues and to learn how Design Tools and Custom
Process can be used to model, simulate and explore multi loop processes.
Reference: “Practical Process Control” Chapter 20-21

1) We learn here how Design Tools and Custom Process can be used to simulate and analyze 2x2
multivariable processes. We pretend that the Multi-Tank process is our "real" process, and we
explore how to create a simulation of it using Custom Process. The procedure detailed here is
important to understand because it is the same one you would follow on a real process in the plant or
lab.

To appreciate how challenging this workshop will be, select Custom Process from Control
Station's main screen and then choose Multi-Loop Process. Observe the process graphic and note
that a 2x2 multivariable simulation requires six models for implementation.
(Aside: A summary of the procedure we will follow in this workshop is:
a) At the design level of operation, collect controller output to measured process variable dynamic
data for loop 1 and loop 2. Also, if appropriate, collect disturbance to measured process
variable dynamic data for both loops.
b) Use Design Tools to fit a first order plus dead time (FOPDT) model to the controller output to
measured process variable data sets. Use the resulting model parameters to compute
controller tuning values for loop 1 and loop 2.
c) Use Design Tools to fit second order plus dead time (SOPDT) models to both controller output
and disturbance data. Implement the models in multi-loop Custom Process to create a simulation.
d) Implement the controllers in Custom Process and test them in tracking set points and/or
rejecting disturbances. Refine the design and tuning as needed.
e) When satisfied with the controller performance in Custom Process, return to the actual process
for final implementation.)

2) Close Custom Process and start the Multi-Tanks process. This will serve as our “real” process for
the study. The first step outlined above is to collect data at the design level of operation, which in
this study is the measured levels: tank 1 = 3.1 m; tank 2 = 3.0 m. Both disturbance flows are
normally at 1.0 m3/min and rarely change so we will not investigate disturbance modeling here.

Move the Multi-Tanks process to the design level of operation (hint: try a CO1 = 57 %/min and a
CO2 = 60 %/min). With the process at the design level of operation, record the controller output,
measured process variable and disturbance values. We will need these for the simulation:

Loop 1: u1,Design = y1,Design = d1,Design =

Loop 2: u2,Design = y2,Design = d2,Design =

3) Begin saving data to a file called PROC1.TXT. Step controller output 1 up and down from the design
level by about 5%, thus creating controller output to measured process variable dynamic test data.
Let the process show a clear response between steps. Stop saving data after the process returns to
steady state. Repeat by stepping controller output 2 in a similar fashion and saving data in a file
called PROC2.TXT.

47
Hands-on Workshop Series by Douglas J. Cooper
Copyright  2004 by Control Station LLC
All Rights Reserved
Workshop 14: Modeling and Decoupling of Multi-Loop Processes (cont.)
4) Exit the Multi-Tank process and return to the main Control Station screen. At this point we pretend
that we have just returned from the plant with dynamic process data on disk.

Start Custom Process and choose Multi-Loop Process from the list. Once the simulation has
started, click Pause to suspend activity and then navigate to Design Tools. Read in file PROC1.TXT,
fit a FOPDT model to the data and use the model parameters to compute PI tuning values using the
Standard tuning correlation. Record the results.

Repeat for PROC2.TXT. Be sure to label the proper columns

Loop 1: KC = I =

Loop 2: KC = I =

5) Next determine SOPDT models that best fit the dynamic process data and thus can be used to
simulate the process in Custom Process. Click the Select Model button on the Design Tools tool
bar and choose the SOPDT model form.

Read file PROC1.TXT into Design Tools. Label controller output 1 data as the manipulated variable
and measured level 1 as the process variable (use the Edit Data button if you are not sure which
columns of data hold the information you need). Fit a SOPDT model to the data and record the
results as direct loop 11.

Direct Loop 11: K11 = 1,11 = 2,11 = 11 =


CO1  PV1

For this same file, label controller output 1 data as the manipulated variable and measured level 2 as
the process variable. Fit a SOPDT model to the data and record the results as cross loop 21.

Cross Loop 21: K21 = 1,21 = 2,21 = 21 =


CO1  PV2

Read file PROC2.TXT into Design Tools. Label controller output 2 data as the manipulated variable
and measured level 2 as the process variable. Fit a SOPDT model to the data and record the results
as direct loop 22.

