You are on page 1of 10

Chapter 34

Advanced Control for the Plant Floor

Dr. James R. Ford, P. E.

34.1 Introduction and recovering the main distillable products, primarily kero-
sene, heating oil, and lubricants. These processes were ini-
Advanced process control (APC) is a fairly mature body of tially batch in nature. A pot of oil was heated to boiling, and
engineering technology. Its evolution closely mirrors that of the resulting vapor was condensed and recovered in smaller
the digital computer and its close cousin, the modern micro- batches.
processor. APC was born in the 1960s, evolved slowly and The first batch in the process was extremely light (virgin
somewhat painfully through its adolescence in the 1970s, naphtha), and the last batch was heavy (fuel oil or lubricating
flourished in the 1980s (with the remarkable advances oil). Eventually, this process was transformed from batch to
in computers and digital control systems, or DCSs), and continuous, providing a means of continuously feeding fresh
reached maturity in the 1990s, when model predictive oil and recovering all distillate products simultaneously.
control (MPC) ascended to the throne of supremacy as the The heart of this process was a unit operation referred to as
preferred approach for implementing APC solutions. countercurrent, multicomponent, two-phase fractionation.
As Zak Friedman1 dared to point out in a recent article, Whereas the batch process was manual in nature and
the current decade has witnessed tremendous APC industry required very few adjustments (other than varying the heat
discontent, self-examination, and retrenchment. He lists sev- applied to the pot), the continuous process required a means
eral reasons for the malaise, among them: cutting corners of making adjustments to several important variables, such
on the implementation phase of the projects, poor inferred as the feed rate, the feed temperature, the reflux rate, and so
property models (these are explained later), tying APC proj- on, to maintain stable operation and to keep products within
ects to optimization projects, and too-cozy relationships specifications.
between APC software vendors/implementers and their Manually operated valves were initially utilized to allow
customers. In a more recent article2 in the same magazine, an operator to adjust the important independent variables.
Friedman interviewed me because I offer a different expla- A relatively simple process could be operated in a fairly
nation for the APC industry problems. stable fashion with this early process control system.
This chapter traces the history of the development of pro- Over the next generation of process technology develop-
cess control, advanced process control, and related applied ment, process control advanced from purely manual, open-
engineering technologies and discusses the reasons that I loop control to automatic, closed-loop control. To truly
think the industry has encountered difficulties. The chapter understand this evolution, we should examine the reasons
presents some recommendations to improve the likelihood that this evolution was necessary and how those reasons
of successful APC project implementation and makes some impact the application of modern process control technol-
predictions about the future direction of the technology. ogy to the operation of process units today.

34.2 Early developments 34.3The need for process control

The discovery of oil in Pennsylvania in 1859 was followed Why do we need process control at all? The single most
immediately by the development of processes for separating important reason is to respond to process disturbances. If
process disturbances did not occur, the manual valves men-
1. Has the APC Industry Completely Collapsed?, Hydrocarbon
Processing, January 2005, p. 15.
tioned here would suffice for satisfactory, stable operation
2. Jim Fords Views on APC, Hydrocarbon Processing, November of process plants. What, then, do we mean by a process
2006, p. 19. disturbance?

2010 Elsevier Inc. All rights reserved.

doi: 10.1016/B978-0-7506-8308-1.00034-6 619
620 PART | V Controllers, Actuators, and Final Control Elements

