Please download to get full document.

View again

of 14

A hybrid model for human factor analysis in process accidents: FBN-HFACS

A hybrid model for human factor analysis in process accidents: FBN-HFACS
0 views14 pages
All materials on our website are shared by users. If you have any questions about copyright issues, please report us to resolve them. We are always happy to assist you.
  Contents lists available at ScienceDirect Journal of Loss Prevention in the Process Industries  journal homepage: A hybrid model for human factor analysis in process accidents: FBN-HFACS Esmaeil Zarei a,b, ∗ , Mohammad Yazdi c, ∗∗ , Rouzbeh Abbassi d , Faisal Khan e a  Department of Occupational Health and Safety Engineering, Faculty of Health, Mashhad University of Medical Sciences, Mashhad, Iran b  Social Determinants of Health Research Center, Mashhad University of Medical Sciences, Mashhad, Iran c Centre for Marine Technology and Ocean Engineering (CENTEC), Instituto Superior Tecnico, Universidade de Lisboa, 1049-001, Lisbon, Portugal d  School of Engineering, Faculty of Science and Engineering, Macquarie University, Sydney, NSW, Australia e Centre for Risk, Integrity, and Safety Engineering (C-RISE), Memorial University, St. John's, NL, Canada A R T I C L E I N F O  Keywords: Process industriesHFACSHuman reliability assessmentFuzzy AHPBayesian network A B S T R A C T Human factors are the largest contributing factors to unsafe operation of the chemical process systems.Conventional methods of human factor assessment are often static, unable to deal with data and model un-certainty, and to consider independencies among failure modes. To overcome the above limitations, this paperpresents a hybrid dynamic human factor model considering Human Factor Analysis and Classification System(HFACS), intuitionistic fuzzy set theory, and Bayesian network. The model is tested on accident scenarios whichhave occurred in a hot tapping operation of a natural gas pipeline. The results demonstrate that poor occupa-tional safety training, failure to implement risk management principles, and ignoring reporting unsafe conditionswere the factors that contributed most failures causing accident. The potential risk-based safety measures forpreventing similar accidents are discussed. The application of the model confirms its robustness in estimatingimpact rate (degree) of human factor induced failures, consideration of the conditional dependency, and adynamic and flexible modelling structure. 1. Introduction Human factors (HFs) are widely known to be the main causes of themajority of accidents in different industries. Many studies have quan-titatively pointed out the pivotal role of these failures in accident oc-currence; over 90% in nuclear accidents, over 80% in chemical processaccidents, 75–96% in marine casualties, over 70% in aviation accidentsin the European Union (Nivolianitou et al., 2006), 75–96% of occupa-tional casualties in the United States (Wang et al., 2011), and more than94% of accidents in oil and gas refineries in developing countries suchas Iran (Zarei et al., 2016, 2013). Therefore, prevention of HFs plays a major role in decreasing accidents and consequently, in providing safeworkplaces (Noroozi et al., 2013, 2014). Management of human error is increasingly needing attention to reduce the risk related to productionloss, asset damage, environmental pollution, and fatality in variousindustries (Abbassi et al., 2015).Pervasiveness of the HFs in accidents guarantees a requirement toinclude these failures in accident investigations in order to preventfuture similar accidents. Effective prevention of accidents needs theincorporation of HFs into accident analysis models (Wang et al., 2011).There are various methods which have been broadly used to discuss HFsin accidents. Reason's Swiss Cheese model, Systems-Theoretic AccidentModel and Process, Classification of Socio-Technical Systems (Rao,2007; Wang et al., 2011), System Dynamics, AcciMap, Human Factors Analysis and Classification (HFAC), and STAMP (Salmon et al., 2012)are examples of these models. Among the developed models, HFACS isthe most popular and is specifically designed for investigating the HOF'scontribution to an accident. Salmon et al. (2012) suggest that theHFACS approach is more reliable than other methods due its taxonomicnature and it is also more useful in multiple case study analyses. Interms of application, HFACS has been widely used in different industrialsectors