Diagnosing and Solving Common

Control Loop Problems EMPOWERING PEOPLE. DRIVING ASSETS.

Abstract
A substantial percentage of industrial control loops perform poorly. This decreases product quality, lowers
profitability, and makes processes difficult to operate. A few types of problems are commonly responsible for
degraded loop performance. To improve control loop performance, problems need to be accurately diagnosed
and the correct steps taken for solving them. This paper describes the most common problems affecting control
loop performance and how to identify and solve them.

Introduction Poor controller tuning (77%)

Incorrect valve size or trim (51%)
Studies show a surprisingly large percentage of Valve hysteresis (27%)
industrial control loops perform poorly. Ender Valve stiction (26%)
[1] and Bialkowski [2] referred to several studies in Nonlinear process characteristic (18%)
two papers published in 1993. The reported loop Large external disturbances (18%)
performance can be summarized as follows: Incorrect control strategy (15%)
More than 30% of controllers remain in
Note, the individual percentages add up to more
manual.
than 100%, indicating the worst performing
More than 30% of loops actually increase
loops typically had more than one problem. The
variability over manual control due to poor
remainder of this paper will look more closely at
tuning.
15% of loops have design problems. diagnosing and correcting these problems.
30% of loops have equipment problems.
Fewer than 20% of installed control loops Oscillations
perform well. Oscillations are a frequently observed symptom of
poor control performance. The inherent feedback
Based on performance assessments done on in a control loop can cause oscillatory behavior
thousands of control loops, Desborough and Miller under the conditions discussed below. In addition
[3, 4] confirmed control loop performance was on to feedback in the control loop, many process
average still this bad in 2001. In 2008, VanDoren designs include circular flows of mass or heat,
[5] reported the following breakdown of loop which can also provide mechanical feedback and
performance: cause oscillations.
36% were in open loop Control loop oscillations can have several causes. It
10% had poor performance could be due to incorrect controller tuning, control
22% had fair performance valve stiction, external cyclic process disturbances,
16% performed acceptably or cycles introduced through the setpoint. Level
16% performed excellently control loops can also cycle due to control valve
hysteresis.
Common Causes of Poor Control
Control loop performance assessment software
Loop Performance
can automatically detect and diagnose control
The author audited several hundreds of poorly loop oscillations using various techniques [7]. In
performing control loops at 15 processing sites some cases manual diagnosis or verification might
to establish the root causes of degraded control be required. Manually finding the root cause of
performance [6]. The study revealed loops with cycling can be done quite effectively through the
poor performance typically had the following diagnostic steps below.
common problems:
 EMPOWERING PEOPLE. DRIVING ASSETS.

Oscillations Due to a Cyclic Setpoint oscillating positioner was eliminated in the previous
step. The next step would be to discriminate between
The first cause of loop oscillations to test for would overly aggressive tuning and valve problems as the
be a cyclic controller setpoint. This problem can cause of the cycle.
be determined by visually inspecting a time trend
of the controllers setpoint. If the setpoint is not Valve or damper problems can obviously be ruled
visually cycling, this cause can be ruled out. out if the controller output is not connected to such
a device, for example, the primary controller in a
A cyclic setpoint is most often due to incorrect cascade control arrangement. In this case, tuning
tuning settings in a higher-level controller. This should be suspected, as previously stated.
controller should be tuned to correct this problem.
Setpoint cycling in a cascade control arrangement If the controller output drives a mechanical control
can also be due to excessive interaction between the element, the shape of time trends of the oscillating
primary and secondary controllers. This cause can controller output and process variable should be
be verified by comparing the integral-action settings visually inspected to differentiate between control
of both controllers. They should differ by at least valve stiction and overly aggressive tuning as the
a factor of five to minimize cyclic interaction. The cause of cycling. If the oscillation is caused by
same applies for the derivative settings, if used. tuning, the process variable will tend to cycle in
the shape of a sine wave. The amplitude of the
Oscillations Due to a Cyclic Disturbance cycle will normally grow until the process variable,
or Control Valve Positioner controller output, or both become bounded by
physical or programmatic limits.
If the oscillations are not setpoint-driven, the next
step would be to determine if they are generated
within the loop itself or by a cyclic external
disturbance. To test this, a time trend of the process
variable behavior with the controller in manual can
be inspected visually. If cycling is present while
the controller is in manual, it could be caused by
an oscillating control valve positioner or by a cyclic
external disturbance.