A process disturbance is a change in any variable that and control valve opening had to be designed to handle the
affects the flow of heat and/or material in the process. We full range of operation. Otherwise, the vessel could overflow
can further categorize disturbances in two ways: by time or drain out completely. This type of control had no specific
horizon and by measurability. target or set point. At constant inlet flow, the level in the
Some disturbances occur slowly over a period of weeks, vessel would reach whatever resting position resulted in the
months, or years. Examples of this type of disturbance are: proper valve opening to make the outflow equal to the inflow.
Level controllers were probably the first type of auto-
Heat exchanger fouling. Slowly alters the rate of heat
matic process controller developed because of the mechani-
transfer from one fluid to another in the process.
cal simplicity of the entire loop. Later on, it became obvious
Catalyst deactivation. Slowly affects the rate, selectivity,
that more sophisticated control valves were needed to further
and so on of the reactions occurring in the reactor.
automate other types of loops. The pneumatically driven,
Automatic process control was not developed to address linear-position control valve evolved over the early 20th
long time-horizon disturbances. century in all its various combinations of valve body and
Manual adjustment for these types of disturbances would plug design to handle just about any type of fluid condition,
work almost as well. So, automatic process control is used pressure drop, or the like. The development of the automatic
to rectify disturbances that occur over a much shorter time control valve ushered in the era of modern process control.
period of seconds, minutes, or hours. Within this short time
horizon, there are really two main types of disturbances:
measured and unmeasured. 34.5Automatic control valves
The first truly automatic control valves were developed
34.4Unmeasured disturbances to replace manual valves to control flow. This is the easi-
est type of variable to control, for two reasons. First, there
Automatic process control was initially developed to respond is essentially no dead time and very little measurement lag
to unmeasured disturbances. For example, consider the first between a change in valve opening and a change in the flow
automatic devices used to control the level of a liquid in a measurement. Second, a flow control loop is not typically
vessel. (See Figure 34.1.) The liquid level in the vessel is subjected to a great deal of disturbance. The only significant
sensed by a float. The float is attached to a lever. disturbance is a change in upstream or downstream pressure,
A change in liquid level moves the float up or down, such as might occur in a fuel gas header supplying fuel gas
which mechanically or pneumatically moves the lever, through a flow controller for firing a heater or boiler. Other
which is connected to a valve. When the level goes up the less significant disturbances include changes in the tempera-
valve opens, and vice versa. The control loop is respond- ture and density of the flowing fluid. The flow control loop
ing to an unmeasured disturbance, namely, the flow rate of has become the foundation of all automatic process control,
material into the vessel. for several reasons.
The first automatic controllers were not very sophisti- Unlike pressure and temperature, which are intensive
cated. The float span, lever length, connection to the valve, variables, flow is an extensive variable. Intensive vari-
ables are key control variables for stable operation of pro-
cess plants because they relate directly to composition.
Intensive variables are usually controlled by adjusting flows,
the extensive variables. In this sense, intensive variables are
higher in the control hierarchy. This explains why a simple
cascade almost always involves an intensive variable as the
master, or primary, in the cascade and flow as the slave, or
secondary, in the cascade. When a pressure or temperature
LIQUID LEVEL FLOAT controller adjusts a control valve directly (rather than the
flow in a cascade), the controller is actually adjusting the
flow of material through the valve.
What this means, practically speaking, is that, unlike the
intensive variables, a flow controller has no predetermined
or best target for any given desired plant operation. The
flow will be wherever it needs to be to maintain the higher-
OUT level intensive variable at its best value. This explains
why optimum unit operation does not require accurate
OPEN flow measurement. Even with significant error in flow mea-
Figure 34.1 Float-Actuated level control diagram. surement, the target for the measured flow will be adjusted
Chapter | 34 Advanced Control for the Plant Floor 621

(in open or closed loop) to maintain the intensive variable control action. A less important type of control action was
at its desired target. This also explains why orifices are per- developed to handle situations in which the loop includes
fectly acceptable as flow controller measurement devices, significant measurement lag, such as is often seen in tem-
even though they are known to be rather inaccurate. perature loops involving a thermocouple, inside a thermow-
These comments apply to almost all flow controllers, ell, which is stuck in the side of a vessel or pipe. Control
even important ones like the main unit charge rate control- engineers noted that these loops were particularly difficult
ler. The target for this control will be adjusted to achieve to control, because the measurement lag introduced insta-
an overall production rate, to push a constraint, to control bility whenever the loops were tuned to minimize SP-PV
the inventory of feed (in a feed drum), and so on. What error. For these situations, a control action was developed
about additional feed flows, such as the flow of solvent in that reacts to a change in the rate of change of the PV. In
an absorption or extraction process? In this case, there is a other words, as the PV begins to change its trajectory with
more important, higher-level control variable, an intensive regard to the SP, the control action is reversed, or puts
variablenamely, the ratio of the solvent to the unit charge on the brakes, to head off the change that is coming, as
rate. The flow of solvent will be adjusted to maintain a best indicated by the change in trajectory. The control action was
solvent/feed ratio. Again, measurement accuracy is not criti- based on comparing the rate of change of the PV over time,
cal; the ratio target will be adjusted to achieve the desired or the derivative of the PV. Hence the name of the control
higher-level objective (absorption efficiency, etc.), regard- action: derivative.
less of how much measurement inaccuracy is present. In practice, this type of control action is utilized very
Almost all basic control loops, either single-loop or little in the tuning of basic control loops. The other type of
simple cascades, are designed to react, on feedback, to change that produces an offset between SP and PV is an SP
unmeasured disturbances. Reacting to unmeasured distur- change. For flow control loops, which are typically adjusted
bances is called servo, or feedback control. Feedback control to maintain a higher-level intensive variable at its target,
is based on reacting to a change in the process variable (the a quick response to the SP change is desirable; otherwise,
PV) in relation to the loop target, or set point (the SP). The additional response lag is introduced. A flow control loop
PV can change in relation to the SP for two reasons: either can and should be tuned to react quickly and equally effec-
because a disturbance has resulted in an unexpected change tively to both PV disturbances and SP changes.
in the PV or because the operator or a higher-level control Unfortunately, for intensive variables, a different closed-
has changed the SP. Lets ignore SP changes for now. loop response for PV changes due to disturbances vs. SP
changes is called for. For temperature, and especially pres-
sure, these loops are tuned as tightly as possible to react to
34.6Types of feedback control disturbances. This is because intensive variables are directly
related to composition, and good control of composition is
So, feedback control was initially designed to react to essential for product quality and yield. However, if the oper-
unmeasured disturbances. The problem in designing these ator makes an SP change, a much less aggressive control
early feedback controllers was figuring out how much and action is preferred.
how fast to adjust the valve when the PV changed. The first This is because the resulting composition change will
design was the float-type level controller described earlier. induce disturbances in other parts of the process, and the
The control action is referred to as proportional because goal is to propagate this disturbance as smoothly as possible
the valve opening is linearly proportional to the level. The so as to allow other loops to react without significant upset.
higher the level, the more open the valve. Modern DCSs can provide some help in this area. For
This type of control may have been marginally accept- example, the control loop can be configured to take pro-
able for basic levels in vessels, but it was entirely deficient portional action on PV changes only, ignoring the effect of
for other variables. The main problem is that proportional- SP changes. Then, following an SP change, integral action
only control cannot control to a specific target or set point. will grind away on correcting the offset between SP and PV.
There will always be offset between the PV and the desired However, these features do not fully correct the deficiency
target. To correct this deficiency, the type of control known discussed earlier. This dilemma plagues control systems to
as integral, or reset, was developed. This terminology is this very day and is a major justification for implementa-
based on the fact that, mathematically, the control action to tion of advanced controls that directly address this and other
correct the offset between SP and PV is based on the calcu- shortcomings of basic process control.
lus operation known as integration. (The control action is
based on the area under the SP-PV offset curve, integrated
over a period of time.) The addition of this type of control 34.7Measured disturbances
action represented a major improvement in feedback con-
trol. Almost all flow and pressure loops can be controlled The other major type of disturbance is the measured dis-
very well with a combination of proportional and integral turbance. Common examples are changes in charge rate,
622 PART | V Controllers, Actuators, and Final Control Elements

