PROCESS CONTROL

Process Engineers:
Take Control!

Process control is too important

to be left to control engineers.
Cecil L. Smith, And, dealing with control issues is
Cecil L. Smith, Inc.
easier than you might think.

W ith the ever growing complexity of

today’s processes and the increas-
ing demands on process perfor-
mance, the role of process engi-
neers in the design and analysis of process con-
trol strategies has to change. This is true both at
the initial design stage and in the analysis of
to some form of digital controls. The available
technology includes distributed control sys-
tems (DCSs), programmable logic controllers
(PLCs), personal-computer-based controls,
and microprocessor-based single-loop con-
trollers. The selection of the type of technolo-
gy and the specific supplier frequently has
control problems in an operating plant. been a trying exercise akin to a holy war.
Two factors are driving the change: Most firms have no interest in repeating this
1. Today’s emphasis on shortening the de- endeavor.
sign cycle is placing a lot of stress on the tradi- Usually, the decisions are based on issues
tional approaches to developing the process and that have little to do with process control. The
instrumentation diagram (P&ID). two features that receive the most attention are
2. Corporate rightsizing has reduced, if not the operator interface and the alarm package.
eliminated, the cadre of in-house control engi- (There is no shortage of instant experts on
neers. Those with the most experience were these two topics.) Naturally, the suppliers re-
often the first to qualify for early retirement. spond to this attention, and now are introduc-
Some were immediately hired as consultants, ing features such as three-dimensional graph-
but this pool of expertise becomes diluted with ics. What this has to do with process control is
time. obscure. In reality, these all are system issues,
These two factors should force changes that not process issues.
are long overdue. Today, the various approaches and even the
The integration of process design and pro- available products are largely equivalent. If the
cess control was deemed to be desirable as long job can be done with one, it can be done with
ago as the 1960s. Process engineers and control any of them. So, why do we need to spend time
engineers tend to mix about as well as oil and on the system issues? The time has come to re-
water, however. They remain mixed as long as emphasize the process issues. This involves two
you agitate; but as soon as the agitation stops, activities:
they quickly separate. Design. The key is for the P&ID to truly re-
By now, most companies have progressed flect the nature of the process. Deficiencies in

Chemical Engineering Progress August 2000 19

PROCESS CONTROL

the P&ID cannot be overcome by ap- ables and the dependent variables de- Control configurations are domi-
plying automated tuning, expert sys- pend upon the values of the manipu- nated by single-loop control, where a
tems, or the like. lated variables and the disturbances. controlled variable is maintained at its
Evolution. Even with the best of The controlled variables must be target by using only one manipulated
efforts, the initial P&ID will not be maintained at or near a target, al- variable. The control logic is largely
perfect. In fact, there are advantages though occasionally a range of values the traditional PID control equation.
to starting simple and then enhancing for one or more controlled variables Although the “shrink wrap” has been
the diagram to address the real defi- is acceptable. Although targets are significantly enhanced through the
ciencies, not the “what if…” and “it not provided for the dependent vari- progression from pneumatic to elec-
would be nice to…” issues. The con- ables, constraints often must be con- tronic to digital technology, the PID
trol strategy also needs to evolve with sidered. The values of the manipulat- control equation in today’s latest-and-
improved understanding of the pro- ed variables are at the discretion of greatest (whatever you deem that to
cess, with new measurement technol- the control system, but are possibly be) is the same as in the large-case
ogy, etc. subject to constraints. The values of pneumatic controllers from the 1940s.
Understanding the process is the disturbances are determined by The improved shrink wrap offers
truly the key to success in both of factors other than the control system. some opportunities, however, one of
these endeavors. Who best under- To maintain each controlled vari- which will be discussed subsequently.
stands the process? The process en- able at its target, the number of ma- Single-loop PID control is ade-
gineers (if not, we are really in trou- nipulated variables must at least quate for a very high percentage
ble). It then follows that the process equal the number of controlled vari- (90% or more) of the controlled vari-
engineers are in the best position to ables (a “square” system). In most ables in most plants. Engineers pay
address the design and evolution of process applications, there are more homage to the “keep it simple” prin-
the process controls. manipulated variables than con- ciple, but, sometimes, much like
trolled variables (a “fat” system), politicians honor a “no new taxes”
The practice and this creates opportunities for pledge. With all of the fancy features
of process control optimization. of modern control systems, the temp-
The reaction of some process engi- Be careful with the terms CV and tation to play with the toys is irre-
neers to the notion that they should MV. They could mean controlled sistible. The result is “creeping ele-
address the control issues may resem- variable and manipulated variable. gance,” the most obnoxious effect of
ble cardiac arrest. This response But, alternative terms for controlled which is configuring an excessive
seems most pronounced among those variable are process variable, PV, and number of process alarms.
who, while at university, took the measured variable, which often is The nature of the control strategy
process control course, which usually designated as MV. To further confuse primarily depends upon the relation-
is a course in mathematics. Process things, the controller output some- ships between the controlled vari-
engineers do need an introduction to times is called the CV. (In this article, ables and the manipulated variables.
process control — but not this type of we’ll limit abbreviations to the famil- There are two aspects of these rela-
introduction. iar P&ID and PID (proportional-inte- tionships, both of which are deter-
Traditional process-control cours- gral-derivative) controller.) mined by the process:
es place great emphasis on topics
such as Laplace transforms. Manuals ■ Figure 1.
from control system suppliers occa- The essence of
process control. Dependent
sionally express relationships in the Variables
Laplace domain, but they do so for
notational convenience. What a prac- •••
titioner needs to know about Laplace
transforms can be taught in 15 min-
utes. Courses that spend more time Manipulated Controlled
Multivariable
than this on Laplace transforms are Variables Process Variables
•••

