摘要详情

ID / 提交时间

218 / 2025-06-14 21:58:18

标题

Meta-Learning-Based Fault Diagnosis Method for Multi-Rotor Unmanned Aerial Vehicles

关键字

Fault diagnosis,Multi-rotor UAV,RflyMAD,Meta-learning,Few-shot learning

主题及专题

4、System Monitoring, Fault Diagnosis and Prediction

状态

终稿

作者

Can Guo / National University of Defense Technology

Jingwei Gao / National University of Defense Technology

摘要

Addressing the critical challenges of scarce fault samples in multi-rotor UAVs operating under complex conditions and diagnostic difficulties from multi-source sensor fusion, this study develops a meta-learning-based framework for few-shot fault diagnosis. Utilizing the RflyMAD dataset from Beihang University's Reliable Flight Control Group, we systematically analyze sensor and actuator fault characteristics. A hierarchical meta-feature extraction network is proposed, integrating temporal feature learning and cross-sensor attention mechanisms. The framework implements three meta-learning approaches: Model-Agnostic Meta-Learning (MAML) for parameter adaptation, Memory-Augmented Neural Networks (MANN) for pattern retrieval, and Prototypical Networks (ProtoNet) for metric-based classification. This methodology provides a novel pathway for handling data-scarce fault diagnosis scenarios, demonstrating the potential to improve diagnostic efficiency while reducing dependency on large-scale fault data.