Analysis of Hazard Identification Methods in Process Industries Using Analytic Network Process Technique (ANP)

Background and aims: Hazard identification is a critical factor to ensure safe design and operation of systems in the process industries. Process industries are one of the most complex systems, with a variety of equipment, control systems, and executive procedures. In these industries, the use of hazardous materials as raw materials or products is quite common. Interactions between technical components, human factors, and organizational and managerial issues can lead to defects and accidents. Therefore, ensuring safety in the process industries can be a very complicated task. The process industries had led to many accidents with very severe consequences. For example, the explosion and fire of the Piper Alpha oil platform with 167 people killed in 1988, the explosion and fire at Esso gas in Longford, with 2 deaths in Australia in 1988, the explosion of the BP refinery with the death of 15 people in Texas in the year 2005 and Deepwater Horizon in 2010 could be mentioned. In order to prevent the occurrence of accidents or reduce the likelihood of occurrence and the severity of the consequences, various techniques have been developed to identify the hazards. However, due to the complexity of the unique conditions of process industries and resource constraints, always the most appropriate technique for identifying hazards should be used. Hazard identification and risk assessment are the implications of system safety since the mid-twentieth century, with the emergence of an action-oriented approach to safety. Since then, several methods have been developed to identify the hazard and evaluate risk in various manufacturing processes. Various generations of these techniques have been presented and each one is trying to provide the best formula or mental patterns to assessors to identify the hazards. Typically, the choice of hazard identification method is based on the frequency of application of that method in a particular industry and the degree of its acceptance among the experts in that industry. But sometimes the situation is not clear and the decision is somewhat difficult. In this situation, experts face a multi-criteria decision-making problem. Multi-criteria decision-making methods are used in situations where there are many alternatives and criteria. Selecting an inappropriate method to identify the hazards in the process industry could lead to a large number of hazards and waste of resources. Therefore, the aim of this study is to identify effective criteria in selecting the most appropriate method for identifying hazards in process industries and determining the most appropriate method in this industry. Methods: In this study, to select the most appropriate methods for identifying hazards in process industries, Preliminary Hazard Analysis (PHA), Hazard and Operability Study (HAZOP), Subsystem Hazard Analysis (SSHA), System Hazard Analysis (SHA), Operability & Support Hazard Analysis (O&SHA), Fault Tree Analysis (FTA), Energy Trace and Barrier Analysis (ETBA), Software Hazard Analysis (SWHA), Failure Mode & Effects Analysis (FMEA), Management Oversights & Risk Tree (MORT), Chang Analysis (CA), and Job Safety Analysis (JSA) techniques were used. Also, in order to assess the hazard identification methods, ‘the cost of implementation of the technique, user-friendly features, flexibility, implementation time of the technique, the human resources required to perform safety analysis, the possibility of using the technique in the most phases of the system life cycle, history of using technique in similar industries, technique’s logic, experience of analysis team, reliability and depth of the analysis of the technique, dependence on information and data, and equipment needed to implement the technique’ criteria were used based on the opinions of the research team. In order to select important criteria from a set of criteria, a panel of experts with a work experience of at least 5 years comprised of 9 PhDs in occupational health, 3 PhDs in chemistry, 1 specialist in risk management, 6 MScs in HSE and 1 PhD in environment were used. Expert’s opinions about the importance of each of the criteria were gathered through a questionnaire composed of the five "Very good, good, moderate, weak and very weak" spectrum. Finally, using one sample t-test in SPSS 16 software was used to determine the important criteria according to the expert’s opinion. In order to determine and select the most appropriate method for identifying hazards in the process industry, according to experts’ comments based on the existence of an internal relationship between the evaluation criteria of the methods, ANP technique was used. Initially, in order to obtain the weights of each criterion, internal relations and alternatives, a panel of experts with a work experience of at least 5 years, consisting of 5 PhDs in occupational health, 1 risk management specialist, 2 PhDs in chemistry and 2 MScs in HSE were selected. In the following, the number of paired comparisons in each questionnaire was calculated to determine the weight and