Risk assessment of chemical process considering dynamic probability of near misses based on Bayesian theory and event tree analysis

With the development of modern automatic control systems, chemical accidents are of low frequency in most chemical plants, but once an accident happens, it often causes serious consequences. Near-misses are the precursor of accidents. As the process progresses, near misses caused by abnormal fluctuation of process variables may eventually lead to accidents. However, variables that may lead to serious consequences in the production process cannot update the risk in the life cycle of the process by traditional risk assessment methods, which do not pay enough attention to the near misses. Therefore, this paper proposed a new method based on Bayesian theory to dynamically update the probability of key variables associated with process failure risk and obtain the risk change of the near-misses. This article outlines the proposed approach and uses a chemical process of styrene production to demonstrate the application. In this chemical process, the key variables include flow rate, liquid level, pressure and temperature. In order to study the dynamic risk of the chemical process with consideration of near misses, according to the accumulated data of process variables, firstly the abnormal probability of the variables and the failure rate of safety systems associated with the variables were updated with time based on Bayesian theory. On the basis of the dynamic probability of key process variables, an event tree of possible consequences caused by variable anomalies was established. From the logical relationship of the event tree, the probability of different consequences can be obtained. The results show that the proposed risk assessment method based on Bayesian theory can overcome the shortcomings of traditional analysis methods. It shows the dynamic characteristics of the probability of different near misses, and achieves the dynamic risk analysis of chemical process accidents.     

Modeling of BP Texas City refinery incident

This paper presents detailed modeling results of the BP Texas City refinery incident. Three different approaches and explosion modeling tools were used to study the event. The results predicted by all three approaches are similar and all approaches identified a hazard potential comparable to what was witnessed on March 23, 2005. This confirms that quantitative risk assessment (QRA) has the ability to model a realistic scenario, and is therefore useful in safety measure design and emergency preparedness decision making to improve overall safety performance. Had QRA been conducted during a management of change (MOC) decision-making process, personnel trailers likely would not have been sited in such close proximity to the process units. The resulting severe consequences would then not have occurred. This work also aims to emphasize the importance of QRA in process safety management.

The paper presents the authors’ perception of the sequence of events involved in the incident based on the published literature available at the time of writing. It also assesses potential consequences for the perceived sequence of events using a variety of consequence assessment tools. In doing so, the analysis illustrates how this incident could have been prevented in spite of many operational difficulties. The observations and commentary presented in this paper are intended solely for the purpose of process safety enhancement on the basis of the lessons learned. BP has published its own detailed report; the incident is also the subject of a recent investigation by the US Chemical Safety and Hazard Investigation Board, with the CSB's final report being available at http://www.csb.gov/index.cfm?folder=completed_investigations&page=info&INV_ID=52 (as of April 2007).     

The importance of near miss reporting to further improve safety performance

Dealing with accidents implies that such events have in common the potential to affect people and the environment in a significant way. Therefore, all parties involved in industrial risk management processes, i.e. industry, regulatory authorities, public as well as scientific and technical institutions, are well aware of the importance of considering and analysing such type of events for the purposes of accident prevention. Also, the methods of Quantitative Risk Assessment (QRA) have large experience in numerically expressing the various degrees of risk related to accidents. On the other hand, the topic of including `near misses' (i.e. any event which could have escalated to an accident) in safety management systems with the aim to prevent major accidents and the occurrence of similar events in the future is relatively new. Although its importance has more and more been recognised in the last few years, it is not yet a commonly accepted fact that near miss reporting and investigation of near misses should be an integral part of a safety management system in industrial facilities. In the European Council's new `Seveso II Directive' 96/82/EC, there is—in addition to the mandatory requirements of major accident reporting—an explicit recommendation to report near misses to the Commission's Major Accident Reporting System (MARS) on a voluntary basis. In this paper, examples of current experience in the chemical industry with the collection and analysis of data on near misses are presented and discussed with regard to industry-wide conclusions. In addition to this more qualitative discussion, quantitative arguments are put forward regarding the impact of near misses on risk estimates derived from QRA.     

