technical

monograph
49
Control Valve Impact on
Loop Performance

Floyd D. Jury
Technical Performance Consultant

Control Valve Impact on Loop Performance

THE CONTROL OBJECTIVE
A manufacturing company is in business to realize a profit
through the sale of a quality product, where a quality
product is defined as one that conforms to specifications.
A deviation from the established specifications can mean
lost revenue due to wasted materials and resources, the
reprocessing of off-spec product to meet specifications, or
selling off-spec product at a lower price.
Therefore its a reasonable assumption that the ultimate
goal of every manufacturer is to avoid any deviation, or at
best, to minimize any deviation and its impact on the
manufacturing process. To that end, control systems
typically are put in place to yield products that are on
spec, time after time, without variation.

industry is standard deviation (), which often is simply

referred to as sigma.

= 2 =

( x )

n1

Equation 1: Standard Deviation (Sigma)

In laymans terms, standard deviation can be visualized

simply as the average deviation from the mean.
Mean Value =
PV Distribution

UNDERSTANDING THE STATISTICS OF VARIATION

Within a manufacturing environment the expectation is
that machines will operate as required and materials will
be within their own tolerance limits.

+/-1-Sigma
+/-2-Sigma
+/-3-Sigma

Figure 2.

However, reality shows that there are unique causes of

variation, such as equipment malfunctions, fouling or
operator errors. Even raw materials may have minor
variations in characteristics and performance.
As each machine, process and raw material varies
randomly within its own tolerance band, the cumulative
effect on the overall process is a band of performance
values (PV) that form a distribution about some average.
The distribution of data typically follows a normal or
Gaussian frequency distribution, which is the familiar
bell-shaped curve shown below.
The process variable values are plotted along the
horizontal axis, and the probability of occurrence of each
particular value is plotted in the vertical direction.
Probability of
Occurrence
Mean Value =
PV Distribution

Measured Process Variable

Figure 1.

The closer the data points are to the mean value (i.e., the
set point.), the more frequently they will tend to occur.
Conversely, those data points which represent larger
deviations from the mean will tend to occur with less
frequency.

Thus, sigma () is a statistically derived parameter that

describes the spread of the data about the mean value.
The larger the value of , the greater the spread.
The area under the PV distribution curve represents the
percentage of the total population between any two points
on the curve. Thus, sigma is a parameter which tells how
much of the total population is contained in a given region
centered about the mean value:
1-sigma contains 68.26% of the total population
2-sigma contains 95.45% of the total population
3-sigma contains 99.73% of the total population
In performance testing, 2-sigma (2) is the metric used
most often to indicate valve performance and to compare
one valve against another in terms of reducing process
variability.
For example, assuming a manufacturer wishes to ensure
that at least 95% of product meets the minimum
specification, the set point of the process must be
adjusted to at least a 2-sigma distance above the lower
limit specification as shown in the figure below.
Lower Limit
Specification

Set Point
PV Distribution

2-Sigma

On average, about the same number of deviations will be

above the mean value as will be below it, thereby forming
a curve that is symmetrical.
Mathematicians have several basic measurements or
metrics they can use to describe the spread of the data,
which is the amount of variation about the mean value.
The metric most commonly used in the process control

2-Sigma

Figure 3.

This adjustment of set point will ensure that only 2.275%

of the finished product will fail to meet the specifications
on the first try. (Note that the tail of products above the 2sigma limit on the upper end also meet the specification.)
While this setting will produce acceptable results, it also
means that 97.725% of the finished product is being

produced at quality levels that exceed the lower limit

specification. This may cost significant dollars in
excessive catalysts, raw materials and energy use.
However, the manufacturer can reduce the degree to
which products must exceed the specification limit by
decreasing the spread of the distribution ( i.e., going to a
lower value of sigma as shown in the following diagram.)

In general, process variance can be reduced by:

Maintaining stable loops
Minimizing the effects of load disturbances
Avoiding unwanted changes in loop gain
Reducing the effects of nonlinearities in the loop
Proper valve maintenance
THE IMPACT OF NONLINEARITIES

Lower Limit
Specification
Set Point
PV Distribution
2-Sigma

2-Sigma

Set Point

Reduced PV Distribution

2-Sigma2-Sigma

Figure 4. Reduced process variability

This is referred to as a reduction in variance or process

variability.
This reduction in process variability allows the
manufacturer to move the process set point closer to the
lower limit specification, which;
Produces a more uniform product
Improves manufacturing efficiencies
Improves profitability
Thus, reducing process variability is really the
manufacturers goal.
Manufacturers tend to understand well that spending
resources on elaborate systems of instrumentation are
often justified to achieve this type of reduction in process
variability. What is not as well understood is that
significant reductions in process variability can be
achieved simply by choosing the right control valve for the
application and ensuring that the control loop is tuned for
the most effective reduction of process variability.
Surveys have shown that as many as 80% of the control
loops in plants are not producing the level of product
uniformity of which they are capable.
The reason for this is twofold:
The right process control equipment is not being
utilized.
The process control equipment that is installed
is not being operated at optimum performance.
Research data show that process variability can be
reduced by as much as 50% by the straightforward
application of:
Good engineering design practices
Intelligent selection of valves and instruments
Careful system installation
Good instrument maintenance
Proper loop tuning methods

What does "linear" mean in the process control context?

