UT researchers, together with Emerson subject matter experts, are using predictive analytics to infer stream compositions in real-time, and they described a system for developing and deploying models, and evaluating inferred measurement performance, during a session entitled "Utilizing Predictive Analytics for Inferred Measurements" this week at the Emerson Global Users Exchange in Denver.
Collaboration Elevates Expertise
"Over the years Emerson Process Management has worked with the researchers and technicians at UT Austin, and it has been great opportunity to have an actual site close to our development center to work with," said Mark Nixon, lead of the applied research group at Emerson Process Management. "Quite a bit of the work has to do with separation research, with our part providing help with control and measurement technology such as this inferred measurement in the innovative DWC distillation process."
"We added a lot of wireless to the column, including 27 WirelessHART devices," explained Nixon. "This enabled wireless control of four liquid flow loops and four temperature loops using Emerson PIDPlus technology. A DeltaV system was installed in 2001, and pretty much every new innovation in DeltaV has shown up at this site."
There are several different research areas. "Two 18" columns can be used together or standalone can be used for distillation, CO2 absorption/stripping or liquid-liquid extraction," said Robert Montgomery, technician at the University of Texas. "We use them for verification of separation, scale-up and trouble-shooting of processes."
"Just as an engineer builds up control strategies using a set of control modules and layered interfaces, we are breaking down the analytics in a similar way." Emerson's Mark Nixon discussed ongoing research that will enable processors to more easily use data analytics to infer compositional measurements—without waiting for lab results.
The DWC in particular, is used to separate three components in a single operation, said Bailee Roach, grad student at the University of Texas. "With standard distillation techniques, it requires two distillation columns, two reboilers, two condensers," she said. "With the DWC, you can separate the third component from the same column. Not only can you save on equipment cost, you can save energy as well. We estimate a 30% to 50% energy and equipment cost saving by using a DWC."Explore the Data
"In the past we really relied on off-line analysis of column samples, but it could take up to 2.5 hours to know where we were in the process," said Roach. For more real-time control adjustments, they're now relying on inferred measurements based heavily on data analytics.
To facilitate visualization and analysis of column data, the researchers are using the Seeq data analysis platform, which is designed specifically for bringing together time-series data such as that generated by the DWC. "It gave us the ability to explore combinations of lab and process data.
The research team is also pioneering the development of a new "analytics studio" for the DeltaV platform. "Just as an engineer builds up control strategies using a set of control modules and layered interfaces, we are breaking down the analytics in a similar way," said Nixon. "You have analytical control blocks that can be used to gather, clean and manipulate raw data, then generate predictive models." Operators, in turn, can use the models to predict stream composition without waiting for lab analysis.
Although inferred measurements are not yet a replacement for lab measurements, they do provide an early indication on where the process is headed, Nixon said. The combination of Seeq software with the DeltaV system provides a platform to capture and explore data, develop and deploy data-based models for product property prediction as well as fault detection, identification and diagnosis.