An oscillating positioner can easily be confirmed

by inspecting the device in operation while the
controller is in manual mode. Cyclic valve stem Figure 1. Controller cycling due to tuning - the controller
output saturates at its upper and lower limits
movement should be visible and periodic air
releases from the positioner audible. Overaggressive tuning refers to excessively high
static or dynamic gains in the controller. This can
Finding the root cause of a cyclic disturbance
be the result of setting the controller gain too high,
requires the engineer to look at factors outside the
integral time too short, or derivative time too long.
loop. Knowledge of the process and control system
If the control loop cycling is due to overaggressive
design and the use of plant-wide oscillation analysis
tuning, the controller should be properly tuned to
software can help pinpoint the root cause of the
correct this problem. See the section on tuning
cycling. The process may have to be physically
further down in this document.
inspected for unmeasured disturbances such as
steam traps. Oscillations Due to Control Valve Stiction
Oscillations Due to Overaggressive Another common cause of cycling in control loops
Tuning is control valve or damper stiction. Stiction is
short for Static Friction. It means once the valve
If the oscillations are not induced through the
position becomes stationary, it tends to stick in that
setpoint or an external disturbance, they must be
position. Then more force is required to induce
generated from within the loop. It is assumed an
 EMPOWERING PEOPLE. DRIVING ASSETS.

valve movement than to sustain the movement. Control Valve Problems

This has the undesirable effect of once the static
friction is overcome, the valve breaks free and Four control valve problems commonly affecting
overshoots the desired position. control loop performance are stiction, hysteresis,
incorrect valve sizing and incorrect valve trim.
If the cycling is caused by stiction, the controller Unless eliminated, these problems can make
outputs cycle often resembles a saw-tooth wave, controller tuning almost impossible. Control valve
while the process variable may look like a square stiction was already discussed above. The other
wave or irregular sine. This is referred to as a problems are discussed below.
stick-slip cycle.
Hysteresis
A valve with hysteresis behaves as if there is a
dead band between the control signal and the
valve position. Every time the controller output
changes direction, the dead band first has to be
traversed before the valve physically starts moving.
Hysteresis may be caused by play in mechanical
linkages, excessive friction in the valve, an
undersized actuator, or a defective positioner. Many
variable-frequency drives have a built-in, adjustable
Figure 2. After a setpoint change the control loop cycles hysteresis function that will have the same effect as
due to control valve stiction a mechanical dead band.

Stiction can also be detected by placing the controller Hysteresis can affect control performance in three
in manual and periodically making small (0.5%) step possible ways. First, because of the dead-band, the
changes in the controller output. A time trend of process variable cannot be controlled very precisely
the process variable can then be inspected to see if under the influence of disturbances. A loop with
the valve responds to every step made in controller hysteresis will display almost the same symptoms as
output or only to every few steps. The latter will sluggish controller tuning.
indicate a sticky control valve.
Second, a level control loop will continuously
Stiction might be caused by an over-tight valve stem oscillate after a setpoint change in the presence
seal, sticky valve internals, an undersized actuator, of hysteresis. This is as a result of the overshoot
or a sticky positioner. It can sometimes be solved (discussed earlier), combined with the control loops
by lubricating the valve stem while stroking its inability to immediately do an effective reversal of
position between limits. Most often, stiction can the excessive control action due to the dead band.
only be solved by servicing the control valve and/
Finally, control loop instability can occur if
or its positioner.
controller tuning is attempted without consideration
The author has found in some cases, lengthening of hysteresis. During a tuning test on a loop with
the integral time setting of the controller (or using hysteresis, part of the controller output change can
a lower integral gain depending on the integral unit be absorbed by the dead band and result in less
of measure) can reduce the tendency of a control valve movement. The process gain would seem to
loop to cycle due to stiction. This is possibly due be less than it actually is. As a result, the controller
to mechanical vibrations in the pipes and structures gain could be set too high. This problem may go
causing a slow creep in control valve position to unnoticed in the presence of minor disturbances or
balance the excessive force or torque coming from if small setpoint changes are made. However, once
the actuator. However, detuning the controller a significantly large setpoint change or disturbance
will lead to sub-optimal and sluggish control loop occurs, the effect of the hysteresis will be small in
performance and is therefore not the recommended comparison with the controller output change and
solution, but a temporary suppression of the cycle the control loop will be at a high risk of becoming
at best. unstable.
 EMPOWERING PEOPLE. DRIVING ASSETS.