•••

courses in mathematics.
The keys to process control are the
controlled variables, the manipulated
variables, the disturbances, and the •••
dependent variables. These are best
understood from Figure 1. Disturbances
The values of the controlled vari-

20 August 2000 Chemical Engineering Progress

 Dynamic. How rapidly does a due to the normal wear of mechanical chemicals, with far less effective re-
change in the manipulated variable elements. Their impact on loop per- sults. Unfortunately, what initially
translate to a change in the controlled formance is even larger when the seems to be similar can turn out to be
variable? Specific attention should be valve is oversized. Too often, plants not very similar at all. The likelihood
given to dead time. If a change in the are saddled with such problems be- of such mistakes increases as the
manipulated variable has no effect on cause someone tried to save some time allowed for the control engi-
the controlled variable for some peri- money during design and construc- neers to become familiar with the
od of time (the dead time), the perfor- tion. For the same reasons, project process is reduced.
mance of a traditional PID controller managers have led the charge to re- Less experienced control engi-
will suffer. move valve positioners. This era neers are being given less time to de-
Steady state. After all transients hopefully is over, though, with the velop the P&IDs for processes of
have elapsed, what is the ratio of advent of smart valves. ever increasing complexity. The limit
the change in the controlled vari- Smart transmitters and smart of these trends is not a pretty picture.
able to the change in the manipulat- valves should be purchased with no The P&IDs for a complex process are
ed variable? Control engineers refer discussion. Nonidealities such as hys- developed instantly by a young engi-
to this as the sensitivity or gain of teresis and stiction do not occur when neer who just last week was bagging
the process. the final control element is a pump groceries at the supermarket.
Traditional process-control cours- with a variable-speed drive. These are As the design cycle is further
es stress dynamics; the steady-state becoming more common, but may compressed, a more viable and desir-
characteristics, however, usually have never attain preferred status. The able approach is to have the process
the greatest impact. measurement devices are our “eyes” engineers designing the process also
The three most common process to the process; the final control ele- develop the P&IDs. These engineers
characteristics that lead to control dif- ments are our “handles.” Cutting certainly understand the process bet-
ficulties are: costs for either invites trouble. ter than anyone else. Developing the
Nonlinearities (a steady-state char- P&IDs for a new process involves
acteristic). Envision a plot of the Developing P&IDs considerations that usually are more
equilibrium values of the controlled With the traditional approach to familiar to the process designers than
variable as a function of a manipulat- process design, the process engineers to anyone else, especially at this
ed variable; this is referred to as the work up the process flowsheet, from stage of the project. Will the process
process operating line. Tuning diffi- which the control engineers then de- designers get it right every time?
culties arise when this relationship velop the P&ID. Today’s emphasis on Certainly not. But, then, the control
exhibits a strong departure from lin- shortening the design cycle is placing engineers do not get it right every
earity. The traditional PID controller a lot of stress on this approach. Even time either. The question is who can
is linear; model-predictive control when the process flowsheet is deliv- do the job better. The answer is the
technologies are also linear. ered on schedule, the control engi- group with the best understanding of
Interaction (a steady-state charac- neers must quickly gain an under- the process — namely, the process
teristic). This exists when a con- standing of the process, which is no engineers.
trolled variable is influenced by two trivial task for a complex process. A Now, I am not suggesting that pro-
or more manipulated variables, and, late delivery of the flowsheet makes cess engineers develop detailed
conversely, when a manipulated vari- this quite an endeavor. P&IDs, the ones that show practically
able influences two or more con- How do the control engineers go every component of the process con-
trolled variables. about developing the P&ID? Ask trols. Control engineers generally
Dead time (a dynamic characteris- them and you are likely to be told don’t develop these either. Instead,
tic). For processes with significant that “you have to understand the pro- they work up simplified P&IDs that
dead time, the performance of the cess” (or something to this effect). In depict only the key components of
PID controller becomes so poor that reality, the control engineers rely the control strategy. Then, the engi-
alternative model-based technologies very heavily on past experiences, neering contractor adds the necessary
such as dead-time compensation or both in what proved to be successful components to create the detailed
model-predictive control must be and what did not. This approach P&IDs. Process designers certainly
considered. works best in those industries like can develop the simplified P&IDs.
The major nonprocess issue that power generation, paper making, and How do the control engineers go
frequently has a detrimental effect on oil refining that repeatedly use the about developing a P&ID? The first
loop performance is the control valve. same or very similar processes. But, step is always “you have to under-
Nonidealities such as hysteresis and it also is applied to the one-of-a-kind stand the process.” But, hopefully, the
stiction tend to increase with time, processes common in specialty process designers already do (if not,

