System-Reliability Analysis by Use of

Gaussian Fuzzy Fault Tree: Application

in Arctic Oil and Gas Facilities
Masoud Naseri and Javad Barabady, University of Tromsø – The Arctic University of Norway

Summary system vulnerabilities, develop accident scenarios, and assess the

Reliability analysis has various applications in oil- and gas-pro- safety level in facilities. Through an FTA, one postulates that the
cessing facilities, such as identifying the bottlenecks of the system, system itself has failed in a certain way that will be considered as
quantitative risk assessments, improving system availability and a top event. The occurrence of the top event is further described in
throughput capacity, spare-parts planning, and optimizing mainte- terms of the occurrence or nonoccurrence of other intermediate/
nance strategies. Reliability performance of a system can be de- basic events. Some mathematical methods, such as Boolean al-
scribed as a function of operation time and a series of operating gebra, rare-event approximation, and set theory, are then used to
conditions. For this purpose, a range of reliability data is required, estimate the probability of the top event (i.e., system failure) as a
on the basis of which the reliability function can be modeled. One function of intermediate/basic events (i.e., component/subsystem
of the challenges in reliability analysis of Arctic oil and gas facili- failures) (Bauer et al. 2009; Bedford and Cooke 2001; Vesely et al.
ties is lack of adequate reliability data. The available historical data 1981). Failure probability of a basic event can be estimated by ana-
gathered in normal-climate regions may not be appropriate because lyzing the life data that can be available from operational or field
they do not include the effects of harsh Arctic operating conditions experiences, maintenance reports, reliability tests, historical data,
on equipment performance. and handbooks.
In this study, the expert-judgement process is used as a tool to Because the oil and gas industry has less experience in Arctic re-
modify the mean time to failure of the equipment to include the gions compared with normal-climate regions, adequate life data may
adverse impacts of Arctic climate conditions on equipment per- be sparse. Available data from normal-climate regions may not be
formance. However, various sources of bias and uncertainties are suitable for reliability analysis of Arctic oil and gas facilities because
involved in expert judgements. Fuzzy set theory is used to deal of the considerable differences in operating conditions. The harsh
with such uncertainties and their propagation in both the compo- operating environment in the Arctic is commonly described as ex-
nent- and system-level analyses. For this purpose, a methodology tremely low temperatures, winds, snowdrifts, polar low pressures, at-
is presented to perform a Gaussian fuzzy fault-tree analysis for mospheric and sea-spray icing, sea-ice-induced vibrations, seasonal
system-reliability assessments. This methodology is further illus- darkness, and poor visibility resulting from fog and snowstorms. Be-
trated by a case study. cause the reliability performance of Arctic oil and gas facilities is
adversely affected by such an environment (Barabadi et al. 2013; Na-
Introduction seri and Barabady 2013), the corresponding reliability assessments
Reliability is defined as “the ability of an item to perform a re- must be performed in accordance with the adverse effects of harsh
quired function under stated conditions for a stated period of time” operating conditions on the equipment and operation performance.
(ISO 8402:1994). The term “ability” can be expressed quantita- Proportional hazard models (Barabadi and Markeset 2011; Gao
tively with probability, referring to the chance or likelihood that et al. 2010) and accelerated life models (Barabadi 2014) are ap-
an item will perform its intended function. The term “stated condi- plied to include the effects of Arctic operating conditions on com-
tions” emphasizes that an item may perform its intended functions ponent-reliability performance. However, such models rely on an
adequately under one set of conditions (operational environment) extensive range of detailed data that may not be available, particu-
and quite poorly under another set (Stapelberg 2009). larly in the Arctic regions. Thus, the expert-judgement process can
There are several tools, such as reliability block diagram (Gao be applied as an alternative method to cope with this shortcoming
et al. 2010), Markov models (Malefaki et al. 2014), fault-tree models and account for such effects. Expert judgements represent the ex-
(Wang et al. 2013; Yuhua and Datao 2005), and Monte Carlo sim- perts’ state of knowledge regarding a technical question at the time
ulation (Zio et al. 2007), that can be used to describe the system of response. Such judgements are expressions of opinion that are
reliability mathematically as a function of the reliability perfor- based on knowledge and experience. Expert judgment is not re-
mance of its components (Bauer et al. 2009). Fault-tree analysis stricted to the experts’ answer, but includes the experts’ mental pro-
(FTA) is a deductive system analysis that has been used extensively cesses of assumptions, definitions, and algorithms, through which
in quantitative risk assessments and prediction of system-failure the answers are formulated (Ortiz et al. 1991). The concept of expert
probability in various fields, such as nuclear power plants (Purba judgement has been applied in a variety of fields, including nuclear
2014), chemical process plants (Wang et al. 2013), oil and gas in- engineering, meteorological research, aerospace, seismic and envi-
dustry (Yuhua and Datao 2005), and aerospace (Phillips and Diston ronmental risk, and risk and safety analysis of oil and gas operations
2011). An FTA provides a comprehensive and structured approach (Clemen and Winkler 1999; Moon and Kang 1999; Purba 2014). In
to estimate system-failure probability, identify and understand key most of these studies, experts are mainly asked to provide a qualita-
tive idea about the frequency of occurrence of an event. Such quali-
tative words are then converted to fuzzy linguistic variables and,
Copyright © 2015 Society of Petroleum Engineers
consequently, to fuzzy numbers for further quantitative assessments.
This paper (SPE 170826) was accepted for presentation at the SPE Annual Technical In this study, a methodology is presented for system-reliability
Conference and Exhibition, Amsterdam, 27–29 October 2014, and revised for publication.
Original manuscript received for review 9 July 2014. Revised manuscript received for review
assessment on the basis of fuzzy FTA, which is applicable to Arctic
11 January 2015. Paper peer approved 29 January 2015. oil and gas operations. More specifically, the expert-judgement

June 2015  •   Oil and Gas Facilities 85

June 2015 • Oil and Gas Facilities 1
process is used to modify available life data gathered in normal- (a) (b)
climate regions to include the effects of Arctic operating condi- ~ ~
µX(x) µX(x)
tions on reliability performance of components and systems. Fuzzy
set theory is used to deal with the uncertainties involved in ex- 1 1
~ ~
pert judgements. For this purpose, the exact values of mean time X X
to failures are combined with the subjective opinions of the ex- 1
perts, which are converted to Gaussian fuzzy numbers. To develop α
√e

a model for system-reliability analysis and to analyze the corre-

sponding uncertainty propagation, a fuzzified form of FTA is de-
veloped. The proposed methodology is illustrated by a case study 0 0
consisting of a three-phase horizontal separator and its surrounding ~ x x x
Cutα (X )
valves. The remainder of this paper is organized as follows: a short σL σR
introduction to fuzzy set theory and Gaussian fuzzy numbers is pre-
sented. The methodology for performing the Gaussian fuzzy FTA Fig. 1—(a) A typical Gaussian fuzzy number X and (b) its α-cut
is then described. Conclusions are presented after illustrating the set.
methodology by a case study.

