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

net/publication/339024305

Reliability analysis of Pneumatic Control Valve in Process Plant Using

Simulated Failure Data

Conference Paper · September 2019

DOI: 10.1109/ICSIPA45851.2019.8977781

CITATIONS READS

0 1,953

4 authors, including:

Nirbhay Mathur Rosdiazli Ibrahim

Universiti Teknologi PETRONAS Universiti Teknologi PETRONAS
23 PUBLICATIONS 93 CITATIONS 200 PUBLICATIONS 2,052 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Nirbhay Mathur on 13 February 2020.

The user has requested enhancement of the downloaded file.

Proc. of the 2019 IEEE International Conference on Signal and Image Processing Applications (IEEE ICSIPA 2019), Malaysia, September 17-19, 2019

Reliability analysis of Pneumatic Control Valve in Process Plant Using

Simulated Failure Data
Nirbhay Mathur1 Vijanth S. Asirvadam1 Azrina Abd Aziz1 Rosdiazli bin Ibrahim1

Abstract— Performance monitoring is an important part of To keep performing the control valve, its reliability and all
process plants. As it is very well known that process plants its related issues should be considered. Since, Fault detection
consist of many control loops to operate. Although, single and diagnosis has fascinate the researchers, however faults
downtime in a process plant can cause huge loses in production
and financially. Hence it is required to understand the behavior can be major divergence [3].The scholarly community has
of controllers, such as control valves to avoid the downtime responded to valve stiction issue by showing several results
and to reflect the reliability analysis. Valve stiction is the which are referenced in [4]. Stiction produces a vast scope
most common and reoccurring problem in pneumatic control of symptoms because of the vast decent variety of control
valves. Since many existing models define the valve stiction loops. Moreover, the valve operational process, the control
and how to control stiction effect, but how the stiction is going
to affect the reliability of the process and how does stiction parameters and the plant components affects the quality
can occur at different stages of valve opening is still missing of oscillation created by stiction [5]. Research review of
to consider. So, there is a need to understand the reliability current valve stiction recognition strategies mentioned in [6]–
behavior and effects on the whole process plant and also [8] presumed that single planning is not enough to analyze
including the maintenance of the plant. This paper focuses on stiction for all cases.
the prediction of reliability analysis using Weibull distribution
and visualizing the maintainability on the bathtub curve to This paper presents the reliability prediction of control
estimate and calculate the life of control valves. valve performance under several opening and closing of
Keywords− Control Valve, Reliability, Valve Stiction, Weibull valve plug from 10% to 100%. In the first stage, it will detect
distribution. the working behavior of control valve under several loads for
a single tank system. In the second stage, the failure data is
I. INTRODUCTION collected using the simulation for valve stiction. In the fourth
stage, the reliability model is implemented and validated
In today’s digital era of the industrial revolution, Industry to predict up to 95% accuracy of prediction. This research
(4.0) or (I4.0) is the latest trend, which is leading tech- paper is formed as follows sections, Section II will define the
nology to the next level ahead. Global promising of best control valve stiction and factors affecting it. Section III will
delivery output or results had made machine and software define the failure mode and failure of data collection. Section
more complicated and hence, more challenging to sustain IV will discuss the implementation of the reliability model
their reliability. Since it fails in performance or become and followed with section V for results. Finally, Section VI
incapable to continue performing operations as promised by will conclude this research paper work.
manufacturers.
Process plant typical consist of thousands of control loop. II. C ONTROL VALVE AND C ONTROL VALVE S TICTION
To make operation conditions smooth it is required to EFFECTING PERFORMANCES
make control loop design more stable. The quality of the This section will define the working of control valves and
plant operations is directly proportional to the loops with how stiction is effecting in reducing reliability.
their efficiency and performances. Therefore, a platform is
required to keep a steady watch on all plant operations along A. Working of Control Valves
with depending controllers. Since there are many controller As it is well known for process plant consist of hundreds
devices which are assembled to construct a single process of control loop. Each control loop is dedicated to controlling
plant. The most common control used is “control valve”, some important variable in control such as pressure, flow,
which is used to regulate and manipulate the process. A level, temperature, etc. So to keep all these processes go
control valve is amassed by following parts, such as valve smoothly without getting any failure control valve are used
body, valve seat and valve plug. As per some research, 30% to control the output.
of loops in process plant get suffer due to valve non-linearity To decrease the impact of these load disturbance, sensors
problems, which are stiction, hysteresis, and deadband [1]. and transmitters gather data about the procedure variable and
A pneumatic control valve are still utilized most broadly its relationship to some ideal set point. A controller at that
in the process industry, because of their lower cost and point forms this data and chooses what must be done to
effortlessness mainatance [2]. Thus, it is required to develop recover the procedure variable to where it ought to be after
a tool that would notify the early stage of failure with a load unsettling influence happens. At the point when all the
visualization of failure on the bathtub curve(Life expectancy estimating, comparing, and figuring are done, some sort of
curve). definite control component must execute the system chosen

