Risk Assessment of Multi-State Bayesian Network in an Oil Gathering and Transferring System

Abstract

The research objects of conventional risk assessment methods such as fault tree, event tree, etc., are always considered as two ultimate states, one is the complete failure and the other is the perfect state. In this condition, the effect of partial system failure on system performance is not taken into account, which leads to a big difference between calculation results by risk assessment model and actual situation. Multi-state Markov model has been proven to be a versatile tool for modeling multi-state system. It has been effectively used to interpret the degradation relationship between the perfect state and complete failure in the real world. Unfortunately, multi-state Markov model suffers state space explosion, where the generating states grows exponentially with the number of components comprised in the system. For these issues, a multi-state Bayesian network method and its corresponding algorithms for risk assessment were proposed in this paper. In addition, the availability and accuracy of the proposed method were validated by an application example of the oil gathering and transferring system. The results showed that multi-state Bayesian network method was effective to cope with multi-state system by its powerful ability of bidirectional reasoning and probability updating. Compared with multi-state Markov model, the multi-state Bayesian network method is simplified and its computational efficiency is improved. Furthermore, this method could ensure the credibility of risk assessment results as far as possible by taking new information into account and updating the prior beliefs about the accident.