B ac k to B a si c s

Empirical Process
control: Part i
Harold Reed ■ sirius technical services

Good process control requires more than just correctly tuning

the controller. First, you need to understand the parts of the process
to determine the necessary type of control.

U
niversity control courses typically concentrate on the in the control room and pinpoint the ones that are chasing a
mathematical complexities involved in modeling a setpoint or cycling (Figure 1). Issues related to hardware and
control system, but they do not spend much time on system configuration, which are critical to good control, may
the actual equipment used to do the controlling (1). While be at the root of these control issues. This section describes
I learned about transfer functions, LaPlace transforms, and some of the problems that can be caused by hardware and
other technical details in school, when I got my first job, I system configurations. The examples below assume that a
did not even know how a proportional-integral-derivative liquid flow is being controlled by a valve.
(PID) controller worked! In my case, I learned the basics Oversized valves. If a valve is too large for the process,
from an instruction manual for a pneumatic controller. more fluid than is needed will pass through the valve as the
Good process control requires more than just correctly controller opens it, increasing the flowrate past the setpoint.
tuning the controller. It is critical to first understand the As a result, the controller will then close the valve. As the
interactions between the parts of the process to decide what flowrate decreases past the setpoint, the controller will
type of control is needed. This article describes how process start to open the valve again. The flow will then shoot up
equipment can impact process control, the external factors once more and the controller will again close the valve.
that affect controller tuning, and the types of tuning that can This action creates a kind of on-and-off control that pro-
be implemented. 85

80
Equipment problems can affect process control
Tuning a system that is being used for the first time 75
Pressure

requires a lot of experience. Typically, if a control system 70

is being designed for a new process, the design team will 65
specify a standard PID control loop and rely on experienced
60
control engineers to do the initial tuning. The methods
described in this article apply only to a system that is already 55
in operation. 50
Identifying a loop with control problems is usually easy.
Most of the time, you can simply look at the process plots p Figure 1. This process plot shows pressure cycling in a stripping column.

38 aiche.org/cep June 2021 CEP

Copyright © 2021 American Institute of Chemical Engineers (AIChE).
Not for distribution without prior written permission.
Rev 6/7/21
duces a series of bumps in the valve plot. The amount of hysteresis that occurs in the valve is
Figure 2 is a process plot for a system that needs at least called its deadband. Sometimes deadband can be lowered by
42 ft3/hr of steam to achieve the desired liquid temperature at tightening the linkage between the valve stem and the actua-
the exit of a heat exchanger. When the flowrate approaches tor. But if the hysteresis cannot be corrected mechanically,
42 ft3/hr, the temperature approaches its setpoint and the a valve positioner may be needed. Valve positioners act as a
controller begins to open the valve to maintain the liquid mini control system for the valve, measuring the position of
temperature. However, the valve is oversized, so when it is the valve stem and adjusting the actuating pressure to move
just 8% open, too much steam flows into the exchanger. The the valve to the desired location. A standard valve actua-
temperature rapidly increases and the valve closes. When the tor (i.e., without a positioner) adjusts the actuating pressure
liquid’s exit temperature begins to fall, the cycle repeats. to a set value and assumes that the valve will reach the
Even if the valve does not close completely but oper- desired location.
ates at less than 20% open, the valve gain (i.e., change in Hysteresis can also be overcome by the integral function
output/change in input) will be large, which will cause a in a PID controller, which continues to move the valve in the
large change in the process variable (PV) for a small change desired direction until the setpoint flow is reached.
in the controller output (OP). For example, opening the Stiction. Valves may stick as they open and close; this
valve from 50% to 51% is a change of only 2% of the valve is called stiction. As the signal increases, the valve does
setting. Increasing the valve from 20% to 21% is a change not move until the pressure is high enough to overcome the
of 5% of the valve setting. This larger gain will make the resistance. When it does, the valve jumps to the new posi-
controller’s job more difficult, because it has to control the tion. Then, as the signal decreases, the valve sticks again
output more precisely. until the pressure is low enough that it jumps in the opposite
To function properly, valves should be sized to be direction. This often occurs in valves with packed stems
30–70% open during normal operation. This value depends when the packing is compressed too tightly.
on the range of flowrates needed by the process. If the If the stiction is severe, it will produce a series of steps
desired operational flowrate is always the same (i.e., no in the plot of the actual flow as the valve suddenly opens or
changes in production rate), the valve could be sized to oper- closes to jump to the new position. In many cases, stiction
ate at 70–80% open under normal conditions. This design is not severe enough as to produce this stepwise pattern.
will enable selection of the smallest and cheapest valve for Instead, valves stick in a position until the controller applies
the application. enough pressure that the valve suddenly moves. The valve is
Hysteresis. When valve opening or closing lags behind typically then able to move around well until the controller
the signal and the valve position ends up in a different place stops moving it and it sticks again (Figure 3).
than intended, it is called hysteresis. If a valve that produces In the plot in Figure 3, the flow is initially about
100 gal/min of flow at 50% open is opened to 60% and then 77 gal/min and falling, but the controller wants the flow to
lowered back to 50%, it should give a flow of 100 gal/min increase to 79 gal/min. Starting at two minutes, the control-
again. If, for example, it now produces 105 gal/min, the ler increases the output to the valve from 30% to 38% and
valve is suffering from hysteresis. nothing happens. Then, the valve stem breaks loose from

60 100
Flowrate 90 Flowrate
50
80
%Open or Flowrate, gal/min
%Open or Flowrate, ft3/hr

70
40
60
30 50
40
20
30
Controller Output 20 Controller Output
10
10
0 0
0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100
Time, min Time, min

p Figure 2. If a valve is oversized, too much fluid will pass through it as the valve p Figure 3. If stiction is severe, it produces a step pattern in the process flow
opens, which will trigger the controller to close the valve. The flow will cycle up plot. However, it more often produces the pattern of peaks and valleys shown here
and down as it tries to achieve the setpoint. as the valve sticks, overcomes the resistance, moves easily, and then sticks again.