A linear dynamic element responds to its input signal in a
uniform fashion over its entire operating range with
unchanging dynamic behavior. It does so regardless of
the size or amplitude of a change in the input signal. If
the time constant of a step response remains the same
for step inputs with amplitudes of 0.01%, 1%, and 100%,
then the element is considered "linear." If the time
constant varies with the amplitude of the input signal, the
element is "nonlinear."
Nonlinearities become a major source of problems for
control loop tuning and performance because the
apparent process dynamics to which the control loop
tuning was matched initially, keep changing with process
conditions. As a result, the variability of the loop no
longer remains stable. That is why it is imperative to find
the source of nonlinearities in the process and eliminate
or reduce them. Historically, the control valve is a good
place to start.
Following is a list of typical valve nonlinearities:
Friction
Backlash
Relay dead zone
Spool valves
Shaft windup
Fluid turbulence
Sample time in transmitters and controllers
Processes which change gain with change in
throughput
Valve characteristics that change valve gain with
throughput
Many of these nonlinearities are extremely complicated to
handle or explain mathematically, but the general
observation can be made that they all introduce
undesirable phase shift and gain effects that place
limitations on the performance of the system.
System stability is always a concern, and nonlinearities
can definitely influence stability. However, a more
insidious effect of nonlinearity is the undesirable influence
it has on process variability.
As a result, these nonlinearities cause the need to be
more conservative in tuning, which limits how much the
process variability can be reduced. This is why it is so
important to review the control hardware components of
the loop first to determine what nonlinearities are present.
Then effort can be made to reduce or eliminate them
through more judicious component selection.

Friction

Friction is a force that tends to oppose the relative motion

between two surfaces that are in contact. The friction
force is a function of the force pressing these two
surfaces together and the characteristic nature of the two
surfaces. Friction has two components: static friction
and dynamic friction.
Static friction is the force that must be overcome before
there is any relative motion between the two surfaces.
Stick/slip or "stiction" are colloquial terms that are often
used to describe static friction. Static friction is one of the
major causes of dead band in a valve assembly. Dead
band is the range through which an input signal can be
varied, upon reversal of direction, without an observable
change in the output signal. (Note: This is not the same
thing as dead time.)
Once relative motion begins between the two parts, static
friction no longer applies. Now, dynamic friction is the
force that must be overcome to maintain the relative
motion. Running friction and sliding friction are colloquial
terms that are sometimes used to describe dynamic
friction.
Static friction is nearly always significantly larger in
magnitude than dynamic friction. In other words, it
usually takes more force to get an element to break out of
its static condition and begin moving, whereas less force
is usually required to keep it moving. This means that
static friction is a much more significant problem in
process control than dynamic friction.

changes direction. Slack, or looseness of mechanical

connection is a typical cause.
One of the biggest sources of backlash comes from lost
motion between the power element of the actuator and
the valve closure member. Linear motion valves are not
measurably affected by backlash except when poorly
designed stem connectors existf. In rotary motion valves,
there are several points where backlash can occur.
Linkage connecting the rotary valve to a linear actuator
will have clearance at the joints, and wear will cause the
clearance to increase resulting in more lost motion.
Some rotary valves are fitted with a scotch yoke or rackand-pinion drive mechanism that introduces backlash.
Actuators that have gear trains are also subject to
backlash in the gears.
Relay Dead Zone
A relay is a device that acts as a power amplifier. It takes
an electrical, pneumatic, or mechanical input signal and
produces a large volume of air or hydraulic fluid to the
actuator. The relay can be an internal component of the
positioner or a separate valve accessory mounted
between the controller and the valve.
Most usually, relays are constructed like a miniature
version of a control valve, and they suffer from the same
type of dead band characteristics as a control valve.
Since relays are small and an unimposing part of the
overall valve assembly, they often are overlooked as a
source of dead band that causes process variability.

Process disturbances do not cause errors that continually

move in the same direction. They typically take the form
of random oscillatory disturbances that are continually
stopping and reversing direction. This means that the
control elements also must continually follow this same
stop-and-reverse-direction pattern. Every time the control
element has to stop, static friction can become a factor.

Spool Valve Dead Zone

Many manufacturers of spool valve positioners
intentionally build some dead band into their instruments
to reduce the large amount of steady-state air
consumption. In other words, some change in input
occurs without any corresponding change in output
because a movement of the spool valve is required to
uncover a particular port before any output change can
occur.

While friction cannot be eliminated completely, good

design practices can have a dramatic impact on the
amount of friction present in a control valve. The places
where friction can exist in a control valve include packing,
guides or other sliding surfaces, bearings in rotary-motion
valves, seals in balanced, cage-guided globe valves,
seals in ball valves, and linings in butterfly valves.
Actuators have friction in their guide bushings, piston
seals, stem seals, and in motion conversion devices.
Friction can be particularly significant if there is
misalignment between stems and bearings or other
sliding surfaces.

Shaft Windup
Shaft windup results when one end of a rotary valve shaft
turns and the other does not. This typically occurs in
rotary style valves where the actuator is connected to the
valve closure member by a relatively long and small
diameter shaft. While seal friction in the valve holds the
control element end of the shaft in place, rotation of the
shaft at the actuator end is absorbed by twisting of the
shaft until the actuator input transmits enough torque to
overcome the friction.