Chemical Engineering Progress August 2000 21

PROCESS CONTROL

inappropriate P&IDs will be the least boilup, not by the overhead flow or technology cannot be applied blindly,
of our problems). Based upon this un- the reflux. however. The trick is to understand
derstanding, proceed as follows: • Check for dynamic separation. what these numbers are telling us
• List the controlled variables. Interaction is a problem only for about the process and our ability to
Process variable transmitters are loops with approximately the same control it.
required for each of these, al- response speed. If one loop is much It is not necessary to eliminate all
though, occasionally, a controlled faster than another, the fast loop must interaction; it only is necessary to re-
variable will be computed from be tuned first, and then the slow loop duce the degree of interaction to the
other measurements. can be tuned. The slow loop may not point where the loops will deliver
• List the manipulated variables. function when the fast loop is in man- satisfactory performance. Introduc-
A final control element is required for ual, but this is best addressed by im- ing ratios, sums, function generators,
each. From an instrumentation per- proving the reliability of the fast etc., also can prove beneficial in this
spective, the manipulated variable is loop. endeavor.
the signal from the control system to • Modify the menu of manipulat- Model-predictive controllers have
the final control element. But, for ed variables. You can introduce ra- been suggested as the ultimate solu-
control valves and pumps with vari- tios, sums, differences, function gen- tion to such problems. Unfortunately,
able speed drives, process engineers erators, etc., to replace one manipu- their implementation requires consid-
can consider the manipulated variable lated variable by another. For exam- erable effort by specialists knowl-
to be the flow. Where flow measure- ple, the stack oxygen from a combus- edgeable about the technology. Com-
ments are feasible, flow controllers tion process depends upon the fuel parable expertise also is required to
should be installed. From the instru- flow and the air flow. The stack oxy- keep them running. As originally pro-
mentation perspective, the manipulat- gen, however, really is a function of posed, model-predictive control was
ed variable is the set point of the flow the air-to-fuel (or fuel-to-air) ratio. an enabling technology that permitted
controller, but, from a process per- Instead of considering the manipulat- the benefits of plant optimization ef-
spective, the manipulated variable is ed variables to be the air flow and the forts to be realized. There certainly
the flow. fuel flow, consider the manipulated are applications where model-predic-
• Assess the degree of influence variables to be the fuel flow and the tive control is the preferred (and pos-
of each of the manipulated variables air-to-fuel ratio (or the air flow and sibly only) solution, but always iden-
on each of the controlled variables. the fuel-to-air ratio). Applying this tify the benefits (in economic terms)
For the initial cut, only a qualitative approach to a process requires sub- before proposing a model-predictive
assessment is required. Consider dy- stantial insight into the behavior of controller.
namic aspects, especially dead time, the process. There is no methodology • Check for departures from lin-
as well as steady-state aspects. For for determining where ratios, sums, earity. A change in the manipulated
each manipulated variable, identify etc., would be beneficial. But, cre- variable will prompt a change in the
all controlled variables that it signifi- ativity here can be very rewarding. controlled variable, the ratio being
cantly affects. Conversely, for each • Address the multivariable is- the sensitivity of the process. Will a
controlled variable, determine all ma- sues. For the remaining controlled given change in the manipulated vari-
nipulated variables that significantly variables, the potential for serious in- able lead to a small change under cer-
affect it. A matrix arrangement often teraction exists; this is not necessarily tain conditions, but to a large change
is convenient. the case, however. Multivariable is- under other conditions? This will lead
• Identify the simple loops. The sues can be approached in two ways: to tuning problems in PID loops, and
best situation is when a controlled Process understanding. If you un- to a degradation in the performance
variable is significantly affected by derstand the process well enough, of a model-predictive controller. A
only one manipulated variable; here, you should be able to sort out the in- change of 2:1 in the process sensitivi-
use single-loop PID control (except teraction issues. Most control engi- ty will degrade the performance of
for some special cases that we’ll dis- neers take this approach, but their re- the controls, but usually not to a suf-
cuss later). When the controlled vari- sults are not always perfect, which re- ficient degree to justify expending ef-
able is affected far more rapidly by flects deficiencies in their understand- fort to address it.
one manipulated variable than by any ing of the process. Control techniques are available to
other, single-loop PID control usually Quantitative measures of interac- address this problem, but each has its
proves adequate. For example, dy- tion (specifically the relative gain). limitations:
namic considerations dictate that the Especially when models provide the Cascade control. You must identify
bottoms composition for a distillation basis for the process design, obtain- a dependent variable that has a linear
column must be controlled by manip- ing quantitative values for the degree effect on the controlled variable. Such
ulating either the bottoms flow or the of interaction is relatively easy. This a dependent variable may not exist.

