The document discusses minimizing the use of model predictive control (MPC) systems and instead relying more on regulatory controls at the distributed control system (DCS) level. It argues that MPC systems often underperform and are more complex and expensive to maintain than DCS controls. For many distillation column applications, the standard DCS controls can achieve similar or better performance than MPC without added complexity. Less than 15% of installed MPC systems are actually providing benefits over what standard DCS controls can deliver. Simpler regulatory controls at the DCS level may be sufficient for many applications currently using MPC.
Apc more on apc designs for minimum maintenance (2009)
1. HPIN CONTROL
ALLAN KERN, GUEST COLUMNIST
kernag@yahoo.com
More on APC designs for minimum maintenance
Perhaps Zak Friedman struck the perfect balance in his series
of editorials,' because while Dr. I^iour thinks Zak went too far in
his suggestions to prune muUlvariable control (MPC) matrices, ' I
would say he didn t go far enough. For many MPCs, 1 would go a
step further and consider abandoning the MPC altogether in favor
of well-designed DCS-level regulatory controls. This will often
result in improved economic and control performance.
This may sound like strong medicine to the MPC genera-tion,
but in my opinion, the practical limitations of MPC over
the past 20 years have gone greatly under-reported. In receni
years, concerns are finding voice regarding "maintenance" and
"sustainability" issues, but a closer look will reveal that many MPC
applications have underperformed .since day one, nor withstand-ing
the great tradition of "successful" project completion rituals.
Fig.l shows typical distillation column regulatory controls.
MPC, whether minimalist as Dr. Friedman recommends, or more
"maximalist" as has been the industry habit, has litde more to con-tribute,
except expense and operational complexity. Some of the
common breakages between the conventional wisdom and actual
practice of MPC-based distillation control are as follows:
• Q)lumn pressure is rarely a suitable MV. It is best kept steady to
preserve the composition profile on which overall operation depends.
At most, pressure can "float" on the condenser to capture the Iienefits
of minimized pressure or to gracefiilly handle a cooling limitation.
This behavior is inherent in the controls depicted in Fig. 1.
• MPC typically includes feed rate as a disturbance variable.
This is inherent in heat-to-feed ratio control and this feedforward
can he applied to the reflux, too, if desired.
• Inferentials, while commonly perceived as a value-added
component ofMPC, are easily incorporated into regulatory con-trols
without MPC. For example, an inferential-based top com-position
controller can be cascaded to, or simply replace, the cop
temperature controller.
• The controls in Fig. 1 respond naturally in the correct man-ner
to saturated reflux, pressure or steam valves, something MPC
practitioners often suggest is value-added.
• Many MPCs are configured for dual top and bottom tem-perature
control, but few if any succeed. In my experience, they
behave in the same unstable manner as regulatory controls con-figured
this way and are quickly defeated by clamping either the
reflux or reboiier MV.
This discussion uses the example of distillation columns, but
a closer look at other common MPC applications leads to similar
conclusions. Most handles, overrides and feedforwards reflected
in MPC matrices, while impressive and promising on paper, prove
impractical or unwanted in operation. In most cases, they are soon
"defeated" by adjusting A/K and controlled variable iCV) limits,
ulting in the well-knnwn condition of high service factor, but
utilization. The small numher of remaining "active" models,
cases, could be much more easily captured with basic
Près,
ourd.
Typical distillation column controls
DCS-level controls, sans all the MPC hoopla.
While MPC remains a sound and often tantalizing technology
in principle, its practical applicability and succes.s rate are not nearly
what popular wisdom might have you believe. I would hazard a
guess that less than 15% of installed MPCs are earning money by
doing something regulatory controls can't do bfttt-r. That means
there's a whole lot of unnecessar)' MPC activity going on, HP
Dr. Friedman's response.
Mr. Kern has a valid point about the balance between APC
complexity, maintainability and potential versus real benefits. The
quest to simplify APC should also consider what can be done by
advanced regulatory control.
When the application dynamics and constraints are simple,
advanced regulatory control can deliver benefits. Do remember,
however, that when an application has multiple constraints and
complex dynamics, implementing it in the DCS does not make
it simple and maintainable. Simplify the problem definition first,
and then choose the appropriate application platform.
LITERATURE CITED
' Friedman, V. 7,.. HP In Omirul Pans 1-3, Hydrocarbon Procesting. June-
August mm.
^ Latour, V. R., HP In Control Pan 1, Hydrocarbon Processing. October 2009.
The a u t h o r has 30 years ot process control experjenct?. including over a decade
as MPC group leader at a ma|or Middle East refinery, and has authored numerous
articles on process control effectiveness. He is a professional engineer (¡nacUve), a
graduate of the University of Wyoming and a senior member of ISA.
LMBER2009 HYDROCARBON PROCESSING