An Assessment Model of Approach Risk Based on QAR Data and Mutual Information Method

Unstable approaches can easily lead to typical consequence events. This study develops a quantitative risk assessment model to evaluate the risks associated with unstable approaches. Quick access recorder (QAR) data and approach risks are analyzed. Key QAR flight parameters indicative of unstable approaches are selected as monitoring indicators. Twelve monitoring indicators are identified to reflect the state of unstable approaches. The Borda count method is used to rank the monitoring indicators. Based on the ranking results, the study calculates how much each monitoring indicator influences unstable approach events. Potential severe consequences of unstable approaches are analyzed to identify typical consequence events. A risk assessment model is constructed based on the mutual information method from information entropy theory, incorporating the following improvements: ①The mutual information method and the Borda count method are integrated to define a weight that comprehensively reflects the monitoring indicators. This approach overcomes the limitations of using either method in isolation for weight determination. ②Laplace smoothing is utilized to handle the zero-frequency problem in the dataset. Information loss is mitigated, and a necessary complement is provided to the mutual information method, particularly for scenarios characterized by limited sample sizes. ③The correlation between consequence events is considered, and the base risk value is adjusted accordingly. The model is validated using a case study. The results show that using QAR data collected from Airline A in 2019, the model assesses the risk values of runway excursion, CFIT and hard landing, and loss of control in-flight as 4.609 5, 2.062 8, and 0.146 8, respectively. This risk ranking is consistent with the data proportion ranking published by the International Air Transport Association. Indicating that the model results align with actual operational situations. The model's risk rankings are found to be consistent across different aircraft type and years. This consistency is observed when comparing data from Airlines A and B. One hundred experiments are simulated under four different environments. The results show that the risk value trends and distributions share similar characteristics. The consistency between the risk rankings in the simulated and real environments reaches 90% overall. The risk of runway excursion fluctuates with changing conditions. The high-risk value of loss of control in-flight may indicate a serious safety event. The risk of CFIT and hard landing shows little fluctuation, with a uniform distribution, indicating a moderate and predictable risk.