Dynamic risk assessment using failure assessment and Bayesian theory

To ensure the safety of a process system, engineers use different methods to identify the potential hazards that may cause severe consequences. One of the most popular methods used is quantitative risk assessment (QRA) which quantifies the risk associated with a particular process activity. One of QRA's major disadvantages is its inability to update risk during the life of a process. As the process operates, abnormal events will result in incidents and near misses. These events are often called accident precursors. A conventional QRA process is unable to use the accident precursor information to revise the risk profile. To overcome this, a methodology has been proposed based on the work of Meel and Seider (2006). Similar to Meel and Seider (2006) work, this methodology uses Bayesian theory to update the likelihood of the event occurrence and also failure probability of the safety system. In this paper the proposed methodology is outlined and its application is demonstrated using a simple case study. First, potential accident scenarios are identified and represented in terms of an event tree, next, using the event tree and available failure data end-state probabilities are estimated. Subsequently, using the available accident precursor data, safety system failure likelihood and event tree end-state probabilities are revised. The methodology has been simulated using deterministic (point value) as well as probabilistic approach. This Methodology is applied to a case study demonstrating a storage tank containing highly hazardous chemicals. The comparison between conventional QRA and the results from dynamic failure assessment approach shows the significant deviation in system failure frequency throughout the life time of the process unit.     

Accident prevention approach throughout process design life cycle

The accident rate in the chemical process industry (CPI) has not been decreasing although majority of accident causes have been identified and could have been prevented by using existing knowledge. These recurring accidents show that the existing knowledge has not been used effectively. In this paper, accident knowledge learned from earlier accident analyses are utilized to predict the common design errors during chemical plant design. An accident prevention approach throughout process design life cycle is proposed for a safer design consideration where designers are guided to identify common design errors, accident contributors and critical points to look for. The accident prevention approach has been applied to analyze the BP Texas City Refinery Explosion and Fire tragedy.     

Dynamic accident modeling for high-sulfur natural gas gathering station

Dynamic accident modeling for a gas gathering station is implemented to prevent high-sulfur natural gas leakage and develop equipment inspection strategy. The progress of abnormal event occurring in the gas gathering station is modeled by the combination of fault tree and event sequence diagram, based on accident causal chain theory, i.e. the progress is depicted as sequential failure of safety barriers, then, the occurrence probability of the consequence of abnormal event is predicted. Consequences of abnormal events are divided into accidents and accident precursors which include incidents, near misses and so on. The Bayesian theory updates failure probability of safety barrier when a new observation (i.e. accident precursors or accidents data) arrives. Bayesian network then correspondingly updates failure probabilities of basic events of the safety barriers with the ability of abductive reasoning. Consequence occurrence probability is also updated. The results show that occurrence probability trend of different consequences and failure probability trend of safety barriers and basic events of the safety barriers can be obtained using this method. In addition, the critical basic events which play an important role in accidents occurrence are also identified. All of these provide useful information for the maintenance and inspection of the gas gathering station.     

Development of a database for accidents and incidents in the Greek petrochemical industry

The present paper describes the development of a database that comprises all incidents from the Greek petrochemical industry for the period 1997–2003. This database includes industrial incidents, accidents, operational accidents and near misses from all petrochemical sites in Greece and Cyprus. The design of the database has been conceived in a user-friendly way with additional possibilities for its further use, such as: statistical analysis of the data, calculation of safety indicators, accident reports and human factors analysis. The database allows the various participating industries to compare the analysis of indicators in their own installations with the national average, as the database comprises data from the entire Greek petrochemical industry. Special care has been given to include data from near misses too.     

Operational risk assessment: A case of the Bhopal disaster

Accidental releases of hazardous chemicals from process facilities can cause catastrophic consequences. The Bhopal disaster resulting from a combination of inherently unsafe designs and poorly managed operations is a well-known case. Effective risk modeling approaches that provide early warnings are helpful to prevent and control such rare but catastrophic events. Probability estimation of these events is a constant challenge due to the scarcity of directly relevant data. Therefore, precursor-based methods that adopt the Bayesian theorem to update prior judgments on event probabilities using empirical data have been proposed. The updated probabilities are then integrated with consequences of varying severity to produce the risk profile.This paper proposes an operational risk assessment framework, in which a precursor-based Bayesian network approach is used for probability estimation, and loss functions are applied for consequence assessment. The estimated risk profile can be updated continuously given real-time operational data. As process facilities operate, this method integrates a failure-updating mechanism with potential consequences to generate a real-time operational risk profile. The real time risk profile is valuable in activating accident prevention and control strategies. The approach is applied to the Bhopal accident to demonstrate its applicability and effectiveness.     