CEP June 2021 aiche.org/cep 39

Copyright © 2021 American Institute of Chemical Engineers (AIChE).
Not for distribution without prior written permission.
 B ac k to B a si c s

the grip of the packing and opens to 38%, which suddenly open. These types of valves are suitable for applications
increases the flow from 75 gal/min to 90 gal/min. The where you might need many times the normal flow in an
controller immediately starts trying to close the valve to get emergency. For example, the equal-percentage valve could
the flow back below 80 gal/min. The unstuck valve is able to be used to control the temperature of a reactor. The valve
move freely in response to the controller, bringing the flow might normally operate at 30% open, but if the temperature
back down to around 75 gal/min. The valve stops moving starts to rapidly increase, the valve can be opened to 70%,
and becomes stuck when it is about 30% open at 23 min. providing a high flowrate of cooling water to the jacket to
The controller starts trying to open the valve once more and lower the reactor’s temperature.
the cycle repeats. Linear valves provide good control across a broad range
To identify if stiction is a problem, look for a sudden of flows. They are suitable for systems that need to operate
increase in the flowrate after the OP has been increasing for at a range of production rates.
a while. If an increase in OP causes the flowrate to increase Interactions between pumps and valves. There are many
slowly, the problem is just a long response time in the different types of pumps. Some are positive displacement
control loop, even if the flowrate initially does not respond pumps, such as lobe, tube, and piston pumps. Others are
for several minutes. If the flowrate remains unchanged for centrifugal pumps, which spin the liquid around in a cham-
several minutes and then suddenly increases, the problem is ber to create the pressure needed to make it flow through a
a sticking valve. pipe. Centrifugal pumps present a special kind of problem
A valve suffering from stiction is difficult to tune and for control systems because as the flow increases, the driving
usually requires a mechanical solution if the amount of pressure that provides the flow decreases (Figure 5).
variation it produces is a problem for the process. Lubricat- Consider a process that requires a flowrate of
ing the valve stem may eliminate stiction. Slightly loosening 20 gal/min and the pressure in the pipe is 30 psig. A valve
the nut that confines the packing to reduce the pressure on could be sized to achieve this flowrate with a pressure drop
the stem may also work, but the valve needs to be monitored of 15 psi (i.e., half of the line pressure) at 50% open. This
to ensure there is no leakage around the stem. seems reasonable, but if a centrifugal pump provides the
Wrong valve type. The three main types of control valves pressure to drive the flow, it may not be able to achieve the
include quick-opening, linear, and equal-percentage valves higher flowrates required during upsets (e.g., startups, shut-
(Figure 4). Quick-opening valves provide a large opening at downs, changes in production rate). Assume that the 30 psig
low signal values; at higher signal values, the opening size pressure is equal to a head of about 66 ft of liquid. The
increases more slowly. Equal-percentage valves do the oppo- pump generates lower head pressures at higher flowrates.
site — they provide a small opening at low signal values, Therefore, opening the valve to increase the flowrate from
and as the signal increases, the opening size increases more 20 gal/min to 25 gal/min would lower the available head
quickly. Linear valves open in proportion to the signal value from 66 ft to 53 ft — a decrease of 20%. If the valve has to
throughout the range. operate at 90–100% open to achieve the desired flowrate, the
If a quick-opening valve is run at 25% open, control will process will be very difficult to control.
be difficult because a small change in the valve position will A large, equal-percentage valve with a low pressure
produce a large change in the flow. The ideal quick-opening drop would be an appropriate valve for this service because
valve controls best in the range of 40–70% open. The ideal the increasing rate at which this valve opens would help to
equal-percentage valve controls best in the range of 30–60% offset the decreasing pressure provided by the pump. For

100
90
90 80
80 Quick
70
Opening
70 60
Percent Open, %

Head, ft

50
60
Linear 40
50
30
40 20
30 10
Equal
20 Percentage
t Figure 4. These 0
plots show the 0 5 10 15 20 25 30
10 Flowrate, gal/min
idealized function of
0 quick-opening, equal-
0 10 20 30 40 50 60 70 80 90 100
percentage, and linear p Figure 5. This typical head vs. flowrate curve for a centrifugal pump illustrates
Stem Position, % control valves. that at high flowrates the achievable head drops off.

40 aiche.org/cep June 2021 CEP

Copyright © 2021 American Institute of Chemical Engineers (AIChE).
Not for distribution without prior written permission.
an existing system, the control issues could be addressed
by recommending a larger valve trim or a larger impeller When a piece of equipment has control
for the pump. Modifying the pump with a larger impeller
would flatten the pump curve at 20 gal/min and could help to problems, first look at the inputs to the system.
stabilize operation. However, this will likely increase energy control problems in a column or reactor
costs, as centrifugal pumps typically operate most efficiently could be due to problems upstream.
in the region to the right of the flat portion of the head curve.

