The Methodology of Dynamical Uncertain Causality Graph for Risk Assessment

With the development of technology, engineering systems are growing rapidly in their scale and complexity, which increases the severity of accidents and the urgency of risk assessment. Although lots of efforts have been done in risk analysis, their applications are limited when complex systems present the characteristic of multi-domain information synthesis and complete risk data is hard to obtain. To solve these problems, a risk assessment method based on DUCGs (Dynamical Uncertain Causality Graphs) is presented in this paper. With the advantage of DUCGs' flexibility, high tolerance for information's inaccuracy and incomplete information processing ability, the method based on DUCGs has lots of advantages: (1) Risk models in different fields can be established separately and the whole model can be obtained by merging the nodes shared by different models, which allow analysts to establish their risk model respectively. (2) Different from DBNs (Dynamic Bayesian Network) that need complete probability information, DUCGs only need probabilities related to risk status and they can alleviate the information incomplete problem to some extent. (3) The probability matrixes used in DUCGs are more consistent with people's cognitive habits, which makes it friendlier for analysts and promotes the integration of expert experience and data. Finally, the practicality and superiority of DUCGs are verified in a case study.