Automated Analysis of Aviation Safety Issues: Hyper-Scale AI PoC

Overview

Recently, there has been a global increase in unexpected major aviation accidents, such as mid-air collisions between aircraft, aircraft crashes due to bird strikes, and passenger fatalities caused by increased turbulence due to climate change.

The International Civil Aviation Organization (ICAO) has established international standards requiring countries to operate systems that collect, integrate, and analyze aviation data for systematic national aviation safety management, enabling its use in data-driven decision-making frameworks.

Accordingly, Korea is revising relevant laws and regulations to implement international standards and has entrusted the Aviation Safety Technology Institute, a specialized agency, with the task of supporting national aviation safety policies based on aviation safety data analysis.

However, the current practice of analysts at the specialized agency manually processing and analyzing diverse data and information directly results in significant time consumption and operational inefficiency.

Therefore, to enhance safety productivity, this project sought to promote the transition to automation for tasks such as AI-based aviation safety data processing, safety issue identification, and document generation.

To prevent aviation accidents and enhance public trust, it is now more crucial than ever to utilize AI to proactively identify, analyze, and eliminate potential risk factors threatening safety, thereby ensuring the safety of national air traffic.

Project description

For model training, numerous articles related to aviation accidents were collected from SKYbrary, an international aviation safety website. An aviation accident classification system and risk factor categories were established and compiled into a database.

This project secured a large volume of high-quality Q&A sets to enable generative AI to rapidly analyze vast amounts of information and provide accurate answers, and preprocessed the data into a format understandable by LLM. Through continuous technical review and optimization, this project developed an automated aviation safety issue analysis app to validate its automation performance. Key services include: inputting documents related to aviation accidents into the app triggers translation and summarization, automatically classifies the accident type, and provides associated risk factors.

Additionally, it provides risk factor information in Excel file format to enable secondary analysis by specialized analysts, and incorporates functionality to present airline accident articles similar to the input document.

Key technological features

• Performed technical verification for automating aviation safety issue analysis tasks
• Define Use Case, then optimize data preprocessing, embedding, RAG/LLM, and prompt engineering → Execute PoC

Expected outcomes

Through this project, the feasibility of transitioning to an AI-based aviation safety management system was confirmed. Previously, data collected from multiple agencies was manually integrated and analyzed by specialized analysts. However, by validating the performance of generative AI-based automated aviation safety issue analysis, it is expected to reduce the data processing and analysis workload for safety managers and data analysts at airlines, airports, and national agencies, thereby enhancing productivity.

Furthermore, it is anticipated that the automatic analysis function based on big data will identify potential risk factors and establish proactive preventive measures, thereby enhancing the nation's aviation safety level and the public's trust in aviation safety.

Launch year: 2024

Attribution

Case study courtesy of the National Information Society Agency, Republic of Korea, Apolitical's Content Partner. Sourced from Use Cases of Public AI Service, Vol.1,  https://eng.nia.or.kr/site/nia_eng/ex/bbs/View.do?cbIdx=31975&bcIdx=28997&parentSeq=29040