978-1-7281-3377-5/19/$31.00 ©2019 IEEE 351

Proc. of the 2019 IEEE International Conference on Signal and Image Processing Applications (IEEE ICSIPA 2019), Malaysia, September 17-19, 2019

Fig. 1: Control Valve arrangements in process plant. Image source (http://in.ari-armaturen.com/in/home.html)

by the controller. Fig.1 shows how the control valve is placed

and used to control the rate of flow in a line.
The control valve typically comprises of the valve body,
the bonnet, the valve seat and valve plug or interior trim
parts, an actuator (a pneumatic, water driven or electrically
controlled gadget that provisions power and movement to
open or close a valve), valve shaft which associates with the
actuator, fixing course of action between the valve stem and
the bonnet and extra valve accessories. Fig. 2 demonstrates
a diagrammatic representation of a single seat two-port
pneumatic valve.
Fig. 3: Valve stiction behavior [9]

III. C ONTROL VALVE FAILURE M ODE

To justify the failure mode of a control valve, there are two

aspects to be considered. The first strategy can be considered
as the method which is discussed and appraised to define
the technique for failure mode. The second approach can
focus on the application of failure mode and classification to
control valves for a specific evaluation [11]. The two most
likely critical faults are:
• The valve stiction and,
Fig. 2: Working of penuamtic control valve. • The valve plug wear
In control valves, the nonlinear phenomenon is the main
B. Stiction Symptoms source to degrade the process and that nonlinearity causes
from friction phenomena. Few formation of failure process
Stiction is a situation which describes a valve stem po- in control valve are such as stick-slick motion, hysteresis
sition response in increasing (overshoot) controlled output (backlash), actuator control saturation, dead band [10]. As
which can happen in flowing three stages [5]. per the research finding, nonlinearity with first-order linear
• The position of the valve remains unchanged while the dynamic model is the most critical situation in formation of
valve input steadily increase, stiction [13].
• The valve stem starts movement abruptly by jumping
from one point to another, IV. M ETHOD
• The valve stem moves smoothly according to increase
in the input signal. The method section will define how the reliability for
Fig 3. shows the typical input-output behavior of a valve control valve can be estimated by applying algorithm and
under stiction. visualizing on bathtub curve.

352
Proc. of the 2019 IEEE International Conference on Signal and Image Processing Applications (IEEE ICSIPA 2019), Malaysia, September 17-19, 2019

