A comprehensive flight performance evaluation model based on flight parameters with comparison to subjective and AI assessments

The study aimed to develop a comprehensive flight performance evaluation model based on flight parameters, covering the entire flight and applicable to normal and abnormal conditions. Thirty-seven pilots performed one normal traffic pattern flight and one single-engine failure emergency flight using a Cessna-172 simulator. The complete flight was divided into distinct phases - takeoff, climb, cruise, descent, approach/landing, and emergency, with evaluation metrics defined for each phase. The analytic hierarchy process was employed to determine the weights of flight phases and evaluation metrics. Two flight instructors provided ratings of performance after reviewing video recordings of the flights. ChatGPT generated five sets of performance scores based on the flight data. Intraclass correlation coefficient and correlation analyses indicated good consistency across multiple evaluation sources. Significant correlations were found among model-derived scores, instructor ratings, and ChatGPT-generated scores. These findings demonstrate that the model is reliable, and potentially applicable to real-world flight training and operations.