such as maritime (Akyuz et al., 2016; Akyuz and Celik, 2014), civil construction (Garrett and Teizer, 2009; Hale et al., 2012; Wong et al., 2016; Xia et al., 2018), aerospace (Liu et al., 2013), mining (Patterson and Shappell, 2010), nuclear (Yoon et al., 2017, 2016), and chemical processes (Theophilus et al., 2018; Xie and Guo, 2018). The validation and reliability of HFACS in the human errors model is de-monstrated by Øien (2001) and it is shown to be widely used for the 15 August 2018; Received in revised form 22 November 2018; Accepted 22 November 2018 ∗ Corresponding author. Postal address: No.18, University Street, Department of Occupational Health and Safety Engineering, Mashhad University of MedicalSciences, Mashhad, Iran. ∗∗ Corresponding author. Centre for Marine Technology and Ocean Engineering (CENTEC), Instituto Superior Tecnico, Universidade de Lisboa, 1049-001, Lisbon,Portugal.  E-mail addresses:, (E. Zarei), (M. Yazdi). Journal of Loss Prevention in the Process Industries 57 (2019) 142–155Available online 27 November 20180950-4230/ © 2018 Elsevier Ltd. All rights reserved.    purpose of accident investigation (Olsen and Shorrock, 2010; Shappell et al., 2007; Wiegmann and Shappell, 2001). HFACS models are de- veloped as an appropriate tool in order to prevent accidents (Rao, 2007)and for human factors investigation (Ren et al., 2008). However, suchabove mentioned studies have significantly focused on complex con-ceptual model construction and have indicated that quantitative riskanalysis cannot be the only candidate for further human and organi-zational factors analysis. Thus, developing qualitative HFACS intoquantitative HFACS can improve the process of accident investigation.As an example, it can be reported that using both Bayesian network(BN) and HFACS can provide quantitative interrelationships betweenprobable occurrence and computation of numerical values (Wang et al.,2011). The prior integrated application of BN and HFACS is representedby Luxhoj and Coit (2006), where the BN is constructed based onHFACS taxonomy. Authors further explained that it is a vital require-ment in order to model risk and its uncertainty, and to present newmethods for assessing hazards. It is suggested that using a combinationof fuzzy set theory and BN can be helpful due to uncertainty handling inthe case of observed evidence relevant to the variables (Zarei et al.,2019).In some studies, causal relationships of accidents are demonstratedusing BN for human and organizational factors analysis (Grabowskiet al., 2009), and also for human, process and mechanical failures inprocess systems (Zarei et al., 2017, 2019). In these studies, fuzzy se- mantics and integral value method are applied to quantify the condi-tional probability table (CPT). However, CPT based on multi-experts’elicitation process seems to be insufficient. To overcome this limitationin the present study, an extension of fuzzy analytical hierarchy process(FAHP) is used to compensate ambiguity and uncertainty in the multi-experts’ elicitation.In addition, in human reliability analysis, dependency assessment isan important issue in risky large complex systems (Chen et al., 2017). Arobust critical analysis is also required in order to accurately reveal thegreatest contributing factors to an accident scenario. It can be ex-tremely helpful in designing a successful strategy to prevent accidents.Nevertheless, these essential matters are taken, to a lesser degree, inaccident literature. Therefore, to improve the existing models, thispaper presents an extension of accident analysis model by integratingfour approaches including HFACS framework, Fuzzy set theory andexpert elicitation, FAHP and BN. The final outcome of this model helpsto analyze the accident under different uncertainties and consider adynamic structure.The rest of this paper is presented as follows. In Section 2 the de-veloped methodology is presented, while in Section 3 application of themethodology with results and discussion are provided. Finally, theconclusion along with the potential risk based safety measures to pre-vent similar accidents in future are demonstrated. 2. The developed methodology This section illustrates an overview of the methodology used todevelop a dynamic Fuzzy Bayesian Network-HFACS model called FBN-HFAC. The model is built considering HFACS framework, fuzzy logic,intuitionistic fuzzy AHP, and BN to analyze accidents under un-certainty. Fig. 1 presents the details of each step. The proposed modeltaps into the joint capabilities of HFACS and BN for the purpose of investigating HFs in accidents of process systems.  