Analysis of emergency decision-making patterns in civil aviation risk events based on the observe-decide-act model

Aviation emergencies are critical to flight safety, yet the core decision-making patterns and their dynamics remain underexplored empirically. This study leverages large-scale ASRS data to analyze decision-making patterns and their association with event outcome severity in aviation fire scenarios. Integrating the Observe-Decide-Act (ODA) decision model, this research develops a data-driven framework combining large language model (LLM)-based deep semantic extraction, multi-level clustering, and dynamic sequence analysis to systematically examine 300 ASRS fire incident reports. This multi-level analysis reveals that emergency decision-making characteristics show a strong association with the final event outcome severity. At the element level, the study identifies core ODA decision elements, demonstrating that their functional importance and coordination vary systematically with the event's severity level. At the pattern level, it uncovers and characterizes six operationally distinct ODA decision-making patterns, with their usage frequencies exhibiting a strong association with the classified severity level. At the dynamic level, it reveals that decision processes generally transition from exploratory behaviors to a subsequent convergence, with this convergence being more pronounced in high-severity event outcomes. Collectively, these findings provide a comprehensive depiction of how aviation emergency decision-making patterns are associated with scenarios of varying event severity. This study offers an empirical understanding for human decision-making in high-risk environments, which is essential for optimizing pilot training and emergency procedure design.