Real time risk monitoring in a processing system using Bayesian networks

Abstract

Safety and risk are essential components of process industries. The research objective of this thesis is to develop a method to measure and monitor safety in terms of real-time risk of a process system failure. The risk monitoring concept was developed using event trees and Bayesian networks. Process instrument data such as flowrate was used as a basis for the risk probability calculations. The risk monitoring methodology was developed and applied to the Williams Geismar reboiler rupture and fire in 2013. The risk level of the reboiler was examined based on the original design prior to failure and an updated design based on recommendations made by the CSB. The accident probability was decreased by 96% and risk level was reduced by 76.9%. By plotting the risk of a process overtime, future projections of risk can be predicted and action can be taken to prevent accidents before they could occur.