Objectives The causal factors underlying anomalies during carrier-based aircraft launch and recovery are often rare and unpredictable. To address the hallucination issues commonly encountered by large language models (LLMs) when generating these anomaly causes, we proposes a scenario generation for carrier aircraft driven by large–small model collaboration (SGCAD) method.

Methods First, a knowledge base for carrier-based aircraft launch and recovery is constructed by integrating professional literature and retrieval-augmented generation (RAG) techniques, creating a dataset of normal operation descriptions. These normal descriptions serve as templates, which, combined with scenario-specific prompts, guide the large language model to generate potential anomaly causes. A smaller model is then employed to discriminate between reasonable and unreasonable anomaly causes. Finally, the large model undergoes fine-tuning using direct preference optimization (DPO) and is iteratively refined to progressively increase the proportion of reasonable anomaly scenarios.

Results Experimental results show that, after multiple iterations of the SGCAD method, the proportion of reasonable anomaly causes in the generated dataset reaches 94%, effectively mitigating hallucination issues and significantly improving the rationality and realism of the generated content. Expert evaluations further confirm that the generated anomaly causes encompass diverse and complex scenarios, utilize standardized terminology, and adhere to physical laws.

Conclusions The proposed approach provides a valuable reference for analyzing anomalies during carrier-based aircraft launch and recovery operations.