A Sound Based Method for Fault Classification with Support Vector Machines in UAV Motors

Abstract

In this study, a machine learning-based method is proposed for Brushless DC (BLDC) motors used in unmanned aerial vehicles (UAV). Shaft failure, magnet failure, propeller failure, and bearing failure are common failures in BLDC motors. These fault types are created on UAV engines. Sound recordings were taken from the engines for each failure type. While collecting the dataset, the motors were run at a constant speed. First of all, sound data was collected for the sound engine. Then, fixed time-length audio recordings were taken for 4 fault classes at a constant speed and a data set was created. This dataset consists of five classes. In order to reduce the data size in these sounds, Average Filter, Average Polling, and Normalization processes were applied, respectively. Then, the Chi2 Method was used for feature selection. In the next step, the Support Vector Machine (SVM) algorithm is used to classify the obtained features. In classification, 96.70% accuracy was calculated with the Cubic SVM algorithm.