A Dynamic Safety Assessment Method for Cascading Effects in Complex Systems Based on Uncertainty Quantification

The safety of complex systems, such as in aerospace, is severely threatened by cascading effects, the evolution of which is fraught with dynamics and uncertainty. Traditional safety assessment methods are often static or deterministic, struggling to characterize the time-varying nature of failure propagation and the combined impact of epistemic and aleatory uncertainties. To address this, this paper proposes a novel method for the dynamic safety assessment of cascading effects, capable of quantifying multi-source uncertainty. This paper proposes a dynamic assessment framework for Cascading Effects Analysis (CEA) that integrates Dynamic Bayesian Network (DBN) and Dempster-Shafer Theory (DST). First, a system functional dependency model is constructed to identify critical components and potential failure propagation paths. Second, DBN is employed to model the temporal evolution of the cascading effect. Innovatively, DST is introduced into the DBN's parameter learning and inference stages. Belief functions are used instead of traditional probability distributions to represent epistemic uncertainty arising from expert knowledge, small datasets, and model inaccuracies. Dempster's rule of combination is applied to handle conflicting or incomplete evidence from various sources, enabling a more robust estimation of component failure probabilities. Finally, a dynamic inference algorithm updates the belief about the system's safety state, achieving a dynamic safety assessment of cascading effects. The proposed method is validated through an application to an flight control system, simulating the propagation of cascading effects following an initial soft sensor fault. The results demonstrate that, compared to traditional probabilistic methods like standard DBN, our approach better distinguishes between "uncertainty" and "ignorance." When fault information is ambiguous or conflicting, it provides more conservative and informative assessment intervals, avoiding overconfident estimations of failure probabilities. The method identifies high-risk propagation paths earlier, providing robust theoretical and technical support for dynamic risk management and fault diagnosis in high-safety requirement systems like those in aerospace.

Keywords: Cascading Effects, Complex Systems, Dynamic Safety Assessment, Uncertainty Quantification, Dynamic Bayesian Network, Dempster-Shafer Theory.