Breaking from Tradition:

Model Based Control vs. PID

AUTHORS:
STEVE NIXON AND MEGAN KORNOWA-WEICHEL

ADVANCED CONTROL SOLUTIONS, INC.

8750 RESOURCE PARK DRIVE
SYLVANIA, OHIO 43560

Page 2

Breaking with Tradition

W H Y

M O D E L -

B A S E D

N O T

P I D ?

H O W

C A N

I M P R O V E
P R O C E S S
C O N T R O L ?

New ACSI model-based controllers are outperforming PID controllers in

glass applications.
The ACSI-integrated process controller quickly
responds to process disturbances, minimizes job change time and reacts
quickly to stabilize temperature variations. The benefits to you are
lower production costs, improved product quality, and ultimately
increased profits.

Controlling glass temperature is key to

achieving optimum glass viscosity and gob
weight.
Even the smallest temperature
variations can negatively impact the quality
of the finished product and result in lost
production time. If we can model zone
temperatures and reduce job change time,
we can achieve tighter control.

Benefits of
Model-Based Control

Reduced process
settling time on startups
or process changes
Increased product
consistency and quality
Reduced energy use
Operation closer to
specifications
Simplified operation and
maintenance

In order to understand the process, we must

first know the basics. The Process Variable
(PV) is the property being measured by the
control system, such as temperature,
pressure, or level measurements.
The
Setpoint (SP) is the desired value for the PV.
The Output is the action the PID loop takes to correct an error, such as
changing valve positions, damper positions, charger speeds, or motor
speeds.
The ACSI model-based controller effectively controls melter and
forehearth temperatures by modeling the existing process.
The
controller creates models for each control/process variable and feed
forward input. These ideal models allow the system to anticipate
changes needed to maintain consistent glass temperature.

Breaking with Tradition

Page 3

What is the Difference Between PID and Model-Based

Control?
Traditional PID Control relies on an Engineer determining the proper
values to be used for the Proportional (P), Integral (I), and Derivative (D)
constants. The engineer must then continuously evaluate the process
and tune the constants. Model-Based Controllers are able to learn how a
process responds to changes, and in turn, they can automatically make
the tuning adjustments that would traditionally require an engineer.
The table below illustrates key differences between these two types of
control.
PID Control
Does not recognize dead
time
Requires manual control
Uses feedback only

Model-Based Control
Recognizes and manages dead
time
Adapts automatically to process
changes
Uses feed forward as well as
feedback

Consider the following everyday example to compare and contrast PID

and Model-Based Control.
You have a very delicate recipe that requires the liquid to remain at
90. The PID controller may be compared to a new trainee, and the
Model-Based controller can be compared to an experienced chef. Each
gas stove alerts the chef when they exceed the temperature limit.
If the trainee follows the process used by a PID controller, he pours in
the liquid and begins to turn up the gas on the stove. He continues to
turn up the gas until he receives feedback that the temperature has
exceeded 90. Because he does not understand the concept of dead
time (the time before a change is observed) he overshoots the desired
temperature. He then reacts to the situation and manually adjusts the
gas higher and lower while the temperature oscillates, moving above
and below the setpoint. He eventually reaches 90.
In the mean time, the experienced chef realizes that there will be a
period of time between when he turns up the gas and the time the
change is measured by a thermometer. He accounts for the dead time
and prevents oscillation, arriving at 90 with ease.

Breaking with Tradition

Page 4

Now the recipe calls for the temperature to be raised to 100. The
trainee has tuned his controller to maintain 90; therefore, he will have
to slowly raise the temperature in order to avoid overshoot. The chef
has years of experience and has learned how the stove responds. He
knows exactly how much heat must be added to raise the temperature
10; therefore, he arrives there much faster.
The next step is to add more water to the pot. The trainee will again
guess how fast the water should be added and manually adjust the
amount of gas. The chef will take into account feed forward inputs,
such as the temperature of the water being added, density of the water,
and ambient temperature in the room to predict exactly how quickly he
can add the water and how much additional heat is needed. He will
always be able to automatically adapt to changes in the rate at which he
pours or the ambient temperature and still maintain 100.
The experience of the chef allows him to predict the behavior of his
stove and the contents in the pot. He can quickly and effectively finish
his recipe, and in turn, the restaurant is able to serve customers in a
timely manner. In much the same way, effective processes allow glass
plants to run with fewer upsets and less lost production time.

Breaking with Tradition

H O W

I S

T H I S

A P P L I E D

T O

G L A S S
M A N U F A C T U R I
N G ?