22 August 2000 Chemical Engineering Progress

 Adaptive control. An understand- below its constraint, the temperature some of which are not even contem-
ing of the process is necessary to de- in the reactor would be controlled. plated at the time. You are educated
velop the relationship upon which the But, should the pressure reach the to solve process control problems.)
adaptive controller is based. Sched- constraint, the controls switch from As processes become more
uled tuning is a simplistic form of controlling the temperature to con- complex, the latter alternative be-
adaptive control, but it also requires trolling the pressure. Such logic is re- comes more attractive. Indeed,
some process understanding to be ferred to as override control. today, process engineers generally
formulated properly. If violation of a constraint is a haz- are in the best position to address
Self-tuning controllers. These po- ard to people or equipment, it is the these problems.
tentially can respond to slow drifts in responsibility of the safety system to The customary initial report of a
process sensitivities, due to equip- take whatever action is appropriate process control problem is a phone or
ment wear, fouling of heat-transfer should the constraint be violated. For e-mail message that, in effect, states
surfaces, etc. They, however, cannot the pressure reactor, the pressure con- that some part (or perhaps all) of the
respond to rapid changes in process straint specified to the process con- controls do not work. If the initial re-
sensitivities. trols is less than the setting on the port gives any useful information at
The best approach is to find a ma- pressure relief device. It is the pres- all, consider yourself to be fortunate.
nipulated variable to which the con- sure relief device, not the process As with any type of problem, the
trolled variable is linearly related. controls, however, that ultimately is first step in its solution is to accurate-
• Assess the impact of distur- responsible for protecting people and ly define the problem. It may be with
bances. Where a disturbance has a equipment. the routine performance of the con-
major influence on a controlled trols or with how the controls reacted
variable, the performance of the Control problems to a specific event. In either case, we
control loop will suffer. There are in operating plants need to address two separate, but re-
two options: If the P&ID is appropriate, the lated, issues:
1. Provide control logic that plant startup team should be able to 1. What did the controls actually
rapidly responds to the disturbance. successfully commission the controls, do?
The following are possibilities: which includes such activities as con- 2. What was expected of the
Cascade control. You must identi- troller tuning. When difficulties are controls?
fy a dependent variable that is affect- encountered, there are several possi- Presumably “it doesn’t work”
ed by the disturbance more rapidly bilities. One is that the P&ID is not means that there is a difference be-
than the controlled variable is. Such a correct — that is, the control strategy tween these two.
dependent variable may not exist. does not properly reflect the charac- A good starting point is to un-
Feedforward control. The distur- teristics of the process. Basically, derstand what the operations per-
bance must be measured and logic in- there are two alternatives for address- sonnel expected the control system
corporated into the controls to take ing this: to do. This brings up a variety of
corrective action before the distur- 1. The control engineers can be- possibilities:
bance affects the controlled variable. come familiar with the process. Even • The controls responded differ-
Measuring the disturbance frequently if such individuals are on staff and ently than they were designed to do.
is the major obstacle. currently are available, becoming fa- There is something in the implemen-
2. Eliminate the disturbance at its miliar with a complex process will tation of the controls or associated
source. This often is the preferred ap- take some time. If an outside expert is hardware (measurement device and
proach. When the raw materials are retained, it is usually necessary to re- final control element) that must be
natural products, however, some vari- veal proprietary technology. corrected.
ability is inevitable. 2. The process engineers can be- • The controls responded as they
• Consider constraints on the de- come familiar with the principles of were designed to do, but this re-
pendent variables. There is one situa- process control. This is where the sponse was not appropriate. The
tion where the value of a dependent “just-in-time” education that is possi- P&ID needs a thorough analysis.
variable becomes of concern, namely, ble with computer-based technologies • The controls responded in an
when it exceeds limits known as con- has distinct advantages. (Note the use appropriate manner. Operations per-
straints. The process controls often of the term “education.” Training is sonnel need a better understanding of
include logic specifically designed to learning to execute a predetermined what control actions are appropriate.
respond should the value of the de- sequence of steps. You are trained to In this article, we’ll focus on the
pendent variable approach the con- change the tire on a car. Education in- first two possibilities.
straint. For example, consider a pres- volves mastering principles that can Determining what the controls ac-
sure reactor. Provided the pressure is be applied to a variety of situations, tually did involves two aspects:

Chemical Engineering Progress August 2000 23

PROCESS CONTROL

1. What were the process Does the measured value on which use such tricks. These can be very
conditions? the control action is based reflect the difficult to find.
2. What was the response of the current conditions within the process? In most cases, the next step is to
process controls? If the answer to this question is no, remove all smoothing.
It is essential to determine these then all other issues are irrelevant. Final control element. When the
two with a high degree of confidence. “Garbage in, garbage out” certainly final control element is a valve, it very
Problems with controls can occur applies. well could be the source of the prob-
when the process is operating at or Rarely can such a problem be re- lem. There are a couple of possibilities:
near normal operating conditions. A solved merely by calibrating the 1. The valve is not positioning to
visit to the plant usually is sufficient measurement device. It may have the value provided by the control sys-
to observe these problems firsthand malfunctioned, but, with the built-in tem; or
and to collect additional data. diagnostics in smart transmitters, this 2. The relationship between valve
Problems also can surface, howev- likely would be detected. Another position and flow through the valve is
er, during major upsets or other situa- possibility is that the instrument is very complex.
tions when the process operations are doing what it is designed to do, but is Although no firm data are avail-
far from normal. This is where a good not telling us what we really want to able, probably around one-third of all
historian is very useful. Especially know. For example, a temperature control problems are due to some de-
during periods of severe process up- transmitter is indicating the tempera- ficiency associated with the control
sets, the recollections of operations ture of a probe that usually is insert- valve.
personnel can be muddled. The histo- ed in a thermowell. What we really Changing the valve position af-
rian is unlikely to provide all of the want to know is the temperature of fects the flow through the valve,
information required. Key variables the process fluid surrounding the which, in turn, affects the process.
may not be recorded at all, others thermowell. These are not necessari- Thus, the valve flow characteristics
may be recorded at such a slow rate ly the same. Process engineers gener- impact the behavior of the controls.
that fast events are missed, and some ally are most adept at understanding In applications where the flow
of the data may not be accurate (mea- such issues. through the valve is high at times, but
surement devices can provide bad in- Smoothing and filtering also must low at other times (such as utility and
formation to the historian). The infor- be examined. An excessive degree of batch processes), the valve character-
mation from the historian generally filtering or smoothing on input sig- istics can significantly impact the per-
has to be supplemented by personal nals impairs the performance of the formance of the controls. If a flow
recollections. But, the time-stamped controls. The usual approach is that if measurement is available, the valve
data retrieved from the historian at the input is “bouncing” too much, characteristics can be determined by
least are indisputable. It usually is en- then increase the degree of filtering. stroking the valve and noting the flow
lightening to understand what they Instead of correcting the root prob- at various valve positions. Otherwise,
mean. lem, it is easier to hide it with heavy the valve characteristics have to be
There are five distinct possibilities filtering. calculated from the inherent valve
for the source of the problem with the In practice, very few inputs re- characteristics (supplied by the valve
controls: quire any filtering. The first step is to manufacturer) and the pressure-drop
1. a problem with the measure- identify all sources of smoothing. relationships for the other compo-
ment device; There are three possibilities: nents of the flow system. The data for
2. a problem with the final control 1. within the control system — such calculations are more readily
element, especially if it is a control the value of the smoothing coefficient available to process engineers than
valve; is always readily available; anyone else.
3. a problem with the controller 2. within the measurement device Although considerable attention
hardware; — with smart transmitters, the value has been directed to the problem,
4. an inability of the process to of the smoothing coefficient is readily oversizing of control valves continues
perform as expected; or available; with conventional transmit- to be common. Two issues usually
5. a mismatch between the char- ters, it is considerably more difficult arise with an oversized valve:
acteristics of the controls and those of to determine; or 1. At large valve openings, an
the process. 3. between the process and the oversized valve has little effect on
Let’s examine each of these. measurement device (such as partial- flow, which definitely will cause
Measurement device. The potential ly closing the block valve in the liq- problems for the controls.
of a problem with the measurement uid line connecting a displacer level 2. The sensitivity of flow to valve
device must be addressed at an early transmitter to the process) — instru- position increases as the valve is
stage in the analysis of the problem. ment technicians do not hesitate to oversized. Thus, nonidealities within