cooling water temperature, steam header pressure, fuel gas process instability so as to operate closer to constraints, why
header pressure, heating medium temperature and ambient not implement APCs that accomplish that goal? (Note that
air temperature, where instruments are installed to mea- this type of APC strategy creates a measured process distur-
sure those variables. The first 20 years of the development bance; we are going to move a major independent variable
of APC technology focused primarily on using measured to push constraints, so there had better be APCs in place to
disturbance information for improving the quality of con- handle those disturbances.) Especially when the goal was
trol. Why? Modern process units are complex and highly to increase the average unit charge rate by pushing known,
interactive. The basic control system, even a modern DCS, measured constraints, huge benefits could often be claimed
is incapable of maintaining fully stable operation when for these types of strategies. In practice, these types of strat-
disturbances occur. APC was developed to mitigate the egies were difficult to implement and were not particularly
destabilizing effects of disturbances and thereby to reduce successful.
process instability. This is still the primary goal of APC. While all this development work was focused on react-
Any other claimed objective or direct benefit is secondary. ing to changes in measured disturbances, the problems cre-
Why is it so important to reduce process instability? ated by unmeasured disturbances continued to hamper stable
Process instability leads to extremely conservative opera- unit operation (and still do today). Some early effort also
tion so as to avoid the costly penalties associated with focused on improving the only tool available at the time, the
instability, namely, production of off-spec product and proportional-integral-derivative (PID) control algorithm, to
violation of important constraints related to equipment life react better to unmeasured disturbances.
and human safety. Conservative operation means staying One of the main weaknesses of PID is its inability to
well away from constraint limits. Staying away from these maintain stable operation when there is significant dead
limits leaves a lot of money on the table in terms of reduced time and/or lag between the valve movement and the
yields, lower throughput, and greater energy consumption. effect on the control variable. For example, in a distilla-
APC reduces instability, allowing for operation much closer tion column, the reflux flow is often adjusted to maintain a
to constraints and thereby capturing the benefits that would stable column tray temperature. The problem arises when
otherwise be lost. the tray is well down the tower. When an unmeasured feed
As stated earlier, early APC development work focused composition change occurs, upsetting the tray tempera-
on improving the control systems response to measured dis- ture, the controller responds by adjusting the reflux flow.
turbances. The main techniques were called feed-forward, But there may be dead time of several minutes before the
compensating, and decoupling. In the example of a fired change in reflux flow begins to change the tray tempera-
heater mentioned earlier, adjusting the fuel flow for changes ture. In the meantime, the controller will have continued
in heater charge rate and inlet temperature is feed-forward. to take more and more integral action in an effort to return
The objective is to head off upsets in the heater outlet the PV to SP. These types of loops are difficult (or impos-
temperature that are going to occur because of these feed sible) to tune. They are typically detuned (small gain and
changes. In similar fashion, the fuel flow can be compen- integral) but with a good bit of derivative action left in as a
sated for changes in fuel gas header pressure, temperature, means of putting on the brakes when the PV starts return-
density, and heating value, if these measurements are avail- ing toward SP.
able. Finally, if this is a dual-fuel heater (fuel gas and fuel Some successes were noted with algorithms such as
oil), the fuel gas flow can be adjusted when the fuel oil flow the Smith Predictor, which relies on a model to predict the
changes so as to decouple the heater from the firing upset response of the PV to changes in controller output. This
that would otherwise occur. This decoupling is often imple- algorithm attempts to control the predicted PV (the PV with
mented as a heater fired duty controller. both dead time and disturbances included) rather than the
A second area of initial APC development effort focused actual measured PV. Unfortunately, even the slightest model
on controlling process variables that are not directly mea- mismatch can cause the controller using the Smith Predictor
sured by an instrument. An example is reactor conversion to become unstable.
or severity. We have been particularly successful in this area with
Hydrocracking severity is often measured by how much development of our smart PID control algorithm. In
of the fresh feed is converted to heating oil and lighter its simplest form, it addresses the biggest weakness of
products. If the appropriate product flow measurements are PID, namely, the overshoot that occurs because the algo-
available, the conversion can be calculated and the reactor rithm continues to take integral action to reduce the offset
severity can then be adjusted to maintain a target conversion. between SP and PV, even when the PV is returning to SP.
Work in this area of APC led to the development of a related Our algorithm turns the integral action on and off accord-
body of engineering technology referred to as inferred ing to a proven decision process made at each controller
properties or soft sensors. execution. This algorithm excels in loops with significant
A third area of APC development work focused on push- dead time and lag. We use this algorithm on virtually all
ing constraints. After all, if the goal of APC is to reduce APC projects.
Chapter | 34 Advanced Control for the Plant Floor 623

