Classification of Electric Motors Faults Using Fourier-Based Features and Self-Organizing Maps

This research has been carried out to improve the identification of faults in electric motors using data processing techniques. The acoustic and vibration data were divided into smaller overlapping segments to allow a more detailed analysis of frequency and time characteristics. Fast Fourier transform (FFT) and short-time Fourier transform (STFT) frequency analyzes were used to identify defects within one-second interval. A comparison with previous studies showed that this approach significantly increases the accuracy of defect identification. The self-organizing map model was used for complex data clustering and multidimensional analysis to identify hidden patterns, and it performed successfully. Also, the models of the nearest neighbor, support vector machine, decision tree, naive bays, and random forest showed high accuracy in separating the classes. The results show the effectiveness of FFT and STFT techniques in identifying frequency patterns, and better results can be achieved with proper pre-processing and accurate selection of frequency ranges and overlapping windows. This approach not only accurately detected various faults but also evaluated the severity of faults well, which has a high potential for use in predictive maintenance and monitoring systems.