Copula-based models for risk analysis of process systems with dependencies

Abstract

With the increasing integration of heat and mass and the complexity of process systems, process variables are becoming strongly interdependent. Ignoring these dependencies in process safety modelling is unreasonable. The present work addresses this dependency challenge. It proposes two simple yet robust risk models for process safety analysis. The first model is the copula-based bow-tie (CBBT) model, which revises the traditional bow-tie (BT) model by considering dependencies among the causes and failures of safety barriers. Copulas are used to simulate hypothetical dependent joint probability densities. The proposed model, along with classical BT analysis, is examined under a case study of the risk analysis of a typical distillation column. Comparing the results from both approaches in terms of the estimated probability of a potential hexane release scenario, it is shown that the dependencies of process units’ malfunctions can increase the likelihood of accident scenarios to a significant extent. Further, to explore the mechanisms behind the impact of such dependencies, the effect of dependencies on the two most basic logic gates is also analyzed. The next model developed is the copula-based Bayesian network (CBBN), which integrates linear dependence modelled by a Bayesian network (BN) and non-linear dependence by copulas. It provides more reliable estimation of accident probability when applied to real cases. Sensitivity analysis identifies the factors that play important roles in causing an accident. A diagnostic analysis is also performed to find the most probable explanation for the occurred event. Results match the accident investigation report and thus prove the effectiveness of the proposed model.