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

net/publication/332673134

Fault Tree Analysis for Control Valves in Process Plants by using R

Conference Paper · March 2019

DOI: 10.1109/CSPA.2019.8696008

CITATIONS READS
2 294

4 authors, including:

Nirbhay Mathur Rosdiazli Ibrahim

Universiti Teknologi PETRONAS Universiti Teknologi PETRONAS
13 PUBLICATIONS   15 CITATIONS    150 PUBLICATIONS   732 CITATIONS   

SEE PROFILE SEE PROFILE

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

Fractional-order controllers for Process Plants View project

Vector Born, GIS, Dengue View project

All content following this page was uploaded by Nirbhay Mathur on 19 July 2019.

The user has requested enhancement of the downloaded file.

 2019 IEEE 15th International Colloquium on Signal Processing & its Applications (CSPA 2019), 8 -9 March 2019, Penang, Malaysia

Fault Tree Analysis for Control Valves in Process

Plants by using R
Nirbhay Mathur Vijanth S. Asirvadam
Department of Electric and Electronic Department Department of Electric and Electronic Department
Universiti Technologi PETRONAS Universiti Technologi PETRONAS
Tronoh, Perak, Malaysia Tronoh, Perak, Malaysia
Email: [email protected]

Azrina Abd Aziz Rosdiazli Ibrahim

Department of Electric and Electronic Department Department of Electric and Electronic Department
Universiti Technologi PETRONAS Universiti Technologi PETRONAS
Tronoh, Perak, Malaysia Tronoh, Perak, Malaysia

Abstract—An industrial plants assembled with many compo- Since, when there is talk about autonomous and digital-
nents, standby systems, sensors, controlling components to make ization of system, it directly comes in mind about Big Data
a smooth operation of plants. Losses produced by unplanned handling. R statistical software is a platform where problems
downtime may lead to major problems. Although, control valves
which are used to control the rate of flow for pressure or liquid is for Big Data, prediction algorithms, and visualization can be
an essential part of the plant operation. Since to reduce the losses done at a very great extent. This paper focuses on a qualitative
or downtime damage, high reliability and standby availability are analysis technique “Event Tree Analysis ”by using R statistical
required to monitor. Hence, it requires some prediction which can software.
express the likelihood of the events in the terms of quantitative
methods. This research performs the prediction for reliability on II. M ETHODOLOGY & I MPLEMENTING R
the control instruments and its availability that is visualized for
a quantitative perspective and estimation of the consequences A. Fault Tree Analysis
of these failure depends on the control valve assembled in a Fault-tree analysis is one of the most usual approach to
process plant. This research will give practical assessments of
the reliability of a control valve for better actions and higher define failure or to analysis the failure of system. Below
reliable redundancy. By using R statistical software and packages mentioned is the flow chart process for constructing Fault
this research will be more vibrant and will be visualized to a very Tree Analysis:
extent level. Package “FaultTree ”will be used in this paper to
demonstrate and to explore the use of R software and it packages
usages in many manners. This paper will demonstrate how Fault
Tree Analysis can be done in process plants and how R statistical Hazard Classification
software holds importance in representing it.
Kewwords− Reliability, Fault Tree Analysis, Valve Stiction, R
software.

I. INTRODUCTION
In present era of revolution in industries, Industry 4.0 (I4.0) Identification of necessary
is the latest trend and the aim of this era is to convert all hazard cause factors
industry process autonomous. It is required to upgrade and
validate the old process with the latest and current technolo-
gies. In a process plant, which is engraved with hundred’s
of loops attached to each other and each loop handles the
individual process. So, it becomes very challenging to predict
the reliability of the process plant which is dependent on Localisation of Hazard causes
another control instrument. The control valve is the common
actuators for the process regulation and manipulation. Accord-
ing to the study, about 30% of all control loop suffers from
valve nonlinearity problems, such as stiction, hysteresis and
deadband [1]. Although failure cant be neglected in any way,
System Failure Tree construction
every engineered object is unreliable as it will fail sooner or
later [2].

978-1-5386-7563-2/19/$31.00 ©2019 IEEE 152

 2019 IEEE 15th International Colloquium on Signal Processing & its Applications (CSPA 2019), 8 -9 March 2019, Penang, Malaysia
Operating principal of control valve:

Fig. 2. Single seat two-port pneumatic globe valve

As shown in Fig 2, the fluid flow in pushing against valve

plug and tending to keep the plug off the valve seat. Due
to the movement of fluid, the difference in pressure (P1) and
downstream (P2) of the valve can be notified with a difference
(4P). Since due to this mechanism, valve seat movement is
some time increased or decreased to control flow which cause
Fig. 1. Symbols used in FTA and their meanings [5] stiction and due to that failure in operation can be noticed.
Implementing R statistical software for FTA:
R is a language and environment for statistical computing
In Stage 1: Hazards are notified which involves general
and graphics. R provides a wide variety of prediction and
hazard or harm to human hazard, breaking hazards to other
graphical techniques based on a wide range of available
subclasses and then breaking these subclasses to another
packages [6]. To conduct this research, package FaultTree is
subclass until hazard can be identified and which can be
used to design fault tree analysis. This package has many
localized to specific points in the plant.
functions such as
Stage 2: This stage identifies the factors that cause failure,
ftree.make which is the top event of failure occurrence.
event or combination of events that can cause a hazard.
Stage 3: It localizes factors places within the plant.
Stage 4: It connects all events and sub-events in form of a 1 #To to satrt the top event
2 tree1 <- ftree.make(type="or",reversible_cond=TRUE, name="
tree to specify the hazard and failure of an operation. Control Valve Fail")

It is important to understand the significance of fault tree Listing 1. Step 1

and its nature of work. The fault tree is not the best suitable As shown in step 1 type is to define gate type,
system for all system failures but performs best with some reversible cond = T RUE to define the condition of gate
applications. Fault tree construction is started from top event ”True” or ”False” and name to enter the event failure name.
or first failure which corresponds to system failure.
As shown in Fig 1, the primary event symbols which are
used to construct Fault Tree, which is categorized on different 12 #Defining atributes or types of failure
tree1 <- addLogic(tree1, at=1, type="and", name="Fails to
priorities and representation and also represents the Gate open fully")
symbols for the output of fault occurrence [3].
Listing 2. Step 2
Now, moving on to the next section which is a case study on
which Fault Tree Analysis (FTA) will be implemented using the following code in step 2 is used to define the divisions or
R statistical software. categorization for the failure types, this can be followed by
numbers of types to construct the tree.
B. Case Study: Control Valves
A control valve consists of following parts such as valve
1 #Defining atributes or types of failure
body, bonnet, the valve plug, and valve seat, an actuator, valve 2 tree1 <- addLatent(tree1, at=2, mttf=25.817,mttr=0.0387,
spindle which connects to the actuators, sealing arrangement inspect=1/16, name="Corrosion")

between the valve stem and the bonnet and additional valve Listing 3. Step 3
accessories as shown in Fig. 1 [4].

153
 2019 IEEE 15th International Colloquium on Signal Processing & its Applications (CSPA 2019), 8 -9 March 2019, Penang, Malaysia

Fig. 3. Visualization of failure events and Fault Tree Analysis of control valve using FaultTree package

addLatent modifies an existing fault tree with the addition display under Optionally, the ID of a sibling event under an
of a latent component event. The usage of addLatent and OR gate for vertical alignment of the component node in the
arguments of the functions are as shown in step 4: graphic display.
tag A very short identifying string (typically 5 characters or
1 addLatent(DF, at, mttf, mttr=NULL, pzero=NULL, inspect=NULL less) uniquely identifying a basic event for minimal cutset
, evaluation.
2 display_under=NULL, tag="", label="",
3 name="", name2="", description="") label An identifying string for the logic gate. Use of label in
ftree.make defines the convention for rest of tree construction.
Listing 4. Step 4 name A short identifying string (typically less than 24
DF A fault tree dataframe such as returned from ftree.make characters).
or related add... functions. name2 A second line, if needed for the identifying string
at The ID of the parent node for this addition. label.
mttf The mean time to failure. It is the user’s responsibility description An optional string providing more detail for the
to maintain constant units of time. basic event.
mttr The mean time to repair (restore). This should only after putting all the values to the specific path of tree and
be provided when system remains at risk, while repair is in node, tree can be displayed by using command as shown in
progress, as it modifies probability by factor ”pzero”. It is step 5:
the user’s responsibility to maintain constant units of time.
pzero A depreciated argument. Any value provided will be 1 ## The tree can be displayed in the browser using the
silently ignored. The pzero value is calculated from mttr following commands:
when provided. 2 ftree2html(tree1, write_file=TRUE)
3 browseURL(’tree1.html’)
inspect The time interval between inspections for the dormant
component. (It will be possible upon future development to Listing 5. Step 5
enter the string for the named inspection dataframe object.
In this case the inspection object will be read to get the The Fault Tree Analysis can be viewed as shown in Fig. 3,
inspection interval. An inspection object must be a dataframe which contain enough information to generate a meaningful
with interval and duration columns. If a positive duration graphics display. The Fault Tree event calculates for the fail
value other than zero is found it is taken that the system is at rate which indicates the performance of equipment can be
risk during the time of inspection. An unavailable probability effected for mentioned cause of failure. Along with that repair
calculated as duration/(interval+duration) will be added to the time for each cause can be calculated as shown in lower event
fractional downtime. of the Fault Tree. This information can be useful for system