2.1. The HFACS framework Human factors analysis and classification (HFACS) is a broad humanerror analysis framework. HFACS is based heavily upon James Reason'sSwiss cheese model of human error which investigates active and latentfailures within an organization that culminate in an accident at fourlevels, including 1) organizational influences, 2) unsafe supervision, 3)preconditions for unsafe acts and 4) unsafe acts. It is a comprehensivehuman error framework and comprises 19 causal categories and at least69 subcategories within four levels of human failure. The HFACS fra-mework is illustrated in Fig. 2 (Shappell and Wiegmann, 2001). In applying the HFACS framework to a studied accident, the acci-dent scenario is first clearly defined and all conditions under which theaccident occurred are determined. Thereafter, through holding varioussafety meetings in the presence of different experts, HFACS are utilizedto identify various human failures from operators (level 1) which areactive, to organization failures (level 4), which are latent, as shown inFig. 1. These experts include the supervisor and senior supervisor of theproject, HSE (health, safety, and environment) manager, contractorconsultant, safety and fire supervisor and manager. The unsafe acts,that is, level 1, comprise active failures performed by operators thatdirectly result in the accident, and it is these that are most visible toinvestigators. When failures at level 1 are determined, the analyzershould move on to explore failures at level 2 (preconditions for unsafeacts) that effect the failures of level 1. This process continues up tofinally identifying the failures in the last level which indicate organi-zational influences. Accordingly, experts are finding human errors indifferent levels, which have a direct and indirect role in accident oc-currence.  2.2. Fuzzy Bayesian Network In this section, fuzzy logic, expert judgment, and intuitionistic fuzzyAHP are utilized for estimating the impact rate of identified failures inorder to more effectively deal with the associated uncertainty in theanalysis. In addition, FBN is used for deductive reasoning and criticalanalysis of root events.  2.2.1. Fuzzy theory and expert elicitation In this study, the expert judgment method, as a scientific consensustechnique, is utilized to obtain the impact rate of human errors whichare identified using HFACS framework. An integration of fuzzy settheory and subjective opinions can help the assessors determine theimpact rates of each human error. Various methods are available toaggregate experts' opinion such as fuzzy priority relations, game theory,arithmetic averaging operation, max–min Delphi method, and simi-larity aggregation method (SAM). However, Liu et al. (2014) noted thatthere is no definite way to explain which of the techniques has super-iority over the others. Additionally, Yazdi and Zarei (2018) representedsome advantages of SAM technique comparing it to others available inorder to compute the failure probability of basic event in fault treeanalysis when applied in real-field case study. Thus, in this study, toobtain the impact rate of human errors, SAM method, which considersboth homogenous and heterogeneous groups of experts, is employed.The qualitative terms used in this study are collected by a combinationof fuzzy triangular and fuzzy trapezoidal number to aggregate the ex-perts’ opinions (Yazdi et al., 2017b). Furthermore, a heterogeneousgroup of experts was selected for computing the impact rate of humanerrors, as each expert opinion has an individual weight based on hisbackground and expertise. (Celik et al., 2010; Lavasani et al., 2015; Ramzali et al., 2015; Yazdi, 2017a; Yazdi et al., 2017a). Accordingly, for qualifying the measurements, the importance of an expert's opinionwas measured to deal with the uncertainty and lack of sufficient data. Inorder to score the quality of an expert's opinion, the following char-acteristics of an expert were considered: job profession, job experience, job tenure, education level and age. There are numerous methods togive a specific weight to the employed expert such as geometrical oraverage simple mean. Although both types of the mentioned methodsare useful, they still suffer from several drawbacks which cannotproperly be used for expert weighing because these approaches do notprovide sufficiently high subjectivity. In this regard, AHP, srcinallyintroduced by Saaty (1977) is utilized for this purpose which has highcapability to deal with the complexity of decision problems. It is usedhierarchy of decision layers due to breaking the problem into several  E. Zarei et al.  