Fuzzy Set Theory exp  − ( x − x )2 / ( 2 2 )  , for x < x

  M L  M
In classical set theory, a set is defined as a collection of objects  X ( x ) =  . ..................(1)
exp  − ( x − x M ) / ( 2 R )  , for x ≥ x M
 2 
2
or elements out of a universal set that share common properties

or characteristics. In that regard, an element receives a member-
ship degree of unity, if and only if that element belongs to the set; Because the element xM (known as the mean value of X)  has a mem-
otherwise the membership degree will be zero (i.e., the element bership grade of unity, the fuzzy number X can be considered as the
does not belong to the set). For example, considering the clas- fuzzified form of the crisp number xM. Fuzzy numbers can be de-
sical set concept, if the failure probability of a component before scribed effectively with the important concept of the α-cut set, which
a certain time is 2%, its membership grade is unity and all other facilitates the fuzzy arithmetic operations. The α-cut set of the fuzzy
failure-probability values have a membership degree of zero be- number X is defined as a crisp set of elements x, with membership
cause they are not included in the set F = {2%}. However, the clas- grades being greater than or equal to some threshold  ∈( 0,1). The
sical set may reach its limit where the property that distinguishes α-cut set of X is mathematically expressed as X =  x ∈   X ( x ) ≥  
the members from nonmembers is ambiguous and vague because
of, for instance, some uncertainties. While in classical set theory, (Dubois and Prade 1980). A Gaussian fuzzy number X can also be
a sharp, crisp, and unambiguous boundary distinguishes the mem- defined using its crisp α-cut set X = ( x L , x R ) , x L ≤ x R , where xαL
bers and nonmembers for any well-defined set of entities, fuzzy and xαR are given as x L , R =  X−1 ( ) (Hanss 2005). Figs. 1a and 1b
set theory accepts partial memberships. On the basis of fuzzy set show a typical Gaussian fuzzy number and its crisp α-cut set, respec-
theory, which was introduced by Zadeh (1965), it is allowed to tively. The method of determining the α-cut set of a Gaussian fuzzy
have an element that at the same time belongs to a set and does not number is given in Appendix A.
belong to that set. The degree at which the element belongs to the
set is assigned by a membership function (Chen and Pham 2001; Extension Principle and Fuzzy-Number Arithmetic. By use of
Dubois and Prade 1980). The higher the membership grade, the the extension principle introduced by Zadeh (1965), the domain
more the element belongs to the set. Considering the aforemen- and range of an ordinary function from ordinary sets are extended
tioned example, to account for the uncertainties, one may also con- to fuzzy sets. The extension principle provides a general method
sider the failure probability values of 1.88 or 2.02% as members for extending the crisp or nonfuzzy mathematical concepts to deal
of F = {2%}, but with a membership degree of less than unity, for with fuzzy numbers and fuzzy functions (Dubois and Prade 1980).
instance 0.95. This principle can be used when one wants to fuzzify a function or
Expert opinions can be formulated with the fuzzy set theory. to include the associated uncertainties in the function parameters,
The impacts of an Arctic operating environment on equipment re- and thus evaluate the uncertainty propagation. In this study, the ex-
liability performance may vary based on a number of factors, in- tension principle is used to fuzzify the expert opinions, combine
cluding equipment type, equipment function, equipment location on them, and then form the reliability or failure-probability function
the platform, and the severity of the weather conditions. In this re- for the system and its components.
gard, exact quantification of all such impacts in the form of single- The extension principle is defined as follows (Zadeh 1965):
point values is not feasible. Thus, experts prefer to present their Suppose that G is an ordinary function G :  n   that maps an
opinions by use of a range or quantiles of a distribution to reflect the element (x1, x2, …, xn) to the element y = G(x1, x2, …, xn). Addi-
uncertainties associated with their ideas. Then, fuzzy set theory can tionally, let X i ⊆  be a fuzzy set defined by a membership func-
be used to combine the expert opinions and their associated uncer- tion  X i ( xi ) , xi ∈ , i = 1, 2, ..., n. Then, using the concept of α-cut
tainties, and finally, to present the failure probability in fuzzy form. set, the membership function Y ( y ) , y ∈  of fuzzy set Y ⊆  with
Mathematically, a fuzzy number X is a convex normalized fuzzy ( )
Y = G X 1 , X 2 , ..., X n is defined as
{ }
set of the real line , which is defined as X =  x,  X ( x )  , x ∈  ,
where  X ( x ) ∈[ 0,1] is the membership grade of the element x in (
Y = G X1, , X 2, , ..., X n , )
X . The membership function  X ( x ) is piecewise continuous, and =  y ∈  y = G ( x1 , x2 , ..., xn ) , x1 ∈ Xi , . ..........................(2)
there is exactly one x M ∈, where  A ( x M ) = 1 (Dubois and Prade
1980). There are various types of fuzzy numbers, some of which
are of particular interest because of the specific behavior of their Gaussian Fuzzy Fault-Tree Analysis (FTA)—Methodology
membership functions such as triangular, trapezoidal, Gaussian, This study proposes a methodology consisting of two phases.
exponential, and quadratic. A Gaussian fuzzy number X over a During Phase I, the system-reliability model is developed for the
universal set S is defined as a fuzzy number whose membership base area, where the life data are available for the system and its
function is characterized by a normalized and, in general, asym- components. The operating conditions in the base area are consid-
metrically parameterized Gaussian function, given as (Hanss 2005) ered normal. During Phase II, the developed model in Phase I is

2
86 Oil and
Oil and Gas
Gas Facilities  
Facilities •   June 2015
• June 2015
 Define the system and its
external boundary
Step I-1: System
identification and
description Define the components/sub-
systems and internal boundaries

Gather failure data corresponding to - Handbooks

Step I-2: base operating environment - Maintenance
Component-level reports
- Reliability
analysis
Estimate failure probability of the tests
components/subsystems - Historical data

Identify the functions of the system, subsystems

and components, and their interactions

Step I-3: Fault-tree

construction Identify the top event

Construct the fault-tree diagram

Develop the fault-tree model - Boolean algebra

Step I-4: Estimation and rare-event
of system-failure approximation
probability Estimate the failure - Set theory and
probability of the system minimal cut-set

Fig. 2—Procedure for Phase I: system reliability modeling for the base area.

