Fault Detection of FW-UAV Based on Spatiotemporal Feature Networks

In fixed-wing unmanned aerial vehicle (FWUAV), anomalous states can propagate errors internally due to the closed-loop nature of flight control systems. This complicates the differentiation of sensor signal variations caused by different component failures, hindering fault detection models from accurately capturing hidden fault features. As a result, FW-UAV may struggle to perceive and interpret critical flight parameters from onboard sensor data, affecting accurate flight state cognition. To address this, we propose a multi-scale fault detection method. Gated Graph Convolutional Networks (GatedGCN) analyze topological relationships between adjacent sensors to extract spatial features, while a CNN-Transformer network captures temporal patterns for precise fault identification. Experiments using real flight data show that this method significantly improves FW-UAV flight state cognition accuracy.