Risk analysis of atmospheric and vacuum distillation unit using Bayesian networks

The accidents occurred in chemical plants often regard as low frequency and high consequence. It is necessary to raise the risk analysis for the petrochemical system to help people to find the weakest process in the whole system thus people can strength the process to improve the safety. In this paper, a methodology by using Bayesian Networks (BNs) to give a model for a chemical plant has been raised. According to the harm extend, the methodology classifies the events into three layers, cause, incident, and accident. Then the application of the methodology is illustrated by analyzing an atmospheric and vacuum distillation unit. The model identifies the most possible cause when an accident happened. After that, mutual information and variety of beliefs are calculated in order to find the most sensitive event of an accident. The study gives suggestions to people of identification the most relevant and weakest point in the plant.