Machine Learning-Based Airborne Prognostics and Health Management (PHM): Development Status and Trends

Advanced machine learning (ML) is fundamentally transforming airborne prognostic and health management (PHM), redefining maintenance paradigms to enhance operational safety and reduce costs. This review comprehensively analyzes the status and trajectory of ML-based airborne PHM. It systematically traces model development from classical SVM (relevant for small-data scenarios) to foundational deep learning like CNN (for fault diagnosis) and LSTM (for RUL prediction). The analysis then advances to state-of-the-art Transformer (for global dependencies) and GNN architectures (modeling system topology). The review also evaluates real-world data challenges, such as imbalance and scarcity, and explores solutions like generative models and transfer learning. Finally, it outlines future trends, including hybrid physics-informed models (Digital Twins, PINNs) and deployment strategies (Federated Learning, Edge Computing). We conclude that the ultimate barrier to adoption is not algorithmic capability but regulatory certification. Explainable Artificial Intelligence (XAI) is asserted as the critical technology to bridge this "trust gap,"providing the V&V evidence necessary for certification under standards like DO-178C.