Piston actuators with very large, high-pressure O-Rings

can contribute significant friction, particularly after they
have been in service for awhile. O-rings typically are
lubricated when first installed, but after only a few hours
of operating in a typical high temperature environment,
they tend to get stiff, and the friction factors can increase
by as much as 400 percent or more.
Backlash
Backlash is the general name given to a form of dead
band that results from a temporary discontinuity between
the input and the output of a device when the input

This phenomenon results in dead band since there is no

output during this input motion. Of course, if the input
signal was to reverse direction, the shaft would simply
untwist and begin twisting up in the opposite direction
until the friction could be overcome. Until then, there
would be no change in the output at all.
Dead band is one of the worst nonlinearities facing the
process control engineer. When a load disturbance
occurs, the process variable will deviate from the set point
by some small percentage. This deviation will attempt to
initiate a corrective action through the controller and back
through the process. Unfortunately, if this initial change
in controller output percentage is smaller than the

process dead band, it produces no corresponding

corrective change in the process variable. Only when the
controller output has changed enough to progress
through the dead band does a corresponding change in
the process variable occur.
Any time the controller output reverses direction the
controller signal must again pass through the dead band
before any corrective change in the process variable will
occur. The presence of dead band in the process
essentially ensures that the process variable deviation
from the set point will have to continue to get larger and
larger until it is big enough to get through the dead band.
Only then can a corrective action occur.
Since most control actions for regulatory control consist of
small changes (e.g., 1% or less), a control valve with
excessive dead band may not even respond to many of
these small changes. At a minimum, a well-engineered
control valve should be able to respond to signals of 1%
or less to provide effective reduction in process variability.
Ideally, the valve should be able to respond to signals as
low as %. Unfortunately, it is not uncommon for audits
to show a high percentage of valves with dead bands
ranging from 2% to as high as 5%.
Regardless of its source, dead band in a valve assembly
is undesirable when it comes to maintaining good process
variability. This should be self-evident, because when a
disturbance produces an error in the process variable, the
valve assembly cannot correct for the error until the error
becomes large enough to get through the dead band of
the valve and produce a corrective action.
DEAD BAND MEASUREMENT
In any practical sense, it is not possible to measure the
effects of all these dead band sources individually. What
is important to know is the cumulative effect on the
process performance. This brings up two extremely
important points:
1. The cumulative effect of all the dead band
sources acting together must be measured.
2. The cumulative dead band in a realistic live
process must be measured.

Load Valve

Test
Valve

Flow
Flow
Meter
Meter
Pump

The Setup
Figure
5: Open-loop, step-test flow loop

Water is used in a 4-inch loop to ensure fast process

response and eliminate compressibility effects. Emerson
also has other loop sizes for testing smaller control
valves. For the purposes of this paper only the larger
loop for high flows is discussed.
The pump is programmed to provide constant flow at
either of two different flow levels, one being a simulated
high flow process condition and the other a simulated low
flow process condition. The load valve is a high
performance, Fisher, cage-guided, sliding-stem valve that
is fixed in the appropriate position for the selected load
flow condition. The load valve is a permanent fixture of
the loop, along with the pump, the flow meter, and the
controller.
A turbine meter is used to measure the flow to ensure
accuracy and speed of response. Notice, however, that
the flow meter signal is only measured at the point
labeled (B) in Figure 5. This signal is not sent back to the
controller to regulate the flow. This is an open-loop test
setup. The reason for this will become clear later.
The test valve represents the piece of hardware to be
tested for dynamic performance. Once it has been
thoroughly tested, it will be removed from the loop and
another test valve will be inserted and subjected to
identical tests. This may be another Fisher valve model
or it may be a valve from another manufacturer. It makes
little difference. They are all treated the same. When the
test valve is installed for testing, it is installed, adjusted,
and calibrated carefully and precisely according to the
manufacturer's instructional literature that came with the
valve. This is to ensure that the valve has a legitimate
opportunity to perform as it was designed.
The controller is a Fisher digital valve controller that is
capable of sending a pure step signal to the test valve
ranging in magnitude from as small as .125% or smaller
to a full 100%.
The Objective
The objective is to determine if the complete valve
assembly is capable of responding to small disturbances
in the process variable and if it can make the necessary
corrective actions without interference from dead band in
the valve assembly. To determine this, the valve
assembly must be installed in the loop under normal flow
conditions.

A B

Controller

Figure 5 shows a standard test system that Emerson

uses to perform cumulative dead band measurement
tests on a realistic liquid-flow process. The test loop
provides a method for measuring the dead band
performance of a variety of test valves under realistic
dynamic load conditions. It is imperative that these tests
be done in a consistent manner so that direct
comparisons can be made from one test valve to another
under identical conditions.

To provide precise measurable, controlled test conditions,

the controller is placed in manual (hence, open-loop) so
that a series of small step inputs of variable size can be
applied to the valve assembly.

The step inputs are a series of two steps up and two

steps down at levels of 0.5, 1.0, 2.0, 5.0, and 10 percent
as shown in Figure 6. During this test, the input signal is
recorded, the stem position signal is recorded, and the
process variable (flow) signal is recorded on the same
chart for comparison.
This test does not make a precise measurement of the
dead band with this test, but it easily and accurately
brackets the magnitude of the dead band by comparing
the process variable (flow) response to the input signal. If
the process variable movement faithfully follows the twostep input, then the dead band of the assembly is less
than the percent input applied. For example, the process
variable may not faithfully respond to the 0.5% signal, but
does respond faithfully to the 1.0% signal. This indicates
that the dead band of the assembly is somewhere
between 0.5% and 1.0%, which is normally expressed as
being less than 1.0%.
It is important to recognize that the process variable (not
the stem position) is the important parameter to observe
when making dead band measurements. Some
manufacturers incorrectly use bench test stem position
measurements as the key parameter when reporting dead
band of their valves.
With some valves, such as rotary valves, it is possible for
the actuator stem position to change without any
corresponding change in valve movement. From a
process variability point of view, simply changing the stem
position will not make a correction to the process variable
until the valve disk actually moves. For rotary valves, the
majority of the valve assembly dead band is usually in the
valve shaft, bearings, and disc seals. These are often
highly affected by the process pressure and flow and
would be totally ignored in a bench test dead band
measurement test.
The Procedure

