[ Follow Ups ] [ Post Followup ] [ Control Consulting ] [ FAQ ]
Posted by Mike McCarty on February 10, 2000 at 19:59:19:
When can you stick with advanced PID control schemes on your process, and when do you need to go with a full MPC application?
The best applications for Multivariable Predictive Controllers (MPC) are coupled processes (highly interactive), with difficult dynamics (meaning significant dead-times and long lag times), and a significant potential economic benefit to be gained by pushing the unit to multiple constraints. The first two issues make it difficult to configure PID control systems to do the job properly, even with proper PID designs, Feed Forward configurations, Lead/Lag Compensators, Ratio controllers, etc. Although much process control improvement and stabilization can be done with these types of changes. Also, MPC systems are really only economically justified when allowed to push the unit to operating constraints - control stabilization is not enough. When there are multiple constraints to push the unit against, an MPC controller will do a better job than DCS override controllers that are designed to push one process variable.
Highly coupled processes, even as "simple" as a distillation column controlling top and bottom temperatures using reflux and reboiler flows are not easy to control with PID controllers. To do this on the DCS system you need to either seriously detune one of the controllers to reduce the amount of interaction, or install a decoupler configuration so that each PID output essentially moves both flows and the effect of both flows are taken into account on both PID controllers. It is questionable how a decoupler is simpler than a 2x2 MPC controller. And the decoupler can't address a DP limit to prevent flooding on this column.
Process dynamics with a lot of dead time and long lags are inherently difficult for PID's to control tightly. To help a feedback-only PID controller on this type of system, you would need to add Feed Forward information to help prevent unmeasured disturbances to the variable it is controlling. However, it still can not be tuned as aggressively as it could be on a fast responding process. Slow dynamics are also difficult for operators to control. An operator can not push his (slow) unit aggressively towards its true limits, because it is too easy to overshoot the target. As a result, process limits and product specifications must be approached slowly since there is less of a penalty on an operator for running under a spec limit than overshooting in his zeal to push the unit.
The final requirement for a good MPC application is an economic incentive to push the unit toward multiple constraints. An MPC that just controls the unit at operator entered setpoints, while perhaps doing a better job than PID controllers, will not make very much money. Also, an override PID controller on the DCS system can do a good job of pushing one manipulated variable to one or more constraints. However, this gets more complicated when you have an incentive to push more than one manipulated variable. For instance, an application to maximize feed and minimize pressure on our simple distillation column - while honoring the limits of feed hydraulics, column flooding, and pressure control valve position. An override PID controller could push either one of these MVs up to any of those limits. But an override system would have trouble pushing both feed and pressure, and would definitely not be able to figure out the economic trade-off between feed and pressure when the column is at a flood limit but has room on both feed hydraulics and pressure valve position.
The final point is that to achieve these economic benefits properly, the MPC system should be configured with an optimizer driver (LP or QP) that can take into account price economics and process gains and limits. Configuring an MPC to "push" the unit by setting a few CV limits higher than the controller can achieve will cause it to push some of the MVs all the time - but this may not be the most economical solution when different or unusual operator limits are in effect. Most of the popular MPC technologies now offer some type of economic optimizer.
And the next final point would be that these MPC applications are complicated, and difficult to configure properly to achieve good unit control with sustainable economic benefits. The experience of the control engineers who are doing the modeling and configuration is much more important to the ultimate success of the project than the specific technology used.