24 August 2000 Chemical Engineering Progress

the valve have a larger effect on con- mines what is possible. And, it occa- The developer of a P&ID does not
trol system performance when the sionally does this in some rather sub- need to know how to tune a PID con-
valve is oversized. tle ways that only can be appreciated troller. But, when analyzing a prob-
Occasionally, the converse is en- from a thorough understanding of the lem loop in a plant, don’t assume that
countered. As the capacity of the pro- process itself. Process engineers are a loop is well-tuned, even if the in-
cess is increased, flows also must in- clearly in the best position to sort this strument technicians contend that it
crease. This can lead to some control out. is. The analysis of control problems
valves being operated with the by- Controls not matched to process. in an operating plant must involve an
pass partially open at all times. The performance of the controls is assessment of how well the controller
Changing the opening of the bypass determined by two factors: is tuned.
valve will affect the performance of 1. the control structure, as repre- The first step is to determine the
the controls. sented by the P&ID; and tuning objective for the loop. Figure
As the mechanical components 2. tuning of the controls to the 2 illustrates two possibilities for the
wear, actuators exhibit hysteresis. process. response to a change in the set point:
The packing required to prevent leaks The tuning issue must be assessed one being the quarter decay ratio (that
around the valve stem resists move- first. is, the ratio of the first peak overshoot
ment of the valve stem, leading to a Tuning. Tuning is the procedure by to the second is 4:1), and the other
behavior called stiction. Both of these which the characteristics of the con- being critically damped (little or no
usually result in a cycle in both the troller are adjusted so that the con- overshoot). The tuning coefficients
controller output and the measured troller is “in tune” with the process. are given in Table 1.
variable. When a flow measurement Tuning techniques have been avail- Instrument technicians are taught
is available, the presence of these able since the 1940s. When digital that good performance is a response
phenomena can be easily detected. computers were first applied to pro- with a quarter decay ratio. That the
Are there times when the controller is cess control in the 1960s, automated quarter-decay-ratio criterion does
changing its output but the flow is not tuning was one of the promises. Yet, provide superior performance is illus-
changing? If so, then the actuator is despite considerable work and the trated by the responses to a distur-
not positioning the valve to the value availability of a number of commer- bance or load change (tuning coeffi-
specified by the controls. cial products, most loops still are cients are the same as in the previous
Controller hardware. In the days tuned with the traditional trial-and- example) shown in Figure 3.
of pneumatic and electronic analog error or “knob twiddling” approach. A good indication of loop perfor-
controllers, hardware malfunctions
within the controller itself were po- ■ Figure 2.
tential problems that had to be con- Two tuning 170
sidered. But, today’s digital hardware approaches.
generally is either working perfectly 4.8˚F
165 Decay Ratio = 1/4
or not working at all. With the self-di-
agnostic features normally incorpo- 1.2˚F
Temperature, ˚F

rated into computer-based industrial 160

control products, problems with the Set Point
hardware usually are very obvious.
155
Both pneumatic and electronic Critically Damped
analog equipment are susceptible to
significant errors in the calibration of 150
one or more of the controller tuning
adjustments. With digital controls,
145
tuning-adjustment calibration errors -15.0 0.0 15.0 30.0 45.0 60.0
are simply not possible.
Process. The basic question is
“Can the process actually do what the
controls are expected to make it do?” Table 1. Tuning coefficients for Figure 1.
With the “our system can do anything Controller Gain, KC Reset Time, TI
you want” sales hype from manufac- %/% min
turers, people sometimes have unreal- Quarter decay ratio 2.0 5.0
istic expectations for the process con- Critically damped 0.5 5.0
trols. The process ultimately deter-

Chemical Engineering Progress August 2000 25

PROCESS CONTROL

mance is the maximum deviation ■ Figure 3.

from the set point (the magnitude of Quarter decay ratio 155
the control error at the peak). For our provides better
response.