modified by expert judgements and fuzzy set theory to develop the t

F ( t ) = P (T ≤ t ) = ∫ f ( x ) dx, ..................................................(3)
system-reliability model for the target area (i.e., an Arctic region), 0
where the life data are sparse or not available. The aim of this phase
is to include the subjective opinions of experts about the poten- where f(x) is the probability-density function of the continuous
+∞
tial impacts of the harsh Arctic operating environment on reliability random variable T, such that ∫ f ( x ) dx = 1 (Verma et al. 2010).
performance of the system and its components. 0
Assuming a one-parameter exponential distribution, the proba-
Phase I: System-Reliability Modeling for the Base Area. The bility-density function can be written as
aim of Phase I is to perform an FTA for the base area to estimate
the system reliability as a function of each component’s reliability f ( t ) =  exp ( − t ), ..................................................................(4)
or failure probability. For this purpose, a set of steps must be fol-
lowed, as illustrated by Fig. 2. where λ is the failure rate of the component. For an exponential
Step I-1: System Identification and Description. A system is probability-density function, failure rate is constant and is defined
defined as a “set of interrelated elements considered in a defined as the inverse of mean time to failure (MTTF) (i.e., λ = 1/MTTF).
context as a whole and separated from their environment” (IEC Substituting Eq. 4 into Eq. 3 gives a description for component-
60050-151, 2001). The elements of a system may also be broken failure probability F(t) as
down to subsystems and components. Internal boundaries are used
to establish a limit of resolution and to determine in how much  t 
F ( t ) = 1 − exp  − . ......................................................(5)
detail one should study the system. Additionally, to decide what  MTTF 
factors could influence the system function, or to determine what
aspects of the system performance are of concern, one needs to es- Mathematically, reliability is expressed as the probability that a
tablish the external boundaries of the system in question (Bedford component fails at a time greater than or equal to a specified time t
and Cooke 2001; Vesely et al. 1981). [i.e., R(t) = P(T ≥ t) = 1 – F(t)]. In some contexts, reliability func-
Step I-2: Component-Level Analysis. The aim of this step is to tion is known as survival function.
develop the reliability or failure-probability function for each com- Step I-3: Fault-Tree Construction. A fault-tree model is
ponent. To perform this task, a set of life data is required, to which simply described as a graphic model of the various parallel and
a theoretical distribution is fitted. Such detailed data can be ac- sequential combinations of faults that will result in the occurrence
quired from handbooks, maintenance reports, reliability tests, and of the top event. The faults can be associated with hardware fail-
historical data. Let T be a random variable representing the time to ures, human errors, or any other relevant event that can lead to
failure of a component. The probability that the component fails the top or an intermediate event (Vesely et al. 1981). To develop
before time t is called failure probability or unreliability. As per the a fault tree, a number of symbols are generally used to describe
probability terminology, failure probability F(t) is the same as the events and their combinations. The primary events of a fault tree
cumulative distribution function of the random variable T, which are those that have not been further developed. Such events can
is given by be categorized into basic event, undeveloped event, conditioning

June 2015 • Oil and Gas Facilities 3

June 2015  
June 2015 •   Oil and
• Oil and Gas
Gas Facilities
Facilities 3
87
 Step II-1: Expert Selection
OR AND
Step II-2: Expert-opinion elicitation

Step II-3: Fuzzification of expert opinions

A B A B

Fig. 3—OR-gate and AND-gate illustration of two basic events Step II-4: Aggregation of fuzzified expert opinions
A and B.

event, and external event. A basic or initiating event, represented Step II-5: Developing fuzzy failure-probability
by a circle, requires no further development and thus signifies the function for components/subsystems
appropriate limit of resolution. In addition to events, a fault tree
consists of various logic gates, of which AND-gate and OR-gate Step II-6: Performing fuzzy FTA to estimate the
are illustrated in Fig. 3 as two basic types of fault-tree gates. An system-failure probability
AND-gate shows that the output fault occurs only if all the input
faults occur, while an OR-gate shows that the output fault occurs Fig. 4—Procedure for Phase II: system reliability modeling for
only if at least one of the input faults occurs (Bedford and Cooke the target area.
2001; Vesely et al. 1981).
Step I-4: Estimation of System Failure Probability. There term “expert.” An expert can be defined as “a person who has a
are different methods to estimate the probability of the top event background in the subject matter at the desired level of detail and
of a fault tree. The rare-event approximation uses the concept of who is recognized by his/her peers or those conducting the study as
Boolean algebra and its associated rules. To have a more-precise being qualified to solve the questions” (Meyer and Booker 1991).
estimation, one can use the simplified Boolean expression of the O’Hagan et al. (2006) state simply that “an expert may, in principle,
fault tree to determine the possible minimal cut-sets. A minimal just mean the person whose judgements are to be elicited.” Selec-
cut-set is the smallest combination of basic events, which, if they tion of experts is a major issue in expert-judgement studies. This is
all occur, will result in the top-event occurrence; and if one of the because the term expert is open to different interpretations because
failures in the cut set does not occur, then the top event will not experts are defined by subjective expressions such as “having a
occur by that combination (Vesely et al. 1981). Set theory and its desired level of detailed background,” “being recognized by their
associated rules can be further applied to the specified minimal cut- peers,” and “being qualified.” Additionally, on the one hand, it is
set. The required formulas to estimate the probability of an OR-gate advantageous to select a group of experts with wide background,
or AND-gate are given by (Verma et al. 2010; Vesely et al. 1981): but on the other hand, the analyst may be under pressure to exclude
some of the experts who are perceived as being less experienced
F ( A and B ) = F ( A ∩ B ) = F ( A ) F ( B ).....................................(6) (Bedford and Cooke 2001).
Step II-2: Expert-Opinion Elicitation. “Elicitation” is de-
and fined as the process of obtaining the subjective opinions of ex-
perts through specifically designed methods of communication,
F ( A or B ) = F ( A ∪ B ) = F ( A ) + F ( A ) − F ( A ) F ( B ), ............(7) such as surveys, interviews, group meetings, and questionnaires
(Meyer and Booker 1991). Elicitation may be performed in quali-
where F(A) and F(B) are failure probabilities of A and B, respec- tative or quantitative forms. In quantitative form, experts are asked
tively. One can also describe the AND- and OR-gates by use of the to express their subjective opinions about a parameter in the form
reliability terminology. of, for instance, a single-point or distribution estimation, an abso-
FTA is recommended in this study because it is a powerful tool lute rating, an interval scaling, and a ratio scaling (Cooke 1991;
in complex-system analysis, especially where dependent failures or Svenson 1989).
operational loops are present. The other advantage of constructing a In this study, experts are asked to provide their opinions on the
fault tree is to determine the minimal cut-sets by applying Boolean al- degree of decrease in MTTF of various components operating in
gebra concepts. Identifying the minimal cut-sets helps to understand the target area as a fraction of the corresponding MTTF in the base
and identify all possible system-failure scenarios. Once the minimal area. To include the uncertainty in expert judgements, experts are
cut-sets are determined, one may perform the system failure or reli- required to express their opinions in the form of 5, 50, and 95%
ability analysis with the reliability block-diagram concept because, in quantiles (i.e., eliciting the quantiles of a distribution). Uncertain-
some cases, it may reduce the number of calculations considerably. ties and biases are important concepts that one needs to take into
account while performing the elicitation step. Several studies list
Phase II: System-Reliability Modeling for the Target Area. The three categories of biases that can be introduced to the study at
aim of Phase II is to include the adverse effects of the Arctic oper- the elicitation step, including structural biases, motivational biases,
ating environment on the system-reliability performance. For this and cognitive biases (Benson and Nichols 1982; Meyer and Booker
purpose, the FTA performed during Phase I needs to be modified 1991; Ortiz et al. 1991; Otway and Winterfeldt 1992).
with the subjective opinions of experts that are aggregated with the Structural biases occur when experts are influenced by the way
concept of fuzzy set theory. A number of experts are asked to give in which a problem is formulated or by the level of detail in the
their opinions on the degree of decrease in MTTF of the compo- study specified by the analyst. For instance, asking for quantiles of
nents. Such opinions will be converted to Gaussian fuzzy numbers a distribution may lead to different judgements if the distribution
and will be further combined with appropriate aggregation methods parameters were elicited.
on the basis of the extension principle. More specifically, a set of Motivational biases occur when the experts may benefit from
steps must be followed, as illustrated by Fig. 4. the results of the study, or when experts express their opinions to
Step II-1: Expert Selection. Expert selection is the first step in please the interviewer or analyst. For example, if the aim of a study
the expert-judgment process, and refers to choosing an appropriate is to show and quantify the differences between the equipment-reli-
number of reliable experts. Various definitions are available for the ability performance in Arctic- and normal-climate regions, experts

