Failure analysis using functional model and Bayesian network

Abstract

A class of functional model known as multilevel flow model (MFM) is used to represent a pilot scale heat exchanger system. MFM is effective in representing chemical process qualitatively through graphical representation, but lacks the ability to quantify the impact of successes or failures of process events, and is not able to quantitatively distinguish between steps in a goal and their contributions towards achieving the main goal. To address this issue, the MFM is converted into its equivalent fault tree (FT) model to accommodate logical sequence of events along with the needed quantifications. The FT model is then converted into Bayesian network (BN) model to facilitate updates of probabilities. Using Hugin 8.1 software, the BN model is simulated to investigate the response of the process when subjected to various faults. The results highlight the capability of the model in detecting process faults and in identifying the associated root causes, thus pointing to the potentials of the proposed strategy in modeling complex chemical processes for higher level functions in plant operations such as facilitating alarm system and fault diagnosis.