Preprints of the 8th IFAC Symposium on Advanced Control of Chemical Processes

The International Federation of Automatic Control

Furama Riverfront, Singapore, July 10-13, 2012

Valve backlash and stiction detection in integrating processes

M. Farenzena and J. O. Trierweiler

Group of Intensification, Modelling, Simulation, Control and Optimization of Processes (GIMSCOP)

Department of Chemical Engineering, Federal University of Rio Grande do Sul (UFRGS)
Rua Luiz Englert, s/n CEP: 90.040-040 - Porto Alegre - RS - BRAZIL,
Fax: +55 51 3308 3277, Phone: +55 51 3308 4163
E-MAIL: {farenz, jorge}@enq.ufrgs.br

Abstract: Valve nonlinearities are responsible for producing limit cycles in the loop and consequently
decreasing plant performance. In the case of an integrating plant, once the loop oscillation is detected, the
engineer should distinguish between stiction and backlash, but no method is available in the literature to
address this class of problem. In this paper, it is proposed a data-driven method to distinguish between
loop stiction and backlash for integrating processes. The closed-loop pattern provides enough information
to verify when either stiction or backlash is available. The efficacy of the proposed technique is
corroborated by simulation case studies.
Keywords: Valves, Backlash, Static friction, Integrators, Hysteresis.

problems: backlash, deadband, leakage, and blocking.

1 INTRODUCTION
In the last twenty years, both academia and industry have
been focus on methods to evaluate in real time control loop
performance. Many industrial surveys have provided the
reasons for this concern: around 60% of all industrial loops
perform poor (Bialkowski, 1993), and improve their
performance means reduce loop variability and increase plant
reliability. Among the most frequents loop illnesses, the
valve has one major disorder: around of 30% of all valves
have any degree of damage, being responsible for increase
significantly loop variability. Two of the most frequent valve
injuries are stiction and backlash.
Stiction imposes limit-cycles in the loop, increasing its
variability and inserting plant-wide periodical disturbances,
what explains why the large majority of the valve related
works is focus on stiction. Backlash also increase loop
variability, however in the case of open-loop stable plants,
backlash only reduces the control loop performance, but does
not insert limit-cycles, as stiction. On the other hand, when
the plant is an integrator, a limit-cycle similar to stiction is
inserted.
If a limit-cycle is detected in a given loop, it is important to
distinguish between stiction and backlash, because backlash
is an invertible nonlinearity, what allows a perfect
compensation without valve replacement. On the other hand,
stiction nonlinearity is more difficult to compensate. This is
the scope of this work: to propose a data-based method to
distinguish between stiction and backlash, in the presence of
limit-cycles, for integrators.
In the literature, backlash detection has not received
significant interest. Hagglund (2007) has proposed a method
to detect backlash for stable loops. Moreover, techniques to
quantify and to compensate backlash are also introduced in
the same work. Ling et al. (2007) have introduced a
nonparametric statistical method to diagnose four valve
However, stiction is not considered. Several works follow the
same direction, whose scope is to distinguish among several
valve problems, but they do not aim to distinguish between
stiction and backlash. Bocaniala et al. (2003) have proposed a
fuzzy classification solution for fault diagnosis of valve
actuators. Dstegr et al. (2006) have suggested a graph
method in which each fault is described by a three levels of
knowledge. Tudoroiu and Zaheeruddin (2005) have analyzed
HVAC systems and methods to detect and to diagnose faults,
including backlash, based on frequency and spectral analysis.
This paper is segmented as follows: in section 2 a brief
description about the backlash and stiction phenomena will
be exposed. In section 3, the method here introduced will be
described. In section 4, the proposed method will be applied
in a set of cases studies, corroborating its validity. The paper
ends with the concluding remarks.
2 STICTION: MODEL AND COMPUTATION
2.1 Deadband: definition
According to Fisher-Rosemount (1999), deadband is defined
as the range through which an input signal can be varied,
upon reversal of direction, without initiating an observable
change in the output signal. Dead band is the name given to
a general phenomenon that can apply to any device. One
cause of valve deadband is backlash, whose definition is the
general name given to a form of dead band that results from
a temporary discontinuity between the input and output of a
device when the input of the device changes direction. Slack,
or looseness of a mechanical connection is a typical
example.
The phase plot of input-output behavior of a valve where the
backlash is seen can be described as Fig. 1. It has two