Hysteresis can be detected from process data Nonlinear Valve Characteristic

by computer software or by manually doing a
hysteresis test. For the manual test, the controller The relationship between the valve position
output is changed in increments of about 5% and (or percentage valve travel between closed and
the process is given time to settle out after each open positions) and its CV (or flow rate, under a
step. Two steps are made in one direction, and a constant pressure differential) is called its inherent
final step is made in the opposite direction. The characteristic curve. Most valves come with one of
process variable should line out at the same level three distinct characteristic curves: quick opening,
after the first and final steps. If not, hysteresis is linear, and equal percentage [8].
present in the control loop.

Hysteresis can be reduced substantially by adding

a positioner to a control [8]. If a valve already has
a positioner, the control valve and/or positioner
need to be serviced.

Improper Valve Sizing

Valves should be sized to obtain full flow at
around 75% of travel, depending on the valve
style. If control valves are oversized it results in
a high process gain, meaning the same change
Figure 3. Valve characteristic curves
in controller output will have a larger effect on
the process. A high process gain by itself is not A valves characteristic curve is determined under
a problem (the controller gain can be set low to laboratory conditions with a constant pressure
compensate for it), but if the control valve has any differential across it. However, once it is installed in a
positioning deficiency like stiction or hysteresis, the process, the characteristic curve can be significantly
effect of the deficiency is amplified by the process distorted by the process own characteristics. The
gain. Increasing the size of control valves directly new relationship between flow and valve position is
reduces the quality of control. called the valves installed characteristic.
Less common than oversized control valves are For the majority of control applications, it is
undersized ones. In the latter case, the controller important to have the installed characteristic curve
output will often reach its upper limit with the valve be as linear as possible. If it is not, the process gain
fully open, but still the process can not be brought will vary with valve position and the tuning settings
to setpoint. Note, this could also be a limitation of will not work optimally across the full valve travel.
the process, especially if the process variable shows
no significant change above 90% of controller The installed characteristic of a control valve can be
output. obtained by recording the flow rate at various valve
positions and plotting these on a curve. The degree
Oversized and undersized valves can be detected of nonlinearity can be determined by dividing the
automatically with control loop performance steepest gradient on the plot by the shallowest.
assessment software or historical process data can Some loop diagnostic and tuning software packages
be manually inspected. Incorrectly sized control can help with calculating the degree of nonlinearity
valves should be replaced with properly sized valves. of a control valve.
Before undersized control valves are replaced, the
engineer should ensure it is in fact the valve causing If the degree of nonlinearity of the installed
the bottleneck in flow and not something else in characteristic of a control valve is more than two,
the process controller tuning issues can be expected. The
loop will tend to be more unstable at the steepest
After changing a control valve with a differently gradient on the installed characteristic curve and
sized one, the controller should be retuned to more sluggish at the flattest gradient.
match the new process gain.
 EMPOWERING PEOPLE. DRIVING ASSETS.

change slightly. The controller should be properly

tuned as described below to correct overshoot.

