Abstract: Aircraft carrier flight deck aviation support is a critical process for generating and sustaining the sortie capability of carrier-based aircraft. Its high-risk, strongly constrained, and highly coupled characteristics impose stringent requirements on the safety, reliability, and interpretability of intelligent support systems. Focusing on decision support for aircraft carrier flight deck aviation support, this paper systematically reviews the research foundations of rule constraints, situational awareness, and scheduling optimization, and analyzes the key challenges faced by large models in this domain, including trustworthy output, multimodal cross-domain fusion, multi-scenario task adaptation, and multi-process causal reasoning. Centered on two representative scenarios, namely pre-launch collaborative support involving deck towing, weapon loading, refueling, and catapult-window coordination, and contingency replanning during recovery under arresting-gear abnormalities or parking-position congestion, this paper further summarizes task objectives, input modalities, constraints, output forms, validation loops, and evaluation indicators, thereby clarifying how large models can be integrated with practical support workflows. The former scenario emphasizes resource coordination under efficiency and safety constraints, whereas the latter focuses on risk convergence and recovery-oriented scheduling under disturbances; together, they cover routine organization and contingency reconstruction in flight deck aviation support. On this basis, this paper proposes a large-model-based decision support framework for high-safety-level aviation support scenarios, forming a closed-loop path of generation, verification, screening, and confirmation. First, physical mechanisms, operating regulations, spatial boundaries, safety intervals, and process dependencies are embedded into the model generation process, while rule engines, collision detection, formal verification, discrete-event simulation, and digital-twin deduction are combined to realize external verification. Second, for heterogeneous information such as flight plans, deck surveillance videos, target-position time series, voice commands, equipment states, engineering geometric constraints, and operation logs, a generalized multimodal unified representation and encoding method is constructed to support situational understanding, event association, and task-state representation. Third, retrieval augmentation, lightweight fine-tuning, prompt templates, and scenario recognition are combined to form a domain-knowledge-driven cross-scenario adaptation mechanism, improving the model’s transferability among launch, recovery, refueling, weapon loading, maintenance, and contingency-handling tasks. Finally, for continuous processes such as towing, support, launch, recovery, and maintenance, a process-level causal reasoning and multi-agent collaboration method is constructed to describe the influence chain of local disturbances on downstream tasks, resource reallocation, and process reconstruction. Furthermore, considering shipborne edge-computing conditions, an engineering deployment approach integrating lightweight inference, local knowledge enhancement, rule- and simulation-based verification, and command workflow integration is proposed. Structured outputs are used to represent task objects, state evidence, candidate actions, verification results, risk levels, and human-confirmation status. The results show that large models for flight deck aviation support should not be directly used as execution-command generators, but should instead serve as auxiliary units for candidate-scheme generation, situational interpretation, risk assessment, and process deduction within a closed loop of local verification and human confirmation. The proposed framework can help reduce untrustworthy outputs, improve multimodal situational understanding, cross-scenario adaptation, forward-looking replanning, and engineering integration capability, and provide theoretical support and methodological reference for the safe and controllable application of large models in complex military support systems.