IFAC, 2012. All rights reserved. 320

8th IFAC Symposium on Advanced Control of Chemical Processes
Furama Riverfront, Singapore, July 10-13, 2012
regions: when the controller comes to rest or change its
direction (A), the valves becomes stationary until the
deadband is overcame (B), and then the valve starts the
motion. The deadband magnitude is given by DB.
Fig. 1: Typical phase plot of a valve with backlash.
Comparing the process output for a loop where a valve with
backlash is applied to an open loop stable process and to an
integrator, the behavior is completely distinct. In the first
case, only the process variability increases with the
increasing backlash. Fig. 2 illustrates the process variable
(PV) for four different DB for an open loop stable process,
with a PI controller. The variability (2) is also shown.
Fig. 2: Process variable behavior for open loop stable
processes with variable deadband.
For open loop stable processes, the backlash diagnostics is
difficult, because its effect can be confused with a poorly
tuned controller. However, the differentiation between
stiction and backlash is not difficult, because only stiction
can insert a limit-cycle.
On the other hand, if the process is an integrator, the process
variable becomes oscillatory, spreading this behavior in all
321
loops. Fig. 3 shows the process variable for four SISO loops,
with variable deadband, when the plant is an integrator.
Fig. 3: Process variable behavior for integrating processes
with variable deadband.
Based on Figs. 2 and 3, it is clear the impact of plant open
loop behavior, when backlash is present in the loop. When
the plant is an integrator, then a limit-cycle is imposed,
similar to stiction behavior. It is important to distinguish
between these two valve damages, because backlash non-
linearity can be inverted, i.e. using a simple algorithm as
shown by Hagglund (2007) it can be compensated. On the
other hand, stiction cannot be completely inverted and its
compensation is more difficult (Srinivasan and Rengaswamy,
2008).
In the next section the stiction phenomenon will be briefly
described and a parallel with backlash will be made.
2.2 Stiction definition and model
A valve with stiction has an apparently similar behavior then
backlash, as shows the phase plot, described in Fig. 4
(Choudhury et al., 2008) .
When the valve changes its direction (A), the valve becomes
sticky. The controller should overcome the deadband (AB)
plus stickband (BC), called staticband or apparent stiction.
Then, the valve jumps to a new position (D). The stiction
model consists of these two parameters: S (staticband) and J
(slipjump). Next, the valve starts moving, until its direction
changes again or the valve comes to rest, between D and E.
The main difference between backlash and stiction is the
slipjump, which is not present in backlash. It represents the
abrupt release of potential energy stored in the actuator due to
high static friction in the form of kinetic energy as the valve
starts to move. The magnitude of the slipjump is crucial to
determine the limit-cycle amplitude and frequency.
Generally, the stiction impact over the loop is more
aggressive than backlash, because of the jump. It inserts
limit-cycles in both open loop stable and integrating
processes.
8th IFAC Symposium on Advanced Control of Chemical Processes
Furama Riverfront, Singapore, July 10-13, 2012
Fig. 4: Typical (IO) behavior of a valve with stiction.
In the literature, several models are available to describe
stiction behavior. In this work, the 2 parameters model
proposed by Kano et al. (2004) will be used, because its
simplicity and good agreement with the real stiction behavior.
3 BACKLASH AND STICTION DETECTION
This section describes the method here proposed to
distinguish between stiction and backlash for integrating
processes.
Before, two considerations are taken: the sampling time is
fast enough to describe the process dynamics and the loop
does not have under-damped behavior. In the first case, if the
sampling time is wrongly selected, i.e. slower than necessary,
the method will point stiction in all cases. In the second, if
the controller is fast tuned, the algorithm will detect stiction.
Fig. 5: Backlash response for an integrating process peak
(I) and intermediate points (II) slope.
322
The idea behind the method is very simple. If we carefully
inspect the peaks (I) and the intermediate points (II) of the
response for backlash and stiction phenomena there is a slight
difference. Fig. 5 shows this behavior for backlash and Fig. 6
shows for stiction.
Fig. 6: Stiction response for an integrating process peak (I)
and intermediate points (II) slope.
The slope is the core of the method: in the case of backlash
the angle in the peak (peak) and in the middle (middle) are
significantly different (see Fig. 5), while in the case of
stiction they are similar (see Fig. 6). Thus, the stiction and
backlash can be easily distinguished, through the angles in
the peak and in the middle, as illustrated in Fig. 7.
The procedure for the method to distinguish between stiction
and backlash is following described:
1. PV data is collected and the white-noise removed;
2. The peaks and valleys are identified and the angles
for the peaks (peak) and the value between the valley
and peak (middle) are computed for each.
3. The mean value for each is computed ( and
).
4. The Backlash Index (BI) is computed based on
and :
(1)
8th IFAC Symposium on Advanced Control of Chemical Processes
Furama Riverfront, Singapore, July 10-13, 2012

Tab. 1: Loop parameters used in the stiction diagnostics.

