Fault Diagnosis of Drone Motors Driven by Current Signal Data with Few Samples

Abstract Multirotor UAVs find extensive application across diverse domains, with their motors serving as pivotal components in the UAV power system. The majority of UAV failures and crashes stem from motor malfunctions, underscoring the critical need for research in UAV motor fault diag-nosis. This paper focuses on quadrotor UAVs employing DC brushless motors as the research subject. The study designs fault experiments for three key components—stator, rotor, and bear-ing—based on the structural and operational characteristics of these motors. Addressing chal-lenges such as limited installation space for UAV payloads and redundant components, difficul-ties in sensor installation, high costs, and data acquisition complexities, this research adopts the current signal as the diagnostic signal. It explores a UAV motor fault diagnostic method based on the current signal. Furthermore, the study investigates a motor fault diagnosis method based on the current signal. To overcome challenges arising from the UAV's heightened sensitivity to the health of its components and the limited availability of fault data training samples, traditional machine learning and deep learning methods face difficulties in identifying representative fea-tures under small samples, often succumbing to overfitting risks and resulting in low diagnostic accuracy. To address these issues, a hybrid neural network fault diagnosis model is proposed, integrating the width learning system and convolutional neural network. The width learning system maps the original current signal into a feature space, yielding more robust and repre-sentative sample features. Subsequently, the convolutional neural network undertakes feature extraction and classification tasks. In experiments focused on current signal data-driven small-sample fault diagnosis in UAV motors, the proposed model outperforms other models used for comparison.