Resumen
Generally, airplane upsets in flight are considered a precursor to loss of control in flight (LOC-I) accidents, and unfortunately LOC-I is classified as the leading cause of fatal accidents. To further explore the risk factors, causal relationships, and coupling mechanism of airplane upsets, this study proposed a risk analysis model integrating the Interpretative Structural Modeling (ISM) and Bayesian Network (BN). Seventeen key risk factors leading to airplane upsets were identified through the analysis of typical accident cases and the literature. The ISM approach was used to construct the multi-level interpretative structural model of airplane upsets, which could reveal the causal relationship among various risk factors and risk propagation paths. Then, taking 286 accident/incident investigation data as training samples, a data-driven BN model was established using machine learning for dependency intensity assessment and inference analysis. The results reveal that the interaction among risk factors of fatal accidents caused by airplane upsets is more significant than that of non-fatal accidents/incidents. Risk factors such as pilot-induced oscillations/airplane-pilot coupling and non-adherence to Standard Operating Procedures (SOPs)/neglect of cross-validation have a significant effect on airplane upsets in flight among seventeen risk factors. Moreover, this study also identifies the most likely set of risk factors that lead to fatal accidents caused by airplane upsets. The research results have an important theoretical significance and application value for preventing airplane upsets risk.