As seen in Figure 6, the step inputs are applied in an

alternating series of two steps up and two steps down,
beginning at input levels of 0.5 percent, and then
progressing to 1.0, 2.0, 5.0, and 10 percent. During this
test, the input signal is recorded, the stem position signal
of the test valve is recorded, and the process variable
(flow) signal is recorded on the same chart for
comparison.
Actually, Figure 6 contains the charts for two separate
dead band tests. The top chart illustrates a valve with a
dead band performance ten times better than the valve in
the bottom chart. This is a dramatic illustration of how
dead band can make a major difference in the dynamic
performance of two different valve assemblies.
Notice in the top chart that the process variable (flow) is
responding faithfully to each of the input steps, including
the 0.5% steps. Every time the input moves, the process
variable changes accordingly. There is no way of
knowing how small the dead band really is, but from this
test it can be concluded that it is 0.5% or better for the top
valve.
On the other hand, the valve test illustrated in the bottom
chart shows that the process variable (flow) doesnt
respond at all to either the 0.5% or the 1.0% input steps.
There is finally some response to the 2% steps, however
flow is not responding to all of the steps. It isnt until the
input steps get to 5% that the process variable responds
faithfully to each of the input steps. Again, the precise
dead band is not known, but it must be somewhere
between 2% and 5%. In most cases, this would classify
as a 5% valve assembly.
Noted earlier was that some manufacturers determine the
valve assembly dead band based on stem motion being
the output, rather than process variable. This tends to
make the valves look better than they really are in actual
practice. It should be obvious that stem motion alone
cannot correct for an error in the process unless there is a
corresponding motion of the valve disk.
In the bottom valve pictured in Figure 6, note that the
valve stem responds to the 0.5% signal changes, but
there is absolutely no change in the flow through the
valve. Using stem motion to rate the assembly for dead
band would suggest that the valve was capable of 0.5%
performance, when in reality it is only good for 5%
performance as far as affecting process variability.
DEAD BAND GUIDELINES
These guidelines are based on process performance
measurements using open-loop step tests as indicated in
Figure 6.
Valve Style

Figure 6. Open-Loop Dead Band Measurement Test

Total Dead Band (%)

Sliding-Stem

0.5

Rotary

1.0

Table 1: Dead Band Guidelines for Control Valves

assemblies, particularly when the gear teeth become

worn.

DEAD BAND VERSUS HYSTERESIS

Hysteresis is not an important factor when dealing with
control valves, but it will be treated here anyway because
there is so much confusion about the distinction between
dead band and hysteresis. These two terms are often
lumped together or even used interchangeably as though
they were the same thing, but they are not. They are
really two different phenomena.
Pure Dead Band
Figure 7 illustrates a case where the tested element is
perfectly linear except for some pure dead band.

Pure Hysteresis
Figure 8 illustrates a case where the tested element is
perfectly linear except for some hysteresis.

B
Output

C
A

Output

Input
Figure 8. Pure Hysteresis Element

It is important to note that each change in input does not

cause a corresponding change in the output.

Input

Figure 7. Pure dead band element

The curve in Figure 7 is generated by a quasi-static test

performed by making small incremental input changes to
the device and then waiting to see what change occurs in
the output as the result of each input change. Making a
whole series of these stop-and-go tests and recording the
input versus output for each point generates the curve.
When in the dead band zones of the device (zones A and
C), there is no corresponding change in the output for
each incremental change in the input. Once through the
dead band and the output does start to respond. It does
so in a linear fashion as long as the input signal continues
to change in the same direction.
On the other hand, as soon as there is a change in
direction with the input, the input has to go through a
period of dead band again where there is no
corresponding change in the output (Zone C).
The reason for this dead band zone typically can be
attributed to some kind of looseness in a mechanical
linkage between the input and the output. An example
might be a linkage pin in an enlarged, out-of-round hole.
In this situation, when the input tries to change the output,
nothing happens until the linkage pin comes in contact
with the edge of the hole. The output then begins to
change linearly with the input. This continues until the
input moves in the opposite direction, and the pin must
then once again move through the looseness in the
linkage before anything happens at the output. Some
individuals refer to this type of dead band as backlash.
It is also a common problem in many types of gear

Like the dead band curve, the hysteresis curve in Figure

8 is generated by making small incremental input
changes to the device and then observing the change
that occurs in the output as the result of each input
change. Plotting the input versus output for each point
generates the curve in Figure 8.
It is important to note that there is one characteristic that
clearly distinguishes hysteresis from dead band. With
pure hysteresis there is a change in output for each
change in input. Note however, that the relationship
between the input and output is not a linear one. Further,
when the input is decreasing (B) the relationship is not
the same as when the input is increasing (A).
The reason for the nonlinearity of these curves and for
the separation between the up and down curves is that
some of the input energy is being converted to other
forms of energy (heat, for example) and is not available to
produce a change in output. The separation between the
up and down curve is an indicator of the dissipated
energy called hysteresis.
Typical sources of hysteresis in control valves are such
things as springs, I/P electromagnets, positioner bellows,
actuator diaphragms, etc. A simplistic way to think about
the energy dissipation that occurs with hysteresis is to
visualize the molecular friction internal to the device. For
example, repeatedly and rapidly flexing a metal spring
causes it to become warm. This is due to the energy
being dissipated in the spring due to hysteresis.
The hysteresis effect is particularly pronounced in
transformers for example where the alternating current is
rapidly and frequently changing the magnetic field in the
core.

Applying these same general principles to all of the

devices listed here helps in understanding why they are
sources of hysteresis, but there is one important point of
caution that needs to be pointed out to keep things in
perspective.

