Structure-enabled vibration encoding for single-sensor UAV actuator fault diagnosis under resource constraints

Actuator health is critical to multirotor UAV flight safety. However, high-precision actuator fault diagnosis in the UAV remains challenging under constrained sensing and computational resources. This paper proposes a structure-enabled single-sensor actuator fault diagnosis paradigm based on a vibration encoder integrated into the UAV fuselage. The vibration encoder performs physical encoding at the sensing end, converting highly similar actuator-related vibration responses into more discriminative frequency-domain representations before model-based diagnosis. Theoretical analysis suggests that the encoding mechanism is attributed to the combined effect of unit-scale frequency-selective modulation and network-scale non-equivalent propagation paths. The feature separability is quantified using the average correlation coefficient and the Euclidean distance in the normalized feature space. Experimental results show that the vibration encoder reduces the average correlation coefficient among actuator encoding transmissions from 0.88 to 0.16 and increases the average normalized feature-space distance from 0.44 to 1.17, corresponding to an average distance gain of 2.66 times. Under five-fold cross-validation, lightweight SVM and LR models achieve approximately 98% average diagnostic accuracy and favorable classification stability for both propeller and motor fault tasks, while more complex MLP and 1DCNN models provide only marginal performance gains. Further evaluations show that the lightweight models retain 92%–95% accuracy with only 10% training data and maintain over 90% accuracy at SNRs of at least 3 dB. Real hovering-flight validation further yields 94%–96% accuracy and Macro-F1 for propeller fault diagnosis, with an inference time of about 0.03 ms and a model size of about (Formula presented). These results indicate that the proposed paradigm effectively balances diagnostic precision and resource consumption, providing a resource-efficient and deployable solution for single-sensor UAV actuator fault diagnosis.