Handling data uncertainties in event tree analysis

Abstract

Event tree analysis (ETA) is an established risk analysis technique to assess likelihood (in a probabilistic context) of an accident. The objective data available to estimate the likelihood is often missing (or sparse), and even if available, is subject to incompleteness (partial ignorance) and imprecision (vagueness). Without addressing incompleteness and imprecision in the available data, ETA and subsequent risk analysis give a false impression of precision and correctness that undermines the overall credibility of the process. This paper explores two approaches to address data uncertainties, namely, fuzzy sets and evidence theory, and compares the results with Monte Carlo simulations. A fuzzy-based approach is used for handling imprecision and subjectivity, whereas evidence theory is used for handling inconsistent, incomplete and conflicting data. Application of these approaches in ETA is demonstrated using the example of an LPG release near a processing facility.