4 Oil and Gas Facilities • June 2015

88 Oil and Gas Facilities  •   June 2015
may bias their opinions by giving a wide distribution on the amount malized nonnegative weight for Expert i. However, to develop a
of decrease in equipment MTTF operating in the Arctic, which is in fuzzy relation for these combination rules, the extension principle
favor of the study goal. Alternatively, some other experts working is applied to already fuzzified expert opinions (see Appendix A).
in the equipment-design field (e.g., manufacturers) may argue that Various methods are available to assign a weight for each expert,
the MTTF of an equipment unit operating in the Arctic is statisti- such as assigning equal weights; asking experts to weight them-
cally the same as one operating in normal-climate regions. selves; ranking experts in a specific preference, and then assigning
Cognitive biases are expressed in various ways, including over- weights proportional to ranks; determining weights on the basis of
confidence, anchoring, and availability. Overconfidence occurs the elicited data; using proper scoring rules; and calibrating experts
when an expert has a tendency to be more precise about their prob- on the basis of their performance (Cooke 1991). In this study, the
ability estimates, which may consequently result in presenting equal and experience-based weighting approaches are used.
distributions that are too tight. Anchoring occurs when an expert Step II-5: Developing Fuzzy Failure-Probability Function for
anchors to an original estimate that is generally as defensible as Components/Subsystems. Substituting Eq. 8 into Eq. 5 gives the
possible. Such judgements may be formed on the basis of a known failure-probability function of a component operating in the target
disaster or failure scenario. Availability refers to a cognitive bias in area as
which the frequency of events that are easily imagined or recalled
are likely to be overestimated, while more-common-failure sce-  t 
narios, with less-significant consequences, can be underestimated. F ( t ) = 1 − exp  − . ...............................................(11)
These biases can be described for the experts to help them to re-  (1 −  ) T j , B 
duce the level of such biases and their resulting uncertainties. Al-
ternatively, the analyst may present the questions in such a way as The fuzzified failure-probability function is in fact an extension
to reduce the structural biases. More-detailed discussion regarding of the failure-probability function from ordinary sets into fuzzy
the biases and how to deal with them is presented in Meyer and sets. In other words, the failure probability of a component until a
Booker (1991). certain time is no longer a crisp value, but a fuzzy number that in-
Step II-3: Fuzzification of Expert Opinions. Let T j , B represent cludes the uncertainties caused by the vague and complex effects
the MTTF of component j in the base area. Expert i provides his or of operating conditions on the component reliability performance
her subjective opinion on parameter δ, which is the degree of de- (i.e., parameter δ, the quantiles of which are given by experts), and
crease in T j , B in the form of [δi,L, δi,M, δi,R]. Parameter δ is given as those uncertainties caused by expert judgements. To develop the
a fraction of T j , B using 5, 50, and 95% quantiles, respectively, given fuzzified failure-probability function of a component, one needs
by δi,L, δi,M, and δi,R. In this regard, the MTTF of component j in the to use the fuzzified form of aggregated expert opinions (i.e., Step
target area T j ,T is defined as II-3). The detailed description of fuzzy failure probability and its
membership function is given in Appendix A.
Step II-6: Performing Fuzzy FTA To Estimate the System-
T j ,T = (1 −  ) T j , B. .....................................................................(8)
Failure Probability: The fuzzy FTA can be carried out by substi-
tuting the fuzzy failure probability of each component into the fuzzy
However, because there is usually more than one expert involved in AND-gate and OR-gate. The corresponding membership functions
the studies, expert opinions need to be combined to form a solution can be determined by applying the extension principle and α-cut
for the analyst or decision maker. Such a solution will be further set concept (see Appendix A). While the failure probability of the
used to modify the MTTF data gathered in the base area. To com- top event is a crisp value in classical FTA, in fuzzy FTA, the failure
bine the elicited expert opinions while including the associated un- probability of the top event is a fuzzy number that assigns a mem-
certainties, one can describe the quantiles given by each expert in bership grade for different values of failure probability.
{
the form of a Gaussian fuzzy number ∆ i = i ,  ∆ i (i )  , i ∈( 0,1) . } Case Study
The membership function  ∆ i (i ) is expressed by use of Eq. 1. The
To illustrate the proposed methodology a three-phase, horizontal
quantiles given by the experts are used to determine the parameters
oil and gas separator system is chosen as a case to estimate its reli-
σL and σR in Eq. 1. The detailed procedure to determine such pa-
ability performance under Arctic operating conditions.
rameters and to finally fuzzify the quantiles given by the experts is
presented in Appendix A.
Illustration of Phase I. Step I-1: System Identification and
Step II-4: Aggregation of Fuzzified Expert Opinions. Aggrega-
Description. Fig. 5a shows a typical three-phase gravity-type
tion of expert opinions refers to the procedure by which the expert
horizontal separator. Wellstream enters the separator, where the
judgements are combined by the analyst to provide a basis for the
associated gas and water cut are separated from the oil phase. Gas
decision maker. Axiom-based approaches are mathematical aggre-
and water leave the separator vessel through pressure-control and
gation methods that are based mainly on the linear and logarithmic
water-dump valves, respectively. The separated oil is then routed
opinion-pool principles, of which the former refers to the weighted
to the next-stage separator through the oil-dump valve (Arnold and
linear combination and the latter refers to the weighted geometric
Stewart 2008). Additionally, according to API SPEC 12J (1989),
combination (Bedford and Cooke 2001; Clemen and Winkler 1999;
some pressure-relief valves are required to be installed on the ves-
Cooke 1991). Eqs. 9 and 10, respectively, give the weighted linear
sel. On the basis of these descriptions, the simplified separator
and geometric combination rules in crisp form:
system, its components, and its external boundary are shown in
N Fig. 5b.
 = ∑ wii ................................................................................(9) Internal boundaries are chosen in such a way that each valve will
i =1 be considered a single component without further analysis of its in-
ternal sections and components, such as actuator, control and moni-
and toring devices, seals, seat rings, and valve body. The internal boundary
N
for the separator is also defined in such a way that the vessel and all
 = ∏ iwi , .............................................................................(10) of its internal sections, level- and pressure-control devices, and moni-
i =1 toring instruments are considered as a single component.
Step I-2: Component-Level Analysis. To perform the com-
where δi is the opinion of Expert i, δ is the combined expert opin- ponent-level analysis, the mean-time-to-failure (MTTF) data for
ions in crisp form, N is total number of experts, and wi is the nor- each component are obtained from the Offshore Reliability Data

June 2015  ••   Oil

June 2015 and Gas
Oil and Gas Facilities
Facilities 5
89
 (a) (b)
Gas compression
PC Wellstream High-pressure flare
Gas outlet High-pressure flare
Gravity settling section Mist extractor LV2
PV PS
Inlet diverter Pressure-control
valve
Inlet 1st stage separator (SEP)
LC

Oil and Emulsion LV1

LC
Water Oil 2nd-stage
separator

Water out Oil out

LV3

Level-control
valve Water-treatment system

Fig. 5—(a) Schematic of a three-phase separator with interface level control and weir; (b) system external boundary of a separator
and its valves.

