Comparison of Bearing Fault Classification Using STFT, CWT and 2D-DOST with a Lightweight CNN Model

Bearing fault diagnosis in electric motor is an important research area due to the impact of these elements on the whole kinematic chain the motor is a part of. Because of this, many methodologies have been developed, mostly based on frequency components in vibration signals, with a growing interest in the use of Deep Learning techniques. This work focused on the measurement of stray magnetic flux from an induction motor to detect different bearing conditions by applying time-frequency transforms and a Convolutional Neural Network as classification tool. To expand the knowledge on the area, a texture analysis tool was applied to the spectrograms to study its effects on feature extraction, achieving a comparison of the performance with four case scenarios, improving classification accuracy over 99% with minimal training time increment.