A. Failure Data for Control Valve 2) Weibull MLE: : MLE for two-parameter Weibull dis-
This research is based on simulation since data was tribution must be calculated numerically. It can be computed
collected using simulation in MATLAB environment. In this for both singly censored data, the estimate for the shape
research single tank plant was considered. Data set is listed parameter, β , can be derived as follows:
in two parts, first is the rate of flow, and second is a failure in
∑ri=1 ti lnti + (n − r)ts lnts
β β̂
1 1 r
the rate of flow. After that rate of flow was divided into two g(β̂ ) = − − ∑ lnti = 0 (6)
r i=1
∑ri=1 ti + (n − r)ts
β̂ β̂ β̂
parts, first in which gate valve opening was kept till 30% of
the control valve. In the second, gate wall opening was fixed V. R ESULTS
till 90%, This process was done to understand the working
This section will show the results on the prediction of the
behavior for the initial stage and the final stage of control
reliability in the pneumatic control valve and visualizing it.
valve control ability. In the final stage of data collection,
As shown in Fig 4. it represents the failure data simulated
data censoring process was performed to define the failure
from MATLAB SIMULINK. In here it visualize the censored
occurrence in the rate of flow.
data formation at gate valve opening of 30% and 90%.
B. Analysis Model
TABLE I: Estimated values for β and η for 30% valve
In the process to predict and to obtain the reliability,
operation.
we assumed some parameters from data to compute as a
censored data set. TTF (Time to Fail) measurement was satis- Parameters lower value estimated value upper value
factorily predicted by using 2- parameter Weibull distribution
as described in the result section. Since some assumptions β̂ 2.4118 2.564 2.726402
were made depending on data and different kind of sampling η̂ 15.68169 16.19037 16.71555
performed. Now, calculating the Weibull distribution CDF
(cumulative density function), which can be expressed as:
tβ TABLE II: Estimated values for β and ηfor 90% valve
Pr(T ≤ t; η, β ) = 1 − exp[−( )],t > 0 (1)
η operation.
where, β > 0 is a shape parameters and η > 0 is a scale Parameters lower value estimated value upper value
parameter.In particular, if T has a Weibull distribution,then
Y = log(T )˜SEV (µ, σ ), where σ = 1/β . is the scale pa- β̂ 17.77 19.63 21.69
rameter and µ = log(η) is the location parameter. Lets say, η̂ 92.22 92.89 93.57
T has a Weibull distribution, which can be represented by
T ˜W EIB(µ, σ ), where SEV is the smallest extreme value
distribution. [12]. After representation of data, Weibull values for shape
Now the Weibull cdf,pdf and hf can be written as: and size which are β and η were calculated. Table I and
Table II shows the values for β and η focusing for lower
h log(t) − µ i
estimated, main estimated and upper limit estimated values
F(t : µ, σ ) = Φsev , (2)
σ for reliability prediction. As shown in Fig. 5 (a) it visualizes
1 h log(t) − µ i β  t β −1 t the reliability plot for 30% of the gate valve opening. Here it
f (t; µ, σ ) = Φsev = exp[−( )β ] can also be visualized about bathtub curve. In 5(b) Reliability
σt σ η η η
(3) plot shows the rapid fall in a curve, which mean the reliability
1 h t i 1 β β  t β −1 failure chances are higher than in comparison to 5(a). Fig 6
σ −1
h(t; µσ ) = = = ,t > 0. shows the failure rate based on hazard function depended on
σ exp(µ) exp(µ) η η η
(4) time.
where, Φsev is the standardized smallest extreme value cdf. After computing the values, estimation was validated
by using bootstrapping for 1000 samples to calculate the
1) Maximum Likelihood Estimator: Maximum likelihood estimated values of failure rate. The calculated values as
estimator(MLE) is more assuring to use for reliability es- displayed in Table III and IV. To validate the curve fitting
timations of Weibull parameters than other methods such E(T) function was superimposed to get 95% of accuracy,
as Median Rank Regression (MRR) [12]. To find the MLE which can be visualized in Fig 7. Hence, in Table III and IV,
for complete data, the maximum of the following likelihood shows the limits which define the range for availability and
function with respect to the unknown parameter θ1 , ..., θk is maintainability by looking at the stage of lower and upper.
found: Which also shows the reliability analysis of the control valve
L(θ1 , ..., θk ) = Πni=1 f (ti |θ1 , ..., θk ) (5) in different operation limits.

The main objective of using MLE is to estimates the values VI. C ONCLUSION
of θ1 , ..., θk , that will render the likelihood function as large This paper presents the extraction of the failure data
as possible for given values of t1 , ...,tn . by performing simulation for stiction. The system aimed

353
Proc. of the 2019 IEEE International Conference on Signal and Image Processing Applications (IEEE ICSIPA 2019), Malaysia, September 17-19, 2019

(a) Cumulative Frequency Distribution (CDF) of the (b) Cumulative Frequency Distribution (CDF) of the
failure rate upto gate valve opening 30%. failure rate upto gate valve opening 90%.

Fig. 4: Representation of failure data used.

(a) Reliability plot vs time which visualize the (β ) (b) Reliability plot vs time which visualize the (η)
values for 30% operation of control valve. values for 90% operation of control valve.

Fig. 5: Reliability plot for 2 parameter Weibull distribution

(a) Failure rate plot vs time which shows the shape

parameter (β ) values for 30% operation of control (b) Failure rate plot vs time which visualize the (η)
valve. values for 90% operation of control valve.

Fig. 6: Failure rate plot visualization for Weibull 2 parameters (β ) and (η)

TABLE III: MTTF estimated values with using Weibull TABLE IV: MTTF estimated values with using Weibull
distribution distribution

Parameters lower estimated upper Parameters lower estimated upper

MTTF 14.37472 MTTF 90.3966
MTTF (Bootstrap) 13.92188 14.37472 14.86119 MTTF (Bootstrap) 89.7012 90.3921 91.1456
Bootstrap Statistics Original Bias Std. Error Bootstrap Statistics Original Bias Std. Error
14.37472 -0.006359 0.23487 90.3966 0.00103 0.3764
Mean 14.27391 Mean 90.567

354
Proc. of the 2019 IEEE International Conference on Signal and Image Processing Applications (IEEE ICSIPA 2019), Malaysia, September 17-19, 2019

(a) 95% of prediction fitting (E(T)) for 30% operation. (b) 95% of prediction fitting (E(T)) for 90% operation.