(OREDA) handbook (OREDA Participants 2009), which includes Step I-4: Estimation of System-Failure Probability. By ap-
the failure rate and mean time to repair of a wide range of equip- plying the concept of minimal cut-sets, the failure-probability func-
ment installed on oil and gas production facilities on the Norwe- tion of the system can be modelled as
gian continental shelf, except the Barents Sea. There are various
types of failure data reported in the OREDA handbook, such as FSystem = F ( S ∪ LV 1 ∪ LV 2 ∪ LV 3 ∪ PS ∪ PV ). ...................(12)
degraded and critical, of which the critical failures “cause imme-
diate and complete loss of an equipment unit’s capability of pro- Performing further simplification with Eq. 7, and substituting
viding its output.” The degraded failures “are not critical, but they the corresponding failure probability of each component from
prevent an equipment unit from providing its output within speci- Table 1, the system-failure probability at time t0 = 2,160 hours
fications” (OREDA Participants 2009). Therefore, in this study, the would be 33.50%, which provides a reliability of 66.50%. Fig.
MTTF data are those related to both the critical and degraded fail- 7 depicts the reliability of the separator system and its compo-
ures. However, one can also consider only the critical failures in nents (SEP, LV1, LV2, LV3, PS, and PV) as a function of opera-
the analyses. The probability of failure for each component is esti- tion time. Because the failure of these components is linked to
mated by use of Eq. 5 at a reference time of t0 = 2,160 hours (ap- system failure by means of an OR-gate, the high failure rate of the
proximately 3 months). Component reliability is also estimated for separator (11 to 20 times more than the failure rate of valves, as
each component as a function of time, as well as at the reference presented in Table 1) has a major negative effect on the system-
time. Table 1 presents the list of components, their identifications, reliability performance. Some measures, such as condition moni-
MTTF data, and the reliability and failure probability at the refer- toring, preventive-maintenance actions, and adding redundancy to
ence time. It is assumed that all level-control valves have identical the system, may be taken into consideration to keep the system re-
failure rates. This assumption stands for the pressure-relief and liability above a desired level.
pressure-control valves.
Step I-3: Fault-Tree Construction. To construct the fault-tree Illustration of Phase II. Step II-1: Expert Selection. In this study,
diagram, it is assumed that if any of the described valves or sep- experts are chosen on the basis of the criteria suggested by Ortiz
arators fail, then the entire separation process fails until the cor- et al. (1991). In this regard, experts collectively should represent a
rective-maintenance tasks restore the failed components to their wide variety of backgrounds and experience. Referring to the pub-
functioning state. On the basis of this assumption, the fault-tree lications of experts and their direct involvement in or consulting
diagram can be constructed as shown in Fig. 6. and managing of research in the related areas could also be a help-

F t R t

Table 1—List of components, their identification, MTTF, and reliability at t0=2,160 hours.

6 Oil and Gas Facilities • June 2015

90 Oil and Gas Facilities  •   June 2015
 Separation 100
system fails 90
80

Reliability (%)
70
OR 60
50
40 System
30 SEP
20 LV1, LV2, LV3
10 PS, PV
SEP LV1 LV2 LV3 PV PS 0
0 200 400 600 800 1,000 1,200 1,400 1,600 1,800 2,000 2,200
Operation Time (hours)
Fig. 6—Fault-tree diagram of the separator and its valves.
Fig. 7—Reliability of the separator system and its components.
ful basis for expert selection. In this study, the key point in expert
experience, which is 40 years. Equal weighting (wi = 0.1667) is
selection is that they must have adequate understanding of Arctic
another approach used in this study, which is one divided by the
operating conditions and their potential effects on the performance
number of experts.
of mechanical systems. Considering these criteria, six experts are
Step II-3: Fuzzification of Expert Opinions. The next step is
chosen from Norwegian academic and industrial sectors, each with
to fuzzify elicited expert opinions, through which the elicited de-
appropriate knowledge of the operating conditions in the North
gree of decrease in MTTF [δi,L, δi,M, δi,R] is converted to a Gaussian
Sea (i.e., the base area) and the Barents Sea (i.e., the target area).
fuzzy number ∆ i . The method to determine the corresponding mem-
Selected experts have expertise in maintenance and reliability en-
bership function for each expert’s data is described in Appendix A.
gineering, process engineering, mechanical engineering, and cold-
The fuzzified expert opinions on parameter δ for both the separator
climate engineering, with an experience ranging from 7 to 40 years.
and valves are plotted in Figs. 8a and 8b, respectively. As can be
The authors selected these experts on the basis of the available
seen, according to the 5, 50, and 95% quantiles of the data given
resources at the time of the study. Additionally, because the primary
by Expert 3, the MTTF of the separator in the target area may de-
goal of this study is to propose the methodology and highlight the
crease by (50, 62.5, 75%). However, on the basis of the member-
necessity of including Arctic operating conditions in system-relia-
ship function of the corresponding fuzzy number, the reduction in
bility assessments, the authors decided to perform the analyses on
MTTF may be as high as 81%, or as low as 43.9%, but with a mem-
the basis of the opinions of these selected experts, and to not exclude
bership degree of 0.05.
the experts with the working experience of 7 and 9 years. However,
Step II-4: Aggregation of Fuzzified Expert Opinions. The fuzz-
those two experts will receive a lower weight compared with those
ified expert opinions are then combined by use of Eqs. 9 and 10 ac-
having more working experience. The results of this study may not
cording to the weights assigned for each expert. For this purpose,
be considered universal because they depend on the expert-selection
four approaches are used in this study, as presented in Table 3.
criteria, number of experts, and varying operating conditions. There-
Fig. 9 shows the combined fuzzy expert opinions by use of these
fore, other studies may develop their own expert-selection schemes
four approaches. For instance, if one chooses arithmetic averaging
and achieve different results. A detailed discussion on selection and
and determines each expert’s weight on the basis of their working
motivating experts is presented by Meyer and Booker (1991).
experience (i.e., Approach II), the MTTF of the separator in the
Step II-2: Expert-Opinion Elicitation. At the next step, a ques-
target area will be 35.2% smaller than in the base area. However,
tionnaire is prepared during which experts are informed about the
this degree of decrease is obtained if one chooses the membership
operating environment in the base and target areas, as well as the
grade of unity. The reduction in MTTF of the separator would be as
MTTF of valves and three-phase horizontal separator. Experts are
large as 51%, but with a membership grade of 0.05.
then asked to provide their subjective opinions on the degree of
Step II-5: Developing Fuzzy Failure-Probability Function for
decrease in such MTTFs, considering that the described equip-
the Components/Subsystems. Having the changes in the MTTF of
ment is planned to operate in the target area. The Johan Castberg
the separator and valves estimated, one can predict the reliability
field, which is located 230 km north of the Norwegian coast in
or failure probability of those components as a function of time.
the Barents Sea, is selected as the target area. The operating con-
For example, Fig. 10a illustrates the separator reliability at t0 =
ditions in this region are much more severe compared with the
2,160 hours in both the base and target areas. As shown in this
southern regions of the Norwegian continental shelf, such as the
figure, after 2,160 hours, separator reliability reduces to 74.17% in
North Sea. The questionnaire used in this study for the expert-elic-
the base area. This reduction is considerably larger in the target area
itation step is presented in Appendix B. Table 2 presents the elic-
because of the adverse effects of Arctic operating conditions on
ited expert opinions, their working experience in years, and both
separator performance. Additionally, as can be seen, four different
the nonnormalized and normalized experience-based weights for
approaches estimate different reductions in separator reliability, of
the experts that will be further used for combining expert opinions.
which the greatest and the least reductions are estimated by Ap-
Nonnormalized weights are determined by dividing the working
proaches II and IV, respectively. By use of the same procedure, the
experience of each expert (in years) by the maximum working

i i,, i,M, i,R

Table 2—Elicited expert opinions and experts’ experience-based weights.