If the overshoot happens only on setpoint changes,

they could be slowed down by using a ramp
function or a first-order filter. Alternatively, the
Internal Model Control (IMC) tuning methods [9]
can be applied. IMC tuning virtually guarantees no
overshoot on setpoint changes.
Figure 4. Different control loop responses due to a Special consideration for level loops
nonlinear installed characteristic of the control valve
Some digital valve positioners can be programmed Under steady load conditions, a level control loop
with its own characteristic curve to linearize a does not require integral action to reach a new
nonlinear installed characteristic. Alternatively, setpoint. However, having integral action present
a linearization curve can be programmed in the anyway causes overshoot following the setpoint
process controller. The obvious, but more costly change. This is normal behavior for a level loop
solution, is to replace the control valve with one and impossible to eliminate through retuning the
having the correct characteristic. controller.
Improper Controller Tuning Sluggish control
As shown before, many poorly performing Sluggish control makes itself evident in large
control loops have incorrect tuning settings. offsets and relatively long recovery times following
Improper controller tuning can cause poor control disturbances and in an uncharacteristically long
performance in three ways: time for a control loop to reach a new setpoint.
Sluggish response to setpoint changes or The speed of response of a control loop is limited,
slow recovery after disturbances. mostly by the dead time in the control loop [10].
Overshooting past setpoint. For this reason, the speed of loop response should
Cycling in process variable and controller always be compared to an achievable value. For
output. example, if a flow control loop reaches a new
setpoint in 30 seconds, while a temperature loop
These problems are easily identified by inspecting takes 15 minutes, it cannot be said the flow loop
a time trend of a control loops setpoint, process performs better than the temperature loop without
variable and controller output. Oscillations accounting for the underlying process dynamics.
have been discussed previously; overshoot and
sluggishness are discussed below. Groundbreaking research by Harris and Desborough
[11] makes it possible for computer software to
Overshoot identify sluggish control loops automatically, while
requiring prior knowledge only of the process
Overshoot of the process variable past setpoint dead time. Their technique compares actual loop
after a disturbance or setpoint change is mostly the performance to minimum variance control.
result of the controllers gain being too high and/
or integral time too short (integral gain too high). Proper Controller Tuning Method
This results in too much controller action, which
in turn causes the process variable to overshoot its Retuning is required to improve the performance
setpoint. of a sluggish controller. The philosophy behind
controller tuning is essentially setting the controllers
Overshoot could indicate the control loop is P, I, and D actions to work in harmony with the
running very close to its stability limits, which is process characteristics. Proper controller tuning is
a serious condition because the control loop can done by using the static and dynamic characteristics
easily become unstable if the process characteristics of the process (gain, dead time, and lag) to set those
 EMPOWERING PEOPLE. DRIVING ASSETS.

of the controller (gain, integral, and derivative). Once new tuning settings are obtained, they should
be entered into the controller and a setpoint change
should be made to verify the expected behavior.
Then the loop performance should be monitored
over several hours or days to ensure the desired
performance objectives are met.

Figure 5. Controller Tuning Philosophy Nonlinear Process

Characteristics
Hence, before a controller can be tuned, it is
necessary to establish the dynamic characteristics Nonlinear behavior in a process makes its control
of the process. This is done through a step test difficult. With a standard controller on a nonlinear
by making small step changes to the controller process, optimal controller response can be obtained
output (or setpoint) and analyzing the response of only at one operating point. The loop could be quite
the process to obtain the process gain, dead time, unstable or sluggish under all other conditions. If
and time constant. the degree of nonlinearity (highest gain divided by
lowest gain) is greater than 2, control problems are
likely to arise. This is also true for varying lag and
dead time. If process characteristics change with
changing operating conditions, we say the process
is nonlinear.

Nonlinear Characteristic Curve

Even in the absence of a control valve, a process has
a certain relationship between control action and
process variable, called the process characteristic
curve. An example would be the process controlled
Figure 6. Determining process characteristics from
manual measurements on step test plot
by the primary control loop in a cascade control
arrangement.
Once the process characteristics are available,
the controller tuning constants can be calculated The process characteristic could be nonlinear,
by plugging the former into a set of tuning which causes the process gain to differ based on
equations. Various sets of tuning equations have controller output. For example, the heat transfer in
been developed, each with its own objectives or a heat exchanger is dependent on the temperature
advantages [12]. differential between the heating and heated fluids.
As the process fluids exit temperature setpoint
Software makes the identification of process becomes closer to the heating/cooling fluids entry
characteristics and the calculation of the controller temperature, the control loop will have a far lower
settings very simple and error-free, as well as gain compared to when the temperatures are far
provides simulations of predicted loop response apart.
and robustness. Using a software application to
optimize controller performance significantly The characteristic curve can be plotted from
improves accuracy and saves time. process variable measurements taken at different
controller output levels. Once the characteristic
Software can also simulate loop response and curve is established, its inverse can be programmed
provides quick fine-tuning capabilities. Note, once in the control system to linearize the control loop.
a control loop is properly tuned, robustness and
speed-of-response are in a trade-off. The loop Gain scheduling can also be applied by
cannot be made faster and more robust at the same programmatically loading different controller gains
time. into the controller based on its output.
 EMPOWERING PEOPLE. DRIVING ASSETS.