Look for external causes first increases to send more of the low-boiling component back
When a piece of equipment has control problems, first down the column to reduce the temperature. This puts more
examine the inputs to the system. Control problems in a load on the bottom of the column and causes the temperature
column or reactor could be due to problems upstream. If, for of the lower trays to decrease. The reboiler control system
example, the feed to a column or the flow of steam to the responds by increasing the boilup and sending more of the
reboiler is fluctuating, then the control system cannot find a high-boiling component back up the column. This increases
good control point and will always be searching for a solu- the temperature of the upper trays and the cycle starts again.
tion. Look for variations in the feed rate, feed composition, To settle this fight between the two controllers, decide
feed temperature, system pressure, steam pressure, chilled which output is most important. Is it more critical to keep the
water temperature, etc. These external variations may be high-boiling component out of the distillate or to keep the
causing the problem, so they should be addressed first. Once low-boiling component out of the bottoms? If keeping the
the upstream variations are fixed, the control problem in the low boiler out of the bottoms is more important, then keep
column or reactor may resolve itself. the bottoms temperature controller as-is and loosely tune the
In one system that I tuned, I found that an entire section reflux ratio controller. The bottom will hold its specification
of the process seemed to have control problems that caused and allow the top temperature to hover around the setpoint.
cycling. I traced it to a column that had a steam load and Any variation in the composition will be evident in the
tray temperature that were constantly oscillating. I checked distillate. When this control method was applied to an actual
the inputs to the column and found that the feed rate varied column, it allowed good automatic control of a system that
significantly. I traced the feed flow variation back to a surge was so unstable it had been run manually for years.
tank that was collecting the bottoms from a stripping column
further upstream. What type of control do you need?
The level in the tank was being tightly controlled at 50%, Most people who tune a controller or use a control pro-
but there was no reason that it had to be held at precisely gram to do the tuning assume that the objective is to achieve
that level. The level could vary as long as the tank did not rapid response on a single device. But this is not always the
empty out or overflow. I loosened the level control on the case. There are four basic types of control:
tank to allow the level to vary between 40% and 60%. The • Rapid response allows the process to vary but attempts
valve would now only make small adjustments as the level to bring the PV back to the setpoint within a short period
rose or fell, which allowed us to smooth out the variations in of time.
the tank’s discharge flowrate and eliminate the big swings in • Slow response can help solve operational issues. A
the column’s feed flowrate. Once the feed to the column was piece of equipment can be set to have loose control, such as
steady, the control problems were eliminated and the cycling the level of the surge tank described in the previous section,
issues downstream were solved. which allows the equipment to absorb upstream variations.
Another source of control problems is the setup of the • Hybrid response, also called cascade control, is used
control system itself. Some equipment may have two or when a control variable that is manipulated to affect a
more control systems. Distillation columns, for example, process value (e.g., temperature, pressure, flowrate) has
typically have a control on the reflux ratio and on the a variable source. Cascade control can be effective, for
reboiler heat load (2, 3). The reflux ratio is adjusted to example, if steam is used to control a process temperature,
control the temperature on an upper tray to meet an over- but the flowrate of steam from the header varies due to
head purity specification, and the reboiler is adjusted to variable header pressure. A main loop reads the temperature
control the temperature on a lower tray to meet a bot- and decides how much steam is needed. It then sends this
toms specification. information to a secondary loop that regulates the steam
This control scheme sounds good in theory, but it cre- flow. The secondary loop has a rapid response and the main
ates a system where the two controllers are fighting each loop has a slow response.
other. If the upper temperature gets too high, the reflux ratio • Tight control may be needed when small deviations in

CEP June 2021 aiche.org/cep 41

Copyright © 2021 American Institute of Chemical Engineers (AIChE).
Not for distribution without prior written permission.
 B ac k to B a si c s

parameters like pH or temperature can cause product loss or

dangerous conditions. The emphasis is not on minimizing
the time of the response but on minimizing the amount of
variation from the setpoint. Usually, increasing the propor-
tional and integral control parameter values can achieve the
necessary level of control response. (Derivative control is
sometimes used for tight control, but it is a more complex
procedure and is not covered in this article.)
More advanced control schemes can be developed that
involve material or energy balances around a piece of equip-
ment or a part of the process. This is necessary in certain
instances, such as to estimate the amount of unreacted raw
material accumulating in a reactor to avoid a runaway reac-
tion. However, such control schemes should not be imple-
mented if a simpler control strategy will work. Complex
systems require that all of the sensors work properly. If
one temperature or flowrate sensor drifts, the whole system
EXPLORING
A NEW
becomes unreliable because the material or energy balance
will be inaccurate.
On the other hand, if a temperature or flow sensor drifts
in a simple PID system, the setpoint can be lowered or raised
in that part of the system and the rest of the process can con-
tinue to run until repairs can be made. If a sensor malfunc-
tions completely, the PID system can be run on manual until
CAREER?
it can be repaired. Simple systems are more rugged, and with
proper tuning, PID systems can control almost any process. Check out our new
in closing
Career Planning Tool
This article describes some of the items that can cause Occupational insights to help
control problems, the types of control that are available, and
some actions to take before starting to tune. you plan your career path
In an article to follow, I will describe the basics of the
PID controller and two different methods that can be used to
tune the controls for live systems. CEP

Literature Cited
1. Alford, J., and G. Buckbee, “Industrial Process Control
Systems: A New Approach to Education,” Chemical Engineering
Progress, 16 (12), pp. 35–42 (Dec. 2020).
2. Herzog, C., “Develop Distillation Controls During Process
Design,” Chemical Engineering Progress, 17 (2), pp. 31–37 careerengineer.aiche.
3.
(Feb. 2021).
Taube, M. A., et al., “Achieve the Benefits of Intensification
through Process Control,” Chemical Engineering Progress,
org/careerinsights
17 (3), pp. 44–49 (Mar. 2021).

