Detecção de falha em motores BLDC por meio de sinais de áudio usando Rede Neural Convolucional e SAC-DM

Abstract

Unmanned aerial vehicles (UAVs), commonly called drones, have been increasingly used to aid human tasks, with many purposes. With the increase of its usage, failures become increasingly common, which can cause damage to human life, material damage, and environmental damage. With this knowledge, a methodology is proposed to identify balance failures in the propellers of BLDC motors used in UAVs. In this work, a non-invasive method is used, which acquires the audio signal emitted by the rotating motor, performs a pre-processing step in which the SAC-DM algorithm (Signal Analysis Based on Chaos Using Density of Maxima) is applied, and then the spectrogram of the pre-processed signals is obtained to apply in a convolutional neural network. To validate the methodology, an experiment was carried out to compare the results obtained, which does not have the step of applying the SAC-DM algorithm. Also, to validate, the result obtained is compared with the results of related works to conclude the benefit of using the SAC-DM. With the experiments carried out, the results of 71.43% accuracy in the classification of audio signals from a motor with an unbalanced propeller are obtained for the SAC-DM experiment. As for the experiment that does not use the SAC-DM, there is an accuracy of 98.21% of accuracy in the classification of these same signals.