(84a) Risk Analysis of Rare Events By Modified Hierarchical Bayesian Modeling (mHBM)

Kumari, P. - Presenter, Texas A&M University
Karim, M. N., Texas A&M University
Risk assessment of rare but catastrophic events in process industry deals with challenges of data scarcity and uncertainty estimation. Data scarcity and uncertainty both are prevalent with operator responses as well as process data. Difficulty presented by sparse data for accounting uncertainty in failure probability of control layers can be solved by Bayesian model. Bayesian model networks when applied with fault tree, event tree etc. deals with the static nature of trees and lack of event occurrence data [1]. A fault tree connects basic events to initiating events and an event tree connects initiation event to consequences, rare events being one of them. However these trees themselves are unidirectional in nature [2]. Also, it is assumed that data collected for risk assessment comes from strictly consistent operating conditions. It leads to unaccountability of uncertainty associated with source-to-source variability in data sources for process data as well as operator responses. The research area of implementing bi-directionality in fault and event trees and estimating uncertainty due to source-to-source variability of data sources has not gained much attention for chemical process industry. According to literature [3], hierarchical Bayesian modeling (HBM) deals effectively with uncertainty presented by source-to-source variability of data sources in off-shore blowout.

The present work integrates the fault and event trees for chemical processes. Bi-directionality in the integrated model is achieved by means of Bayesian network. It enables top-down approach from consequences to basic events, hence making possible prediction of consequence probabilities by number of occurrences of basic events and vice-versa. The integrated model is amalgamated with HBM to deal with the factor that data points for process data and operator data comes from the conditions which may or may not be similar, tackling source-to-source variability and uncertainty in operator responses in process industry. The new technique is demonstrated on the case study of Tennessee Eastman problem (TEP). TEP is modified to include two more control layers of operator action and action effectiveness for consideration of human interaction with the process. Hence this work presents a novel integration model which includes a holistic method for development of a complete structure for risk assessment of rare events combining all sources of uncertainty.

Keywords: Hierarchical Bayesian modeling, fault tree, rare events


  1. He, Rui et al. “A quantitative risk analysis model considering uncertain information” Process Safety and Environmental Protection 118 DOI: 10.1016/j.psep.2018.06.029
  2. Khakzad N, Khan F, Amyotte P. Dynamic safety analysis of process systems by mapping bow-tie into Bayesian network. Process Safety and Environmental Protection, 2013; 91(1):46–53
  3. Yu, Hongyang et al. “A Flexible Hierarchical Bayesian Modeling Technique for Risk Analysis of Major Accidents.” Risk analysis : an official publication of the Society for Risk Analysis 37 9 (2017): 1668-1682