Event and fault tree-based Bayesian network for probabilistic safety assessment of earthquake-induced fire and explosion hazard

Abstract

In nuclear power plant engineering, probabilistic safety assessment (PSA) has been actively studied to evaluate risk due to earthquake events. Recently, the similar PSA framework has been proposed to calculate probability of shut-down of gas plants when earthquake occurred. However, in process plants, earthquakes can also trigger secondary hazards such as fires and explosions, which have been less addressed in seismic PSA despite their potentially catastrophic consequences. These cascading events would cause severe casualties, asset losses, and long-term health impacts by a leak of hazardous substances. To consider such multi-hazard impacts, i.e., earthquake-induced fires or explosions, this work proposes a Bayesian network (BN)-based framework, which is modelled by transforming from fault- and event-tree. For seismic risk, fault tree is constructed to represent the joint operation of constituting equipment, while the top event is defined as a shut-down by earthquake events. Then, the event tree is derived to represent an evolving process from release to final events (i.e., several types of fires and explosions). These constructed trees are transformed into BN, and this process can prevent causal errors when BN is modelled directly. By extending seismic PSA concepts with the traditional fire/explosion event-tree methodology in a unified BN framework, the intended contribution is to enable integrated multi-hazard risk assessment that can account for both seismic and post-seismic accident scenarios. The proposed framework is demonstrated by constructing BN model for earthquake-induced fire and explosion at a gas plant. Then, the inference of the BN model is presented. First, the risk of the multi-hazard on the system is quantified for different hazard levels of earthquake. Second, the contribution of each component to the system failure is evaluated with a retrofit strategy on crucial facilities. By analyzing various accident scenarios, it is showed that the proposed BN model can provide risk-informed decision-making for prioritizing repair and/or retrofitting of structures or equipment in the plant.