Temperature, ˚F
example, the quarter decay ratio gives Decay Ratio = 1/4
a deviation of 4.4°F vs. 7.2°F for crit- 150
ically damped.
Production personnel, however, 145 Critically Damped
often are not comfortable with the de-
gree of oscillations in responses with
140
a quarter decay ratio. When tuned this -15.0 0.0 15.0 30.0 45.0 60.0
way, increasing the sensitivity of any
element in the loop by approximately Time, min
a factor of two results in an unstable
loop. For most production personnel,
■ Figure 4.
this is a little too close for comfort. Many combinations 170
When you are involved in plant oper- of coefficients give
ations, you do not want your phone to quarter decay ratio. KC = 2.0 %/% KC = 0.5 %/%
ring at 2:00 in the morning because 165
TI = 5.0 min TI = 1.8 min
some loop is unstable. You want de- Temperature, ˚F
pendable controls, and you are will- 160
ing to sacrifice performance to en- Set Point
hance dependability. After all, you,
not the instrument technicians enam- 155
ored with quarter decay, will get that
2:00 am call.
150
A widespread malady in con-
troller tuning is for the reset time to
be too short, because of the common 145
-15.0 0.0 15.0 30.0 45.0 60.0
misconception that fast response is
attained through a short reset time. Time, min
This is not the case at all. The con-
tributions of the three modes are as
follows: ■ Figure 5.
Proportional — determines speed Faster response offers 155
benefits.
of response of the loop.
Temperature, ˚F

Integral or reset — forces the loop 150

to line out at its set point (eliminates
offset or droop).
A
Derivative — enhances stability 145
B
margin, which, in turn, permits the
controller gain to be increased to ob- 140
tain a faster response. -15.0 0.0 15.0 30.0 45.0 60.0
Compared to the controller gain,
Time, min
the reset time has little effect on the
speed of response of a loop. To obtain
fast response, make the controller
gain as high as possible, not the reset gain of 0.5%/% and a reset time of with the higher controller gain, not
time as short as possible. 5.0 min gave a critically damped re- with the shorter reset time. The bene-
A major deficiency with using the sponse. Shortening the reset time to fits of the faster response are more
quarter-decay-ratio performance ob- 1.8 min, however, gives a response clearly illustrated by the load re-
jective is that, for PI and PID con- with a quarter decay ratio. Figure 4 il- sponses, as seen in Figure 5 and sum-
trollers, many combinations of the lustrates two responses — both have marized in Table 2. Note that the
tuning coefficients will give a re- a quarter decay ratio, but the respons- faster response (and the one whose
sponse with a quarter decay ratio. In es certainly are not equivalent. maximum deviation from the set
the previous example, a controller The fastest response is the one point is the lesser of the two) has the

26 August 2000 Chemical Engineering Progress

highest controller gain, not the short- TI’ ≈ 0.5 P tan–1 (2π TI /P) (1) trol valve). Flow loops are so much
est reset time. If the reset time is less than half faster than all other loops that there
Many instrument technicians take of this value, the controller should rarely is any incentive to make them
pride in being able to tune a loop be retuned. A good approach is to respond as fast as possible. The typi-
quickly. One approach to do so is to set the reset time to the value com- cal tuning for a flow loop should be a
set the controller gain to some value puted by Eq. 1, and then adjust the controller gain of about 0.2%/%, a
(how this value is obtained is ob- controller gain to obtain the desired reset time of about 0.05 min (3 s),
scure) and then adjust the reset time performance. and no derivative.
to obtain a response with a quarter Table 3 presents the values of TI’ Loops where the flow through the
decay ratio. The technicians consider computed for the two responses pre- control valve is a disturbance to a
the loop to be well-tuned. The usual sented previously. This analysis sug- downstream process. This most com-
result of this approach, however, is a gests that the loop with a controller monly is encountered in level and
controller gain that is too low and a gain of 2.0%/% and a reset time of pressure loops. If the loop is tightly
reset time that is too short, which re- 5.0 min is reasonably tuned, but the tuned, the disturbances to the down-
sults in a slow responding loop. But, loop with a controller gain of 0.5%/% stream process offset the benefits
instrument technicians don’t like to and a reset time of 1.8 min is not. from maintaining the measured vari-
discuss whether a loop whose re- Because 2π TI/P is always posi- able close to its set point. Such loops
sponse is a quarter decay ratio might tive, then have to be tuned with a low controller
not be well tuned. 0 < tan–1 (2π TI/P) < p/2 gain and a long reset time. These
When digital controls are used, and loops also are candidates for nonlin-
there is a way to determine how well 0 < TI’ < (π/2) P ear (error squared or error deadband)
a loop is tuned. As noted earlier, a or, approximately, versions of the PID control equation.
major advantage of digital controls is 0 < TI’ < 0.78 P In reality, the following statements
that there is no error in the tuning pa- If you do not like to compute arc- apply to controller tuning:
rameter adjustments. If the reset time tangents (although with today’s cal- • The value of the reset time
is set to 1.44 min, the reset time is culators, it is hard to understand solely depends upon the characteris-
1.44 min. This statement is rarely true why), the following approximation is tics of the process. This statement
for either pneumatic or electronic satisfactory: also applies to the derivative time.
controllers. TI’ ≈ (0.78 P)/[1 + P/(2π TI)] (2) • Only the value of the controller
Suppose a loop has been tuned so This analysis assumes that the gain depends upon the performance
that the response exhibits a cycle, al- loop should be tuned to respond as objective.
though not necessarily a quarter fast as possible. Most loops should Setting the controller gain to some
decay ratio. Determine P, period of be so tuned, but, as always, there value and then adjusting the reset
the response, in min, and TI, the cur- are exceptions: time to attain the desired performance
rent value of the reset time, in min. Flow loops (specifically those for (the approach often used by instru-
The appropriate value, TI’, for the which the flow meter is immediately ment technicians) is totally contrary
reset time is given by: upstream or downstream of the con- to these statements. The following
approach to controller tuning is more
Table 2. Data for responses shown in Figure 5. consistent with them:
1. With the controller gain set to
Controller Gain, KC Reset Time, TI Maximum Deviation From Set Point some value (usually a low value),
%/% min °F shorten the reset time until cycling is
Response A 2.0 5.0 4.4 observed (the cycle does not have to
Response B 0.5 1.8 6.4 exhibit a quarter decay ratio). Note
the period of the cycle and the current
value of the reset time.
2. Change the reset time to the
Table 3. Values computed for the two responses. value of TI’ computed using the rela-
tionships presented previously.
Controller Gain, KC Reset Time, TI Period, P Appropriate Reset Time, TI’ 3. Adjust the controller gain so
%/% min min min that the response is consistent with
Response A 2.0 5.0 12.5 7.5 the appropriate performance objec-
Response B 0.5 1.8 23.0 5.3 tive (quarter decay ratio, critically
damped, or whatever) for the loop.
P&ID deficiencies. Just because a