Direct Loop 22: K22 = 1, 22 = 2, 22 = 22 =


CO2  PV2

For this same file, label controller output 2 data as the manipulated variable and measured level 1 as
the process variable. Fit a SOPDT model to the data and record the results as cross loop 12.

Cross Loop 12: K12 = 1, 12 = 2, 12 = 12 =


CO2  PV1

48
Hands-on Workshop Series by Douglas J. Cooper
Copyright  2004 by Control Station LLC
All Rights Reserved
Workshop 14: Modeling and Decoupling of Multi-Loop Processes (cont.)
6) Compute the relative gain for this process:

K11 K 22
 
K11 K 22  K12 K 21

Based on this relative gain, what do you expect will be the nature of the control loop interaction?

7) Exit Design Tools and return to multi-loop Custom Process. To create the Multi-Tank process
simulation, click the Process 1 button on the graphic. The Multi-Loop Construct Process and
Disturbance Models input form will open. The SOPDT model is an overdamped linear model so
be sure that model is selected on the menu.

There are four models to enter from step 5 that correspond to the first four tabs on the input form.
Enter the process gain, two time constants and dead time for each model. Take care to enter each
model on its proper tab and each parameter in its proper box. Parameters not used in the model
should be zero.

Custom Process assumes as a default that the controller outputs, process variables and disturbance
variables have a minimum value of 0, a maximum value of 100, and a design level startup value of
50. As in this study, these values are not always appropriate for a simulation. In step 2 we noted
the design level start up values. The minimum and maximum values along with these start up
values are listed in the table below.

Click on the Zeros and Spans tab and enter the maximum, minimum and start up values for each
variable of the process as listed in the table:

CO1 PV1 D1 CO2 PV2 D2


Minimum 0 0 0 0 0 0

Maximum 100 10 4 100 10 4

Startup 57 3.1 1 60 3.0 1

When finished, click on Done to start the simulation. You now are simulating the Multi-Tank
process. Test the open loop dynamic behavior by stepping the controller output and observing the
measured process variable response. Return the process to the design conditions when you accept the
simulation as appropriate.

8) Implement your PI controller for control loop 1, put the set point at 3.0 m (which is very near our
design level of operation) and let the process steady. Leave control loop 2 in manual. Step set point 1
from 3.0 m up to 4.0 m and back again.

Leaving control loop 1 in automatic, implement your PI controller for control loop 2, put both set
points at 3.0 m, and let the process steady. With both controllers in automatic, step set point 1 from
3.0 m up to 4.0 m and back again.

49
Hands-on Workshop Series by Douglas J. Cooper
Copyright  2004 by Control Station LLC
All Rights Reserved
Workshop 14: Modeling and Decoupling of Multi-Loop Processes (cont.)
Do your responses look similar to first two set point steps from the actual Multi-Tank process shown
below (we have not done the decoupling case yet). If your simulation looks similar to that below,
you have successfully used Control Station to simulate this “real” process.

Co n tro l Statio n : M u lti-T an k Pro cess


P roc e s s : M ulti-T a nk P roc e s s L e ve l 1: P ID w / D e c ouple r
L e ve l 2: P ID w / D e c ouple r
L e v e l 1

4. 0
3. 5
3. 0
no decoupling
3. 5 controller 1 only with decoupling
L e v e l 2

3. 0

2. 5

0 200 400 600 800


T im e (m ins )
T uning: G a in = 19. 6, R e s e t T im e = 14. 4, S a m ple T im e = 0. 10
P roc e s s M ode l: G a in(K p) = 0. 066, T 1 = 8. 47, T 2 = 8. 47, T L = 0. 0, T D = 2. 50
D is turb M ode l: G a in(K d) = 0. 039, T 1 = 8. 65, T 2 = 8. 65, T L = 0. 0, T D = 2. 90
T uning: G a in = 17. 7, R e s e t T im e = 16. 2, S a m ple T im e = 0. 10
P roc e s s M ode l: G a in(K p) = 0. 07, T 1 = 9. 30, T 2 = 9. 30, T L = 0. 0, T D = 2. 90
D is turb M ode l: G a in(K d) = 0. 033, T 1 = 9. 00, T 2 = 9. 00, T L = 0. 0, T D = 3. 00

9) (Aside: A multivariable decoupler is a feed forward element where the measured disturbance is the
output signal of the cross-loop controller. To construct a decoupler, we need to identify a process
and cross-loop disturbance model for both loops (four models total). The models we developed in
step 5 are the models we seek.)