34.8The need for models MPC has become the preferred technology for solving
not only multivariable control problems but just about any
By the mid-1980s, many consulting companies and in-house control problem more complicated than simple cascades and
technical staffs were involved in the design and implemen- ratios. Note that this technology no longer relies on tradi-
tation of the types of APC strategies described in the last tional servo control techniques, which were first designed
few paragraphs. A word that began to appear more and more to handle the effect of unmeasured disturbances and which
associated with APC was model. have done a fairly good job for about 100 years. MPC
For example, to implement the constraint-pushing APC assumes that our knowledge of the process is perfect and that
strategies weve discussed, a dynamic model was needed all disturbances have been accounted for. There is no way
to relate a change in the independent variable (the charge for an MPC to handle unmeasured disturbances other than
rate) to the effect on each of the dependent, or constraint, to readjust at each controller execution the bias between the
variables. With this model, the adjustments that were needed predicted and measured value of each control variable. This
to keep the most constraining of the constraint variables can be likened to a form of integral-only control action. This
close to their limits could be determined mathematically. partially explains MPCs poor behavior when challenged by
Why was it necessary to resort to development of disturbances unaccounted for in the controller.
models? As mentioned earlier, many of the early constraint- DMCPlus uses linear, step-response models, but other
pushing efforts were not particularly successful. Why not? MPC developers have incorporated other types of models.
Its the same problem that plagues feedback control loops For example, Pavilion Technologies has developed a whole
with significant dead time and lag. body of modeling and control software based on neural
There are usually constraints that should be honored networks. Since these models are nonlinear, they allow
in a constraint-pushing strategy that may be far removed the user to develop nonlinear models for processes that
(in time) from where the constraint-pushing move is made. display this behavior. Polymer production processes (e.g.,
Traditional feedback techniques (PID controllers acting polypropylene) are highly nonlinear, and neural net-based
through a signal selector) do not work well for the con- controllers are said to perform well for control of these
straints with long dead time. We addressed this issue by processes. GE (MVC) uses algebraic models and solves
developing special versions of our smart PID algorithm to the control execution prediction problem with numerical
deal with the long dead times, and we were fairly successful techniques.
in doing so.

34.10MPC vs. ARC

34.9The emergence of MPC
There are some similarities between the older APC tech-
In his Ph.D. dissertation work, Dr. Charles Cutler developed niques (feed-forward, etc.) and MPC, but there are also
a technique that incorporated normalized step-response some important differences. Lets call the older technique
models for the constraint (or control) variables, or CVs, as a advanced regulatory control (ARC). To illustrate, lets take
function of the manipulated variables, or MVs. This allowed a simplified control problem, such as a distillation column
the control problem to be linearized, which then permitted where we are controlling a tray temperature by adjusting
the application of standard matrix algebra to estimate the the reflux flow, and we want feed-forward action for feed
MV moves to be made to keep the CVs within their limits. rate changes. The MPC will have two models: one for the
He called the matrix of model coefficients the dynamic response of the temperature to feed rate changes and one for
matrix and developed the dynamic matrix control (DMC) the response of the temperature to reflux flow changes. For
control technique. He also incorporated an objective func- a feed rate change, the controller knows that the tempera-
tion into the DMC algorithm, turning it into an optimizer. ture is going to change over time, so it estimates a series of
If the objective function is the sum of the variances between changes in reflux flow required to keep the temperature near
the predicted and desired values of the CVs, DMC becomes its desired target.
a minimum variance controller that minimizes the output The action of the ARC is different. In this case, we want
error over the controller time horizon. to feed-forward the feed rate change directly to the reflux
Thus was ushered in the control technology known in flow. We do so by delaying and lagging the feed rate change
general as multivariable, model-predictive control (MVC (using a simple dead time and lag algorithm customized
or MPC). Dr. Cutlers work led eventually to formation to adjust the reflux with the appropriate dynamics), then
of his company, DMC Corporation, which was eventually adjusting the reflux with the appropriate steady-state gain
acquired by AspenTech. The current version of this control or sensitivity (e.g., three barrels of reflux per barrel of feed).
software is known as DMCPlus. There are many compet- The ultimate sensitivity of the change in reflux flow to a
ing MPC products, including Honeywell RMPCT, Invensys change in feed rate varies from day to day; hence, this type
Connoisseur, and others. of feed-forward control is adaptive and, therefore, superior
624 PART | V Controllers, Actuators, and Final Control Elements