importance of the factors. The way of completing the paired comparison questionnaire was to first explain to each of the experts the parameters of the questionnaire and how to complete the paired comparison using the 9-way suggested range by saaty. It should be noted that each of the experts completed the questionnaires alone. The weight vector of each of the criteria and alternatives was calculated by pair comparisons for each expert. Since a panel of experts has been used in this study for the sake of completeness, the geometric meaning was taken from all of the matrices drawn from the experts. Inconsistency ratio by 0.1 bases for pairwise comparison Matrices was calculated. Subsequently, the formation of the inverted supermatrix, the calculation of the superharmonic matrix and the calculation of the supermatrix Mandar distribution were performed. In this study, Super Decision software version 2.6.0 was used to determine the weight of criteria and alternatives by ANP method. Results: The statistical analysis of one sample T-test was based on the opinions of 20 experts in order to determine the important criteria for participating in the proposed ANP model structure at 95% confidence level, 6 of the 12 criteria were selected. These criteria were ‘experience of analysis team, reliability, and depth of the analysis of the technique, dependence on information and data, history of using the technique in similar industries, flexibility, and the possibility of using the technique in the most phases of the system life cycle’. Criteria obtained from the last stage were pairwise comparison by the experts' opinions. The inconsistency ratio was 0.058. The results showed that the reliability and the depth of the analysis of the technique criterion with a normalized weight of 0.21 has the highest weight and is in the first priority. In the following, based on the obtained weights, ‘the possibility of using the technique in the most phases of the system life cycle (0.206), flexibility (0.201), dependence on information and data (0.106), experience of analysis team (0.189), and history of using technique in similar industries (0.088)’ criteria were prioritized, respectively. In this study, based on expert panel comments, there was an internal relationship between reliability and depth of the analysis of the technique, flexibility, and dependence on information and data criteria. The supermatrix represents the relationships between the components of the network, through which the final weight of the alternatives can be achieved according to the importance of criteria and their internal relations. Accordingly, the normalized weight and the importance of the alternatives (hazard identification methods) were obtained. Based on the results, the HAZOP (0.1396), FMEA (0.1385), ETBA (0.1197), FTA (0.0984), PHA (0.0875), SHA (0.0806), CA (0.0769), O&SHA (0.0735), SWHA (0.0574), MORT (0.0495), SSHA (0.0395) and JSA (0.0389) were the most preferred techniques in order to identify hazards in process industries. Conclusion: One of the biggest problems in the process industries is the selection of the most appropriate method for identifying hazards and risk scenarios, why so determining the correct control measures is conditional on complete identification of the risk scenarios. In general, the purpose of this study was to explain a structured method for selecting a risk identification method in the industry so that experts and analysts take into account the criteria affecting the application of a method and the degree of importance of each criterion in choosing the appropriate technique for their purpose. In this regard, the process industries have been selected as a high-risk industry, which has been significantly developed in Iran. In this industry, a specific method for identifying hazards called HAZOP has been presented, and the results of this study showed that this technique is at the top of possible choices. However, other techniques such as FMEA are also applicable to these industries. Accordingly, the HAZOP, FMEA, FTA, PHA, SHA, CA, O&SHA, SWHA, MORT, SSHA, and JSA methods were identified as the most preferred techniques for identifying hazards in the process industries by experts. Also, the implementation of this study showed that in other industries that have not been provided with a specific method, a systematic selection of hazard identification methods is possible. The results of this study showed that, despite the various criteria for selecting a risk identification method, some of the criteria are more important and their significance and the internal relationship could be estimated using multi-criteria decision-making techniques such as ANP. One of the limitations of this study is extracting hazard identification techniques and evaluation criteria from one source. Another limitation of this study is the use of exact numbers scale in the process of weighing the criteria and alternatives based on verbal expressions. Therefore, it is suggested that the proposed ANP model be solved under fuzzy conditions in future studies in order to eliminate the probabilistic ambiguity and possible uncertainty in verbal expressions.