Fire incident at a refinery in West Malaysia: the causes and lessons learned

Refineries are major hazard installations (MHI) which possess large quantities of hazardous substances. Refineries are characterized by high complexity and tight-coupled organization. Due to the high complexity and level of interaction among subsystems, designers and operators are unable to predict failures at the refinery units. The world has seen many incidents in refineries through leakage, fire and explosions. The consequences of the accidents sometimes extend beyond the boundary of the property and reach the neighboring residents. This paper reviews refinery incidents worldwide and also outlines the causes of a fire incident at a refinery in West Malaysia and the lessons learned.     

Dynamic risk analysis of offloading process in floating liquefied natural gas (FLNG) platform using Bayesian Network

The growing demand for natural gas has pushed oil and gas exploration to more isolated and previously untapped regions around the world where construction of LNG processing plants is not always a viable option. The development of FLNG will allow floating plants to be positioned in remote offshore areas and subsequently produce, liquefy, store and offload LNG in the one position. The offloading process from an FLNG platform to a gas tanker can be a high risk operation. It consists of LNG being transferred, in hostile environments, through loading arms or flexible cryogenic hoses into a carrier which then transports the LNG to onshore facilities. During the carrier's offloading process at onshore terminals, it again involves risk that may result in an accident such as collision, leakage and/or grounding. It is therefore critical to assess and monitor all risks associated with the offloading operation. This study is aimed at developing a novel methodology using Bayesian Network (BN) to conduct the dynamic safety analysis for the offloading process of an LNG carrier. It investigates different risk factors associated with LNG offloading procedures in order to predict the probability of undesirable accidents. Dynamic failure assessment using Bayesian theory can estimate the likelihood of the occurrence of an event. It can also estimate the failure probability of the safety system and thereby develop a dynamic failure assessment tool for the offloading process at a particular FLNG plant. The main objectives of this paper are: to understand the LNG offloading process, to identify hazardous events during offloading operation, and to perform failure analysis (modelling) of critical accidents and/or events. Most importantly, it is to evaluate and compare risks. A sensitivity analysis has been performed to validate the risk models and to study the behaviour of the most influential factors. The results have indicated that collision is the most probable accident to occur during the offloading process of an LNG carrier at berth, which may have catastrophic consequences.     

Lagging or leading? Exploring the temporal relationship among lagging indicators in mining establishments 2006–2017

Problem: Safety management literature generally categorizes key performance indicators (KPIs) as either leading or lagging. Traditional lagging indicators are measures related to negative safety incidents, such as injuries, while leading indicators are used to predict (and therefore can be used to prevent) the likelihood of future negative safety incidents. Recent theory suggests that traditional lagging indicators also possess characteristics of leading indicators, and vice versa, however empirical evidence is limited. Method: The current research investigated the temporal relationships among establishment-level injuries, near misses, and fatal events using injury and employment data from a sample of 24,910 mining establishments over a 12-year period. Results: While controlling for employee hours worked, establishment-level reported injuries and near misses were associated with of future fatal events across the sample of mines and over the time period studied. Fatal events were also associated with increases in future reported near misses, providing evidence of a cyclic relationship between them. Discussion: These findings challenge the strict categorization of injuries, near misses, and fatal events as lagging indicators. Practical applications: Understanding the KPIs that should be used to manage organizational safety, and how they can be used, is of critical practical importance. The results of the current study suggest that, depending on several considerations, metrics tied to negative safety incidents may be used to anticipate, and possibly prevent, future negative safety events.     