Fig. 7: Prediction fitting for 95% accuracy acquired with using Weibull method.

to collect the data and to do reliability analysis. Weibull [6] Horch, Alexander. ”Benchmarking control loops with oscillations and
distribution was used to perform the reliability analysis. stiction.” In Process control performance assessment, pp. 227-257.
Springer, London, 2007.
Specifically, this research divides the data collection at two [7] Rossi, M., and Claudio Scali. ”A comparison of techniques for
major functioning points of control valve. In first situation automatic detection of stiction: simulation and application to industrial
control valve behavior with 30% of gate valve opening data.” Journal of Process Control 15, no. 5 (2005): 505-514.
[8] Jelali, Mohieddine, and Biao Huang, eds. Detection and diagnosis
along with stiction was implemented to recorded failure data. of stiction in control loops: state of the art and advanced methods.
Another, situation was observed at 90% gate valve opening Springer Science & Business Media, 2009.
was considered. In this manner the operation behavior of [9] Choudhury, MAA Shoukat, Mridul Jain, and Sirish L. Shah. ”Stic-
tion?definition, modelling, detection and quantification.” Journal of
pneumatic control valve was understood. Process control 18, no. 3-4 (2008): 232-243.
The results disused about the prediction of the reliability [10] Yame, Joseph J., and Lhoussain El-Bahir. ”Teaching control valve
analysis. However, the results of the Weibull shows reliability friction using modeling and simulation.” age 10 (2005): 1.
[11] Trunzer, Emanuel, Iris Weib, Jens Folmer, Carolin Schrufer, Birgit
prediction. 95% of accuracy fitting was imposed to attaining Vogel-Heuser, Stefan Erben, Stefan Unland, and Christian Vermum.
the best analysis. Hence, the method was able to show the ”Failure mode classification for control valves for supporting data-
reliability prediction, along with indicating the maintenance driven fault detection.” In Industrial Engineering and Engineering
Management (IEEM), 2017 IEEE International Conference on, pp.
required time. Moreover, the failure of the control valve was 2346-2350. IEEE, 2017.
be visualized on the bathtub curve which is the most suitable [12] Meeker, William Q., and Luis A. Escobar. Statistical methods for
cure to define the working-age of instruments. However, reliability data. John Wiley & Sons, 2014.
[13] Mathur, Nirbhay, Vijanth S. Asirvadam, Azrina Abd Aziz, and Rosdi-
research is still required to make understand the reliability azli Ibrahim. ”Visualizing and Predicting Reliability of Control Valves
analysis. Hence, this research will help system engineers to based on Simulation.” In 2018 IEEE Conference on Systems, Process
take correct steps, to avoid unplanned downtime. and Control (ICSPC), pp. 54-59. IEEE, 2018.

ACKNOWLEDGMENT
Author wants thanks Universiti Teknologi PETRONAS
for providing resources. Author also want to thank for
utilizing YUTP(015LC0-125) grant and other co-authors for
providing their support and help in conducting this research.

R EFERENCES
[1] Desborough, Lane. ”Increasing customer value of industrial control
performance monitoring-Honeywell’s experience.” Preprints of CPC
(2001): 153-186.
[2] Brasio, Ana SR, Andrey Romanenko, and Natercia CP Fernandes.
”Modeling, detection and quantification, and compensation of stiction
in control loops: The state of the art.” Industrial & Engineering
Chemistry Research 53, no. 39 (2014): 15020-15040.
[3] Desborough, L. D. ”Control loop economics.” Honeywell proprietary
report (2003).
[4] Di Capaci, Riccardo, and Claudio Scali. ”Review on valve stiction.
Part I: From modeling to smart diagnosis.” Processes 3, no. 2 (2015):
422-451.
[5] Garcia, Octavio Pozo, Alexey Zakharov, and Sirkka-Liisa Jamsa-
Jounela. ”Data and Reliability Characterization Strategy for Automatic
Detection of Valve Stiction in Control Loops.” IEEE Transactions on
Control Systems Technology 25, no. 3 (2017): 769-780.

355
View publication stats