CareerEngineer
HAROLD REED is a process technical engineer at sirius technical
services (theodore, aL). With more than 40 years of experience
in the chemical industry, he has worked for Dow corning, DuPont,
ciba-Geigy, ciba, Novartis, syngenta, and sirius technical services, FUELED BY AIChE
including roles in laboratories, pilot plants, and full-scale production
processes. His areas of interest are process modeling, data analysis,
process control, and experimental design. Reed has a bachelor’s
degree from the Univ. of kansas and a master’s degree from Lamar
Univ., both in chemical engineering. © 2021 AIChE 6120_21 • 05.21

42 aiche.org/cep June 2021 CEP

Copyright © 2021 American Institute of Chemical Engineers (AIChE).
Not for distribution without prior written permission.
 B AC K TO B A SI C S

Empirical Process
Control: Part 2
Harold Reed ■ Sirius Technical Services

Proportional-integral-derivative (PID) controllers can handle most

control problems if properly tuned. This article describes two basic
methods of controller tuning.
Proportional (P). Proportional control responds to the

G
ood process control relies on proper controller tun-
ing. However, you must first understand the parts of error directly. This control method tries to stop the error
the process to determine the necessary type of con- from changing. As the error increases, the valve opening
trol. Part 1 of this article (June 2021, pp. 38–42) described changes in direct proportion to the change. The tuning coef-
some of the factors that may affect controller tuning and ficient may be expressed as a proportional gain or a propor-
how to deal with them (1). Part 2 describes the basics of tional band. If gain is used, increasing the coefficient will
proportional-integral-derivative (PID) control and outlines increase the response. Proportional band has just the oppo-
two methods for tuning live systems without the need for site effect, where increasing the coefficient will decrease the
tuning formulas or knowledge of equipment characteristics. response. Proportional gain is assumed in all of the examples
Instead, the process itself provides the feedback information used in this article.
needed for proper control. Integral (I). Integral control, also referred to as reset con-
trol, responds to the total error. It tries to drive the PV back to
Basic PID control theory the SP and drive the current error value to zero. This method
A control system has three variables: the setpoint (SP),
1,100 45
the process variable (PV), and the output (OP) (Figure 1). SP
The PV is the parameter to be controlled (e.g., tempera- 1,050 PV 40

ture, flowrate, etc.), the SP is the desired value for the 1,000 OP 35
PV, and the OP is the value sent from the controller to the 950 30
PV, gpm

OP, %

control element. 900 25

For ease of description, we will use a flow that is con- 850 20
trolled by a valve on a feed stream. When the value of the
800 15
PV is not the same as the SP, the difference (PV – SP) is
called the error. All of the controller actions are assumed to 750 10

be based on this error value, although some controllers use 700 5

the PV directly instead of the error for control of the deriva- 0 50
Time, min
100

tive action, as discussed in later sections. Next, we review

p Figure 1. This plot shows a typical proportional-integral-derivative (PID) control
the standard PID controller and some of the factors that loop. The three main control variables are the setpoint (SP), process variable (PV),
affect its operation. and output (OP).

42 aiche.org/cep December 2021 CEP

Copyright © 2021 American Institute of Chemical Engineers (AIChE).
Not for distribution without prior written permission.
works by constantly accumulating the error to calculate a in this mode. When the system is put back on automatic
total error value. The effect of the integral control is then pro- control, it will use the current OP and SP values to restart
portional to this total. When the current error goes to zero, the its calculations.
total error stops changing. The integral tuning coefficient may Lag time. Lag time is the most significant problem that
be expressed as repeats/sec or sec/repeat, depending on how occurs in control systems. It is usually easy to tune a system
the PID equation is set up. Repeat is how often the controller that responds quickly to a change in the OP, but if the system
comes back and looks at the error to adjust the OP. has a long lag time, it must be tuned to react slowly enough
When the integral tuning coefficient is expressed as so that cycling is not produced. Long lag time is usually
repeats/sec, the response becomes stronger as the number overcome by using a control setting that is mostly propor-
increases. Alternatively, when the integral tuning coefficient tional with a small amount of integral to bring the PV slowly
is expressed as sec/repeat, the response becomes weaker as back to the SP.
the number increases. The typical units are repeats/sec, but
if you are unsure of the units for your system, make a small Basic control equations
increase in the integral and see if the response gets stronger The type of process control taught at the university level
or weaker. Repeats/sec is assumed in all of the examples typically requires information on the valve type and size, lag
used in this article. times in the control OP and the process response, transfer
Reset windup is a problem that may arise with integral functions for the process, etc. The methods described in this
action. It occurs when the OP goes up to 100% or down article consider all of these elements together as a “black
to 0%. The integral (or reset) function continues to try and box.” The total effect of all of these elements will be seen
increase the OP to above 100% or below 0%. Mathemati- in the response of the control plots to a disturbance that is
cally, this occurs because the error continues to accumu- either provided by the process or created manually.
late in the PID equation. This accumulation will delay the The standard form of the basic PID control equation is:
response when the error begins to fall and keep the OP OP = K × E + I × ΣE + D × ∆E/∆t (1)
from changing until the excess error that has accumulated is where OP is the output, K is the proportional gain constant,
worked off. Reset windup is overcome automatically in most E is the error (PV – SP), I is the integral time constant, D is
modern digital control programs. Once the OP reaches an the derivative time constant, and t is time. The OP is propor-
upper or lower limit, the program stops the accumulation to tional to E, the sum of E (ΣE), and the change in E over time
prevent reset windup. (∆E/∆t). To use PID control in a simulation, it is often easier
Derivative (D). Derivative control responds to the rate of to use the velocity form of this equation. The velocity equa-
change of the error. It tries to keep the slope of the error con- tion is a time derivative of the standard equation:
stant and does so by calculating the change in error vs. time OP2 = OP1 + K × ∆E + I × E + D × ∆(∆E/∆t) (2)
(slope) for the recent time interval. The effect of the deriva- where OP1 and OP2 are the initial and current control
tive control is then proportional to this slope. This provides a outputs, respectively.
form of anticipatory control (rapid response) and can tighten In Eq. 1, it is necessary to provide a value for the initial
the overall control. It can also cause unstable operation if SE to produce the initial OP value. Then, you must keep
not tuned properly. A sudden change in the PV in a system up with the accumulated total error to generate the OP for
with derivative control could cause the OP to go to 100% or each new time increment. The current OP is calculated from
0%, so it is not usually used (2). If you have a process where scratch every time and bears no relation to the previous OP.
there are significant disturbances in the process inputs and Equation 2 uses OP1 as a starting point and calculates the
the PV must be held within tight limits, derivative control change caused by the change in error over the next time
may be needed. interval. Thus, the equation calculates an increment, and
If derivative control is active, a change in the SP will
also be seen as a sudden infinite change in the slope of the
error, which will significantly change the OP. Most control Nomenclature
systems avoid this setpoint change effect by using changes D = derivative time constant
in the PV rather than changes in the error for derivative con- E = error
trol. The effect of derivative control is not part of the tuning I = integral time constant
K = proportional gain constant
discussed in this article.
OP = output
Manual SP and OP changes. If a system is put into OP1 = initial output
manual control, the automatic control is paused. During this OP2 = current output
time, the PV continues to track the actual value. Both the t = time, min
OP and SP may be changed manually by the operator while