Other Nonlinear Process Characteristics However, if external disturbances can be

measured and if they have a predictable effect on
In some cases, process dead time and lag also a control loops process variable, a technique called
change with varying operating conditions. For feedforward control can be implemented.
example, the dead time between a feeder and mass
meter on a conveyor belt will change with the speed
of the belt.

It is quite possible for process gain, dead time, and

time constant to vary when some process operating
condition is changed. In this case, step tests should
be done under different operating conditions
and controller tuning settings calculated for each
condition. These PID settings should then be
programmatically loaded into the controller, based
on the operating condition.

Improving Control Loop Figure 7. Feedforward control added to feedback

Performance with Advanced control
Control Strategies
With feedforward control, the effect of a
Even after the hardware is fixed and the control disturbance is negated by applying a counteracting
loops are properly tuned, the performance of control action as soon as the disturbance is detected.
some control loops may still not meet desired Ideally, the disturbance and control action will
specifications. Feedback control loops have inherent cancel each other out and the process variable will
performance limits that cannot be exceeded remain unaffected.
regardless how well they are tuned. Processes could
be interactive, nonlinear, and inherently slow, which Cascade Control
also limits what a single controller can do.
Similar to the way in which flow loops control
In cases like these, the control performance better if its control valve is fitted with a positioner,
can likely be improved by implementing a more slower level and temperature loops control better if
complex control strategy. These advanced control their underlying process flows are controlled with a
strategies include cascade, feedforward and ratio secondary flow controller.
control, gain scheduling, decoupling, and ultimately,
model predictive control. The two most commonly For example, consider a level control loop on a
applied advanced control strategies are described large tank, of which the controller output goes to
below. a control valve directly. Now consider the valve has
hysteresis or stiction. These defects will cause the
Feedforward Control tank level to oscillate continuously due to reasons
discussed previously. The cycle will have a long
Process disturbances outside the control loop period due to the slow dynamics of the large tank.
can affect its process variable, thereby impacting The underlying flow will also cycle and this may
performance negatively. Regardless of how propagate to upstream or downstream processes,
tightly a feedback control loop is tuned, it can only depending on the design.
initiate a control action after the process variable
has deviated from setpoint. If the process has Now consider the level control loop is provided
long dead time or large lags, it could take an even with an inner flow control loop. If the control valve
longer time to eliminate the effect of a disturbance. had only hysteresis, the cycling will cease because
Straight-line control is never possible with hysteresis does not cause cycling in a flow control
feedback control if disturbances are present. loop. If the valve had stiction, the flow loop will
cycle, but the period of the cycle will be so short it
 EMPOWERING PEOPLE. DRIVING ASSETS.

will have virtually no effect on the level of the large Work Process for Improving
tank. The effect of the fast cycle on the upstream Control Loop Performance
or downstream processes could also be attenuated
to undetectable levels. Optimizing control loops and sustaining long-
term performance can be done very effectively
by implementing this seven-step approach
developed by PAS, Inc.:

