Safety analysis of human-machine interaction of autonomous ships using system theory and complex networks

Maritime Autonomous Surface Ships (MASS) are expected to be deployed widely due to their potential to enhance operational efficiency and reduce costs. Ensuring their success requires a new high level of safety analysis in Human-Machine Interaction (HMI). However, the task is particularly challenging under complex and dynamic maritime conditions. To address this challenge, this study proposes an advanced safety analysis framework that integrates Systems Theory Process Analysis (STPA) with Complex Network (CN) theory, namely STPA-CN, for analysing MASS HMI risks. To improve the precision of network analysis, a two-stage adaptive node ranking algorithm called MI-WLR is developed, which incorporates Mutual Information (MI) theory into the Weight LeaderRank (WLR) structure. The framework contains four components: (1) constructing the MASS HMI risk evolution Network (MHN) based on CN modelling and STPA outcomes; (2) analysing the topological characteristics of the MHN; (3) applying MI-WLR to rank the importance of risk nodes; and (4) conducting robustness analyses to validate the model's effectiveness. The results reveal that system-level hazards and accidents serve as key hubs in the MHN, with human factors, interface design, and environmental conditions exerting diverse degrees of influence. Notably, critical risks are often embedded within the human–machine interface, and targeting high-ranking nodes can effectively disrupt risk propagation pathways. This study fills an important research gap by introducing a novel and scalable framework for MASS HMI safety assessment, providing both theoretical and practical support for risk mitigation and the secure integration of autonomous technologies in maritime operations. • A novel STPA-CN framework is developed to model nonlinear risk propagation in MASS HMI systems. • A complex network is constructed based on STPA-derived risk factors to capture system-wide interactions. • The MI-WLR algorithm integrates mutual information with network topology to rank critical risk nodes. • Robustness analysis confirms MI-WLR's superior accuracy in identifying structurally vulnerable nodes. • Findings offer actionable insights for MASS HMI interface design and targeted risk mitigation strategies.