Application of Bayesian Networks to quantitative assessment of safety barriers' performance in the prevention of major accidents

Process plants are particularly subjected to major accidental events, whose catastrophic escalations, triggered by external factors and characterized by very high impact and low probability, can affect both workers and population in the nearby of a process facility, as well as assets and environment. In the context of major accidents prevention, control and mitigation, safety barriers are widely employed; nevertheless, concerning their quantitative assessment, only availability is generally accounted as a measure of their performances and a frequency-based approach is applied. Recent applications have highlighted the potential of Bayesian Networks, a graphical probabilistic method, in major accidents modelling and prevention. This contribution is aimed at applying Bayesian Networks to quantitative assessment of safety barriers' performance in the context of major accidents prevention within the process industry. The Bayesian approach will be compared with a conventional Event-Tree based one by the application to an illustrative case study, considering a major accident whose occurrence can be prevented by the action of several pertinent technical safety barriers. In the Bayesian approach, safety barriers performance has been assessed by means of specific gates, depending on barriers states and classification. An adequate number of final states has been considered. The conversion of the Event-Tree, key element of the conventional frequency-based approach, into a Bayesian Network has been performed, with the aim to test the ability of Bayesian Networks in representing possible events sequences. Indeed, the potentialities of the Bayesian approach in revising probabilities have been explored by means of two different techniques: probability updating and probability adapting. In probability updating, the information about one of the outcomes is used as an evidence, determining the most probable explanation, leading to that specific final state. In probability adapting, additional information during a time interval, in the form of accident sequence precursors, are inserted in the analysis, in purpose to revise safety barriers' performances and final events probabilities. The results of the case study will highlight the advantages of a Bayesian approach to safety barriers' performance assessment, proving that its application may eventually will turn into a more flexible and realistic analysis of major accidental scenarios, in comparison with conventional techniques.