While hysteresis may be present to some degree in

control valve assemblies, it is usually a VERY MINOR
problem, almost to the point of being negligible.
Consider, for example, how infrequently and slowly the
springs, bellows, diaphragms, etc. in a control valve get
flexed during the course of their normal operation. Can
you imagine that any of these elements are any warmer
than their surroundings due to hysteresis generated heat?
If you're not certain, try touching an actuator spring the
next time you're out in the plant. Believe me, dead band
is a far more serious problem than hysteresis.
DEAD BAND VERSUS RESOLUTION
The small-step, open-loop test is another way to look at
dead band of a valve. It can provide a more precise
measurement, and it also has the advantage of being
able to measure the resolution of the valve assembly.
The test uses the same test setup as shown in Figure 5
except that the test is conducted in a different manner as
shown in Figure 9.
gp ,

23

Small Step Test

Input Tvl
(%) (%)

16

21

15

19

14

17

13

15

12

13

11

11

Cv
22

22

21

21

20

20

19

19

18

18

Flow

Actuator Travel

I/P Input Signal

17

17
500

Data File: (96)01150827

1000

1500

2000

2500

10
3000

DEAD TIME
When most process control operators talk about
controlling the "process," they are talking about
everything in the loop except for the controller (i.e., they
are using the controller to control everything else in the
loop collectively.)
For purposes of loop tuning, these processes can be
represented by a first-order lag response with the addition
of some dead time. Without dead time in the loop, the
response to a step input to an open loop, such as that
shown in Figure 5, would be a smooth exponential
increase to some final value, with no overshoot.

Time (seconds)

Figure 9. Small step test for measuring dead band and

resolution

In this test, the valve assembly is subjected to a series of

uniform step changes in the I/P signal input that are each
.125% in magnitude. This is a quasi-static test with
pauses of 30 seconds after each step.
The bottom input curve represents 20 small input steps in
one direction with about a 2-minute pause and then taking
20 steps in the other direction.
After the change in direction occurs, dead band can be
measured as the number of small steps required at the
input before the flow signal begins to change direction.
For this valve, it would appear that the dead band is
approximately nine small steps, or 1.125%. This is a
much more precise measurement than the technique that
was used earlier.

98%

f (t )

Flow
(%) Cv

23

Resolution on the other hand is a measure of how

sensitive the valve is to incremental changes when
moving in the same direction. On both the decreasing
steps and the increasing steps for this valve, it would
appear that after the valve has gotten through the dead
band, it takes approximately three small input steps
before the valve responds with a step change in the flow.
This valve would be characterized as having
approximately .375% resolution.

63%

Time

Figure 10. First order lag with no dead time

How fast this response occurs would be an inherent

characteristic of the process, which in Figure 5, includes
the test valve. The parameter used to identify the speed
of response of any first-order lag is called "Tau," the time
constant ().
t

f (t ) = A1 e

Equation 2. First Order lag equation with no dead time

To gain an appreciation of what this time constant means,

let the time parameter (t) in equation (2) be equal to the
time constant (). By doing so, the response at that time
will be equal to 0.63A. In other words, at the end of one
time constant of time, the output response will be at 63%
of the final value of the response to the step input. This
suggests that we have a convenient way to measure the
time constant of any process (i.e., simply subject the
process to a step input and then measure the time it
takes for the output response to achieve 63% of the final
steady-state value.) However, it is not quite that easy due
to the presence of dead time.
While on the subject, consider letting the time variable (t)
be equal to four times the time constant (i.e., t = 4.) At
this point the output response will have reached 98% of

its final value. In other words, the response to the step

input is nearly over.
Previously mentioned was that most real processes have
some dead time in addition to the first order lag
characteristics. Dead time is one of the most common
nonlinearities in any process loop.
Dead time, as might be supposed, is time when nothing
appears to be happening due to transport lag times, dead
zones in a relay, long digital sampling times, etc. These
types of dead time will exhibit themselves on the
measured response as follows, where Td is the dead
time.

100%

98%
f (t )
63%

Td

Time

Therefore, a two-step change procedure is used to deal

with the dead band. One step is designed to position the
valve at the dead band and the second step is designed
to take the valve through the dead band.
To ensure that the system is positioned properly to begin
an increasing step input test, the system should be
subjected to an up-down/down-up cycle before
conducting the first actual up-step measurement.
Figure 6 illustrates this kind of two-step test. Figure 12
takes one of the steps from this two-step test and
amplifies the scale to make it easier to read. (Because of
its large size, it is placed at the end of this paper.) For
simplification, notice that only the valve assembly input
signal (i.e., the controller output) and the process variable
(i.e., the flow rate) are shown.

0%

Due to any dead band nonlinearity present, the system is

likely to respond differently to a decreasing step input
than it would to an increasing step input.

T 63
T 98

To measure the dead time, first estimate the time when

the input signal was applied (approximately 239.9 sec)
and the time when the output response appears to begin
(approximately 240.1 sec). The difference between these
two numbers is the dead time (i.e., Td = 0.2 sec.)

The dead time (Td) is the time that occurs after the step

Finding the time constant (tau) first requires finding the

time when the output (flow) reached 63% of its total
change, once it actually started to change. The first step
is to determine how much change actually occurred in the
output flow. Using a straightedge to draw an asymptote
to the flow curve in Figure 12 provides an estimate that
the final flow rate is approximately 65.2%. Since the
change began at 60.4%, this gives a total output change
of 4.8%. Finding 63% of this change gives (0.63)(4.8%) =
3.0%. Adding the beginning 60.4% to this means that the
63% point is reached at a reading of 63.4% on the chart.
This occurs at a time of approximately 242 sec.

input in which there is no output response. This diagram

should make it clear that T63 and are not the same
unless there is no dead time since T63 includes the dead
time.

To determine tau, subtract the time when the output first

began to respond (240.1 sec) from the 63% time (i.e., =
242.0 - 240.1 = 1.9 sec.)