1. First, a process control philosophy is

developed to provide guidelines for
consistency in implementation and
optimization of control loops.
2. Then computer software is used to collect
process data, analyze the performance of
all loops, and rank the loops according to
importance and performance.
3. The next step is to diagnose the problems
of the bad-acting control loops. This is
initially done by the performance assessment
Figure 8. Cascade control software, but certain problems require
human expertise for proper diagnosis.
Once diagnoses are obtained, problems are
In the same way, flow rate disturbances due to addressed accordingly.
pressure fluctuations will be detected and eliminated 4. Control loops with hardware problems
by the new flow control loop before the level is should have these problems addressed
affected. By using cascade control, we add an inner before tuning can be done effectively.
control loop to deal with control problems in a fast 5. Controllers with poor tuning are then step
and effective manner, shielding the primary process tested and new tuning settings calculated,
of its existence. implemented, and performance verified.
6. If performance is still not adequate,
One main criterion for successfully applying advanced control strategies can be
cascade control is the dynamics of the inner or implemented.
secondary control loop must be significantly faster 7. Once the loops have been optimized, the
than those of the outer or primary control loop. final step of periodic monitoring and
The secondary control loop should have a settling reporting is used to ensure any further
time at least five times shorter than the primary performance problems are identified and
control loop. If this criterion is not met, the addressed early.
secondary control loop will not effectively improve
control performance and interaction between the
two loops can be come a problem. To learn more about the 7 steps, please visit www.
pas.com to download the white paper.
Tuning of cascade loops should address the inner
loop first, and once it has been optimized, it is Summary
placed in cascade control mode after which the
outer loop is tuned. Several studies have shown average control loop
performance is very poor. While many engineers
look to controller tuning as the solution for
improving control loop performance, this paper
highlighted several scenarios where tuning is not
the best solution and provided techniques for
diagnosing and fixing the actual problems.
 EMPOWERING PEOPLE. DRIVING ASSETS.

References
[1] D.B. Ender, Process Control Performance: Not as Good as You Think, Control Engineering 40 (10), 1993,
173 186.

[2] W.L. Bialkowski, Dreams Versus Reality: A View From Both Sides of the Gap, Pulp & Paper Canada, 1993,
94 (11), 19 27.

[3] L. Desborough and R. Miller, Increasing Customer Value of Industrial Control Performance Monitoring:
Honeywells Experience, Proc. 6th Int. Conf. on Chemical Process Control (CPC VI), Arizona, USA, 2001,
172192.

[4] L. Desborough, P. Nordh, R. Miller, Control System - Process out of Control, Industrial Computing, August
2001, 52 55.

[5] V. VanDoren, Advances in Control Loop Optimization, Control Engineering, March 2008, 48 52.

[6] J.F. Smuts, Common Loop Problems, Internal Report, ControlServe Ltd., July 1999.

[7] N.F. Thornhill and T. Hagglund, Detection and diagnosis of oscillation in control loops, Control Engineering
Practice 5 (10) (1997) 1343.

[8] Fisher Controls International LLC, Control Valve Handbook, Fourth Edition, Emerson Process Management,
2005.

[9] S. Skogestad, Simple analytic rules for model reduction and PID controller tuning, Journal of Process Control
13 (2003) 291309.

[10] J.F. Smuts, Process Control in Practice, PAS Inc. Training Course Manual, August, 2002.

[11] L. Desborough and T. Harris, Performance assessment measures for univariate feedback control, The Canadian
Journal of Chemical Engineering 70 (1992) 1186-1197

[12] K.J. strm and T. Hgglund, PID controllers: theory, design, and tuning, Instrument Society of America,
1995.

About the Author

Dr. Jacques Smuts is the Business Development Manager for Control Products at PAS, Inc. He is the creator of
PAS ControlWizard, TuneWizard, and Process Analyzer software products, and the Process Control in Practice
training package. Jacques has optimized thousands of control loops in many industries, consulted to industrial
customers across a wide spectrum of processes, and trained several hundreds of engineers and technicians on the
theory and practice of process control and loop optimization.

About PAS
PAS is a leading supplier of software products and consulting services to the process industries worldwide and
has been delivering value to customers since 1993. Solutions offered include Alarm Management, Automation
Configuration Management, Knowledge Management, and Control Loop Performance Management.

PAS offers a comprehensive suite of products and services to support our customers control performance
improvement initiatives through the seven steps above. These include software for monitoring and improving
loop performance, process control training, and consulting and optimization services.

More information is at www.pas.com.