Analysis of Brushless DC Motor Sounds with Machine Learning Methods Using Wavelet Transform Based Features

Abstract

Brushless DC (BLDC) motors are widely used in various applications due to their high efficiency, reliability and low maintenance requirements. The absence of mechanical brushes reduces wear and minimizes maintenance. These features make them preferred in many areas, especially industrial automation, electric vehicles, and robotic systems. Integration of a BLDC motors with machine learning (ML) can significantly increase the efficiency, reliability and performance of these motors. ML algorithms can help detect faults in advance by analyzing the performance data of the motor. ML algorithms, which monitor deviations from the normal operating conditions of the motor, can quickly identify faulty situations. ML can learn the most efficient operating points depending on the operating conditions of the motor and dynamically optimize the speed or other parameters of the motor accordingly. In this study, a method is proposed that enables the detection of mechanical faults in a BLDC motors with sound analysis. With sound analysis, Discrete Wavelet Transform (DWT) based features were extracted from the sound recordings of normal and faulty motors and the obtained features were classified with machine learning methods. Here, the data size is reduced with DWT, unwanted and unimportant coefficients are suppressed. Bagging trees are used to avoid overfitting with extracted statistical features. Bagging tries to balance the overfitting tendency of each tree by combining multiple decision trees and the generalization capacity of the model increases. In addition, since each model is trained independently, it allows parallel calculation. With the obtained model, 89.205% accuracy and 0.821 kappa value were obtained.