See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/345259859

Non-Obtrusive Stiction Detection Methods for Control Systems

Article in Technology Reports of Kansai University · August 2020

CITATIONS READS
0 53

8 authors, including:

Kelvin DE JESUS Beleño Saenz Javier Jiménez-Cabas

Universidad Autónoma del Caribe Corporación Universidad de la Costa
22 PUBLICATIONS 124 CITATIONS 40 PUBLICATIONS 138 CITATIONS

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Solución basada en software para el monitoreo y diagnóstico en línea del desempeño de lazos de control, para aumentar y mantener del buen rendimiento de los procesos
productivos automatizados. View project

Gastric Cancer detection with E.nose View project

All content following this page was uploaded by Javier Jiménez-Cabas on 03 November 2020.

The user has requested enhancement of the downloaded file.

 ISSN: 04532198
Volume 62, Issue 07, August 2020

Non-Obtrusive Stiction Detection Methods for Control

Systems
Leonardo Escobar Davidson1, Stephany Sucerquia Rincones 1, Samir Hadechni Bonett1, Jhon Ramirez
Parra1, Jean Coll Velasquez1, Kelvin Beleño Saenz1, Javier Jiménez-Cabas2, Carlos Díaz Saenz1

Universidad Autónoma del Caribe, GIIM research group, Barranquilla, Colombia 1

Departamento de Ciencias de la Computación y Electrónica, Universidad de la Costa, Barranquilla,
Colombia2

Abstract— Industrial processes play a key role in the production sector. Production demands have forced the
search for strategies such as automatic diagnosis to maintain continuous production with minimized machine
failures. An industrial process provides many measured, controlled, and manipulated variables that associate
nonlinearities and uncertainties, so it is necessary to monitor them, to acquire information about the behavior
of the process. Historical and present information resulting from monitoring is used to implement intelligent
monitoring systems. Within the monitoring scheme is the detection of failures, diagnosis, and restoration of
operating conditions according to process performance criteria [1].

Keywords— Control systems, Hysteresis, Cross correlation, Non-linearity, Curve fitting.

1. Introduction
The diagnostic procedure begins when oscillation is detected in the control loop, the most common origin of
this variation occurs in the actuators, mainly in the control valves because these are exposed to both external
and internal factors during the process, as friction in the valve increases, it moves more slowly, which in some
cases causes it to jam momentarily and suddenly jump to another position, causing unwanted process
oscillations, in order to detect these oscillations is made use of a number of non-invasive detection methods.
With this computational tool you can analyze and diagnose the possible static friction failures that can occur
in the control elements (actuators) and the parameters to be considered according to the type of process [2].

2. Tools and Methods

2.1 Mathematic Models

Three methods were integrated through the MATLAB® application plus the toolbox AppDessigner, in which
they were implemented and developed. The first method called ‘nonlinearity detection and PV-OP pattern
analysis’ is based on the analysis of the graph obtained by overlapping the mv data against the OP data so no
mathematical model was necessary. The second method called 'cross-correlation-based stiction detection'
requires a mathematical model which is, cross correlation function observed in Equation 1.

(𝑓 ∗ 𝑔)𝑖 ≝ ∑𝑗 𝑓∗𝑗𝑔𝑖 + 𝑗 (1)

This function was used to obtain a graph that corresponds to the similarity between two signals, in this case,
of the OP and PV data, during the development of the computational tool in the MATLAB® software, the
function 'xcorr' whose syntaxis is: r xcorr(x,y) was used; where 'x' and 'y' are the PV and OP data.

3961
 Davidson, et.al, 2020 Technology Reports of Kansai University

Figure 1. Cross-correlation between control signal and process output for the case of no stiction (left) and
stiction (right).

For the third method, called ‘diagnosis based on curve adjustment’ was used a numerical iterative method, i.e.
Nonlinear least-square method was used to find the best fitting for the sine curve, and a polynomial adjustment
was used for the triangular curve [10].

Figure 2. Sine adjustment.

Figure 3. Triangular fit.

In MATLAB software®, the ‘polyfit’ function was used for the triangular setting, and the 'lsqcurvefit' function
was used for the sine setting.

2.2 Boundary Conditions

The computational tool developed in the MATLAB AppDesigner toolbox [4], is divided into tabs, the first
3962
 ISSN: 04532198
Volume 62, Issue 07, August 2020

shows an overview of the system to be analyzed (Figure 4), on the other 3 tabs apply the methods mentioned
above.

Figure 4. Overview computational tool.

The data is inserted in the format (.xlsx), with the parameters; time, set point, process variable, and output
consecutively on each column in the table.

2.3 Design
In this window (Figure 5), you can see all the data with which the tool works in addition to being able to see
the graphs of the PV and OP data, at the bottom there is a slider bar to manipulate the maximum range of the
horizontal axis of both graphs, and finally the values of the mean and the standard deviation of both variables
and the possibility to display the normalized or in-filter graphs.