to MPC (the MPC models are static). Note: Dr. Cutler capacity and flow of the external reflux, the overhead vapor
recently formed a new corporation, and he is now offering temperature, and the reflux temperature. The control next
adaptive DMC, which includes real-time adjustment of the back-calculates the external reflux flow required to main-
response models. tain constant IR and then adjusts the set point of the exter-
How does the MPC handle an unmeasured disturbance, nal reflux flow controller accordingly. This type of control
such as a feed composition change? As mentioned earlier, provides a fast-responding, first-level improvement in sta-
it can do so only when it notices that the temperature is not bility by isolating the column from disturbances caused by
where its supposed to be according to the model prediction changes in ambient conditions. There are multiple inputs
from the series of recent moves of the feed rate and reflux. It to the control (the flow and temperatures which contribute
resets the predicted vs. the actual bias and then calculates a to calculation of the internal reflux), but typically only one
reflux flow move that will get the temperature back where its outputto the set point of the reflux flow controller.
supposed to be, a form of integral-only feedback control. Moving up the hierarchy to the advanced supervisory
On the other hand, the ARC acts in the traditional feedback control level, the overhead product composition can be fur-
(or servo) manner with either plain or smart PID action. ther stabilized by controlling a key temperature in the upper
part of the tower, since temperature (at constant pressure) is
directly related to composition. And, since the tower pres-
34.11Hierarchy sure could vary (especially if another application is attempt-
ing to minimize pressure), the temperature that is being
Prior to MPC, most successful APC engineers used a pro- controlled should be corrected for pressure variations. This
cess engineering-based, hierarchical approach to develop- control adjusts the set point of the IRC to maintain a constant
ing APC solutions. The bottom of the control hierarchy, its pressure-corrected temperature (PCT). Feed-forward action
foundation, is what we referred to earlier as basic process (for feed rate changes) and decoupling action (for changes
control, the single loops and simple cascades that appear on in reboiler heat) can be added at this level.
P&IDs and provide the operator with the first level of regula- Further improvement in composition control can be
tory control. achieved by developing a correlation for the product com-
Simple processes that are not subject to significant dis- position, using real-time process measurements (the PCT
turbances can operate in a fairly stable fashion with basic plus other variables such as reflux ratio, etc.), then using this
process control alone. Unfortunately, most process units in correlation as the PV in an additional APC that adjusts the
refineries and chemical plants are very complex, highly inter- set point of the PCT. This type of correlation is known as an
active, and subject to frequent disturbances. The basic con- inferred property or soft sensor.
trol system is incapable of maintaining fully stable operation Thus, the hierarchical approach results in a multiple-
when challenged by these disturbancesthus the emergence cascade design, an inferred property control adjusting a
of APC to mitigate the destabilizing effects of disturbances. PCT, adjusting an IRC, adjusting the reflux flow.
The hierarchical approach to APC design identifies the In general, APCs designed using hierarchical approaches
causes of the disturbances in each part of the process, then consist of layers of increasingly complex strategies. Some of
layers the solutions that deal with the disturbances on top the important advantages of this approach are:
of the basic control system, from the bottom up. Each layer
Operators can understand the strategies; they appeal to
adds complexity and its design depends on the disturbance(s)
human logic because they use a systems approach
being dealt with.
to problem solving, breaking a big problem down into
As an example of the first layer of the APC hierarchy,
smaller problems to be solved.
consider the classic problem of how to control the composi-
The control structure is more suitable for solutions at
tion of distillation tower product streams, such as the over-
a lower level in the control system; such solutions can
head product stream. The hierarchical approach is based on
often be implemented without the requirement for addi-
identifying and dealing with the disturbances. The first type
tional hardware and software.
of disturbance that typically occurs is caused by changes
The controls degrade gracefully; when a problem pro-
in ambient conditions (air temperature, rainstorms, etc.),
hibits a higher-level APC from being used, the lower-
which lead to a change in the temperature of the condensed
level controls can still be used and can capture much of
overhead vapor stream, namely the reflux (and overhead
the associated benefit.
product). This will cause the condensation rate in the tower
to change, leading to a disturbance that will upset column How would we solve the preceding control problem using
separation and product qualities. Hierarchical APC design MPC? There are two approaches. The standard approach is
deals with this disturbance by specifying, as the first level to use the inferred property as a CV and the external reflux
above basic control, internal reflux control (IRC). flow controller as the MV. In this case, then, how does the
The IRC first calculates the net internal reflux with MPC deal with the other disturbances such as the reflux tem-
an equation that includes the heat of vaporization, heat perature, reboiler heat, and feed rate? These variables must
Chapter | 34 Advanced Control for the Plant Floor 625