Copyright © 2021 American Institute of Chemical Engineers (AIChE). CEP December 2021 aiche.org/cep 43
Not for distribution without prior written permission.
 B AC K TO B A SI C S

each new OP2 depends on OP1. This makes it much easier and then cycles to find the solution. Either the proportional
to describe and implement in a spreadsheet or computer pro- gain constant (K) or the integral time constant (I) is too
gram. The starting condition is then set simply by choosing a large, or maybe both. The empirical way to tune in this
value for OP1. situation is to eliminate the integral and then try again. The
Control systems can be direct-acting or reverse-acting. current values are K = 0.7 and I = 0.8. Write down these
In a direct-acting system, an increase in the OP causes an initial conditions and any changes that you make so that
increase in the PV. In a reverse-acting system, an increase if any problems come up, you can always go back to the
in the OP causes a decrease in the PV. If we were applying initial conditions. Also, be sure that you know how to put
chilled water to control a temperature, then increasing the the system in manual control, if needed. If something gets
valve opening (OP) would cause the temperature (PV) to out of control while tuning, put the OP under manual control
decrease and it would be a reverse-acting system. and set it to the average value that it had when you started,
The type of controller action is an important factor in the then wait. After the PV returns to near the SP, set the control
control equations. In these equations, increasing the SP will parameters back to where they were initially and then put the
cause the error to become more negative (i.e., if the SP goes system back on automatic control until it stabilizes.
above the PV, we would think of this as a negative error). Step 1. Eliminate the integral control and see how
For a direct-acting system, when the SP is above the PV, you the system responds. In this case, the integral is measured
want to increase the OP to increase the PV and drive it back in repeats/sec, so we need to reduce the value to zero to
up to the SP, which means the error term in Eq. 2 has to be eliminate the integral action. Figure 3 shows Step 1 of the
positive. This is handled internally by the control system soft- SP change tuning method where you eliminate the integral
ware. If you specify that you want a direct-acting system, the effect and evaluate the effect of a change in SP with the cur-
error will be calculated as (SP – PV). If you specify a reverse- rent K value (i.e., K = 0.7, I = 0.0).
acting system, the error will be calculated as (PV – SP). The pure proportional control only brings the PV up to
about 45% of the setpoint. The overshoot in the original plot
Setpoint change tuning method was caused by having too much integral control. The propor-
The first and most straightforward type of tuning is tional control provides the fastest and most stable response
the setpoint change tuning method. This is a good tun- and is the least sensitive to a lag time in the system response,
ing approach for systems that can be manipulated without so it should be set to provide most of the control action.
significantly affecting the process, such as the feed rate to a Step 2. Increase the proportional control until the PV is
product filter. It will be assumed that a rapid response type roughly 70% of the SP. Typically, the proportional control
of control tuning is desired. should be set so that the PV reaches about 70% of the new
Before tuning, set up the SP/PV/OP plots and allow the SP. If the proportional K value is set too high, it will cause
system to stabilize, then increase the SP by a small amount overshoot and cycling. Figure 4 shows the response when
(Figure 2). Observe the plots and see how they respond. In the K value is raised to 1.5. The proportional control now
this example, the system overreacts. When the SP is raised gets the PV up to 60% or 70% of the new SP.
from 900 gpm to 950 gpm, the valve (OP) opens too much Step 3. Add back in the integral control until the system
responds as desired. Now that we have a good setting for
1,100 45
SP 1,100 45
1,050 PV 40 SP
OP 1,050 PV 40
1,000 35 OP
1,000 35
950 30
950 30
PV, gpm

OP, %

900 25
PV, gpm

OP, %

900 25
850 20
850 20
800 15
800 15
750 10
750 10
700 5
0 50 100 150 200 250 300 700 5
Time, min 0 50 100 150 200 250 300
Time, min
p Figure 2. To begin the setpoint change tuning method, change the setpoint
(SP) and see how the system reacts before any tuning. In this case, the system p Figure 3. In Step 1 of the setpoint change tuning method, eliminate the integral
overreacts. Control parameters: proportional gain constant (K) = 0.7, integral time (I) effect and evaluate the effect of a change in setpoint (SP) with the initial
constant (I) = 0.8. proportional gain constant (K) value. Control parameters: K = 0.7, I = 0.0.