Param Description Value
KC Controller gain [0.7:0.1:1.5 2:0.5:4]
I Controller integral [1:0.2:2.4]
constant
S Staticband [1:0.5:6]
J Slipjump [0.5:0.1:1]S
Process time constant [10:10:100]

Based on Tab. 1, a total of 52800 different scenarios have

been analyzed. In 99.8% of all cases the stiction was
correctly pointed, while in the remaining 0.2%, no conclusion
is stated by the method. None of cases the method has
detected incorrectly backlash.
In the complementary analysis, the backlash should be
diagnosed under a wide variety of loop parameters. The
values used for backlash detection are shown in Tab 2.
Tab. 2: Loop parameters used in the backlash diagnostics.
Param Description Value
KC Controller gain [0.7:0.1:1.5 2:0.5:4]
I Controller integral [1:0.2:2.4]
Constant
DB Deadband [0.5:0.5:6]
Fig. 7: Peak and middle angles for an integrating process Process time constant [10:10:100]
with backlash.
The angles are measured considering the mean and the
In this analysis, in all cases the algorithm detected the
extreme points and the neighbor. If the angles are different, in
backlash correctly. None of the cases the conclusion was null
the case of the peak, the mean angle is taken. If the data is
or misleading.
corrupted by white-noise, the mean between three points is
advised. In the second analysis, the influence of the sampling time will
be evaluated. As previously stated, if this value is high, then
The following rules are used to diagnose stiction:
the backlash detection will be spoiled. Using the same
BI 3 backlash; procedure, the deadband and process time constant will be
varied according to Tab. 2. The controller gain will be set
BI 1.5 stiction; equal to 2 and the integral constant is equal to the process
time constant. Tab. 3 shows the influence of sampling time
3 > BI > 1.5 no decision. (st) in the backlash detection.

These values have been based on a large number of Tab. 3: Percentage of backlash detection with variable
simulations. The main advantages of this method are that it is sampling time.
computationally fast and requires only routine operating data. st % of detection
Moreover, no controller tuning parameters and no
information about the valve stem are necessary. 0.1 100%
0.2 100%
4 CASE STUDIES
0.5 100%
The objective of this section is to evaluate the proposed
method to distinguish between stiction and backlash for 0.75 90%
integrating processes, using only routine operating data. In
the first analysis, the stiction should be diagnosed, 1.0 90%
considering several plants, controllers, and stiction 2.0 78%
parameters, as described in Tab. 1. A SISO loop is used with
a PI controller and integrating plant. The stiction model is 5.0 38%
inserted between controller and plant (Hammerstein model).
In all cases, the sampling time was equal to 0.1.

323
8th IFAC Symposium on Advanced Control of Chemical Processes
Furama Riverfront, Singapore, July 10-13, 2012

Tab. 3 shows the impact of sampling time over the methods LING, B., ZEIFMAN, M. & LIU, M. (2007) A practical
accuracy, where increasing sampling times cause reduction in system for online diagnosis of control valve faults.
backlash detection. However, it is clear that the method does Proceedings of the IEEE Conference on Decision
not require an extremely fast sampling time. and Control.
SRINIVASAN, R. & RENGASWAMY, R. (2008)
Approaches for efficient stiction compensation in
5 CONCLUSIONS process control valves. Computers & Chemical
In this work, we have presented a novel solution to the Engineering, 32, 218-229.
problem of stiction and backlash detection in integrating TUDOROIU, N. & ZAHEERUDDIN, M. (2005) Fault
loops. The detection method is based on process variable detection and diagnosis of the valve actuators in
patterns of a valve with backlash and stiction. The slope in HVAC systems, using frequency analysis. ICIECA
peaks and in the middle of the curve provides the necessary 2005: International Conference on Industrial
information for detection. This procedure can be automated Electronics and Control Applications 2005.
using the proposed Backlash Index
The efficacy of the proposed technique was corroborated
through several simulation case studies. When backlash or
stiction should be diagnosed, a wide variety of plants,
controllers, and stiction or backlash scenarios have been
evaluated. In the first, only stiction is seen in the loop and in
the second only backlash. In both cases, the percentage of
correct detection was around 100%. No incorrect diagnostic
was provided in all cases.

ACKNOWLEDGMENTS
The authors are very grateful for the grants from
PETROBRAS.

REFERENCES
BIALKOWSKI, W. L. (1993) Dreams versus reality: A view
from both sides of the gap. Pulp and Paper Canada,
94, 19-27.
BOCANIALA, C. D., SA DA COSTA, J. & LOURO, R.
(2003) A fuzzy classification solution for fault
diagnosis of valve actuators. Lecture Notes in
Artificial Intelligence (Subseries of Lecture Notes in
Computer Science).
CHOUDHURY, M. A. A. S., JAIN, M. & SHAH, S. L.
(2008) Stiction - definition, modelling, detection and
quantification. Journal of Process Control, 18, 232-
243.
DSTEGR, D., FRISK, E., COCQUEMPOT, V.,
KRYSANDER, M. & STAROSWIECKI, M. (2006)
Structural analysis of fault isolability in the
DAMADICS benchmark. Control Engineering
Practice, 14, 597-608.
FISHER-ROSEMOUNT, A. (1999) Control Valve
Handbook.
HAGGLUND, T. (2007) Automatic on-line estimation of
backlash in control loops. Journal of Process
Control, 17, 489-499.
KANO, M., MARUTA, H., KUGEMOTO, H. & SHIMIZU,
K. (2004) Practical model and detection algorithm
for valve stiction. IN IFAC (Ed.) 7th IFAC
DYCOPS. Boston, USA.

324