be included in the model matrix as additional independent using the least costly combination of manipulated variable
variables. A step-response model must then be developed assets. That is certainly correct, mathematically; that is the
for the CV (the inferred property) as a function of each of way the LP or QP works. In practice, however, any actual
the additional independent variables. This is the way most of optimizing is marginal at best. This is due to a couple of
the APC industry designs an MPC. reasons. The first is the fact that, in most cases, one MV
The hierarchical approach would suggest something dominates the relationships of other MVs to a particular CV.
radically different. The CV is the same because the product For example, in a crude oil distillation tower, the sensitivity
composition is the variable that directly relates to profitabil- between the composition of a side-draw distillate product
ity. However, in the hierarchical design, the MV is the PCT. and its draw rate will be much larger than the sensitivity
The lower-level controls (the PCTC and IRC) are imple- with any other MV. The controller will almost always move
mented at a lower level in the control hierarchy, typically the distillate draw rate to control composition. Only if the
in the DCS. draw rate becomes constrained will the controller adjust a
Unfortunately, the industry-accepted approach to MPC pump-around flow or the draw rate of another distillate prod-
design violates the principles of hierarchy. Rarely, if ever, uct to control composition, regardless of the relative cost
are intermediate levels of APC employed below MPC. There of these MVs. The second reason is the fact that, in most
is no hierarchyjust one huge, flat MPC controller on top cases, once the constraint corners of the CV/MV space are
of the basic controllers, moving all of them at the same time found, they tend not to change. The location of the corners
in seemingly magical fashion. Operators rarely understand is most often determined either by discretionary operator-
them or what they are doing. And they do not degrade grace- entered limits (for example, the operator wants to limit the
fully. Consequently, many fall into disuse. operating range of an MV) or by valve positions. In both
situations, these are constraints that would have been pushed
by any kind of APC, not just a model-predictive controller.
34.12Other problems with MPC So, the MPC has not optimized any more than would a
simpler ARC or ASC.
A nonhierarchical design is only one of the many problems When analyzing the MPC controller models that result
with MPC. The limitations of MPC have been thoroughly from plant testing, control engineers often encounter CV/
exposed, though probably widely ignored.3 Here are a few MV relationships that appear inappropriate and that are
other problems. usually dropped because the engineer does not want that
In many real control situations, even mild model mis- particular MV moved to control that particular CV. Thus,
match introduces seriously inappropriate controller moves, the decoupling benefit that could have been achieved with
leading to inherent instability as the controller tries at each simpler ASC is lost.
execution to deal with the error between where it thinks it is If every single combination of variables and constraint
going and where it really is. limits has not been tested during controller commissioning
MPC is not well suited for processes with noncontinuous (as is often the case), the controller behavior under these
phases of operation, such as delayed cokers. The problem untested conditions is unknown and unpredictable. For even
here is how to model the transitory behavior of key con- moderately sized controllers, the number of possible combi-
trol variables during coke drum prewarming and switching. nations becomes unreasonably large such that all combina-
Unfortunately, every drum-switching operation is different. tions cannot be tested, even in simulation mode.
This means that both the time to steady state and ultimate Control engineers often drop models from the model
CV gain are at best only approximations, leading to model matrix to ensure a reasonably stable solution (avoiding
mismatch during every single drum operation. No wonder approach to a singular matrix in the LP solution). This is
they are so difficult to tune during these transitions. most common where controllers are implemented on frac-
Correcting model mismatch requires retesting and tionation columns with high-purity products and where the
remodeling, a form of very expensive maintenance. MPC controller is expected to meet product purity specifications
controllers, particularly complex ones implemented on on both top and bottom products. The model matrix then
complex units with lots of interactions, require a lot of reduces to one that is almost completely decoupled. In this
babysittingconstant attention from highly trained (and case, single-loop controllers, rather than an MPC, would be
highly paid) control engineers. This is a luxury that few a clearly superior solution.
operating companies can afford. What about some of the other MPC selling points? One
The licensors of MPC software will tell you that their that is often mentioned is that, unlike traditional APC, MPC
algorithms optimize operation by operating at constraints eliminates the need for custom programming. This is simply
not true. Except for very simple MPC solutions, custom
3. The emperor is surely wearing clothes! See the excellent article by
code is almost always requiredfor example, to calculate
Alan Hugo, Limitations of Model-Predictive Controllers, that appeared a variable used by the controller, to estimate a flow from a
in the January 2000 issue of Hydrocarbon Processing, p. 83. valve position, and so on.
626 PART | V Controllers, Actuators, and Final Control Elements