44 aiche.org/cep December 2021 CEP

Copyright © 2021 American Institute of Chemical Engineers (AIChE).
Not for distribution without prior written permission.
K, more integral is added to complete the tuning. Figure 5 SP rapidly and with only minor cycling. Once the tuning is
shows the response when the integral is increased to 0.3. done, it can be tested by moving the SP to higher and lower
The integral now pushes the PV up to barely past the SP and values and seeing how well the PV follows.
allows it to converge rapidly. The tuning process would be different if this had been a
In actual practice, these tests are performed sequentially level control for a tank. In that case, only a loose control is
with stepwise increases and decreases. The response from an needed, so the proportional parameter would be set to give a
increase may be slightly different than the response from a response of 20% to 30% of the SP change, and the integral
decrease, but it is seldom significant for this small range. would be set to some low value to move the level slowly
The full plot of the actual tuning session is shown in back to the SP. The actual control parameters used would
Figure 6. In the actual tuning session, the change in the need to be strong enough to keep the tank level from getting
SP was made from 900 gpm to 950 gpm, and the system too low (i.e., allowing the tank to empty) or too high (i.e.,
stabilized at the new value. Then, I was set to zero, and the tank overflow or high-level alarm activation). This type of
SP was decreased to 900 gpm again. The PV only moved tuning is discussed later in the article.
about halfway toward the SP. At this point, I was changed to
0.5 to force the PV toward 900 gpm. Once the SP was met, Active process tuning method
the control settings were changed so that K = 1.5 and I = 0.0, In many cases, you will not be able to make signifi-
then the SP was increased back to 950 gpm. This time, the cant changes to the SP because it will upset the process.
PV went up by about 70%, which means the K value is now In this case, an analysis of the patterns and interactions
acceptable. The I value was then increased to 0.3 and the SP between the PV and OP may be used to tune the system,
was lowered back to 900 gpm. The PV now came back to the based on an understanding of the proportional and integral
control behavior.
1,100 45
SP Before you begin the active tuning process, it is impor-
1,050 PV 40 tant to understand the objectives of the proportional and the
1,000
OP
35 integral controls. The proportional control merely increases
950 30
or decreases the OP to stop the PV from changing. The pro-
portional control does not try to bring the PV back to the SP;
PV, gpm

it only wants to slow it down and make it stop moving. You

OP, %

900 25

850 20 can think of this control as applying brakes on the system.

800 15
The faster the PV is moving, the harder the system applies
the brakes. If the PV is increasing, the system will try to
750 10
slow down the rise. If the PV is decreasing, the system will
700 5 try to slow down the fall. Once the signal is constant, the
0 50 100 150 200 250 300
Time, min proportional control is satisfied.
p Figure 4. In Step 2 of the setpoint change tuning method, change the propor-
The integral control, on the other hand, does want to
tional (K) value to produce a response that is roughly 70% of the new SP. Control force the PV back to the SP. The further away from the SP it
parameters: K = 1.5, I = 0.0. is, the harder it pushes. As it gets closer to the SP, it pushes
less and less. When it reaches the SP, it is satisfied and stops
1,100 45
SP 1,100 45
1,050 PV 40 SP
OP 1,050 PV 40
1,000 35
OP
1,000 35
950 30
950 30
PV, gpm

OP, %

900 25
PV, gpm

OP, %

900 25
850 20
850 20
800 15
800 15
750 10
750 10
700 5
0 50 100 150 200 250 300 700 5
Time, min 0 50 100 150 200 250 300
Time, min
p Figure 5. In Step 3 of the setpoint change tuning method, add the integral (I)
control back until the system responds as desired. Control parameters: K = 1.5, p Figure 6. This PID plot shows what an actual setpoint change tuning session
I = 0.3. might look like.

CEP December 2021 aiche.org/cep 45

Copyright © 2021 American Institute of Chemical Engineers (AIChE).
Not for distribution without prior written permission.
 B AC K TO B A SI C S

