Application of Multimodal Large Models in Aviation Safety: Automated Airspace Analysis for Towerless Airports

This article focuses on the innovative application of multimodal large models in the field of aviation safety. Addressing the safety challenges faced by towerless airports due to the lack of real-time air traffic control, it proposes using the multimodal capabilities of large language models to integrate multi-source information such as ADS-B data, pilot voice broadcasts, and visual monitoring, achieving intelligent airspace situational awareness and providing an AI-driven monitoring solution for towerless airports to help improve aviation safety levels.

Aviation safety is a core concern of civil aviation. Towerless airports rely on pilots' self-coordination and visual observation due to the absence of real-time controller command, which reduces operational costs but poses potential safety risks. Existing monitoring methods include pilot broadcasts (prone to omissions), visual observation (limited by weather and visibility), and ADS-B data (lack of semantic understanding). How to integrate multi-source information to achieve intelligent situational awareness has become a key issue.

Unique Advantages of Large Models

• Multimodal Understanding: Process text (voice transcription), images (radar/camera), perform cross-modal association, and generate evaluation reports;
• Domain Knowledge Transfer: Pre-train on aviation-related texts and fine-tune to adapt to towerless monitoring tasks.

System Architecture

• Data Acquisition Layer: Integrate ADS-B (real-time position), voice communication (CTAF broadcast converted to text), and visual data (camera/radar/weather images);
• Multimodal Fusion Layer: Time alignment, spatial association, semantic enhancement (e.g., marking approach phases/conflict risks);
• Intelligent Analysis Layer: Situational awareness (aircraft distribution/flight phases), conflict warning, intent inference, and natural language report generation.

Real-Time Guarantee

Adopt stream processing, incremental analysis, model quantization, and edge deployment to reduce inference latency;

Accuracy Guarantee

Ensure result reliability through confidence assessment (manual review for low-confidence cases), rule verification (aviation regulation constraints), multi-model voting, and human-machine collaboration (manual confirmation for key decisions);

Data Privacy Compliance

Implement data desensitization, role-based access control, audit logs, and compliance certification to meet aviation data protection requirements.

Dataset Construction

Based on real operational data from multiple towerless airports, covering different seasons/weather/scenarios (normal, busy, emergency), annotated by professionals;

Evaluation Metrics

Detection accuracy (aircraft recognition/flight phase classification/conflict recall rate), warning effectiveness (coverage of real conflicts/control of false alarms/early warning time), report quality (accuracy/completeness/readability);

Experimental Results

Multimodal fusion significantly improves accuracy compared to single-modal baselines, conflict warning recall rate reaches a practical level, and generated reports are well-received by pilots and controllers.

Auxiliary Decision Support

Help pilots with approach preparation, conflict avoidance, and real-time situational awareness;

Safety Analysis Improvement

Used for incident review, hidden danger pattern recognition, and pilot training simulation;

Future of Automated Monitoring

Expected to develop into virtual controllers, traffic management systems, and automatic emergency response initiation procedures.

Research Contributions

Fill the gap in multimodal LLM applications for towerless airports, propose multi-source data fusion methods, domain fine-tuning strategies, and safety AI design principles; submitted to AIAA 2026;

Current Limitations

Limited data coverage, performance under complex weather conditions to be verified, scarce data on extreme cases;

Future Directions

Expand airport testing scale, optimize real-time performance, achieve multi-airport collaboration, improve human-machine interaction interface;

Summary

This study demonstrates the application potential of multimodal large models in aviation safety, provides an intelligent solution for towerless airport monitoring, and is expected to play a greater role in the future.