What happens when you combine wireless transmitters and valves with model-based predictive control (MPC)? Can you maintain close control of a relatively poorly characterized piece of process equipment like a divided wall distillation column (DWC)? Researchers at the University of Texas at Austin are about to find out.
Built in 1984, UT's 6-in.-diameter, 30-ft.-tall DWC has 19 temperature sensors and 28 stages. Adding a vertical dividing wall prevents the feed from mixing with the product, allows the draw of side streams in addition to tops and bottoms, and offers the potential for saving 30% of the energy for conventional distillation. But, "Adding a wall allows a lot of degrees of freedom and more interactions," said Bailee Roach, PhD graduate student at UT, Thursday morning at Emerson Global Users Exchange 2014 in Orlando, Florida. "It's harder to control."
UT has been working with Emerson to tame the DWC with MPC, and Roach's research will now include the effects of using wireless to control valves and heaters on four liquid flow loops and two temperature control loops. The valves and heaters are set up to allow comparison of wired or wireless measurement and/or wired or wireless throttling. "We're not afraid of wireless for controlling critical processes," Roach said.
The principals behind MPC with wireless are simple, explained Willy Wojsznis, senior technologist, Emerson Process Management. Compared to wired connections, wireless offers longer and sometimes varying scan intervals. "In MPC, a process variable is compared to a predicted value, the error is measured and the model is corrected as needed," Wojsznis said, "Control works on the predicted value."
With wireless, the controller is set to retain the last value until a new value is available. Instead of the one second typical of wired networks, wireless devices typically update at 8, 16 or 32-second intervals to conserve battery life. "So we do not provide a correction until a new value is available," Wojsznis said.
This is not a lot different from what happens when a conventional measurement fails. "In MPC, we simulate a value for up to a defined maximum period," Wojsznis said, which may be as long as 2 minutes or more. A similar approach is taken with lab data—a lab value is held until a new lab reading is available.
"So wireless is not a lot different," Wojsznis said. "We set the input value as constant until a new value is transmitted. After a status change, the status is 'good' for more than one MPC scan. After that, hold constant for up to a maximum period of time."
In a simulation using an MPC function block for the DWC, researchers tested update rates 8, 16 and 32 times longer than the MPC execution rate. Response curves show "a small bump every time the variable changes. Otherwise, it's the same as wired, with no instability," Wojsznis said. "Step response performance is less than 10%, which is acceptable."
Having a good model is more important for wireless, because "a significant part of the time, MPC works on the model with no correction," Wojsznis said. "But in the real world, MPC execution is not once per second, it might be once per minute. So it's not as different as it might seem."
At UT, testing will begin this fall using water, with hydrocarbon trials coming next year, Roach said. "Look for our results at next year's Exchange."