Using your SOPDT models from Step 5, implement decouplers on both loops. Select controller C1,
choose PID with Decoupler from the menu list, enter your process model (CO1  PV1) and
disturbance model (CO2  PV1) on the decoupler menu, and click Done. Repeat for controller C2.
For both loops, leave your original PI tuning values unchanged.

With both loops under automatic PI Control with Decoupler, test the process by stepping set
point 1 from 3.0 m up to 4.0 m and back again. Does your response look similar to the last set point
steps on the above plot?

10) Return to the Multi-Tank process and repeat the three cases shown in the plot above. Verify for
yourself that the Custom Process simulation indeed represents the behavior of this process.

50
Hands-on Workshop Series by Douglas J. Cooper
Copyright  2004 by Control Station LLC
All Rights Reserved
Workshop 15: Dead Time Compensation Using the Smith Predictor
Objective: To design, implement and explore the capabilities of the model-based Smith predictor on
processes with significant dead time.
Reference: “Practical Process Control” Chapter 22

1) Begin this study by constructing a linear process simulation using Custom Process. Click on the
Custom Process button on Control Station’s main screen and then choose Single Loop Process
from the list. When the simulation starts, click the Process button on the graphic. This opens a
“Construct Process and Disturbance Models” menu.

Click the Process Model tab (it should already be active if you have not done any exploring) and
select the Overdamped Linear Model from the list of models available. Enter these process
parameters to create a third order (three time constants) without dead time process:
Process Gain, KP = 0.8
First Time Constant, P1 = 70.0
Second Time Constant, P2 = 50.0
Third Time Constant, P3 = 10.0
Lead Time, PL = 0
Dead Time, P = 0
We will use the default Zeros and Spans in this study, but you should click that tab to confirm that
all parameters range from 0-100 with startup values equal to 50. We will not specify a model
describing the disturbance behavior right now so leave the default values in place.

When you are finished, click Done at the bottom of the menu to begin the simulation of your third
order without dead time process.

2) While saving data to file, perform a dynamic test at the startup operating conditions of the process
(CO = 50 and PV = 50). When the test is complete, navigate to Design Tools, fit a FOPDT model
to the data and record the PI tuning values using the Standard tuning correlations:

FOPDT Model Fit: KP = P = P =

PI tuning values: KC = I =

When you fit a FOPDT model to this third order without dead time process, does the model fit
produce a positive value for dead time? Why?

3) Return to Custom Process and test the set point tracking performance of this controller on your
third order without dead time process. Starting from steady state, make a set point step from 50 up
to 55. When the response is complete, click View and Print Plot. Using the plot options as needed
to help you visualize the response, compute and record the rise time and peak overshoot ratio for
this base case:
trise = POR =

51
Hands-on Workshop Series by Douglas J. Cooper
Copyright  2004 by Control Station LLC
All Rights Reserved
Workshop 15: Dead Time Compensation Using Smith Predictor (cont.)
We will call this peak overshoot ratio and rise time our base case performance (your POR should
be in the range of 15-25% and the measured process variable should settle completely to the new
set point in roughly one cycle of the measured process variable. If this is not true, you should
recheck your design.)

4) Explore the impact of dead time on controller performance. Click on the Process block on the
control loop graphic and add a dead time of 25 to your existing third order process. Click Done to
start the simulation. Do Not change the controller tuning.

Make a set point step from 50 up to 55 and from the screen plot, determine the rise time and peak
overshoot ratio for your controller:

trise = POR =

Comment on the change in these values due to the presence of dead time.

5) Put the controller in manual mode and use Design Tools to design a PI controller for this third
order plus dead time process using the Standard tuning correlations. Record both the PI tuning
values and the FOPDT model parameters computed by Design Tools. Also fit a SOPDT model
that describes the dynamic behavior of this process and record those parameters:

PI tuning values: KC = I =

FOPDT model: KP = P = P =

SOPDT model: KP = P1 = P2 = P =

6) Return to Custom Process and implement your updated tuning values. Make a set point step from
50 up to 55. If the peak overshoot ratio is not similar to the base case (the no dead time process of
step 3), adjust tuning by trial and error until it is. Record the peak overshoot ratio and rise time for
this third order with dead time process:
trise = POR =

Comment on how the dead time has impacted the performance on this properly designed
controller.