The benefits of standardization are often touted. Using internal reflux control. The internal reflux can then be utilized
the same solution tool across the whole organization for all as an MV in a higher-level control strategy (for example, to
control problems will reduce training and maintenance costs. control a pressure-compensated tray temperature as an indi-
But this is like saying that I can use my new convertible to cator of product composition) or an inferred property.
haul dirt, even though I know that using my old battered What about unmeasured disturbances, such as feed com-
pickup would be much more appropriate. Such an approach position, where the control strategy must react on feedback
ignores the nature of real control problems in refineries and alone? As mentioned earlier, we have had a great deal of
chemical plants and relegates control engineers to pointers success with intelligent feedback-control algorithms that
and clickers. are much more effective than simple PID. For example, our
Smart PID algorithm includes special logic that first deter-
mines, at each control algorithm execution, whether or not
34.13Where we are today? integral action is currently advised based on the PVs recent
trajectory toward or away from set point. Second, it deter-
Perhaps the previous discussion paints an overly pessimistic mines the magnitude of the integral move based on similar
picture of MPC as it exists today. Certainly many compa- logic. This greatly improves the transient controller response,
nies, particularly large refining companies, have recognized especially in loops with significant dead time and/or lag.
the potential return of MPC, have invested heavily in MPC, Another successful approach is to use model-based, adap-
and maintain large, highly trained staffs to ensure that the tive control algorithms, such as those incorporated in prod-
MPCs function properly and provide the performance that ucts like Brainwave. Here are some additional suggestions:
justified their investment. Using a hierarchical approach suggests implementing the
But, on the other hand, the managers of many other com- lower-level APC strategies as low in the control system as
panies have been seduced by the popular myth that MPC is possible. Most modern DCSs will support a good bit of ARC
easy to implement and maintaina misconception fostered at and ASC at the DCS level. Some DCSs such as Honeywell
the highest management levels by those most likely to benefit Experion (and earlier TDC systems) have dedicated, DCS-
from the proliferation of various MPC software packages. level devices designed specifically for implementation of
So, what can companies do today to improve the utiliza- robust APC applications. We are currently implement-
tion and effectiveness of APC in general and MPC in par- ing very effective first-level APC in such diverse DCSs as
ticular? They can do several things, all focused primarily Foxboro, Yokogawa, DeltaV, and Honeywell.
on the way we analyze operating problems and design APC When available, and assuming the analyses are reliable,
solutions to solve the problems and thereby improve produc- use lab data as much as possible in the design of APCs.
tivity and profitability. Here are some ideas. Develop inferred property correlations for control of key
As mentioned earlier, the main goal of APC is to isolate product properties and update the model correlations with
operating units from process disturbances. What does this the lab data.
mean when we are designing an APC solution? First, iden- Track APC performance by calculating, historizing, and
tify the disturbance and determine its breadth of influence. displaying a variable that indicates quality of control. We
Does the disturbance affect the whole unit? If so, how? For often use the variance (standard deviation), in engineering
example, the most common unit-encompassing disturbance units, of the error between SP and PV of the key APC (for
is a feed-rate change. But how is the disturbance propagated? example, an inferred property). Degradation in the long-term
In many process units, this disturbance merely propagates in trend of this variable suggests that a careful process analysis
one direction only (upstream to downstream), with no other is needed to identify the cause of the degraded performance
complicating effects (such as those caused by recycle or and the corrective action needed to return the control to its
interaction between upstream and downstream unit opera- original performance level.
tions). In this case, a fairly straightforward APC solution
involves relatively simple ARCs for inventory and load
controladjusting inter-unit flows with feed-forward for
inventory control, and adjusting load-related variables such 34.14Recommendations
as distillation column reflux with ratio controls or with feed- for using MPC
forward action in the case of cascades (for example, control-
If MPC is being considered for solution of a control prob-
ling a PCT by adjusting an IRC). No MVC is required here
lem, apply it intelligently and hierarchically.
to achieve significantly improved unit stability. If the effect
Intelligent application of MPC involves asking impor-
of the disturbance can be isolated, design the APC for that
tant questions, such as:
isolated part of the unit and ignore potential second-order
effects on other parts of the unit. A good example is the IR Is the control problem to be solved truly multivariable?
control discussed earlier. The tower can be easily isolated Can (and should) several MVs be adjusted to control one
from the ambient conditions disturbance by implementing CV? If not, dont use MPC.
Chapter | 34 Advanced Control for the Plant Floor 627