Journal of Loss Prevention in the Process Industries 57 (2019) 142–155 143  layers and is unable to understand subjective practical approach (cap-ability of reflecting the human thinking). The main advantage of AHPcompared with other available methods is the use of signified judgmentin the form of pairwise comparison. However, the conventional AHP iscommonly criticized due to its inability in dealing with inherent un-certainty and vagueness during the decision making procedure. Inconventional AHP, crisp values (e.g. within the 1–100 scale) are usedfor comparison over different criteria. However, in the realistic situa-tions, there is difficulty for decision makers to judge using crisp eva-luation for applying the pairwise comparison matrix. Due to the expertlimited knowledge (subjectivity of the qualitative criteria), the decisionmakers are commonly reluctant and hesitant to represent their in-dividual judgments in a decision making group. Thus, it seems thatconventional AHP is insufficient to clearly capture the significant eva-luation for concluding the priorities in the mentioned situations. Fuzzyset theory is a popular approach to overcome the mentioned challengeson subjectivity of assessment based on expert judgment. Therefore, theconventional AHP is extended to Fuzzy AHP (FAHP), where instead of acrisp value, the set of triangular or trapezoidal fuzzy numbers (asmembership function) have been used in the comparison analysis (Duanet al., 2016; Yazdi, 2018, 2017a; Yazdi et al., 2018; Yazdi and Kabir, 2017). The membership function signifies the degree to which mea-sured elements belong to the preference set. As mentioned earlier,FAHPs use fuzzy numbers for the pairwise comparison step which il-lustrates some shortcomings due to the restriction of fuzzy sets itself.The membership function of the fuzzy set is represented as a single-value function. This means that it cannot be engaged to indicate theobjection proofs and supports at the same time in a variety of practicalcircumstances. Decision makers and experts commonly have someambiguity, vagueness, and uncertainty in order to allocate the pre-ference values to considered criteria, objectives, or alternatives. As anexample, terms of “agreement” and “disagreement” can be indicated foran objective based on fuzzy set numbers, whereas the term “abstention”signifies the hesitation and doubt over the objective. Therefore, in orderto deal with such realistic circumstances and model human thinking,cognition, and perception in a much more extensive way, Atanassov(1986) enhanced conventional fuzzy set into the intuitionistic fuzzy set(IFS) which is presented by (Xu and Liao, 2014) as well.The concept of linguistic expression has a high value in dealing withany circumstances that are to be described in the old model of quan-titative expression and are ill-defined or complex (Zadeh, 1965). Inorder to convert qualitative terms to corresponding fuzzy numbers,Chen and Hwang (1992) represented a numerical approximation. Toacquire this criterion, there are common verbal expressions in thesystem. Chen and Hwang (1992) developed qualitative terms and theircorresponding fuzzy numbers on different scales from scale one, con-taining two verbal terms, to scale eight containing thirteen verbal termsas provided in Table 1. In addition, Yazdi (2017b) discussed the common estimation of human memory competence is seven plus-minustwo patches. This means that the proper number of linguistic expres-sions required for a human to make a suitable judgment is between fiveand nine (Miller, 1956; Nicolis and Tsuda, 1985). For this purpose, Fig. 1.  The proposed human reliability assessment model for process systems.  E. Zarei et al.  