Figure 11. First order lag with dead time

The above diagram provides an opportunity to discuss

terms that frequently cause some confusion. These
terms are Tee-63 (T63), Tee-98 (T98) and dead time (Td).
T63 represents the time in which the response function
reaches 63% of the final response, including the dead
time.

By the same token, T98 also includes the dead time.

T98 represents the time in which the response function
reaches 98% of the final response. Use of T98 as a
measurement parameter should be discouraged not only
for the reasons mentioned previously, but also for the
deleterious effects that dead time has on the first-order
characteristics of the response. In many processes, dead
time can be a rather significant factor compared to the
time constant of the process. In these cases it can begin
to severely destabilize the loop, amplify the disturbances,
cause overshoot, and make measurements such as T98
relatively meaningless.

The T63 time is determined by adding the dead time to the

time constant (i.e., T63 = 0.2 sec + 1.9 sec = 2.1 sec.)

MEASURING DEAD TIME

To ensure further accuracy, this entire sequence should

be repeated at least twice to obtain a total of three upstep measurements.

The same set up shown in Figure 5 can be used to

measure the dead time of the process.
The presence of dead time requires that some
precautions be observed to ensure the most accurate
representation of the process.

It is important to note that because of nonlinearities in the

valve assembly, these parameters will likely be different
for decreasing steps than they are for increasing steps.
Also, accuracy can be improved by taking more than one
measurement in each direction. For this reason, it is
common to make determinations of these parameters for
at least two different up-steps and two different downsteps and averaging the values together to obtain the
most realistic set of parameters.

Three down-step measurements are also obtained during

this same sequence.

Final values for the process dead time (Td) and the
process time constant () are then obtained by averaging
the six measurements (three up and three down) for each
of these two parameters.

RESPONSE TIME GUIDELINES

Valve Size
(inches)
0-2
>2-6
>6-12
>12-20
>20+

(cm)
0-5
>5-15
>15-30
>30-50
>50+

Td
(sec)
0.1
0.2
0.4
0.6
0.8

(sec)
0.2
0.4
0.8
1.2
1.6

Table 2. Guidelines for dead time and time constant

It should be noted that some guidelines in the past have

listed values for T63 instead of . This practice can give
unfair advantage to the equipment being measured.
Since dead time is included in the T63 measurement, any
unused margin in the dead time measurement would
automatically and unfairly accrue to the T63 measurement.
Both dead time and time constant should stand on their
own, independent of each other.
GAIN
Maintaining proper loop gain is the secret of good control.
It is what allows the loop operator to reduce the process
variability by clustering the measured process variable
measurements tightly around the set point as shown in
the lower diagram of Figure 4. It is important not only to
find the proper loop gain, but it is imperative to maintain
that loop gain as steady as possible over a wide range of
operating conditions.
To illustrate what is meant by loop gain, return to the
example used to generate the chart in Figure 12. A 5%
step input was applied at the controller input (point A in
Figure 5), which is shown in Figure 12 as the step input
going from 64.8% to 69.8%. This represents a 5%
disturbance that has been introduced into the steady
state flow rate. The controller and the rest of the
elements in the loop will try to correct for this disturbance.
The loop responds by making a first-order type flow
response (at point B in Figure 5) that has been plotted in
Figure 12. To measure the gain of this loop response,
the apparent final value of the process variable
(approximately 65.2%) is subtracted from the starting
value (approximately 60.4%) to get a total output change
of 4.8%.
Since the input change was 5% in this example, the loop
gain is simply the ratio of the output change over the input
change (4.8/5.0 = 0.96.) In other words, the "sensitivity"
or gain of the loop is such that it removed 96% of the
disturbance in one pass around the loop.

If the loop were to be instantly reconnected at points A &

B in Figure 2, the remaining 4% of the disturbance would
go around the loop again and another 96% would be
removed. This process would continue until all evidence
of the original disturbance was removed or until it became
so small that it was dissipated by pipe friction or other
energy losses in the system.
The lower the loop gain, the longer it takes the loop to
bring the system back into stable control. The ideal, of
course, would be to have the loop gain be equal to a
value of one so that the disturbance is completely
eliminated on the first pass around the loop.
What happens if the loop gain is greater than 1.0? Again
return to the test setup in Figure 5 and put in a 5% step
input at the controller input (Point A). Assume that the
loop responds with a flow change of 10% at Point B. This
results in a loop gain of 2.0. The initial 5% disturbance is
now a 10% disturbance. If the loop was instantly
reconnected as done previously, the disturbance would
go round and round the loop, getting larger and larger
each time.
While this appears bad on the surface, there may be
some extenuating circumstances that will help out. In real
life operations, it is not likely to experience a single,
unidirectional step disturbance in isolation. Instead, there
will be literally thousands of disturbances of various
magnitudes that are both increasing and decreasing in a
random pattern. Figure 13 shows a load disturbance
recorded from a typical plant. Many of these will cancel
each other out. Others will likely be of such low
magnitude that they may be consumed by friction or other
energy losses in the system.

640

620

Mean

Filtered
Flow
(gpm)

600
2

580

560

Mean, = 603 gpm

2 = 26.6 gpm
2 / = 4.4%

540

520

D ata File: (95)01251440

200

400

600

800

1000

1200

Time (seconds)

Figure 13. Typical plant load disturbance

Most control experts agree that a control loop can operate

in a relatively stable mode as long as the loop gain does
not get any higher than 2.0 for a period of time. If it gets
any higher than that, the loop is certain to become
unstable. Keep in mind that the loop will not provide as
good a process variability at this gain setting as it would
at an ideal gain of 1.0.
At the low end of the gain curve effective control requires
that the loop gain should never go below a value of 0.5.
To simplify the application of control valves the term
control range references the range of valve travel over

which a control valve can maintain the installed loop gain

between the normalized values of 0.5 and 2.0.

tests it, as long as it is tested under the same conditions

with constant pressure drop.