June 2015 • Oil and Gas Facilities 7

June 2015  •   Oil and Gas Facilities 91
 1 1

Degree of Membership (µ)

Degree of Membership (µ)

Experts 1 & 4 Experts 1, 3

0.8 Experts 2 & 3 0.8 Expert 2

Experts 5 & 6 Experts 4, 6
Expert 5
0.6 0.6

0.4 0.4

0.2 (a) 0.2 (b)

0 0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
δ (fraction) δ (fraction)

Fig. 8—Fuzzified expert data on the amount of decrease in MTTF of (a) the separator and (b) the valve, in general.

system reliability in the base and target areas on the basis of four
approaches and taking a membership degree of unity. Among the
introduced approaches, Approaches II and III estimate the greatest
and the least reduction in system reliability, respectively. However,
it must be noted that these results are not universal because they
may differ on the basis of various expert opinions, varying oper-
ating conditions, operation location, system production rate, type
Table 3—Approaches used to combine fuzzified expert opinions. of wellstream, and available inspection and maintenance activities.

Conclusion
reliability of pressure- and level-control valves can be described in This study focuses on the application of expert judgement in reli-
fuzzy form, as well. ability prediction of oil and gas topside facilities in Arctic regions,
where adequate life data may not be available. The proposed meth-
Step II-6: Performing Fuzzy Fault-Tree Analysis (FTA) To Es- odology can be used in the design phase for oil and gas operations
timate the System-Failure Probability. Once component reliabili- in the Arctic. The estimated reduction in reliability performance of
ties (or failure probabilities) are estimated, system performance can the equipment can be used for optimizing maintenance and spare-
be analyzed by use of the fuzzified form of Eq. 12 (see Appendix parts provision plans. Moreover, this methodology provides a basis
A). Fig. 10b shows the fuzzy form of system reliability at time t0 for deciding the measures of winterization that need to be applied
= 2,160 hours. As can be seen, the system reliability in the base if the system reliability is below the acceptable level. Additionally,
area is 66.50%. Taking into account the effects of operating condi- as illustrated in this study, fuzzy set theory can be used to aggregate
tions in the target area on system performance, this reliability con- expert opinions, while including and modeling the uncertainties
tinues to decrease, as illustrated by Approaches I through IV. For and their propagation in system analysis. The estimation made by
instance, considering Approach II and membership grades of unity the presented methodology may need to be further modified when-
and 0.05, system reliability reduces to 53.58 and 43.53%, respec- ever new historical or laboratory life data are available.
tively. Choosing a specific membership grade and its corresponding The life data for reliability analysis in the base area are obtained
system reliability depends on the aim of the study, requirements and from the OREDA handbook (OREDA Participants 2009), which
regulations, and the risk perception of the decision maker. However, provides only the constant failure rate of the equipment. Thus, ex-
according to the fuzzy set theory, the higher the membership grade, ponential distribution is the only applicable probability-density
the more the element belongs to the set. One may also defuzzify the function to estimate component reliabilities. Use of detailed life
results to obtain the corresponding crisp values. data from maintenance reports can lead to more-dependable reli-
In addition to the system reliability at a specified time, the pro- ability analysis.
posed methodology is also applicable to developing the system The number of selected experts is also a determining factor in
reliability as a function of operation time. Fig. 11 illustrates the the reliability of the results. Such experts must have adequate un-

1 1
Approach I Approach I
Approach II Approach II
Degree of Membership (µ)

Degree of Membership (µ)

0.8 Approach III 0.8 Approach III

Approach IV Approach IV

0.6 0.6

0.4 0.4

0.2 (a) 0.2 (b)

0 0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
δ (fraction) δ (fraction)

Fig. 9—Combined fuzzy expert opinions on (a) MTTF of SEP and (b) on MTTF of LV1, LV2, LV3, PS, and PV.

8 Oil and Gas Facilities • June 2015

92 Oil and Gas Facilities  •   June 2015
 1 1
Approach I Approach I
Approach II Approach II
Degree of Membership (µ)

Degree of Membership (µ)

0.8 Approach III 0.8 Approach III
Approach IV Approach IV
Base Area Base Area
0.6 0.6

0.4 0.4

0.2 (a) 0.2 (b)

0 0
30 40 50 60 70 80 90 100 30 40 50 60 70 80 90 100
Reliability (%) Reliability (%)

Fig. 10—Fuzzy reliability of (a) the separator and (b) the system in the base and target areas at t0 = 2,160 hours.

100 δi = degree of decrease in MTTF of a component elicited from

Expert i; element of the fuzzy number ∆ i
90 δi,L = 5% quantile of the parameter δ elicited from Expert i
δi,M = 50% quantile of the parameter δ elicited from Expert i
Reliability (%)

80 δi,R = 95% quantile of the parameter δ elicited from Expert i

δi,α = lower bound of the α-cut set of Gaussian fuzzy
L
70 Approach I
number ∆ i
Approach II δi,α = upper bound of the α-cut set of Gaussian fuzzy
R
60
Approach III number ∆ i
Approach IV δj = combined crisp expert opinion on the reduction in MTTF
of component j; element of the fuzzy number ∆ j
Base Area
50
0 200 400 600 800 1,000 1,200 1,400 1,600 1,800 2,000 2,200 δM = mean value of the combined fuzzified expert opinions
Operation Time (hours) ∆ = combined fuzzified expert opinions
Δα = α-cut set of Gaussian fuzzy number ∆
Fig. 11—System reliability in the base and target areas as a func-
∆ i = Gaussian fuzzy number representing the quantiles given
tion of operation time.
by Expert i
∆ i, = α-cut set of Gaussian fuzzy number ∆ i
derstanding of the failure mechanisms of various components and ∆ j = combined fuzzified expert opinions on the reduction in
the effects of Arctic operating conditions on such mechanisms. The MTTF of component j
use of equal weights for experts may not seem realistic because Δj,α = α-cut set of Gaussian fuzzy number ∆ j
one may argue various experts have different levels of expertise λ = failure rate
and thus should receive different weighting factors. In this regard,  X ( x ) = membership function of Gaussian fuzzy number X
other weighting methods that can calibrate expert opinions may σL = left-side standard deviation of a Gaussian fuzzy number
be more useful. Additionally, within the elicitation step, questions σR = right-side standard deviation of a Gaussian fuzzy number
may be asked in different forms to ensure the consistency of expert  = fuzzified failure probability of component j

opinions. Although one needs to establish a series of criteria for 
j

expert selection, expert selection remains a major issue in any ex-  j, = α-cut set of Gaussian fuzzy number  j

pert-judgement process because understanding and quantifying the

competence level of each expert are quite challenging tasks.