Chemical Engineering Progress August 2000 27

PROCESS CONTROL

loop is specified on a P&ID does not that can cope with the changes in pro- dynamics almost always is assured.
mean that, in fact, it will work. The cess characteristics. But, when a temperature controller
control structure indicated on the Dead time. The dynamic charac- provides the set point to another tem-
P&ID might not properly reflect the teristic that presents the most difficul- perature controller, the separation of
process requirements. The symptom ty for the PID control equation is dynamics may not be adequate. Fur-
of this is one or more loops that can- dead time. Most processes exhibit thermore, temperature processes al-
not be satisfactorily tuned. The fol- dead time, but to a modest extent. ways are interacting stages (each
lowing are the common process This results in some deterioration in temperature affects the other temper-
problems that lead to untunable the performance of a PID controller, ature), which complicates the inter-
loops: but not enough to justify implement- pretation of the responses in the inner
• process nonlinearities; ing an alternative approach. loop temperature. These control con-
• excessive dead time; As the dead time rises, the first figurations have to be analyzed by
• improperly nested cascade; and mode to suffer is derivative. The ben- someone very familiar with the char-
• interaction between controlled efits from derivative decrease as the acteristics of the process.
variables. dead time increases. The next mode Interaction. All processes are multi-
It is easy to throw technology at to suffer is proportional. As the dead variable in nature. If ten variables are
such problems, resorting to automat- time goes up, the controller gain must to be controlled, most P&IDs will pro-
ic tuning, expert systems, model-pre- be reduced to maintain an acceptable pose to do this with ten individual loop
dictive controllers, etc. But, until the margin of stability. For long dead controllers that are paired with ten final
problem is understood and the P&ID times, the PID controller essentially control elements. A single-loop con-
revised, the loop will not perform will reduce to a pure reset controller. troller works well in a multivariable
satisfactorily. The controller may operate on auto- environment, however, only when the
Process nonlinearities. A nonlin- matic, but the response is so slow that following statements are true:
ear process exhibits different behav- the controller is practically useless. • The final control element for
ior under different process conditions. For processes with a long dead the loop does not affect the measured
Most processes are nonlinear with re- time, a model-based control technolo- variables for the other loops.
gard to process throughput. Process- gy known as dead-time compensation • None of the final control ele-
es, however, can exhibit nonlinear be- is available and has been routinely ap- ments for the other loops affect the
havior with regard to other variables plied to paper machines since the late measured variable for this loop.
as well. Batch processes are particu- 1960s. Model-predictive controllers When these statements are not
larly notorious in this respect. also provide dead-time compensation. true, the result is loop interaction. A
In such cases, the controller can be There, however, is one prerequisite to small degree of interaction can be tol-
satisfactorily tuned at a point in time. the successful application of dead- erated, but even modest degrees of
But when process conditions shift, time compensation: a good value interaction will degrade the perfor-
the tuning is no longer satisfactory. must be available for the process dead mance of one or more loops. Some-
The logical starting point is to tune time. In paper machines, the dead- times, the interaction can be suffi-
the controller with the process operat- time compensator is, in turn, compen- ciently reduced by incorporating ra-
ing at or near normal (or design) con- sated for the effect of changes in ma- tios, summers, function generators, or
ditions. All processes, though, are chine speed on the dead time. other simple elements into the control
subjected to major upsets from time Improperly nested cascade. In a configuration. For the more severe
to time. How does the controller per- cascade control configuration, one situations, a model-predictive con-
form under the process conditions controller (the outer controller) pro- troller may be the best (or perhaps
during these upsets? We normally vides the set point to another con- only) solution.
learn the answer through operational troller (the inner one). For this con-
experience. This is one source of the figuration to perform satisfactorily,
“it doesn’t work” messages. The first the dynamics of the inner loop must Upcoming AIChE control courses
step is to determine the process con- be faster than the dynamics of the
In early November in Los Angeles, Dr.
ditions that led to the problem. The outer loop. The usual desire is that Smith is scheduled to give courses on “Auto-
controller can be retuned to these the inner loop be at least five times matic Control of Processes,” “Control of Batch
conditions (the “one tuning fits all” faster than the outer loop. An inade- Processes,” and “Distillation Control.” For
approach). This, however, usually quate separation of dynamics leads to more information on these courses, or to re-
leads to some sacrifice in perfor- tuning problems in the outer loop. ceive a catalog on all AIChE courses, contact
mance at the normal operating condi- When the cascade is a temperature Nina Weber, Director of Education Services
tions. The alternative is to develop a controller providing the set point to a ((212) 591–7526; E-mail: [email protected]).
more sophisticated control structure flow controller, such a separation in

