Implementation a Multi-Fault Diagnosis System for Three-Phase Induction Motors Using Machine Learning Techniques

Induction motors (IMs) have been a cornerstone in industrial applications for decades due to their robustness, reliability, and efficiency. However, as the reliance on these motors has increased, so has the necessity for ensuring optimal performance and minimizing downtime. This study applies machine learning (ML) algorithm to diagnose three-phase induction motors with multiple faults. Consequently, the development of advanced fault diagnosis methods has become vital. The research aims to establish a comprehensive detection technique. It includes artificially creating a total of 12 common fault types in induction motors, such as rotor unbalance, stator short circuit, rotor broken bars, and composite failures of three, as the targets for ML diagnosis and judgment. By optimizing the data preprocessing method using the fast Fourier transform (FFT) and validating classifiers with k-nearest neighbors (KNN), the study's results show that the preprocessing method optimized with the FFT and standard deviation can effectively improve the prediction accuracy and reduce the diagnosis time. Moreover, this method exhibits superior performance in real-time fault diagnosis applications. After the FFT processing, KNN accuracy improved from 87.43% to 100%, and classification time reduced by 83.9%, from 97.61 seconds to 15.77 seconds. This method not only boosts accuracy but also shortens classification time, enhancing diagnostic efficiency. FFT significantly improves the model's ability to handle new data.