3963
 Davidson, et.al, 2020 Technology Reports of Kansai University

Figure 5. Overview of data loaded.

In this tab (Figure 6), the curve fitting method is applied, which consists of taking the PV and OP data and
adjusting it to a sine and triangular wave, respectively. One of the common characteristics of the controlled
variable and the manipulated variable is the pattern of its curve, so for this method it is performed a fitting of
PV, to detect the typical signature of stiction and distinguish it from other causes, to both sinusoidal (Figure
2) and triangular (Figure 3) uses a segment of the curve equal to 10% of its total size [12] [11]. If the fit for
the triangular wave is better, then stiction is concluded, otherwise no stiction occurs. By comparing the error
between real and fitted data, an evaluation of the accuracy of approximation and then a stiction index can be
obtained [6].

Figure 6. Curve fitting.

Figure 7. OP vs PV.
3964
 ISSN: 04532198
Volume 62, Issue 07, August 2020

In this tab (Figure 7), the ‘Nonlinearity detection and PV-OP pattern analysis’ method is applied, in which
the PV variable data is plotted against the data of the OP variable. In this method the detection of valve or
process nonlinearity is first carried out using higher-order statistical method-based. Once a nonlinearity is
detected, the data are treated by a wiener filter, and the PV–OP plot, generated from a segment of the data
that has regular oscillations, is used to its cause isolate. A signature of valve stiction is when PV–OP plot
shows cyclic or elliptic patterns. If not, such patterns are observed, it is concluded that there are valve
problems, but these are not due to the stiction [8].

Figure 8. Cross-correlation function.

In this tab (figure 8), the ‘Static friction detection based on cross-correlation’ method consisting of the analysis
of the cross-correlation plot between the OP variable and the PV variable is applied. This was the simplest
method to implement because it is based entirely on the analysis of the cross-correlation function.

If the cross-correlation function between controller output OP and process variable PV is an odd function (i.e.
Asymmetric with the vertical axis), the likely cause of the oscillation is stiction.
• if the cross-correlation function graph is even (i.e. Symmetric with the vertical Axis), then stiction is not
likely to have caused the oscillation. In this case, the oscillation may be due to external disturbances,
interaction, or aggressive tuning of the controller. [3]

The following assumptions are needed to apply this stiction detection method:
• the process does not have an integral action.
• the process is controlled by a pi controller.
• the oscillating loop has been detected as being oscillatory with a significantly large amplitude.

3965
 Davidson, et.al, 2020 Technology Reports of Kansai University

3. Results and discussion

A comparative table (Table 1) was made to determine the three methods to be implemented according to the
established requirements, being selected: Cross-Correlation-based stiction detection, nonlinearity detection
and PV–OP pattern analysis, diagnosis based on curve fitting. Those that met the objective of detecting static
friction in control loops.

Table 1. Comparative table.

Static friction Diagnoses static friction by

Simple statistics and advanced diagnostics.
detection based on nonlinearity and displacement in
Great pressure handling and reliability [9].
cross-correlation actuators.

It is performed first using a higher order

statistical method based on ngi and nli
Diagnoses nonlinearity, analyzing
Nonlinearity indexes, once it is detected or not linearity,
valve movement behavior or
detection and PV-OP the data are processed by a wiener filter
changing the input signal to the
pattern analysis (pvf, opf) and the pvf – opf chart, generated
valve.
from a segment of the data that has regular
oscillations, are used to isolate its cause [5].

The output controller or process

A SP-OP, or SP-PV data adjustment is
variable conforms in parts (each
Diagnosis based on performed to detect the typical static
significant half cycle) to the
curve adjustment friction signature and distinguish it from
triangular and sinusoidal curve
other causes [7].
segments using a ls method

Therefore, the tool that was presented allows to show the results of the detection of oscillations from the
application these methods. The 'cross-correlation-based stiction detection' method detects static friction by
analyzing the cross- correlation function applied to the PV and OP variables, on the other hand, the
'nonlinearity detection and PV–OP pattern analysis' method detects disturbances from analysis and similarity
with cyclic or elliptical patterns, and finally the 'diagnosis based on curve fitting' method that allows static
friction to be detected by adjusting the data to ideal curves with the same properties( period, etc.). Tests
were carried out with different databases found in the sacac repository to verify the optimal functioning of
the computational tool in situations where the process presented, or not, static friction. By implementing the
methods, it was concluded that static friction detection analysis is intuitive, i.e. That their result does not
depend on a value or a constant, therefore text boxes were added in the different interface tabs, which
describe the conditions under which the presence of static friction in the system can be concluded to guide
the user.

