A method for safety analysis of human-machine systems based on dynamic Bayesian simulation

Abstract

Accidents in human-machine systems often lead to serious consequences, so safety analysis is very important for such systems. However, the existing approach to safety analysis of human-machine systems lacks clear delineation of the coupling relationships between human and machine, or provide quantitative analysis. To address these issues, this paper proposes a method for safety analysis of human-machine systems, utilizing dynamic Bayesian network (DBN) and dynamic fault tree (DFT). The core of this method is the identification of human-machine coupling relationships, proposing 10 types of logical relationships and presenting corresponding DFT logic. Then, a conversion method from DFT to DBN is designed to avoid combinatorial explosion in complex FTA calculations. Based on the DBN model, simulation is conducted using Gibbs sampling, which offers higher computational efficiency. Additionally, a method for importance analysis is devised to identify critical nodes that impact the system risk. At last, a case study of refueling mission at space launch site is given to illustrate how to apply the method. Through simulation analysis, the safety risks during the refueling mission are quantitatively assessed, while critical nodes are identified. The results indicate that the dynamic Bayesian simulation method is good in information utilization, dynamic representation, and time performance.