NOTE: Normalized means that the values are expressed

as a ratio of percent rated output versus input spans.
For years the industry has relied upon terms such as
rangeability, turndown ratio, etc. to judge the ability of a
control valve to provide proper control. In actuality, these
terms are meaningless for this purpose since they have
nothing to do with the issue of control. Instead, the term
"control range" directly relates to the issue of good
control.
Theoretically, the loop could be tuned for optimum
performance at some desired set point flow condition.
However, since the gain of a typical loop changes with
flow, process variability optimization becomes difficult.
There is also danger that the loop gain might change
enough to cause instability, limit cycling, or other dynamic
difficulties.

Although there have been various minor modifications

over the years, there are only three basic inherent flow
characteristics which are illustrated in Figure 14.
Before studying these inherent characteristic curves, it is
useful to define what is meant by the term "valve gain."
As always, gain is defined as the ratio of the change in
output over the change in input. From the perspective of
the control valve, the input is a change in the stem
position (i.e., the travel.) The resulting output is a change
in the flow. Thus the valve gain is simply the ratio of the
change in flow divided by the corresponding change in
travel. A close look at Figure 14 shows that this is simply
the definition of the slope of the curve at any point.

To overcome these difficulties, a means must be found to

compensate for these changes in loop gain and maintain
a relatively constant gain over the entire range of flows.
As a minimum, it is important to stay within the range of
0.5 to 2.0. This compensation is often referred to as
"characterization."
By far the best process performance occurs when the
required flow characterization is obtained through
changes in the valve trim rather than through use of cams
or other methods.
VALVE CHARACTERIZATION
The concept of valve characterization has been around
for a long time and may well be one of the most talked
about concepts in the industry. At the same time, it may
well be one of the least understood. There are so many
legends and rules-of-thumb that have developed over the
years that it is hard to tell what is useful and what is not.
First, exactly is meant by valve characterization (i.e., the
things that characterize the flow through a given valve.)
Obviously, the size and style of the valve, as well as its
internal flow geometry make a difference in flow.
Other factors that can make a difference are external,
such as the type of flowing fluid and its properties, the
inlet pressure to the valve, the pressure drop across the
valve, and the valve travel (valve opening.)
INHERENT VALVE CHARACTERISTICS
When a valve manufacturer, such as Emerson, develops
a new valve style, they take a representative sample of
the new valve and perform a valve characterization test
according to established industry standards. During this
test, all the factors previously mentioned are held
constant except for the valve opening. The flow-versustravel results (with constant pressure drop) are plotted to
obtain what is called the inherent characteristic curve.
Inherent means that it is part of the "constitution or
essential character" of the device. In other words, the
valve will always exhibit this same flow characteristic no
matter when or where it is tested, or regardless of who

Figure 14. Inherent flow characteristics

The linear flow characteristic curve shows that the flow

rate is directly proportional to the valve travel. This
proportional relationship produces a characteristic with
constant slope so that with constant pressure drop, the
valve gain will be the same at all flows. The linear valve
characteristic is commonly specified for liquid level control
and for certain flow control applications requiring constant
gain.
The quick-opening flow characteristic provides for
maximum change in flow rate at low valve travels with a
nearly linear relationship with a slope about 1.5 times that
of the linear characteristic for the first 70% of flow
capacity. Additional increases in valve travel give sharply
reduced changes in flow rate, and when the valve nears
the wide-open position, the change in flow rate
approaches zero as it asymptotically increases to full flow
capacity. In a control valve, the quick opening valve plug
is used primarily for on-off service, and is often found in
regulators and relief valves where the valve can establish
significant flow quickly with minimum travel.
This characteristic is sometimes used in applications
where a linear characteristic should normally be specified.
This is not a good idea because of the dramatically wide

variation in the valve gain over its travel range. It goes

from very high gain at lower travel range to virtually zero
gain at high travel ranges. This could make it very
difficult to keep the loop gain within the 0.5 to 2.0 control
range.
With the equal-percentage flow characteristic, equal
increments of valve travel produce equal percentage
changes in the existing flow. The change in flow rate is
always proportional to the flow rate just before the change
in valve opening is made. Reference to Figure 14 shows
that when the equal-percentage valve is nearly closed, a
10% increase in travel produces very little increase in
flow, but at greater flows, each 10% increase in travel
produces greater and greater changes in flow, with a
dramatic increase in flow during the final 10% travel.

The installed characteristic is not the same all the time,

even for the same valve, since it depends entirely upon
the application in which the valve is installed.
This example illustrates that the installed valve
characteristic is both a function of the inherent valve
characteristic and the process characteristics. While in
this instance a linear inherent characteristic was used for
simplicity, exactly the same procedure can be used for
each of the other inherent characteristics.