7) From the controller design menu, select the PID with Smith Predictor controller. A process model
form will open. Enter the FOPDT model parameters from step 5 as the predictive process model
for the Smith architecture.

For PI controller tuning, use the Standard tuning correlations. You should use the FOPDT KP and
P from step 5, but since a Smith predictor can theoretically eliminate dead time, you should use
P = 0.1P for the dead time in the correlation. This requires a hand calculation.

52
Hands-on Workshop Series by Douglas J. Cooper
Copyright  2004 by Control Station LLC
All Rights Reserved
Workshop 15: Dead Time Compensation Using Smith Predictor (cont.)
(Aside: There are no tuning correlations for a PI controller used in a Smith predictor architecture.
Since the Smith predictor theoretically eliminates the impact of dead time on control performance,
we use a theoretical minimum dead time in the correlations to compute PI tuning values. For
commercial equipment, the theoretical minimum dead time is the controller sample time. That is,
one sample time passes between a control action and when the controller receives the next
measurement. Since sample time, T, should be less than or equal to 0.1P, use P = 0.1P and the
process gain and time constant from the FOPDT fit of the process test data in the IMC tuning
correlations.)

Recall the Standard tuning correlation for the PI controller:

Standard Tuning:  C is the larger of 0.1  P or 0.8  P

1 p
KC = I =  P
KP ( P   C )

KC = I =

With these tuning values and the FOPDT process model implemented as the Smith predictive
model, make a set point step from 50 up to 55.

How does the rise time and peak overshoot ratio compare with the values you achieved in step 3.

trise = POR =

Don't forget that even in the perfect case, the rise time must increase by 25 because of the dead
time added to the process.

8) Perform a sensitivity study to determine which of the three FOPDT model parameters most impact
set point tracking performance. To do this, double and then halve first the model gain, then the
time constant and finally the dead time implemented in your Smith predictor model while keeping
the other two constant at the proper values.

After each change, perform a set point tracking experiment. Which of the model parameters most
impacts performance? What is its effect?

9) Replace the FOPDT Smith model with the SOPDT model from step 5. Use the base case tuning
values from step 2 (no dead time) for the PI controller. Repeat the set point tracking experiment.

Does the second order model provide any improvement to control performance? Is there any
impact on the trace of the controller output signal? Is trial and error tuning required to achieve the
design rise time and peak overshoot ratio?

53
Hands-on Workshop Series by Douglas J. Cooper
Copyright  2004 by Control Station LLC
All Rights Reserved
Workshop 15: Dead Time Compensation Using Smith Predictor (cont.)
Optional

10) Click the Disturbance button on the control loop graphic to open the “Construct Process and
Disturbance Models” menu. Enter as a disturbance simulation the same model used for the
process from step 1. Use the PI tuning (no Smith predictor) from step 2 and test the disturbance
rejection performance of the controller by stepping the disturbance variable from 50 to 55.

Click on the Disturbance block on the control loop graphic and add a dead time of 25 to your
existing third order process. Test the disturbance rejection performance of the controller by
stepping the disturbance variable from 50 to 55. Compare the result to that above. Does
disturbance dead time impact disturbance rejection performance?

Select the Smith predictor controller and input the FOPDT process model from step 5 and PI
tuning values from step 7. Repeat the disturbance test and draw a conclusion on the benefit of the
Smith predictor when used for disturbance rejection applications.

54
Hands-on Workshop Series by Douglas J. Cooper
Copyright  2004 by Control Station LLC
All Rights Reserved
Workshop 16: DMC Control of the Gravity Drained Tanks
Objective: To explore DMC design and implementation. Also, to explore how the tuning parameters
impact DMC performance.
Reference: “Practical Process Control” Chapter 23

1) Start the Gravity Drained Tanks process. For this DMC level control design, consider the case
where the design level of operation is a measured liquid level in the lower tank of 5.3 m. The flow
rate of the disturbance stream out of the bottom of the lower tank is expected to be about 2.0
L/min during typical operation. When your controller is put in automatic, the set point should be
at the design level of the measured process variable:

ysetpoint =

2) A DMC control element will be used in this study. The DMC controller element is comprised
tuning parameters and a step response process model that describes controller output to measured
process variable dynamic behavior.