Identification, analysis and dissemination of information on near misses: A case study in the construction industry

Near misses are well-known for providing a major source of useful information for safety management. They are more frequent events than accidents and their causes may potentially result in an accident under slightly different circumstances. Despite the importance of this type of feedback, there is little knowledge on the characteristics of near misses, and on the use of this information in safety management. This article proposes guidelines for identifying, analyzing and disseminating information on near misses in construction sites. In particular, it is proposed that near misses be analyzed based on four categories: (a) whether or not it was possible to track down the event; (b) the nature of each event, in terms of its physical features (e.g. falling objects); (c) whether they provided positive or negative feedback for the safety management system; and (d) risk, based on the probability and severity associated with each event. The guidelines were devised and tested while a safety management system was being developed in a healthcare building project. The monitoring of near misses was part of a safety performance measurement system. Among the main results, a dramatic increase in both the number and quality of reports stands out after the workforce was systematically encouraged to report. While in the first 4 months of the study – when the workforce was not encouraged to report – there were just 12 reports, during the subsequent 4 months – when the workforce was so encouraged – there were 110 reports, all of them being analyzed based on the four analytical categories proposed.     

Performance evaluation of process industries resilience: Risk-based with a network approach

Risk management can be defined as coordinated activities to conduct and control an organization with consideration of risk. Recently, risk management strategies have been developed to change the approach to hazards and risks. Resilience as a safety management theory considers the technical and social aspects of systems simultaneously. Resilience in process industries, as a socio-technical system, has four aspects of early detection, error-tolerant design, flexibility, and recoverability. Meanwhile, process industries' resilience has three phases: avoidance, survival, and recovery, determining the transition between normal state, process upset event, and catastrophic event. There may be various technical and social failures such as regulatory and human or organizational items that can lead to upset or catastrophic events. In the avoidance phase, the upset event is predicted, and thus, the system remains in a normal state. For the survival phase, the system state is assumed to be an upset process event, and the system tries to survive through the unhealthy process conditions or remains in the same state, probably with low performance. In the recovery phase, the system is supposed to be catastrophic, and the emergency barriers are prioritized to show the severity of the consequences and response time, leading to a resumption of a normal state. Therefore, a resilience-based network can be designed for process industries to show its inherent dynamic transition in nature. In this study, network data envelopment analysis (DEA), as a mathematical model, is used to evaluate the relative efficiency of the process industries regarding a network transition approach based on the system's internal structure. First, a resilience-based network is designed to consist of three states of normal, upset, and catastrophic events. Then, the efficiency of each industrial department, which is defined as decision-making units (DMUs), is evaluated using network DEA. As a case study, a refinery that is considered a critical process industry is assessed. Using the proposed model shows the efficient and inefficient DMUs in each of three states of normal, upset, and catastrophic events of the process and the projection onto efficient frontiers. Besides calculating the network efficiency, the performance of each state is extracted to precisely differentiate between DMUs. The results of this study, which is one of the fewest cases in the area of performance evaluation of process industries with a network approach, indicated a robust viewpoint for monitoring and assessment of risks.     

Maintenance-related major accidents: Classification of causes and case study

The potential for major accidents is inherent in most industries that handle or store hazardous substances, for e.g. the hydrocarbon and chemical process industries. Several major accidents have been experienced over the past three decades. Flixborough Disaster (1974), Seveso Disaster (1976), Alexander Kielland Disaster (1980), Bhopal Gas Tragedy (1984), Sandoz Chemical Spill (1986), Piper Alpha Disaster (1988), Philips 66 Disaster (1989), Esso Longford Gas Explosion (1998), Texas City Refinery Explosion (2005), and most recently the Macondo Blowout (2010) are a few examples of accidents with devastating consequences.Causes are being exposed over time, but in recent years maintenance influence tends to be given less attention. However, given that some major accidents are maintenance-related, we intend to concentrate on classifying them to give a better insight into the underlying and contributing causes.High degree of technological and organizational complexity are attributes of these industries, and in order to control the risk, it is common to deploy multiple and independent safety barriers whose integrity cannot be maintained without adequate level of maintenance. However, maintenance may have a negative effect on barrier performance if the execution is incorrect, insufficient, delayed, or excessive. Maintenance can also be the triggering event.The objectives of this article are: (1) To investigate how maintenance impacts the occurrence of major accidents, and (2) To develop classification schemes for causes of maintenance-related major accidents.The paper builds primarily on model-based and empirical approaches, the latter being applied to reports on accident investigation and analysis. Based on this, the Work and Accident Process (WAP) classification scheme was proposed in the paper.     

