BP neural network-based quantitative classification model for safety in experimental flight training

To enhance the safety of test pilots during aviation experimental flights and reduce the risk level of experimental flights, a quantitative analysis model for the safety classification of experimental flight training is developed based on the Back-Propagation (BP) neural network. Adopting a human factor engineering perspective, the factors influencing test flight safety during the experimental flight training process are analyzed. Three key components, namely the human factor, aircraft factor, and environmental factor, are chosen for the analysis. A structured system of quantitative analysis indicators for safety classification during the experimental flight training process is established, focusing on the pivotal elements of human factors, specific experimental training subjects, and environmental conditions. After quantifying the human factors engineering indicators, due to the relatively low similarity in the fused data types, the BP neural network is employed to construct a quantitative analysis model for safety classification in experimental flight training. Following model training and testing, a quantified safety classification level for the experimental flight training process is output. Utilizing customized tactile acquisition devices and a non-contact facial recognition system, the physiological rhythm data of test pilots during the execution of experimental flight training subjects are collected. The analysis focuses on understanding the impact of the complexity and difficulty of training subjects on the psychological and physiological characteristics of test pilots. On this basis, the quantitative metrics and warning indicators are established, allowing for optimization of experimental flight training subjects, improvement of training quality, and the assurance of test flight safety. The experiments indicate that the output deviation of the quantitative classification model for safety in the experimental flight training process is less than 2%. The model’s predicted results closely align with the actual risk levels observed during the training process, showing that the model proposed can be effectively utilized for risk warning during experimental flight training, and thereby enhance the quality of pilot training and provide reliable safeguard for actual test flight safety.