Tuning requires a controller output to process variable dynamic data set. Start saving data to a file
and perform an open loop test by stepping the controller output from 80% down to 75% then up to
85% and finally back to 80%. Let the response become clearly established before making each
change. Turn file storage off when the response is complete.

3) Pause the simulation and navigate to Design Tools. Open the data file and verify that the first,
second and third data columns are labeled as time, manipulated variable and process variable,
respectively. Fit a first order plus dead time (FOPDT) model to the data and record the DMC
controller tuning values:

T= N = P= M= =
Note: Make sure that you pay specific attention to the time units that your data file is saved in and
the time units needed for the sample time on the Gravity Drained Tanks controller design menu.

Also record the FOPDT model parameters for use in constructing the step response process model
portion of the DMC controller. Include the SSE to gauge of goodness of model fit.

KP = P = P = SSE =

Later we will compare the benefit of different models on DMC performance. Select a second
order plus dead time model (SOPDT) and fit it to the process data. Record the results:

KP = P,1 = P,2 = P = SSE =

Based on the SSE, does the SOPDT model show improved capability in describing process
behavior over the FOPDT model?

4) Close Design Tools and return to the gravity drained tanks. Click on the controller icon on the
tanks graphic. At the top of the controller design menu, select the Dynamic Matrix Controller.

55
Hands-on Workshop Series by Douglas J. Cooper
Copyright  2004 by Control Station LLC
All Rights Reserved
Workshop 16: DMC Control of the Gravity Drained Tanks (cont.)
On the left side of the menu, enter the set point from step 1 and T, P, N, M and  from step 3.
Underneath the tuning parameters, click on the Specify Step Response Model button and enter

your FOPDT process model parameters. Leave parameters not included in the model as zero.
Click Done when you are finished inputting the step response model. And, then click Done on the
controller design menu to put the controller in automatic.

5) Test the set point tracking performance of this DMC implementation. Make step changes in the
set point from 5.3 up to 6.3 m and back to 5.3 m again. When the response is complete, pause the
simulation.

6) Enter your SOPDT model parameters for the step response model. Leave parameters not included
in the model as zero. Click Done when you are finished inputting the step response model. And,
then click Done on the controller design menu to put the controller in automatic. Make step
changes in the set point from 5.3 up to 6.3 m and back to 5.3 m again.

The SOPDT models (hopefully) had lower SSE values than the FOPDT models, implying better
model fits of the data. Does the SOPDT models show clearly improved DMC performance? Why
or why not?

7) Return to the dynamic FOPDT step response model parameters (Step 4) and perform a sensitivity
study to determine how each tuning parameter impacts the set point tracking performance. To do
this, first double and then halve the sample time and perform a set point tracking experiment after
each change. Keep all the other tuning parameters and step response parameters at their original
design values. Repeat this procedure for the prediction horizon, model horizon, control horizon
and move suppression coefficients.

Also, perform a sensitivity study to determine how step response model parameters impacts the
set point tracking performance. To do this, first double and then halve the process model gain and
perform a set point tracking experiment after each change. Keep all the tuning parameters and
other model parameters at their original design values. Repeat this procedure for the process
model time constant and finally the process model dead time.

Which of the tuning and step response model parameters most impacts set point tracking
performance? For each model parameter, determine if it is more conservative (less aggressive) to
have it be too large or too small.

8) Explore the benefit of constraints by entering these constraints into the controller design menu:

1.0  u  1.0
80  u  87
1  ŷ  9

What affect do they have on the response of the controller output and the process variable?

9) Explore the benefit of DMC control on disturbance rejection performance. Step the disturbance
from 2.0 L/min up to 2.5 L/min and back again. Does the DMC controller offer any benefit in
disturbance rejection performance.

56
Hands-on Workshop Series by Douglas J. Cooper
Copyright  2004 by Control Station LLC
All Rights Reserved