基于神经网络的民航安全态势评估模型及仿真

民航安全态势评估可以向管理者提供民航安全态势和未来态势变化的信息,帮助管理者作出科学的决策,是预防事故发生的关键。航空器事故征候、事故征候率和灾变的科学预测评估是民航安全态势评估的核心内容,分析选取影响民航安全态势的安全运行因素,尝试建立了民航安全态势评估模型,结合BP神经网络和Elman神经网络进行民航安全态势评估。2008-2010年民航安全态势评估结果如下：民航安全态势整体良好但事故征候较多,其中2009年的民航安全态势相对严峻,需要密切注意民用航空器事故的发生,同时应对2010年民用航空器事故进行高度关注,希望管理者采取积极措施进行事故预防。结果表明,基于神经网络建立的民航安全态势评估模型是可行的,可以作为我国民航安全态势评估的有效工具。     

Analyzing system safety and risks under uncertainty using a bow-tie diagram: An innovative approach

A bow-tie diagram combines a fault tree and an event tree to represent the risk control parameters on a common platform for mitigating an accident. Quantitative analysis of a bow-tie is still a major challenge since it follows the traditional assumptions of fault and event tree analyses. The assumptions consider the crisp probabilities and “independent” relationships for the input events. The crisp probabilities for the input events are often missing or hard to come by, which introduces data uncertainty. The assumption of “independence” introduces model uncertainty. Elicitation of expert's knowledge for the missing data may provide an alternative; however, such knowledge incorporates uncertainties and may undermine the credibility of risk analysis.This paper attempts to accommodate the expert's knowledge to overcome missing data and incorporate fuzzy set and evidence theory to assess the uncertainties. Further, dependency coefficient-based fuzzy and evidence theory approaches have been developed to address the model uncertainty for bow-tie analysis. In addition, a method of sensitivity analysis is proposed to predict the most contributing input events in the bow-tie analysis. To demonstrate the utility of the approaches in industrial application, a bow-tie diagram of the BP Texas City accident is developed and analyzed.     

Rare event risk analysis – application to iceberg collision

To design an engineering system, testing in extreme conditions is at least recommended if not required. There are ambiguities about how to define an extreme state and how to consider it in the design of a system or its operation. The probability estimation of such an event is challenging due to data scarcity, especially in many engineering domains, e.g. offshore development. In this study, available techniques for analyzing the probability of extreme events are examined for their suitability in engineering applications, and a framework is proposed for rare event risk analysis. The framework is comprised of three phases. In the first phase, the outlier based extreme value theory is implemented to estimate the rare event probability. The maximum likelihood criterion is used to estimate the extreme distribution parameters. In the second phase, the rare event is considered as a heavy tail event, and the tail index is estimated through the Hill and the SmooHill estimator. In the third phase, The uncertainty analysis is conducted, and the risk is computed. The proposed methodology is tested for extreme iceberg risk assessment on large offshore structures in the Flemish Pass basin. For this specific case, the estimated design extreme iceberg speed was 4.31 km/h, with an occurrence probability of 3.61E-06.     

A conceptual offshore oil and gas process accident model

The purpose of this paper is to present and discuss an accident prevention model for offshore oil and gas processing environments. The accidents that are considered in this work relate specifically to hydrocarbon release scenarios and any escalating events that follow. Using reported industry data, the elements to prevent an accident scenario are identified and placed within a conceptual model to depict the accident progression. The proposed accident model elements are represented as safety barriers designed to prevent the accident scenario from developing. The accident model is intended to be a tool for highlighting vulnerabilities of oil and gas processing operations and to provide guidance on how to minimize their hazards. These vulnerabilities are discussed by applying the 1988 Piper Alpha and the 2005 BP Texas City disaster scenarios to the model.