Chemical Engineering November 2010 - 51

Plant
DCS
Process monitoring
Dynamic
optimization
Historian
Simulator
New
set
points
DYIL
model
Figure 1. These
typical connections
between a simulator
and DCS are possible
today, and are no longer
restricted to high-end
integrations
Steady state (SS)
optimization
Simulator
SS
model
Dynamic
optimization
Simulator
SS
model
Analysis program
(Excel, LP, etc.)
Trouble
shooting
information
Simulator as OTS or online future
predictor - advanced process
control (APC) in a water tank. A
mixed-integer linear-program (MILP)
controller was developed to optimize
control of a valve in a water purification
system. Four large tanks were
connected in series, with a single
control valve in the series. Liquid
level differences provided the driving
force for flow between the tanks. Excessive
operation of the valve would
cause increased turbidity in the water,
complicating water treatment, so the
valve needed to change position at a
slow rate. The MILP was developed
with the goal of predicting the future
setpoint and moving the valve earlier
than with traditional control.
A dynamic model of the network
ers. Such behavior was experienced
when the second branch was brought
online; manual tuning was required.
A dynamic simulation for the twoloop
system was developed. Historical
data were used to validate the
model. The model was tested to see
if it produced the same response that
occurred during 24 hours of actual
operations. The use of historical data
led to a high level of confidence that
the simulation model could be used to
predict system response.
The first step for this project was to
make a steady-state process simulation
model that performed hydraulic calculations
to solve pressure and flowrates
simultaneously. Next, a dynamic simulation
of the model was run, using the
proportional-integral-derivative (PID)
terms from the actual equipment. Historical
data for end-user demand were
incorporated into the model, allowing
calculations at each time step to predict
pressures based on actual flow. The
control system reacted to the pressures
throughout the network. Historical
data on valve positions, controller output
and pressures in the network were
compared with the simulated results.
As a final step, the third branch was
added to the model and used to develop
the control logic. Upon commissioning,
the behavior was found to be
well predicted by the model.
Online software sensor. A chemical
producer integrated a steady-state
simulator as a " software sensor " for its
control system. Real-time data from
the DCS was sent to the simulator
model. The model performed calculations
using the sensor data as input
specifications, and results were sent
back to the DCS as tag values. Results
included temperatures in the
process, conversion at a reactor and
recommended makeup water for the
final column. This information was
displayed to the operator panel, and
recorded in the data historian.
Use of a process simulator as a sensor
significantly reduced the amount
of required instrumentation, and
the number of analytical samples
required. Use of the simulator as a
software sensor led to improved control
of the process, reducing off-specification
product.
Data reconciliation and online
sensor validation in a heat exchanger
network. A company
wanted to calculate fouling factors
throughout a heat exchanger network.
The control system provided
values from the flowmeters and thermocouples
in the network.
The company's engineers developed
a simulation to perform rigorous heat
exchanger calculations, based on geometry
of the exchangers. Data reconciliation
was used to adjust the simulation
to best fit the sensor data. From
this baseline case, the fouling factors
were calculated by using current sensor
data and adjusting performance of
the exchangers. Additionally, the results
of reconciliation suggested that
sensors might need calibration.
was developed, including the APC
control logic. A custom interface was
programmed for an operator training
system, using the model as a calculation
engine.
Using data from the DCS, the model
was used to " predict the future " based
on current conditions and assumption
of typical demands for the next several
hours. This approach improved operator
confidence in the control system;
before changing control to manual,
the operator could simulate the future
and see what might happen if he or
she allowed the control system to do
its work.
How to make the connection
Develop a simulator model. Developing
a high-fidelity process simulation
model is the first step. A high-fidelity
simulation is one that is able to
calculate results similar to observed
results, based on rigorous specifications
and geometry calculations. For
example, using outlet temperature
as a heat exchanger specification is
a low-fidelity approach; using a heat
transfer coefficient (or calculating it
rigorously from geometry and fluid
properties) and calculating outlet
temperature is a high-fidelity approach.
For pumps, a high-fidelity
model specifies a performance curve
(head developed versus volumetric
flowrate). For control valves, a highfidelity
model specifies valve Cv (sizing
term) and position.
When a valve is opened on a highChemiCal
engineering www.Che.Com november 2010 35
http://www.Che.Com
Chemical Engineering November 2010

Table of Contents for the Digital Edition of Chemical Engineering November 2010
