Artificial Neural Network for Rotor Fault Diagnosis in an Asynchronous Traction Motor Using Magnetic Induction Parameters

This paper presents a hybrid approach to the diagnosis of the rotor condition in an asynchronous traction motor. This approach is based on the analysis of magnetic induction parameters. Using experimental data obtained in CSV format, a model was constructed in Python environment employing an artificial neural network. Methods: In the initial phase, data preparation included numerical value standardisation and the elimination of extraneous noise. Subsequently, spectral analysis of magnetic induction signals was performed using the Fast Fourier Transform (FFT) to extract diagnostically significant frequency components. Results: The neural network architecture has been implemented using Keras and comprised two hidden layers. The model was trained on a dataset comprising 100 samples, and it has demonstrated high accuracy (up to 98%) on test data. A comparative analysis of magnetic field characteristics was carried out for defective and non-defective cases. The findings have demonstrated that the proposed approach is capable of effectively detecting rotor cracks while minimizing the necessity for complex measurement setups. Practical significance: The work has been carried out exclusively through the utilization of open-source tools.