A Fault Detection Method Based on Embedded Feature Extraction and SVM Classification for UAV Motors

Abstract

In this study, a lightweight method has been proposed for the early detection of faults in UAV motors. To test our method, balance, magnet, propeller, and bearing failure have been created in 820KV UAV motors. UAVs can have single or multi motors. Therefore, helicopter, duocopter, tricopter, and quadcopter models have been created. Sounds have been collected by installing healthy and faulty motors on these models. Sound datasets consisting of five classes have been obtained for four cases (helicopter, duocopter, tricopter, and quadcopter models). Feature extraction has been performed on the collected sound signals using Mel-frequency Cepstral Coefficients (MFCC) method. Support Vector Machines (SVM) are used to perform fault classification on selected features. In the proposed method, 100% accuracy has been computed for helicopter and duocopter models. In the tricopter and quadcopter models, 99.06% and 90.53% accuracy have been calculated, respectively. Since the proposed method is based on sound, it can detect many types of faults. In addition, since our method is light-weight, it is in an architecture that can work in real-time in embedded systems.