Small Sample Fault Diagnosis Based on Intrinsic Characteristic Encouraged Network for Unmanned Aerial Vehicle Sensors

Fault detection and classification (FDC) of industrial sensors are commonly featured with rare fault cases and demanding diagnostic accuracy. Data-driven diagnostic strategies including signal processing and neural networks prove to be more prominent compared to traditional ways. Most data-driven studies on fault diagnosis pay less attention to small sample situations, which potentially erodes the practical application value. In this article, an intrinsic characteristic encouraged network (ICEN) is structured to address the aforementioned problem in the area of unmanned aerial vehicle (UAV) sensor fault diagnosis. The model of UAV sensor output with multiple fault types is realized for the set-up of elaborate small sample datasets. Specifically, gyroscopes and accelerometers are chosen as the subject sensors, and the generated samples are grouped accordingly. The designed ICEN framework incorporates the signal processing algorithm into a boosted encoder-decoder structure to strengthen the feature compactness of samples and achieve the fault diagnosis with limited data. After parameter optimization, the ICEN framework is trained and tested depending on the variety of sensors. Through comparative analysis of both software and hardware platforms, it is demonstrated that the proposed ICEN framework claims higher accuracy and faster speed in the field of FDC of UAV sensors when faced with small sample situations.