Nomenclature Acknowledgments
f(x) = probability-density function The authors would like to thank all the anonymous experts for their
F(x) = failure probability or unreliability function contribution to this study.
Fj = failure probability of component j
N = total number of experts
R(x) = reliability function References
T j , B = mean time to failure (MTTF) of component j in the base API SPEC 12J—Specification for Oil and Gas Separators. 1989. Wash-
area ington, D.C.: American Petroleum Institute (API).
T j ,T = MTTF of component j in the target area Arnold, K. and Stewart, M. 2008. Surface Production Operations: Design
wi = normalized weight for Expert i of Oil Handling Systems and Facilities, third edition, Vol. 1. Burl-
x = element of Gaussian fuzzy number X ington, Massachusetts: Surface Production Operations, Gulf Profes-
xM = mean value of Gaussian fuzzy number X sional Publishing.
X = Gaussian fuzzy number Barabadi, A. 2014. Reliability Analysis of Offshore Production Facilities
Xα = α-cut set of Gaussian fuzzy number X under Arctic Conditions Using Reliability Data from Other Areas. J
XαL = lower bound of the α-cut set of Gaussian fuzzy number X Offshore Mech Arct Eng 136 (2): 020901. OMAE-13-1043. http://
XαR = upper bound of the α-cut set of Gaussian fuzzy number X dx.doi.org/10.1115/1.4026240.
δ = combined expert opinions in crisp form; element of the Barabadi, A. and Markeset, T. 2011. Reliability and Maintainability Perfor-
Gaussian fuzzy number ∆ mance under Arctic Conditions. International Journal of System As-

June 2015 • Oil and Gas Facilities 9

June 2015  •   Oil and Gas Facilities 93
 surance Engineering and Management 2 (3): 205–217. http://dx.doi. Purba, J.H. 2014. A Fuzzy-Based Reliability Approach to Evaluate Basic
org/10.1007/s13198-011-0071-8. Events of Fault Tree Analysis for Nuclear Power Plant Probabilistic
Barabadi, A., Naseri, M., and Ratnayake, R.M.C. 2013. Design for the Safety Assessment. Ann Nucl Energy 70 (August 2014): 21–29. http://
Arctic Conditions: Safety and Performance Issues. Proc., 32nd In- dx.doi.org/10.1016/j.anucene.2014.02.022.
ternational Conference on Offshore Mechanics and Arctic Engi- Stapelberg, R.F. 2009. Handbook of Reliability, Availability, Maintain-
neering, Nantes, France, 6–14 June, OMAE2013-10287, http://dx.doi. ability and Safety in Engineering Design, first edition. London:
org/10.1115/OMAE2013-10287. Springer-Verlag.
Bauer, E., Zhang, X., and Kimber, D.A. 2009. Practical System Reliability, Svenson, O. 1989. On Expert Judgements in Safety Analyses in the Pro-
first edition. Hoboken, New Jersey: John Wiley & Sons. cess Industries. Reliab Eng Syst Saf 25 (3): 219–256. http://dx.doi.
Bedford, T. and Cooke, R. 2001. Probabilistic Risk Analysis: Foundations org/10.1016/0951-8320(89)90095-1.
and Methods, second edition. Cambridge, UK: Cambridge Univer- Verma, A.K., Ajit, S., and Karanki, D.R. 2010. Reliability and Safety En-
sity Press. gineering, first edition. London: Springer Series in Reliability Engi-
Benson, P.G. and Nichols, M.L. 1982. An Investigation of Motivational neering, Springer-Verlag.
Bias in Subjective Predictive Probability Distributions. Decision Sci- Vesely, W.E., Goldberg, F.F., Roberts, N.H. et al. 1981. Fault Tree Hand-
ences 13 (2): 225–239. http://dx.doi.org/10.1111/j.1540-5915.1982. book, first edition. Washington, D.C.: U.S. Nuclear Regulatory Com-
tb00145.x. mission.
Chen, G. and Pham, T.T. 2001. Introduction to Fuzzy Sets, Fuzzy Logic, Wang, D., Zhang, P., and Chen, L. 2013. Fuzzy Fault Tree Analysis for Fire
and Fuzzy Control Systems, first edition. Boca Raton, Florida: CRC and Explosion of Crude Oil Tanks. J Loss Prev Process Ind 26 (6):
Press LLC. 1390–1398. http://dx.doi.org/10.1016/j.jlp.2013.08.022.
Clemen, R.T. and Winkler, R.L. 1999. Combining Probability Distributions Yuhua, D. and Datao, Y. 2005. Estimation of Failure Probability of Oil
from Experts in Risk Analysis. Risk Anal 19 (2): 187–203. http:// and Gas Transmission Pipelines by Fuzzy Fault Tree Analysis. J
dx.doi.org/10.1023/A:1006917509560. Loss Prev Process Ind 18 (2): 83–88. http://dx.doi.org/10.1016/j.
Cooke, R.M. 1991. Experts in Uncertainty: Opinion and Subjective jlp.2004.12.003.
Probability in Science, first edition. New York: Oxford Univer- Zadeh, L.A. 1965. Fuzzy Sets. Information and Control 8 (3): 338–355.
sity Press. http://dx.doi.org/10.1016/S0019-9958(65)90241-X.
Dubois, D. and Prade, H. 1980. Fuzzy Sets and Systems: Theory and Appli- Zio, E., Marella, M., and Podofillini, L. 2007. A Monte Carlo Simulation
cations, first edition, Vol. 144. New York: Mathematics in Science and Approach to the Availability Assessment of Multi-State Systems with
Engineering, Academic Press, Inc. Operational Dependencies. Reliab Eng Syst Saf 92 (7): 871–882.
Gao, X., Barabady, J., and Markeset, T. 2010. An Approach for Prediction http://dx.doi.org/10.1016/j.ress.2006.04.024.
of Petroleum Production Facility Performance Considering Arctic In-
fluence Factors. Reliab Eng Syst Saf 95 (8): 837–846. http://dx.doi.
org/10.1016/j.ress.2010.03.011.
Hanss, M. 2005. Applied Fuzzy Arithmetic: An Introduction with Engi- Appendix A
neering Applications, first edition. Stuttgart, Germany: Springer Gaussian Fuzzy Form of δ. Let the value of parameter δ be given
Berlin Heidelberg. by Expert i in the form of [δi,L, δi,M, δi,R], where δi,L, δi,M, and δi,R are
IEC 60050-151, 2001: International Electrotechnical Vocabulary - Part 5, 50, and 95% quantiles, respectively. To fuzzify the parameter δ
151: Electrical and magnetic devices. Geneva, Switzerland: Interna- with a Gaussian fuzzy number, one needs to determine the standard
tional Electrotechnical Commission. deviations of the normal distributions corresponding to the elicited
ISO 8402:1994—Quality Management and Quality Assurance—Vocabu- quantiles. This can be achieved by solving the cumulative density
lary. 1994. Geneva, Switzerland: International Organization for Stan- function of a normal distribution for σi,L and σi,R, as given by
dardization (ISO).
Malefaki, S., Limnios, N., and Dersin, P. 2014. Reliability of Maintained
Systems under a Semi-Markov Setting. Reliab Eng Syst Saf 131 (No-
vember 2014): 282–290. http://dx.doi.org/10.1016/j.ress.2014.05.003.  ( )
 i , L = i , L − i , M /  2erf −1 ( −0.9 )  , for  < i , M
 
 . ............ (A-1)
Meyer, M.A. and Booker, J.M. 1991. Eliciting and Analyzing Expert Judge-
ment - A Practical Guide, first edition, Vol. 5. London: Knowledge-
( )
 i , R = i , R − i , M /  2erf ( +0.9 )  , for  ≥ i , M
 −1

Based Systems, Academic Press.