pushing entirely. The action of the control system is the cant upward slope at that point. At the top of the PV peak at
result of the balance between the pushing of the integral and 24 min, the OP curve shows a significant downward slope.
the braking of the proportional controls. This valve movement is from the integral part of the equation
With this in mind, to tune an active system, first set up alone, so this controller has strong integral control.
the control plot (Figure 7). An analysis of the wave patterns When is the integral control satisfied? Integral control
in Figure 7 shows that this is a direct-acting control system. is based on the current error value, so it goes to zero when
In a direct-acting system, an increase in the OP causes an the error is zero. This occurs when the PV equals the SP.
increase in the PV. The increase in the OP between 10 min For example, look at the point where the PV curve crosses
and 20 min causes the flow (PV) to stop falling and go back the SP curve at 11 min, 20 min, and 29 min in Figure 7. The
up. In a reverse-acting system, an increase in the OP causes slope of the PV curve is large, but the slope of the OP curve
a decrease in the PV. at those points is almost flat. This shows that the propor-
This poorly tuned system can now be analyzed for tional part of the control is very weak, so most of the control
improvement. When the PV of a system moves far past the is being done by the integral. (It is recommended to look at
SP and cycles, it is called an underdamped system. It could two or three points to evaluate the slope because the chang-
be overreacting because it is tuned too tightly or because the ing process values coming into the system will also affect
balance between the proportional and the integral controls the plots.)
is off. Step 2. Slowly adjust the control constants until the
Step 1. Determine which control type has a stronger system responds as desired. In this case, the integral appears
effect on the system. Remember that the change in the to be too strong and is causing cycling. We could try apply-
proportional control is proportional to the change in error ing more brakes (stronger proportional) or reducing the
and the change in the integral control is proportional to the push (weaker integral). In this case, we will try to weaken
error itself (Eq. 2). When the proportional control is satis- the integral and see what happens. The current values are
fied, it will stop applying the brakes, and all of the remaining K = 0.5 and I = 0.5. The first adjustment will be to make
control action is from the integral. In the same way, when I = 0.4 at 45 min (Figure 8). Once again, be sure to write
the integral control is satisfied, it will stop pushing the PV down the initial conditions and any changes that you make.
toward the SP, and all of the remaining control action is from The PV peaks at 60 min and 70 min (where the effect
the proportional. of proportional control is zero) show that the slope in OP
When is the proportional control satisfied? Proportional generated by the integral control is lower than before we
control is based on a change in the error, so it will be satisfied decreased I, but still fairly strong. The slope of the OP curve
when the PV stops changing. This occurs when the slope of at the 73 min point where the effect of integral control is
the PV plot equals zero. The slope equals zero at the top and zero shows that the proportional control is still relatively
bottom of the peaks in the PV curve. For example, look at the weak. Next, we set I = 0.3 at 100 min in the second adjust-
PV trough at 15 min in Figure 7; the OP curve has a signifi- ment to reduce its effect even more (Figure 9).
The PV low points at 113 min and 149 min show that the
1,000 35 slope in OP caused by the integral control is still significant,
950
SP
PV 30
but getting smaller. In addition, the deviation from the SP
OP 1,000 35
900 25 SP
PV, gpm

PV
OP, %

950 30
850 20 OP
900 25
PV, gpm

800 15
OP, %

850 20
750 10
800 15
700 5
0 50 100 150 200 250 300 750 10
Time, min
p Figure 7. Analyze the control plot of the system before beginning the active 700 5
0 50 100 150 200 250 300
process tuning method. This system is a direct-acting control system, which
Time, min
means an increase in the output (OP) causes an increase in the process variable
(PV). Look for when the PV stops changing (e.g., 15 min, 24 min) to determine p Figure 8. In Step 2 of the active process tuning method, a small adjustment is
when the proportional control is satisfied. For example, at the top of the PV peak at made at 45 min to reduce the integral control. Control parameters: K = 0.5, I = 0.4
24 min, the OP curve shows a significant downward slope. This valve movement is at 45 min. The PV peak at 60 min and the PV trough at 70 min (where the effect of
from the integral part of the equation alone, so this controller has strong integral proportional control is zero) show that the slope in OP generated by the integral
control. Control parameters: K = 0.5, I = 0.5. control is lower, but still fairly strong.

46 aiche.org/cep December 2021 CEP

Copyright © 2021 American Institute of Chemical Engineers (AIChE).
Not for distribution without prior written permission.
seems to be decreasing. At some points, the PV is holding be increased. Figure 11 shows the control achieved when the
at the SP instead of moving past it. Now, we will try making parameters are doubled to K = 1.0 and I = 0.4 at 240 min.
I = 0.2 at 170 min in a third adjustment (Figure 10). The response of the valve is faster, and the variations of the
At 175 min and 199 min, we can now see that the slope PV are smaller with these tighter control parameters. Once
in OP is even smaller. In addition, you can see that when a the tuning is complete, follow the plots for a few hours to
deviation occurs, the control system brings it back to the SP verify the results. You should see that the variations are now
and holds it there until the next deviation comes. For exam- much smaller under standard conditions.
ple, a bump begins at 170 min, and the PV returns to the SP Figure 10 shows a system tuned for rapid response
at 182 min and remains there until the next bump. Another and Figure 11 shows a system tuned for tighter control. If
bump occurs at 194 min, and the PV is brought back to the this were a surge tank intended to even out the variations
SP by 207 min. The system is now slightly overdamped from the upstream process, it would need to be tuned more
because the PV plot does not go much past the SP line. For loosely. This is achieved by making it primarily a propor-
this example system, the optimal I value is likely between tional control system with slight integral control to move the
0.2 and 0.3. This tuning could also have been accomplished PV slowly back toward the SP.
by increasing the K value and leaving the I value constant. As an example, we can look at a surge tank with two
Once the system is tuned, if tighter control is needed, takeoff streams. One stream takes small batches out for
then both K and I should be increased together. If each is purification and recycle and the other uses level control to
made more responsive but offset by the other, the whole provide a feed to the downstream process (Figure 12). The
system will react faster. However, the tighter the control, the level controller has settings of K = 0.1 and I = 0.01. As the
more likely it is that a change in SP or a variation in the pro- tank slowly empties, the level control valve closes, but does
cess will cause the control system to overreact and become not move rapidly enough to drive the level back up to the SP.
unstable. Therefore, there is a limit to how much K and I can Batches of liquid are removed until a new batch of feed is
1,000 35 1,000 35
SP SP
950 PV 30 950 PV 30
OP OP
900 25 900 25
PV, gpm
PV, gpm

OP, %
OP, %

850 20 850 20

800 15 800 15

750 10 750 10

700 5 700 5
0 50 100 150 200 250 300 0 50 100 150 200 250 300
Time, min Time, min
p Figure 9. The integral control is slightly decreased at 100 min to reduce its p Figure 11. This plot shows the active process tuning method results where K
effect even more. Control parameters: K = 0.5, I = 0.3 at 100 min. The low points and I are doubled, producing a tighter control. Control settings: K = 1.0, I = 0.4 at
in the PV at 113 min and 149 min show that the slope caused by the integral is still 240 min.
significant, but getting smaller.
1,000 35
1,000 35 SP
SP 950 PV 30
950 PV 30 OP
OP 900 25
PV, gpm