A flow characteristic should not be considered inherent to

a particular style of valve because many globe style
valves, for example, have internal trim parts (plugs and
cages) which can be easily changed to produced any of
the characteristics shown in Figure 14. This is not so
easy for other valve styles, however. For example, rotary
valves, such as high-recovery ball valves and butterfly
valves typically will exhibit inherent equal-percentage
characteristics and there is no way to modify trim parts to
change that. One caveat applies here. It has been
common practice in the industry to select line-sized
butterfly valves in many applications. As a result, these
valves are typically over-sized, which means small
movements of the valve disk will produce large changes
in flow. Thus, for all outward appearances, the valve acts
like a quick-opening valve.
INSTALLED VALVE CHARACTERISTICS
In actual service, the pressure drop across the valve is
likely to vary all over the place. To understand how
varying pressure drop can change the valve's flow
characteristics, consider a simple example of a linear
inherent characteristic valve.
Figure 15 illustrates a number of linear flow versus travel
curves for constant pressure drops across the valve.
Logic suggests that at any given valve travel, the highest
pressure drops across the valve would give the greatest
flow. The steeper the curve, the higher the pressure drop
as indicated in the drawing. Since the pressure drops are
held constant throughout the travel, the flow gain remains
constant. The result is linear flow curves throughout the
range regardless of the pressure drop, as long as the
pressure drop is held constant.
To illustrate how varying the pressure drop can affect the
flow characteristic, assume that the same linear valve
body is installed in a process where the valve pressure
drop (P) decreases with load flow. In Figure 15, a locus
of points has been plotted that defines the flow rate as a
function of travel. These points are found by selecting the
appropriate P curve for each increment of travel.
Connecting this locus of points generates what is known
as the installed valve characteristic curve. Note that this
installed valve characteristic curve is not linear, even
though the valve itself is inherently linear.

Figure 15. Installed characteristic curve for linear valve with

decreasing pressure drop

INSTALLED VALVE GAIN

In the discussion of inherent valve characteristics it was
stated that flow gain (flow/travel) was simply the slope of
the inherent characteristic curve. This is true for the
installed gain as well. In most cases, it is actually the
installed valve gain that is of most interested. In Figure
15 the installed gain can be obtained directly from the
measured slope of the installed valve characteristic
Installed Flow
Characteristic and Gain
4" 8560 / 1052(33) /
Load Valve at

Valve Travel
1000

10

20

30

40

50

60

70

80

90

800

Flow 600
(gpm)
400

Experimental
Flow

200
0
3

Gain

Gain
(%Flow/%Inpu

2
Gain Spec

1
0

10

20

30

40

50

60

70

80

Valve Travel

Figure 16. Installed flow characteristic and installed gain curves

90

tend to fall somewhere in between the two extremes of

conventional butterfly valves and globe valves.
curve. In fact, this is often how the installed gain is
obtained for the entire process loop. A flow versus valve
travel measurement is made in a loop similar to Figure 5,
the data analyzed and the slope (valve gain) calculated at
each valve travel. This data is then plotted on a graph
similar to Figure 16. The top curve is the installed

characteristic flow curve. The dots represent actual

measured data sample points and the solid line is the flow
curve calculated from the measured data sample points.
Once the installed characteristic flow curve is determined,
the curve is used to calculate the slope (installed gain) at
each travel point. Installed gain versus travel is plotted
below the installed characteristic flow curve.
On the lower graph, there are three horizontal lines. The
middle line shows the ideal gain of 1.0. The other two
lines, at gains of 2.0 and 0.5 show the upper and lower
limits of the control range that was defined earlier.
Figure 16 represents an eccentric disk valve with an
installed flow characteristic curve that is far from linear,
and the installed gain deviates considerably from the ideal
value of 1.0. The effectiveness of good control is lost
when the installed process gain gets outside the control
range.
Note that the eccentric disk valve in this process only has
a control range of about 20 degrees of travel (from 25
degrees to 45 degrees.) In the higher travel region,
above the control range, the installed flow curve has
flattened out, and the installed gain is approaching zero.
There is virtually no ability to control at all in this region.
Wide ranges in valve travel can be made with essentially
no change in flow.
Conversely, at the lower end of the valve travel, below the
control range, the installed gain is tending to get very
high. This makes the loop so sensitive to changes that
loop instability is a real threat.
What this means is that the eccentric disk valve has been
incorrectly applied in this process. Two things could be
wrong. The first is that the wrong valve style has been
used. The second is that the valve may be incorrectly
sized for the application. Oversizing of valves is often a
major problem in the industry, partly due to a tendency to
use line-size valves.
It cannot be overemphasized that valve style has a major
impact on the installed process gain and control range. In
general, globe valves tend to have the widest control
range and the best performance in terms of process
variability. Test results show that globe valves have
nearly triple the control range of conventional butterfly
valves.
Conventional butterfly style valves tend to have the
narrowest control range and the lowest performance in
terms of process variability reduction. Other valve styles

EFFECT OF VALVE SIZING ON CONTROL RANGE

AND INSTALLED PROCESS CHARACTERISTICS
As stated, proper sizing of the control valve also plays an
important role in achieving a wide control range and a
good installed process characteristic.
It is possible to take a globe valve capable of achieving a
control range of 70% of its travel range and reducing it
down to only 20% of its travel range simply by oversizing
it. This is what often happens when line size valves are
used instead of using one size smaller with swages.
Regardless of its inherent flow characteristics, when a
valve is oversized, the installed flow characteristics tend
to appear more like a quick opening valve. As soon as
the disc leaves the seat the result is a lot more flow than
normally would be expected. The valve doesn't have to
travel very far before it provides all the flow that is
needed. Thus, the control range is very small, and the
sensitivity or flow gain is very high.
A valve is much more effective and controls much more
smoothly when the control range covers a larger
percentage of the available travel.
SUMMARY
There are several key points to remember when working
to achieve effective loop performance:

Control valves can play a major role in reduction

of process variability in the loop.

The importance of dead band and dead time

cannot be over emphasized when it comes to
reducing process variability.

Proper valve selection means managing the

installed gain by matching the valve to the
process. For example, conventional butterfly
valves have a very narrow control range
compared to other valve styles.

70

70

69

69

68

68

67

Controller Output (%)

67
Flow Rate

66

(%)

66

(%)

65

65

64

64

63

63

62

62

61

61

60
60
239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254

Time (seconds)
Figure12. Dead time and time constant measurement

D351615X012