Pre-launch Fault Detection Based on Spatiotemporal Features of Multiple Test Cases

Fault diagnosis speed is crucial for the pre-launch fault detection of flight control systems. To save the time costs, test cases only acquire the corresponding signals of some pieces of equipment for different test tasks. Therefore, the fault detection model has difficulty in analyzing the unknown topology relationships of missing test points and hidden fault features. This study proposes a novel fault detection method based on a spatiotemporal feature extractor. The topological relationship between the test points in different test cases is analyzed by graph attention networks. The temporal correlations of dynamic spatial features are captured by a transformer encoder. The experimental results with the ground simulation platform verify the feasibility and effectiveness of the proposed method.