900 25
OP, %
PV, gpm

850 20
OP, %

850 20
800 15
800 15
750 10
750 10
700 5
700 5 0 50 100 150 200 250 300
0 50 100 150 200 250 300 Time, min
Time, min
p Figure 12. In this example, loose control is desired in a surge tank with two
p Figure 10. In a third adjustment, the integral control is decreased even further. takeoff streams. Loose control is achieved by making it primarily a proportional
Control parameters: K = 0.5, I = 0.2 at 170 min. The PV returns rapidly to the SP and control system with slight integral control to move it slowly back toward the
good control is now achieved. setpoint. Control settings: K = 0.1, I = 0.01.

CEP December 2021 aiche.org/cep 47

Copyright © 2021 American Institute of Chemical Engineers (AIChE).
Not for distribution without prior written permission.
 B AC K TO B A SI C S

provided, and the level goes back up. The discharge of this method to try and tune the secondary controller based on
tank (represented by the %OP), which feeds another part how it responds to the SP from the primary controller. Make
of the process, is much more stable than if the system were small tuning adjustments in this loop so that the secondary
tightly controlled to maintain the level. controller more closely follows the moving SP.
4. Once the primary controller has been set to manual,
Cascade control tune the secondary controller as described above using the
Cascade control is another important type of process con- setpoint change tuning method. Make small changes in
trol, and perhaps one of the most difficult to tune. This type the OP of the primary controller and watch the results. By
of control is necessary when the parameter used to provide making a positive change followed by a negative change,
the control varies with time. A good example is the steam you can make them average out and not disturb the process
flow to a distillation column reboiler. If the steam pressure is too much.
constant, a temperature control loop can adjust a valve on the 5. Once the secondary controller is tuned and tracking
steam line directly and it will work fine. However, the pres- well, put the whole system back in automatic mode and see
sure in a steam header will often vary based on the demand how it performs.
from several processes that are using it at the same time. 6. Now proceed to tune the primary controller using the
There can even be pressure variations from the steam boiler active controller tuning method. In some cases, the system
operation itself. In this case, a separate control must be used may now be so stable that this method cannot be used. In
to adjust the steam flowrate. A temperature controller controls that case, use the SP change method to test the primary
the SP for this flowrate, i.e., the OP from the temperature controller loop and see if it needs further tuning. Remember
controller is sent to the steam flow controller as a SP. This that when tuning this loop, you should tune it to react slowly
means that the temperature controller is the primary control- and allow time for the secondary controller to catch up.
ler and the steam flow controller is the secondary controller. Since cascade control introduces a longer lag time, use more
The steam flow controller adjusts the valve to produce the proportional and less integral control in the primary control-
flow requested by the temperature controller. This will com- ler to improve stability.
pensate for any variations in the pressure of the steam. 7. Monitor the system for several hours to ensure
For these control systems, the secondary controller that changes in the process conditions do not cause sta-
should be tuned for tight control and the primary controller bility problems.
should be tuned for loose control to avoid cycling. The goal
of the secondary controller is to move rapidly to the target In closing
while the primary controller moves that target slowly. If the Simple PID controllers can handle most control prob-
target is moved too fast, the secondary controller will not be lems if properly tuned. First, determine what kind of control
able to keep up. By the time it reaches the target, the PV has you need: rapid, tight, loose, or cascade. Then, use the meth-
already gone too far and the primary controller is moving the ods described in this article to achieve the desired control.
target back in the other direction. This leads to cycling as the The active process tuning method is particularly suitable for
system constantly searches for a solution. sensitive systems, while the setpoint change tuning method
Take the following steps to tune cascade control systems: is more suitable for rugged systems. CEP

1. Set up two PID plots as shown previously — one for

the primary controller and one for the secondary controller.
Monitor the SP that the primary controller provides to the Literature Cited
secondary controller and find the average value. Also, find 1. Reed, H., “Empirical Process Control: Part 1,” Chemical Engi-
the average value for the OP of the secondary controller. neering Progress, 117 (6), pp. 38–42 (June 2021).
This is the actual control element for the system. If a prob- 2. Rohani, S., “Coulson and Richardson’s Chemical Engineer-
ing: Volume 3B: Process Control,” Fourth Ed., Butterworth-
lem occurs during tuning, put the secondary controller on Heinemann, Oxford, U.K. (2017).
manual and set its OP to this value to bring the system back
under control.
2. Record the control variable values (K and I) for both HAROLD REED is a process technical engineer at Sirius Technical
Services (Theodore, AL). With more than 40 years of experience
loops for future reference. in the chemical industry, he has worked for Dow Corning, DuPont,
3. If the system does not have much variation, put the Ciba-Geigy, Ciba, Novartis, Syngenta, and Sirius Technical Services,
including roles in laboratories, pilot plants, and full-scale production
primary controller on manual and set its OP (the SP for the processes. His areas of interest are process modeling, data analysis,
secondary controller) at its average value. This should keep process control, and experimental design. Reed has a bachelor’s
degree from the Univ. of Kansas and a master’s degree from Lamar
the system operating in its normal range. If the system does Univ., both in chemical engineering.
have significant variation, then use the active process tuning

48 aiche.org/cep December 2021 CEP

Copyright © 2021 American Institute of Chemical Engineers (AIChE).
Not for distribution without prior written permission.