154
 2019 IEEE 15th International Colloquium on Signal Processing & its Applications (CSPA 2019), 8 -9 March 2019, Penang, Malaysia
engineers or maintenance team to avoid downtime or complete The position error can be denoted by
failure of the plant.
After looking at tree, columns of dataframe can be filtered e = i−h (1)
or viewed by following step 6
where i is the normalized input signal (0 ≤ i ≤ 1) and h is
the normalized valve travel (0 ≤ h ≤ 1), then the following
1 tree1 <- ftree.calc(tree1) equation describes the time-delay
2
3 tree1[,1:8] 
max(e, 0) for z < −zmax
Listing 6. Step 6 dz 
= min(e, 0) for z > zmax (2)
dt 
and the output of these code can be represented in step 7, 
e for |z| ≤ zmax
where the variable z can be viewed as a relative pressure (it
1 > ftree2html(tree1, write_file=TRUE) is the time-delay state variable). From above equation, it state
2 ID Parent CFR PBF CRT MOE Cond Label Name
3 1 -1 -1.0000 -1.0000 -1.0000 0 0 that when absolute valve of variable z is below a certain limit
Control Valve Fail value zmax the valve stem does not move.
4 2 1 -1.0000 -1.0000 -1.0000 0 0 Fails
to open fully
5 3 1 -1.0000 -1.0000 -1.0000 0 0 Fails to Deadband and hysteresis
open partially
6 4 1 -1.0000 -1.0000 -1.0000 0 0 Fails
to close fully Dead band is caused by mechanical backlash and static
7 5 2 0.03873 0.00270 0.03870 0 0
Corrosion friction. Backlash is lost motion due to slop in valve linkage.
8 6 2 0.00478 0.00050 0.00478 0 0 Dirt/ Since, the control error can be computed as
Debris trapped
9 7 4 0.00478 0.00028 0.00478 0 0 seat (
10 8 4 0.00478 0.00022 0.00478 0 0 line e − eedead sign(e) for|e| > edead
presure too high e= (3)
11 9 4 0.00478 0.00017 0.00478 0 0 0 for|e| ≤ edead
Mechanical
12 10 3 0.20000 0.00410 0.00136 0 0 where edead is a half from the measured dead band.
Damage

Listing 7. Step 7 Dynamics

A numerous things can be reviewed from available information
In here dynamics is defined by a first-order system with a
to predict and to visualize the reliability of control valve and
velocity limitation as
modes of its failure.
( e
dh min(| |,Vmax )sign(e) quand|z| > zmax
III. CONCLUSIONS = τ (4)
dt 0 quad|z| ≤ zmax
In this research paper, a brief review is made over appli-
cation of Fault Tree Analysis by using R statistical software.
where, h is the valve stem position,τ is the time constant
It was shown how R software can be used in creating Fault
and vmax is the maximum relative speed. Need to note that
Tree and what kind of functions are available for “Fault-
velocity is set zero when the time-delay state variable z is
Tree”package. The visualization of the tree was plotted to
inside the ‘delay-time zone’.
understand the failure modes in a control valve and to predict
the MTTF, MTTR. A nice concept can be developed by using
R software between reliability prediction and visualization the ACKNOWLEDGMENT
failure modes. Since the results show enough evidence for the Author wants thanks Institute of Autonomous System
use of R software and its application. and Universiti Teknologi PETRONAS for providing re-
sources. Author also want to thank for utilizing FRGS grant
APPENDIX
(FRGS/0153AB-L27) and other co-authors for providing their
Stiction Model to get failure data: support and help in conducting this research.
A valve stiction model was simulated in MATLAB to create
stiction effect to visualize the plant operation. Based on the R EFERENCES
most critical nonlinearities, a second-order linear dynamic [1] Desborough, Lane. ”Increasing customer value of industrial control
model can be added with nonlinearities. performance monitoring-Honeywell’s experience.” Preprints of CPC
(2001): 153-186.
[2] Jelali, Mohieddine. ”Estimation of valve stiction in control loops using
Friction and delay separable least-squares and global search algorithms.” Journal of Process
Control 18, no. 7-8 (2008): 632-642.
[3] Vesely, William E., Francine F. Goldberg, Norman H. Roberts, and
The time -delay cannot be represented as it is done David F. Haasl. Fault tree handbook. No. NUREG-0492. Nuclear
classically because it depends on the position control error. Regulatory Commission Washington dc, 1981.

155
 2019 IEEE 15th International Colloquium on Signal Processing & its Applications (CSPA 2019), 8 -9 March 2019, Penang, Malaysia
[4] Prasad, M. Hari, G. Rami Reddy, A. Srividya, and A. K. Verma.
”Reliability prediction of control valves through mechanistic models.”
Journal of Mechanical Engineering Research 2, no. 3 (2010): 52-57.
[5] Lee, Wen-Shing, Doris L. Grosh, Frank A. Tillman, and Chang H.
Lie. ”Fault Tree Analysis, Methods, and Applications A Review.” IEEE
transactions on reliability 34, no. 3 (1985): 194-203.
[6] https://www.r-project.org/about.html.

156

View publication stats