A Bayesian Network to Ease Knowledge Acquisition of Causal Dependence in CREAM: Application of Recursive Noisy‐OR Gates

Abstract

Cognitive Reliability and Error Analysis Method (CREAM) is a common Human Reliability Analysis (HRA) method of second generation. In this paper, to improve the capabilities of CREAM, we propose a probabilistic method based on Bayesian Network (BN) to determine control mode and quantify Human Error Probability (HEP). The BN development process is described in a four‐phase methodology including (i) definition of the nodes and their states; (ii) building the graphical structure; (iii) quantification of BN through assessment of the Conditional Probability Tables (CPT) values and (iv) model validation. Intractability of knowledge acquisition of large CPTs is the most significant limitation of existing BN model of CREAM. So, the main contribution of this paper lies in its application of Recursive Noisy‐OR (RN‐OR) gate to treat large CPTs assessment and ease knowledge acquisition. RN‐OR allows combination of dependent Common Performance Conditions (CPCs). Finally, a quantitative HEP analysis is applied to enable more precise estimation of HEP through a probabilistic approach. Copyright © 2016 John Wiley & Sons, Ltd.

Human reliability analysis for manned submersible diving process based on CREAM and Bayesian network

“…To build a BN, we need to analyze the task development steps and determine the risk factors as variable nodes firstly then analyze the relationship between the factors and use directed edges to form a network framework and establish network node relationships. Finally, we estimate probabilities of the root nodes and then reason about the conditional probability table of the entire network …”

“…According to the CPCs and associated linguistic terms of basic CREAM (), the CPCs' level and corresponding performance reliabilities and weighting factors of diving process can be obtained, as shown in Table .…”

“…Finally, we estimate probabilities of the root nodes and then reason about the conditional probability table of the entire network. 28 After the task steps are listed, hazard identification and risk analysis are performed using historical data 29,30 and decision methods to identify possible human errors and explore their forming factors. At the same time, we use the what-if analysis to analyze how the operations of submarines may be affected when uncertainties change.…”

“…The semantic range of each CPC level corresponds to (0, 50), (10, 90), and (50, 100) obtained by Formula (2), as shown in Figure 3. According to the CPCs and associated linguistic terms of basic CREAM (Appendix C), 28 the CPCs' level and corresponding performance reliabilities and weighting factors of diving process can be obtained, as shown in Table 3.…”

Lifecycle risk assessment of a technological system using dynamic Bayesian networks

“…To complete this segment of the DBN model, it is required to identify the nodes relationship after determining model nodes. The interactions between nodes in organizational and human factor levels were investigated by reviewing the underlying literature of HRA and safety management methods . Furthermore, the knowledge of the domain expert was utilized to verify extracted relationships.…”