Are significant dead time and lag present such that the might imagine a situation in which incredibly sophisticated
model-based, predictive power of MPC would provide automation solutions would eventually eliminate the need
superior control performance compared to simple feed- for direct human involvement.
forward action? Thats not what this author sees for the next 40 years.
Look at the model matrix. Are there significant interac- Why not? Three reasons. First, the more sophisticated auto-
tions between many dependent variables and many inde- mation solutions require process models. For example, MPC
pendent variables, or are there just isolated islands of requires accurate MV/CV or DV/CV response models.
interactions where each island represents a fairly simple Real-time rigorous optimization (RETRO) requires accurate
control problem that could just as easily be solved with steady-state process models. Despite the complexity of the
simpler technology? underlying mathematics and the depth of process data analy-
How big does the controller really have to be? Could the sis involved in development of these models, the fact is that
problem be effectively solved with a number of small these will always remain just models. When an optimizer
controllers without sacrificing the benefits provided by a runs an optimization solution for a process unit, guess what
unit-wide controller? results? The optimizer spits out an optimal solution for the
model, not the process unit! Anybody who believes that the
Applying MPC hierarchically means doing the
optimizer is really optimizing the process unit is either quite
nave or has been seduced by the same type of hype that
Handle isolated disturbance variables with lower-level accompanied the advent of MPCs and now seems to pervade
ARC. For example, stabilize and control a fired heater the world of RETRO.
outlet temperature with ARC, not MPC. The ARC can When well-designed and -engineered (and -monitored),
easily handle disturbance variables such as feed rate, RETRO can move operations in a more profitable direction
inlet temperature, and fuel gas pressure. and thereby improve profitability in a major way. But the
Use the lower-level ARCs, such as the fired heater out- point is that models will never be accurate enough to elimi-
let temperature just mentioned, as MVs in the MPC nate the need for human monitoring and intervention.
controller. The second important reason involves disturbances.
Use intensive variables, when possible, as MVs in the Remember that there are two types of process disturbances,
controller. For example, use a distillate product yield, measured and unmeasured. Despite huge advances in pro-
rather than flow rate, as the MV in a controller designed cess measurement technology (analyzers, etc.), we will
to control the quality of that product; this eliminates unit never reach the point (at least in the next 40 years) when all
charge rate as a disturbance variable. disturbances can be measured. Therefore, there will always
be target offset (deviation from set point) due to unmeasured
disturbances and model mismatch. Furthermore, it is an
34.15Whats in store for the absolute certainty that some unmeasured disturbances will
next 40 years? be of such severity as to lead to process upsets, condi-
tions that require human analysis and intervention to pre-
A cover story that appeared in Hydrocarbon Processing vent shutdowns, unsafe operation, equipment damage, and
about 20 years ago included an interview with the top auto- so on. Operators may get bored in the future, but theyll
mation and APC guru of a major U.S. refining company. still be around for the next 40 years to handle these upset
The main point proffered by the interviewee was that the conditions.
relentless advances in automation technology, both hard- The third important reason is the weakest link argu-
ware- and software-wise (including APC), would lead one ment. Control system hardware (primarily field elements)
day to the virtual control room in which operating pro- is notoriously prone to failure. No automation system
cesses would no longer require constant human monitor- can continue to provide full functionality, in real time,
ingno process operators required. Im not sure how this when hardware fails; human intervention will be required
article was received elsewhere, but it got a lot of chuckles to maintain stable plant operation until the problem is
in our office. remedied.
Process control, whether continuous or discrete, involves Indeed, the virtual control room is still more than 40
events that are related by cause and effect. For example, a years away. Process automation will continue to improve
change in temperature (the cause event) needs a change in with advances in measurement technology, equipment reli-
a valve position (the effect event) to maintain control of the ability, and modeling sophistication, but process operators,
temperature. Automation elevates the relationships of the and capable process and process-control engineers, will
cause and effect events by eliminating the requirement for a still maintain and oversee its application. APC will remain
human to connect the two. an imperfect technology that requires for its successful
With the huge advances in automation technology application experienced and quality process engineering
accomplished in the period of the 1970s and 1980s, one analysis.
628 PART | V Controllers, Actuators, and Final Control Elements

The most successful APC programs of the next 40 years APC solutions will make greater use of inferred proper
will have the following characteristics: ties,and these will be based on the physics and chemistry
of the process (not artificial intelligence, like neural
The APC solution will be designed to solve the operating
problem being dealt with and will utilize the appropriate
The role of operating company, in-house control engi-
control technology; MPC will be employed only when
neers will evolve from a purely technical position to
more of a program manager. Operating companies will
The total APC solution will utilize the principle of
rely more on APC companies to do the grunt work,
hierarchy; lower-level, simpler solutions will solve
allowing the in-house engineers to manage the projects
lower-level problems, with ascending, higher-level,
and to monitor the performance of the control system.
more complex solutions achieving higher-level control
and optimization objectives. APC technology has endured and survived its growing
An intermediate level of model-based APC technology, pains; it has reached a relatively stable state of maturity, and
somewhere between simple PID and MPC (Brainwave, it has shaken out its weaker components and proponents.
for example), will receive greater utilization at a lower Those of us who have survived this process are wiser, hum-
cost (in terms of both initial cost and maintenance cost) bler, and more realistic about how to move forward. Our
but will provide an effective means of dealing with mea- customers and employers will be best served by a prag-
sured and unmeasured disturbances. matic, process engineering-based approach to APC design
The whole subfield of APC known as transition manage- that applies the specific control technology most appropri-
ment (e.g., handling crude switches) will become much ate to achieve the desired control and operating objectives
more dominant and important as an adjunct to continu- at lowest cost and with the greatest probability of long-term
ous process control. durability and maintainability.