Chemical Engineering October 2011 - 6

Letters
Process automation: PID versus MPC
In his article Today's Process Automation Worlds (CE,
August 2011, pp. 30-38), Mr. Cecil Smith states that the
derivative mode in a proportional-integral-derivative (PID)
controller in automatic tuning would require model-based
approaches to consistently tune the mode, and then he
goes on to indicate that the statement itself raises another
question: " Once a process model is available, why not use
a single-loop MPC [model predictive control] instead of
PID? " .
MPC is NOT a substitute for regulatory controls.
MPC for a single-loop is a Smith predictor, and that was
invented over 20 years before the MPC; and the performance
of a single-loop model-based controller just can't be
better than that of a properly tuned PID.
It is important for your readers to understand that
MPC is not a substitution for regulatory controls.
Best regards,
Sigifredo NINO
Sr. Process Control Consultant
Author responds:
Target Optimal
Performance
Engineers inherit air systems where reliability, integrity
and countless devices determine the fate of productivity.
SPX Pneumatic Products' Legendary Class products
feature AMLOC® energy management technology,
moisture hungry DE-4 desiccant beads, low-pressure
drop, RS-485 communications, and application
engineered non-lubricated valves.
Enjoy demand-matched energy savings, rugged
durability, reliable performance and long life. Specify
SPX Legendary air and gas treatment systems for the
toughest applications in the nastiest of environments.
Process testing is routinely used only in the two examples
cited by Mr. Nino:
1. MPC as applied to multivariable interacting processes,
usually in conjunction with optimization
endeavors: PID proved to be ineffective for implementing
changes proscribed by optimization, which
spurred the development of MPC to completely replace
PID for this purpose.
2. Smith predictor or dead-time compensator as applied
to processes such as commodity paper machines. PID
is retained, but dead-time compensation provides the
improvement in control performance.
It is interesting that where process testing is being
applied, model-based control either replaces or supplements
PID.
SPX Flow Technology
Tel | 352 | 873 | 5793
pneumatic.products.sales@spx.com
www.pneumaticproducts-spx.com
www.spxft.com
Circle 49 on p. 78 or go to adlinks.che.com/35070-49
6 ChemiCal engineering www.Che.Com oCtober 2011
Properly tuning PID is a challenge, a major factor being
that PID is usually installed on slow processes, such as
reactor temperatures where bumping the process for the
purpose of tuning will not be tolerated for too long. We
have known how to tune PID based on models from process
tests since the late 1960s, but this has not become
routine practice and probably never will.
Although not the primary market for today's commercial
packages, MPC methodology is capable of regulatory
control. PID relies on a simplistic predictor based on the
current rate of change of the process variable. MPC relies
on a model-based predictor. For valid process models, the
latter delivers superior performance. For applications
such as reactor temperature control, MPC may not be
the answer (reaction mechanisms and heat transfer are
nonlinear), but we will eventually have to come up with
something that can meet the ever-increasing demands for
improved control performance from the process chemists.
PID will continue to be used in over 90% of our loops.
But for the demanding applications, it is time we start
seriously thinking outside the PID box.
Cecil L. Smith
© SPX 2011
http://www.pneumaticproducts-spx.com http://www.spxft.com http://adlinks.che.com/35070-49 http://www.Che.Com
Chemical Engineering October 2011

Table of Contents for the Digital Edition of Chemical Engineering October 2011
