Advancing Aircraft Maintenance Through Artificial Intelligence: A Comprehensive Review and Future Perspective

Abstract

Objectives This paper identifies trends in AI for aviation maintenance. Aiming to uncover untapped benefits, find research gaps, and provide insight into the current state of research and active applications. Methods The method of research is a manual review of research papers, following the structured PRISMA method. Research papers are collected and filtered, before key data points are extracted through reading and keyword search. Papers selected were required to be related specifically to aviation maintenance, written in English, and published in the current landscape of AI, with increasing adoption of generative AI (year >= 2018). Findings Findings reveal that AI research geared towards aviation is an emerging topic, with a general year-over-year increase in the number of research papers published. The research reveals several findings: Nuanced topics such as ethical considerations in the use of AI in aviation are yet to be understood completely, and require further exploration, with relatively few papers (n = 29) focusing on explainability and even fewer (n = 3) focusing primarily on the development of xAI; The study of practical applications of AI in maintenance, and research into complex systems that utilise AI-driven subsystems, such as Computer Vision, Digital Twins and Robotic Inspection, have already started industry use and show promise for further enhancements in the future. These systems trend high for research in recent years (n = 14, n = 13, and n = 10, respectively); Few papers research the creation of datasets (n = 2), and a significant number (n = 53) use the C-MAPSS synthetic dataset, suggesting that a gap exists in research and development for datasets specific to aviation maintenance. Application/Improvements The paper provides insight on research and trends surrounding aviation maintenance, possible gaps, and possible avenues for further research. There is a risk of bias due to research method limitations.