Quantitative Risk Assessment for Autonomous Vehicles: Integrating Functional Resonance Analysis Method and Bayesian Network

In autonomous vehicles (AVs), intricate functional-level couplings exist among the components. Accidents can occur even when all functions are operating normally, as subtle performance variabilities in these functions can aggregate through these couplings, leading to functional resonance. The aim of this study is to identify, analyze and quantitatively assess the safety issues caused by these complex interactions in AVs and to propose appropriate risk management strategies to improve vehicle safety. Commonly used modern methods of risk assessment, such as system-theoretical process analysis and accident mapping, struggle to capture this resonance in AVs and lack quantitative analysis. To this end, this paper proposes a quantitative risk assessment method that integrates functional resonance analysis method (FRAM) with Bayesian network (BN) to reveal the complex interactions and quantify risks within AVs. Initially, a FRAM model is constructed to characterize the function couplings of a system, which are subsequently aggregated into functional resonance chains to identify potential hazards. Then, these functional resonance chains are used to develop a BN model for quantitative assessment of system risk. A case study of an automatic emergency braking (AEB) system on an open-source vehicle is conducted to verify its effectiveness. The results demonstrate that the proposed approach not only identifies functional resonance but also effectively quantifies risks in the AEB system.