Research on Intelligent Simulation and Decision Support for Carrier-based Aircraft Takeoff and Landing Abnormal Cases Based on Large Models

Objectives The launch and recovery of carrier-based aircraft are characterized by the rarity of incidents, their unpredictability, and severe consequences. As a result, flight deck commanders are unable to accumulate effective experience solely through case studies of such incidents. In contrast, large models, which possess extensive prior knowledge and generative capabilities, have already been successfully applied across various aspects of daily life due to their simplicity of operation and strong logical reasoning abilities. By utilizing large models to simulate different incident scenarios and decision-making processes, a substantial number of emergency decision-making cases for launch and recovery operations can be generated, thereby assisting in training flight deck commanders to enhance their emergency response capabilities. Methods The evolution of carrier-based aircraft launch and recovery incidents under different decisions and conditions, given specific causative factors, has been studied. First, a launch and recovery knowledge base was constructed based on resources such as carrier-based aircraft operation manuals. By integrating additional domain-specific knowledge, the outputs of the large model were constrained within the scope of carrier-based aircraft launch and recovery operations. Subsequently, through prompt engineering, the large model was guided to generate latent trends representing potential evolutionary trajectories based on the given causative factors. Once the latent trends were obtained, a decision-making model augmented with launch and recovery knowledge was employed to make decisions on these latent trends. This step further eliminated unreasonable development paths and updated the incident's state. Results By conducting multiple iterations within the proposed framework, a complete set of derivative incidents can be obtained for a given causative factor. Furthermore, through the simulation of various causative factors and decision-making schemes, a comprehensive database of contingency cases can be constructed. Conclusions Relevant experiments demonstrate that the proposed framework effectively leverages the extensive prior knowledge and robust logical reasoning capabilities of large models to address the scarcity of carrier-based aircraft contingency incidents. Additionally, it provides learning examples to support commanders in enhancing their emergency decision-making skills.