Page 5

The everyday situation described above can be translated to glass

manufacturing using the example of a forehearth zone.
During installation an ACSI engineer uses a detailed procedure to
teach the controller how to model the process. For example, the
controller learns that for every 10% it puts in, a certain reaction occurs.
This information is used to take the best series of actions to control the
temperature. The better the controller is trained, the better the
model, and the more effectively the controller will manage the
temperature. The controller continues to learn and refine itself in order
to have the most reliable roadmap when the next change is required.
It is then able to understand the process, adapt to changing conditions
or production rate changes, and predict the effect a certain situation
will have on the process all without requiring an engineer to tune the
loop.

Managing Dead Time

PID controllers do not understand dead time and cannot take it into
account when controlling temperature. Often, the result is oscillation,
where the temperature overshoots the setpoint, corrects, goes below
setpoint, corrects, and goes back above. Model-based control does
understand dead time and takes it into account when making
adjustments. During the learning process, a model-based controller
learns what the typical dead time is and uses that knowledge to know
how long to wait before continuing with adjustments.

Feedback vs. Feed Forward

PID controllers use feedback to control glass temperature, where modelbased controllers use both feedback and feed forward input. Using the
example of a forehearth, a PID Control system will have sensors at the
exit of each forehearth zone. Each sensor feeds information back to the
PID controller, which then adjusts the heating or cooling to bring the
temperature back to setpoint. Because glass has already been flowing
through the forehearth, the system must always try to catch up with the
desired temperature. This process has a long recovery time during
which production values decrease. With PID the error is not detected
until the glass reaches the zone exit. At this point, a problem already
exists, and the controller must try to catch up to the correct
temperature. In addition, the engineer must learn how the system
reacts to different conditions and tune the loop accordingly.

Page 6

Breaking with Tradition

In contrast, Model-Based Control can use both feedback and feed

forward input to predict what will happen and make the adjustments
needed to remain at setpoint. Feed forward provides the controller with
information that allows it to anticipate the changes that will be needed
to maintain consistent glass temperature. The controller can look at the
incoming temperature as feed forward and analyze how it will affect the
process. If it knows the temperature will be too hot when it gets to the
zone entrance, it can reduce the heating to avoid the error. Instead of
waiting for feedback from an instrument, model-based control analyzes
the feed forward input and makes the adjustment before receiving the
feedback.

Managing Job Change Time

In order to maximize profit potential, it is essential to minimize job
change time. New glass temperature setpoints must be achieved as
quickly as possible with minimum overshoot. It is difficult to achieve
both objectives simultaneously with a standard PID controller. One of
two scenarios is likely to occur. The first scenario is quickly raising the
glass temperature. In this scenario, the temperature often overshoots
the optimum SP and must be adjusted back to SP. The second option is
to gradually raise the temperature, which requires a long period of time.
Both scenarios typically require several hours to stabilize the
temperature during which production operates at less than optimal
parameters.
With model-based control, once the optimum process is modeled, the
controller can predict what actions are needed to reach setpoint quickly
without overshoot.
Disturbances

Targets

Model Based Manip. Input

Controller

Updated Model/
Disturbances

Controlled Output
Process
Measurements

Model

Disturbance
Estimation and/or
Model Adaptation

Breaking with Tradition

Page 7

Model-Based Control

PID Control

Models the existing process, learns,

adapts, and predicts.

Works on error from setpoint.

Requires an operator to learn what
action needs to be taken, take a
reading, then act.
A sensor at the outlet of each
forehearth zone measures glass
temperature as it exits the zone. The
sensor then relays data to the PID
controller, which adjusts heat to bring
temperature back to SP. As glass
travels through each chamber, the
controller continues to play catchup.
There is a long recovery time during
which production values decrease.
Requires the operator to react to
readings and tune the loop.

Models the existing process by creating

models for each control/PV and feed
forward input. The ideal model allows
the system to anticipate changes
needed to maintain consistent glass
temperature. First, the ACSI engineer
teaches the controller while onsite at
startup. It then continues to learn.
Adapts to process and production rate
changes automatically for better
control without loop tuning.
Models feed forward inputs and
updates control actions to quickly
stabilize temperature variation.
The controller can predict control
actions required to drive the glass
temperature to SP quickly without
overshoot.

Glass temperature can be raised

quickly, but the temperature
overshoots the optimum SP and must
be adjusted back to SP.

Continues to refine itself, so that it

has the most efficient roadmap the
next time a change is needed.

Glass temperature is achieved with

gradual rise, which requires a long
time to reach SP.

Calculates dead time and takes it into

account when making adjustments.
Can look at incoming temperature as
feed forward and analyze how it will
affect the process. For example, if
the glass temperature will be too hot
when it arrives, it can back off and
avoid the error.

Does not understand dead time.

Without feed forward, the error is not
detected until the glass reaches the
zone exit. At this point, a problem
already exists.