民航飞机预测性维修数据驱动与混合方法综述

A Review of Data-Driven and Hybrid Approaches to Predictive Maintenance for Civil Aircraft

摘要: 预测性维修技术在民航业是保障运行安全、降低维修成本与减少非计划停场损失的关键技术手段，正推动着民航机务维修模式由定时预防性维修向数据驱动、混合模式维修转型。围绕算法研究方法、混合建模方式以及实际工程应用三个方面，本文对民航飞机预测性维修的相关研究成果和发展现状进行整理与总结。本文从数据、方法及挑战三个层面总结民航预测性维修研究进展，梳理飞行记录数据、机载健康监测数据、维修日志以及运行环境参数等多源数据，分析CNN/LSTM、Transformer和图神经网络等剩余寿命预测方法，并介绍物理信息神经网络与数字孪生在该领域的应用情况，最后讨论该领域所面临的数据标注不足、跨工况泛化能力有限、模型可解释性欠缺以及工程部署困难。

Abstract: Predictive maintenance technology is a key technical means to ensure operational safety, reduce maintenance costs, and minimize AOG losses in the civil aviation industry. It is driving the transformation of civil aviation aircraft maintenance models from scheduled preventive maintenance to data-driven, hybrid maintenance. This paper organizes and summarizes research findings and current developments in predictive maintenance for civil aircraft, focusing on algorithmic research methods, hybrid modeling approaches, and practical engineering applications. The research progress of predictive maintenance in civil aviation is summarized from three aspects: data, methods, and challenges. First, it sorts out multi-source data such as flight record data, on-board health monitoring data, maintenance logs, and operating environment parameters. Secondly, it analyzes remaining lifetime prediction methods such as CNN/LSTM, Transformer, and graph neural networks. Thirdly, the application of physical information neural networks and digital twins in this field is introduced. Finally, this paper discusses the insufficient data labeling, limited generalization ability across operating conditions, lack of model interpretability, and difficulties in engineering deployment faced in this field.