FW-UAV fault diagnosis based on knowledge complementary network under small sample

Fixed Wing Unmanned Aerial Vehicles (FW-UAVs) are prone to faults when performing a variety of tasks, which can lead to mission failure and even pose a safety risk. These faults can be recorded by mission-specific time-series flight data, but are very limited. Traditional methods are usually difficult to process these data, which poses a huge challenge to FW-UAV fault diagnosis (FD). To address this problem, this paper proposed a novel Heterogeneous Deep Multi-Task Learning (HDMTL) framework with adaptive sharing and knowledge complementation for FW-UAV FD. Specifically, we first capture the temporal and spatial features in the flight data through sub-networks respectively. Then, we design a novel attention-based adaptive sharing strategy. The sharing strategy aims to transfer relevant knowledge to different sub-networks to improve their prediction accuracy through knowledge complementation. Finally, extensive experimental results show that HDMTL is significantly competitive with currently popular methods. The code and data are available at https://github.com/YizongZhang/HDMTL.