Moon, J.H. and Kang, C.S. 1999. Use of Fuzzy Set Theory in the Aggrega-
tion of Expert Judgments. Ann Nucl Energy 26 (6): 461–469. http:// Having σi,L and σi,R determined, the Gaussian membership func-
dx.doi.org/10.1016/S0306-4549(98)00073-5.
Naseri, M. and Barabady, J. 2013. Offshore Drilling Activities in Barents tained by use of Eq. 1, as
{ }
tion of the fuzzy number ∆ i = i ,  ∆ i (i )  , i ∈( 0,1) can be ob-

Sea: Challenges and Considerations. Proc., 22nd International Con-

ference on Port and Ocean Engineering under Arctic Conditions,
Espoo, Finland, 9–13 June,
(
exp  −  −  ) / ( 2 ) , for  < 
2 2
O’Hagan, A., Buck, C.E., Daneshkhah, A. et al. 2006. Uncertain Judge-   i,M i,L i,M
 ∆ i ( ) =  . ........... (A-2)
ments: Eliciting Experts’ Probabilities, first edition. West Sussex,
(
exp  −  − i , M ) / ( 2 ) , for  ≥ 
2 2
England: Statistics in Practics, John Wiley & Sons, Ltd.   i,R i,M

Ortiz, N.R., Wheeler, T.A., Breeding, R.J. et al. 1991. Use of Expert Judg-
ment in NUREG-1150. Nucl Eng Des 126 (3): 313–331. http://dx.doi.
org/10.1016/0029-5493(91)90023-B. To write the Gaussian membership function with the α-cut set
Otway, H. and Winterfeldt, D. 1992. Expert Judgment in Risk Analysis ( )
concept ∆ i , = i ,L , i ,R , one can equate the membership function
and Management: Process, Context, and Pitfalls. Risk Anal 12 (1):  ∆ i ( ) with α, and solve the resulting equation for δ, as given by
83–93. http://dx.doi.org/10.1111/j.1539-6924.1992.tb01310.x.
OREDA Participants. 2009. OREDA (Offshore Reliability Data) Handbook,
fifth edition. Trondheim, Norway: OREDA Participants.
 =  − −2 2 ln 
Phillips, P. and Diston, D. 2011. A Knowledge Driven Approach to Aero-  i ,L i,M i,L
space Condition Monitoring. Knowledge-Based Systems 24 (6): 915–  . ................................................ (A-3)
i ,R = i , M + −2 i , R ln 
2
927. http://dx.doi.org/10.1016/j.knosys.2011.04.008.

10
94 Oil and
Oil and Gas
Gas Facilities  
Facilities •   June 2015
• June 2015
 Table B-1—The questionnaire used for eliciting expert opinions.

To combine the fuzzified expert data by use of the arithmetic    

and geometric averaging rules, the following equations can be  t 
=  Fj ∈( 0,1) Fj = 1 − exp  −  ,  j ∈∆ j ,  . ........ (A-9)
used, respectively:   1 −  j(T j,B 
 ) 
N To perform the fault-tree analysis on the basis of the fuzzy set
∆ = ∑ wi ∆ i ......................................................................... (A-4) theory, the failure probability of each component must be fuzzified
i =1
with Eqs. A-8 and A-9. The top-event probability of the fault tree
and will be further estimated by fuzzifying the AND-gate and OR-gate
probabilities, as given by
N
∆ = ∏ ∆ iwi , ........................................................................ (A-5) 
  
AND-Gate = 1 2 ............................................................... (A-10)
i =1

and
where ∆ i is fuzzified opinion of Expert i, ∆ is the combined ex-
pert opinions in fuzzy form, N is the total number of experts,     
and wi is the normalized nonnegative weight for Expert i. The  OR-Gate = 1 +  2 − 1 2, .............................................. (A-11)

membership function of the combined fuzzified expert data

{ }
∆ =  ,  ∆ ( )  ,  ∈( 0,1) is then obtained with the extension prin- where  , j = 1, 2 is the fuzzy failure probability of component j.
j
The membership functions of   
ciple and α-cut set concept, given by AND-Gate and  OR-Gate can be obtained
by use of the extension principle and the α-cut set concept:
N
 N

∆  = ∑ wi ∆ i , =  ∈( 0,1)  = ∑ wii , i ∈∆ i ,  ................ (A-6)  AND-Gate, = 1,  2,
i =1  i =1 
=  F ∈( 0,1) F = F1F2 , F1 ∈ 1, , F2 ∈  2, ....... (A-12)
and

 OR-Gate, = 1, +  2, − 1,  2,

N
 N

∆  = ∏ ∆ iw,i =  ∈( 0,1)  = ∏ iwi , i ∈∆ i ,  , ................. (A-7)
i =1  i =1  =  F ∈( 0,1) F = F1 + F2 − F1F2 , F1 ∈ 1, , F2 ∈ 2, 

where Δi,α is the α-cut set of the fuzzy number ∆ i . .......................................... (A-13)

Fuzzified Failure Probability and Fuzzy FTA. Once the fuzzi-

fied expert opinions on the amount of decrease in the mean time to Appendix B
failure of component j are combined, the failure-probability func- See Table B-1 for the questionnaire used to elicit expert opinions.
tion of that component can be expressed in fuzzy form by

 t  Masoud Naseri is a PhD-degree student in the Department of Safety

 = 1 − exp  −
 , ............................................ (A-8)
j
 (
 1 − ∆ j T j , B ) and Engineering at the University of Tromsø—The Arctic University of
Norway. His research interests include cold-climate engineering, risk
and safety analysis of Arctic offshore operations, operation and mainte-
nance, and analytical and simulation methods in reliability, availability,
where its membership function is obtained by use of the extension
and maintainability analysis of oil and gas facilities. Naseri has au-
principle and the α-cut set concept as
thored or coauthored eight technical papers presented in international
conferences or published in journals. He holds BSc and MSc degrees
 t  in petroleum engineering from the Petroleum University of Technology
 j , = 1 − exp  − 
( )
 1 − ∆ j , T j , B 
and Sahand University of Technology, respectively. Naseri is a member
of SPE.

June 2015 • Oil and Gas Facilities 11

June 2015  
June 2015 •   Oil and
• Oil and Gas
Gas Facilities
Facilities 11
95
Javad Barabady is a professor of Safety and Technology in the Department coauthored 29 technical papers presented in international conferences or
of Safety and Engineering at the University of Tromsø—The Arctic Univer- published in journals. Barabady holds a PhD degree in operation and main-
sity of Norway. He has worked for more than 5 years as a lecturer at Azad tenance engineering from Luleå University of Technology, Luleå, Sweden; a
University of Shahrood, Iran. Barabady’s research interests include reliability master’s degree in mining engineering from the University of Tehran, Iran;
and risk analysis, operation and maintenance engineering and planning, and a bachelor’s degree in mining engineering form Azad University of
and production-assurance analysis and management. He has authored or Shahrood, Iran.

12
96 Oil and
Oil and Gas
Gas Facilities  
Facilities •   June 2015
• June 2015