in control, the issue arises of what defines a truly “advanced”
control. Advanced control involves distributed and supervisory control
above the basic level of regulatory control. Closed loop PID controllers
provide basic regulatory control. Also, simple cascades of a process
variable to flow control,
such as temperature setting to flow control, are regulatory controls.
Also, there now exist a category of controls that were previously
known as “advanced” that are now generally described
as augmented regulatory controls. These include nonlinear level
controllers, dead time compensators, pressure-compensated temperature
controllers, and cascades to a non-flow secondary control variable.
Today’s advanced process controls usually involve composition
and constraint control. They can be implemented as a number of individual
controllers or as part of a large multivariable model-based predictive
controller. The recent trend has been towards larger controllers
and away from numerous smaller controllers. By integrating controls
into a single controller, interactions across an entire unit or
plant can be better managed. These large model-based controllers
might be built in a modular fashion with use of sub-models that
can be deactivated if problems with sensor data occur.
In the early 1990’s,
the typical multivariable predictive controller handled about 12
controlled or constraint variables using about 6 manipulated variables
or control valves. Recently, successful applications have been reported
that look at 400 variables and move 200 valves.
In the future, neural network models and expert systems will be
gradually integrated into both advanced
control and optimization applications.
As the ability to precisely control the plant
improves, interest naturally shifts to making sure that one
has the right control objective. Exact control to the wrong target
does not satisfy any process control engineer. Optimization involves
any method to determine the feedstock selection or process operating
characteristics that obtain maximum profit and/or minimum cost.
The interaction of linear program
models used for off-line plant-wide optimization with on-line