Journal of Loss Prevention in the Process Industries 57 (2019) 142–155 144  transformation scale of six which includes 5 verbal expressions is se-lected to fulfil the intellectual assessment for obtaining impact rate of human errors in the present study. Table 1 presents total fuzzy mem-bership numbers in the form of trapezoidal numbers. The reason of thisselection is based on cognitive and human thinking which prefers toexpress subjectivity through five verbal expressions and has been dis-cussed earlier. In addition, reliability of the scales has been validated indifferent types of industrial applications such as oil and gas (Omidvariet al., 2014), aerospace (Yazdi et al., 2017a), marine (Helvacioglu and Ozen, 2014), and medical (Kabir et al., 2018). The linguistic expres- sions in Fig. 3 are in the form of both triangular and trapezoidal fuzzynumbers. It should be noted that it is possible to transform all the tri-angular fuzzy numbers to corresponding trapezoidal fuzzy numbers.The membership function can be defined for both trapezoidal andtriangular fuzzy numbers as follows:a) In triangular form,  A ˜  = (a 1 , a 2 , a 3 )  µ x  x aa x aa x a x a ( )0,,,0, i x aa aa x a a 11 22 33 12 133 2 =<> (1)b) In trapezoidal form,  A ˜  = (a 1 , a 2 , a 3 , a 4 )  µ x  x aa x aa x aa x a x a ( )0,,1,,0, i x aa aa x a a 11 22 33 44 12 144 3 =<> (2)Assume that each expert,  E  l  ( l =1, 2,…,  m ) expresses his/her Fig. 2.  The HFACS framework used in this study (Shappell and Wiegmann, 2001).  E. Zarei et al.  Journal of Loss Prevention in the Process Industries 57 (2019) 142–155 145  viewpoints about a specific attribute in a certain context by use of apredefined set of qualitative terms. The qualitative terms are convertedto the corresponding fuzzy numbers. The applied procedure is pre-sented in Fig. 4., and provided in detail using the following 6 steps. Step 1.  Computing the degree of similarity (degree of agreement). S (R ˜ , R ˜ ) uv u v   is defined as opinions (degree of agreement) between eachpair of experts,  E u  and  E  v  . According to this, consideration for S (R ˜ , R ˜ ), uv u v   A˜ = (a 1 , a 2 , a 3 ) and  B  = (b 1 , b 2 , b 3 ) being the twostandard triangular fuzzy numbers, and the degree of agreementfunction of S is defined by using Equation (3). S A B J a b i ( ˜ , ˜) 1 141,2,3 ii i 14 === =  (3)where J is the number of fuzzy set members, meaning that triangularand trapezoidal should be 3 and 4, respectively. Step 2.  When ( A˜ ,  B˜ ) [0, 1], the greater the value of S ( A˜ ,  B˜ ) has thehigher similarity between two experts with respect to fuzzy numbers, A˜ and  B˜ . Accordingly, for two standard trapezoidal fuzzy numbers, inthe above equation, computing the Average of Agreement (AA) degree AA(E ) u  of an expert's opinions can be done using equation (4).  AA E mS Ru Rv ( ) 11( ˜ , ˜ ) uu vv J  1 = =  (4) Step 3.  Computing the Relative Agreement (RA) degree,  RA(E ) u  of allexperts.  E u m as RA E  AA E  AA E  ( 1,2, , ) ( ) ( )( ) u uuumu 1 = … = =  (5) Step 4.  Estimating the Consensus Coefficient (CC) degree,  CC(E ) u  of expert opinions,  E ( u 1, 2, . . ., J): u  = CC E W E RA E  ( ) ( ) (1 ) ( ) u u u = +  (6)The coefficient is presented as a relaxation factor of the untakenprocedure satisfying (0 ≤ ≤ 1). It illustrates the importance of W Table 1 Linguistic expressions and their corresponding fuzzy numbers (Chen and Hwang, 1992). Linguistic Expressions Scale 1 Scale 2 Scale 3 Scale 4 Scale 5 Scale 6 Scale 7 Scale 8None (0,0,0.1)Very low (0,0,0.2) (0,0,0.1,0.2) (0,0,0.1,0.2) (0,0,0.2) (0,0.1,0.2)Low-Very (0,0,0.1,0.3) (0.1,0.2,0.3)Low (0,0,0.2,0.4) (0.1,0.2,0.3) (0,0,0.3) (0,0.2,0.4) (0.1,0.25,0.4) (0,0.2,0.4) (0.1,0.3,0.5)Fairly low (0,0.25,0.5) (0.2,0.4,0.6) (0.2, 0.35,0.5) (0.3,0.4,0.5)Mol. Low (0.4,0.45,0.5)Medium (0.4,0.6,0.8) (0.2,0.5,0.8) (0.3,0.5,0.7) (0.3,0.5,0.7) (0.3,0.5,0.7) (0.3,0.5,0.7) (0.3,0.5,0.7)Mol. High (0.5,0.55,0.6)Fairly High (0.5,0.75,1) (0.4,0.6,0.8) (0.5,0.65,0.8) (0.5,0.6,0.7)High (0.6,0.8,1) (0.6,0.8,1,1) (0.6,0.8,1) (0.7,1,1) (0.6,0.75,0.9) (0.6,0.75,0.9) (0.6,0.8,1) (0.5,0.7,0.9)High-Very High (0.7,0.9,1,1) (0.7,0.8,0.9)Very High (0.8,1,1) (0.8,0.9,1,1) (0.8,0.9,1,1) (0.8,1,1) (0.8,0.9,1)Excellent (0.9,1,1) Mol.: More or less. Fig. 3.  Transformation of scale six. Fig. 4.  The steps of the fuzzy aggregation procedure.  E. Zarei et al.  Journal of Loss Prevention in the Process Industries 57 (2019) 142–155 146
We Need Your Support
Thank you for visiting our website and your interest in our free products and services. We are nonprofit website to share and download documents. To the running of this website, we need your help to support us.

Thanks to everyone for your continued support.

No, Thanks

We need your sign to support Project to invent "SMART AND CONTROLLABLE REFLECTIVE BALLOONS" to cover the Sun and Save Our Earth.

More details...

Sign Now!

We are very appreciated for your Prompt Action!