28 August 2000 Chemical Engineering Progress

The role of tuning; tuning must be appreciated, ential equation; the dynamic charac-
for process engineers however, by those who resolve con- teristics of processes are described by
Process control issues can be di- trol problems in the plant. differential equations. Engineers do
vided into two categories: Input/output (I/O) elements. This not need to solve either. The control
Systems issues. These include con- is limited to those aspects of mea- systems solve the PID control equa-
figuring the hardware, building points surement devices and final control el- tions; various simulation packages
(for inputs, outputs, control points, ements that affect control system per- are available to solve the differential
etc.), creating graphic displays for the formance. Measurement device selec- equations that describe the process.
process operator, and so on. This tion (thermocouples vs. RTDs vs. py- But, in process control, it is necessary
“grunt work” can be done with very rometers vs. …) and valve sizing are to work with differential equations.
little, if any, understanding of the pro- not required. Some familiarity with This is where the Laplace transform
cess. The systems issues are ade- filtering and smoothing is needed so enters; as noted previously, differen-
quately covered by training courses that the engineer will not hesitate to tial equations are very conveniently
(yes, training, not education) offered remove them. represented in the Laplace domain.
by suppliers. Process engineers cer- Split range. This handles situations It is understandable why a process
tainly can do this if they are so in- where, under some conditions, a vari- engineer’s initial reaction would be to
clined, but these tasks should be left able is controlled using one manipu- attempt to avoid the Laplace trans-
to others to free the process engi- lated variable, but, under other condi- form entirely. The above topics can
neers’ time for the more important as- tions, using a different manipulated be presented without using the
pects of the control system. variable. Most plants have a couple Laplace transform at all. Laplace ex-
Process issues. At the design of such applications. pressions, however, actually make the
stage, these center on creating the Override control. This technology presentation both easier and clearer.
P&IDs for the process controls. Dur- is employed to address the constraints In the final analysis, it is better to join
ing and after startup, these involve re- on production operations. The track- them than to fight them.
solving problems with the process ing issues must be understood. But, joining them does not require
controls. Process issues are the im- Cascade control. The advent of becoming an expert in the subject.
portant aspects of the control system, digital controls has led to an increase Many of the topics (such as partial
and it is crucial that they be done cor- in the application of cascade control. fraction expansion) taught in college
rectly. Both endeavors require a sub- Process engineers are in the best posi- courses never are used in the field.
stantial understanding of the process tion to recognize opportunities for What a process engineer needs to
— so, process engineers are in the this technology, and, probably, will know about Laplace transforms can
best position to do them. have to tune most of the temperature- be conveyed in 15 min. Instead of
Yet, a lot of process engineers do to-temperature cascades. trying to avoid the subject entirely, it
not feel comfortable about addressing Ratio control. This simple technol- is best to invest the 15 min. Then,
the process issues. Many of these en- ogy can be understood quite quickly. you will not be intimidated by the
gineers probably owe their misgiv- Feedforward control. Most process Laplace expressions that occasionally
ings to the “process control” course engineers need only an introduction appear in the manuals provided by
they endured at university — it more to this topic, although a few should control system manufacturers. And, it
than likely emphasized mathematics have an in-depth exposure. only takes a little knowledge of
and gave the wrong impression about Multivariable control. All process Laplace transforms to keep the con-
the practice of process control. engineers must understand how inter- trol engineers honest. CEP

To address process issues requires an action can lead to problems in control

appreciation of the following concepts: systems, and how to recognize when
Process characteristics. The focus interaction is the source of controller
should be on how steady-state sensi- tuning difficulties. Some process en- C. L. Smith is President of Cecil L. Smith,
tivities and simple process dynamics gineers must know how to quantita- Inc., Baton Rouge, LA ((225) 761-4392;
(integrating vs. nonintegrating, time tively assess the degree of interaction Fax: (225) 761–4393; E-mail:
[email protected]), a firm
constants vs. dead times, etc.) impact and how to develop control configu- specializing in automation technology.
control loop performance. rations that have acceptably low de- He is the author or co-author of 14
PID control equation. Topics in- grees of interaction. Topics such as books and over 100 articles. He
clude the role of the three modes, the model-predictive control can be left received BS, MS, and PhD degrees
in chemical engineering from
options available in digital systems to the control experts, though. Louisiana State University.
(especially tracking features), and Let’s conclude by putting the
tuning. Actually, P&IDs can be de- proper perspective on mathematics.
veloped with very little understanding The PID control equation is a differ-

Chemical Engineering Progress August 2000 29