Thrust-level dependent vibration diagnostics of UAV propeller using fast Fourier transform and K-nearest neighbour

Abstract

A big majority of UAV failures link to its motors, propellers, or any other mechanical parts. Early diagnosis of problems related to these parts can play a crucial role in preventing incidents. In this study, vibration data from damaged and undamaged UAV propellers is captured using a three-axis accelerometer. Then, raw vibration data features are extracted using the fast Fourier transform in order to analyse the frequency components of the signal. These features are used to train a machine learning model using the K-nearest neighbour algorithm and obtain good performance with an accuracy of 81.90% in classifying damaged and undamaged propellers in a total of three different thrust levels. Results are promising in diagnosing abnormal behaviours on propellers using vibration data and diminishing propeller-related failures.