4. Conclusions
The computational tool has a versatility when processing each of the methods and is very easy to use. On the
other hand, its function is to detect and analyze whether the closed loop control system under study has static
friction, which we can conclude from the results of each method. Which are classified as follows cross-
correlation corresponding to the analysis of the graph obtained by overlapping the mv data against the OP
data, the method of nonlinearity detection and PV-OP pattern analysis is intended to graph the data of the
PV variable against the data of the OP variable, determined if it has static friction, as last would be the method
of diagnostics based on the adjustment of curves that aims to adjust the PV-OP or PV-PV data, in this way to
determine whether there is static friction or distinguish it from other causes. Predictive Model Control
3966
 ISSN: 04532198
Volume 62, Issue 07, August 2020

(MPC) is a useful tool to control processes that handle a large number of input and output variables, usually
implemented in chemical and industrial plants, it is especially effective in continuous data readings so it could
be a great Contribution to future updates of this computational tool [13]. The results obtained with the
methods were validated with the Simulink development tool, which is a visual programming environment,
which works in the MATLAB programming environment®, thus validating theoretical and practical. When
a stable system was simulated in the environment, compare, and verify that the values obtained in the GUI
are accurate. Otherwise, it was guaranteed that they would not be invasive so as not to alter the processes, the
tool describes the performance of the driver. On the other hand, stable system tests were carried out to
validate and verify each of the methods. As stiction normally occurs in control valves, the work area is
normally oil and hydrocarbon industries and to transport various substances the use of pipelines is required,
therefore a correct control of the use of these is required. in conjunction with the valves to achieve superior
performance, it is recommended to use a predictive control strategy to solve possible problems related to the
correct control and operation of pipelines [14].

5. References
[1] Thornhill, nf and horch, a. (2007). Advances and new directions in the detection and diagnosis of
disturbances throughout the plant. Control engineering practice, 15 (10), 1196-1206.

[2] C. Pryor, "autocovariance and power spectrum analysis derive new information from process data.,"
control eng, vol. V 29, no. N 11, pp. 103–106, 1982.

[3] Thornhill, nf and hugglund, t. (1997). Oscillation detection and diagnostics in control loops. Control
engineering practice, 5(10), 1343-1354.

[4] Verification, validation, and testing: MATLAB and Simulink solutions. (s. F.). MATLAB &
Simulink. Retrieved 10 March 2020, from https://la.mathworks.com/solutions/verification-validation.html.

[5] M. Jelali, control performance management in industrial automation. London: springer, 2013.

[6] Escobar Davidson, L, Rincones Sucerquia, S.S, Diaz Sáenz, c., & Jiménez Cabas, j. (2020, May).
Computational tool for the detection and diagnosis of oscillations in a control system. Universidad
Autonoma Del Caribe.

[7] Tarantino R., Szigeti F., Colina E., (2000) “Generalized Luenberger Observer-Based Fault-Detection
Filter Desing: An Industrial Application”. Control Engineering Practice. Julio, pp. 665-671.

[8] Thornhill, nf and horch, a. (2007). Advances and new directions in the detection and diagnosis of
disturbances throughout the plant. Control engineering practice, 15 (10), 1196-1206.

[9] N. F. Thornhill and t. H'gglund, "detection and diagnosis of oscillation in control loops," control eng.
Pract., vol. 5, no. 10, pp. 1343–1354, 1997.

[10] E. Naghoos, "oscillation detection and causality analysis of control systems", era.library.ualberta.ca,
2016. [online]. Available: https://era.library.ualberta.ca/items/57ba6990-7ddc-4b58-8555-
b9ea5ec4b79d/view/b06afee3-316c-48e6-b8b6-f08c3d7a9ae4/naghoosi_elham_201607_phd.pdf. [accessed:
31- aug- 2019].
3967
 Davidson, et.al, 2020 Technology Reports of Kansai University

[11] Jelali, M. (2012). Control Performance Management in Industrial Automation Assessment,

Diagnosis, and Improvement of Control Loop Performance. 1st ed. [eBook] Available at:
https://www.springer.com/gp/book/9781447145455.

[12] Takahashi, s., tachibana, k., & saito, t. (1991). U.s. patent no. 5,043,862. Washington, dc: u.s. patent
and trademark office.

[13] Borrero-Salazar, A. A., Cardenas-Cabrera, J. M., Barros-Gutierrez, D. A., & Jiménez-Cabas, J. A.

(2019). A Comparison Study of Mpc Strategies Based on Minimum Variance Control Index Performance.

[14] Cardenas-Cabrera, J., Diaz-Charris, L., Torres-Carvajal, A., Castro-Charris, N., Romero-Fandiño, E.,
Ruiz Ariza, J. D., & Jiménez-Cabas, J. (2019). Model Predictive Control Strategies Performance Evaluation
Over a Pipeline Transportation System. Journal of Control Science and Engineering, 2019.

This work is licensed under a Creative Commons Attribution Non-Commercial 4.0

International License.

3968

View publication stats