Imbalance fault detection in BLDC motors through audio signals using the density of maxima and CNN

• Yuri C. Gouveia UFPB
• Alisson V. Brito UFPB
• Tiago P. Nascimento UFPB
• Jorge Gabriel Ramos UFPB
• Abel Lima Filho UFPB

Abstract

Unmanned Aerial Vehicles (UAVs), commonly known as drones, are increasingly used for various tasks. However, increased usage leads to more failures, posing risks to humans and the environment. This study identifies propeller imbalance faults in Brushless DC (BLDC) motors of UAVs. It compares two fault detection methodologies based on chaotic behavior: the alreadyknown CNN approach, and the SAC-DM+CNN approach. Both methods involve acquiring audio signals, generating spectrograms, and training a Convolutional Neural Network (CNN) model. Results show that the CNN approach achieves 98.18% accuracy, while the SAC-DM+CNN approach achieves 71.43% accuracy in classifying propeller balance faults.

Keywords: Brushless DC, spectrograms, Convolutional Neural Network, chaotic behavior