|
Build Virtual Sensors with
Process Modeler Put Them On-Line with
Process Intellect
Take Action
 |
Not
Finding What You are Looking For?
Tell Us!
(that's how we improve) |
|
| Please note this page is ancient. The current version of Intellect is 3.0 and is found at our IntelliDynamics site. |
Oil
Production:
The virtual sensor plot on the right shows the estimated (blue)
and actual (red) production from an  oil platform.
It uses a variety of surface measures, mostly casing, tubing and flow line
pressures, to estimate total liquid production. Data shown is
"out-of-sample" (actual vs. estimated production on data not used to build the
virtual sensor) and reflects true performance (98.83% accurate). This is
not the most accurate model possible, but one built in about 10 minutes work
using the
Process Modeler.
This virtual sensor, now built, does not rely on any flow measure hence forward.
Distillation:
This virtual sensor shows the estimated (blue)
and actual (red) distillation
product property downstream from a distillation column.
A model of this type is in use 24 hours a day, 7 days a week
estimating product performance between sensor readings and
while the sensor is removed. Data shown is
"out-of-sample" (actual vs. estimated production on data not
used to build the virtual sensor) and reflects true
performance (99.59% accurate).
Fluidized Bed Drying:
Moisture from Lab Results
This virtual sensor shows the estimated (blue)
and actual (red) percent moisture
(water) in product coming out of a fluidized bed dryer.
The remarkable thing about this virtual sensor is that it uses
NO WATER INFORMATION as an input, but just screw feed and
conveyor amps, dryer temperatures and pressures. Data
shown is "out-of-sample" (actual vs. estimated moisture on
data not used to build the virtual sensor) and reflects true
performance (98.1% accurate). The available data is
limited in quantity which is why you only see 12 data points
in the out of sample results.
NOx Emissions
This virtual sensor shows the estimated (blue)
and actual (red) NOx emissions
per million BTUs for a gas and oil fired power plant. Data
shown is "out-of-sample" (actual vs. estimated moisture on
data not used to build the virtual sensor) and reflects true
performance (93% accurate). Now granted the customer has
an NOx emissions sensor in the stack and can calculate the NOx
Per MM BTU, but this virtual sensor can back up the physical
sensor in case of fouling or failure and the difference
between the virtual sensor and the physical sensor can be
monitored for abnormalcy, leading to alerts for investigation
or repair.
How Do You Build
and Implement Virtual Sensors?
You gather data that relate to the
predicted result. You
then build and validate a mathematical model of the relationships using
Process Modeler, our
Virtual Sensor builder (request
a free evaluation here). You then place the model on-line using
Process Intellect.
Process Intellect supports unlimited sensor implementations and can estimate
properties sub-second, up to the performance limitations of the computer.
Note: All data shown here is
REAL, but has been modified so as to not reveal any customer information.
|
