Fault information prediction of unmanned aerial vehicles based on temporal-graph convolutional network
Z.Z. He, Yijun Yao, Fei Jiang, Zhilin Wu, Shaohua Zhang · 2025
The intricate mechanical structure of unmanned aerial vehicles (UAVs) poses challenges in detecting minor faults that can lead to severe consequences. During the flight of UAVs, fault information can be acquired through the utilization of multiple sensors. When time series data collected from multi sensor is represented in Euclidean form, the intrinsic connections among the individual data points are disregarded. Utilizing a graph structured representation for non-Euclidean data allows for capturing spatial correlations among sensors. Additionally, most neural network models are not suitable for handling non-Euclidean data. In this paper, a fault information prediction method of UAV based on a Temporal-Graph Convolutional Network (T-GCN) is proposed to address the above problems. Firstly, the collected time series data is preprocessed and transformed into graph structured data according to the correlation between different data. Secondly, the proposed method leverages the strengths of both Graph Convolutional Networks (GCN) and Gated Recurrent Units (GRUs), extracting spatial structural relationships among data nodes using GCN. Then, it captures the temporal dependencies of the data using GRUs. The prediction is performed by using the previous n historical moment data to forecast the next t moment data. The effectiveness of the proposed method is validated through several fault experiments on UAV propellers. Furthermore, experiments have revealed that the acceleration in the Y-axis direction of UAVs has a more significant impact on the prediction results.