Machine Learning–Based Fault Detection for Condition Monitoring of a Three-Phase Induction Motor Using Current Signature Analysis

In developed nations, motors use between 40 and 50 percent of the total capacity produced. Induction motors need special care since they are prone to errors. There are a few systems that can identify motor defects using heat, current, and vibration analysis, but they are costly, invasive, and not appropriate for small businesses. Current Signature Analysis can be used to identify most motor defects. In this study, broken bar faults in induction motors (IMs) are analyzed using stator current and voltages to assess machine learning-based methodologies. For both healthy and faulty motors, the discrete wavelet transform (DWT) is used to retrieve the features. The experimental setup used LVDAC systems to extract stator current and voltage signals from the motor through EMS Software. Discrete Wavelet Transforms (DWT) have been applied to these signals to extract the required frequency components of those signals through MATLAB. Different machine learning models are trained to evaluate the performance for broken rotor bar defect diagnosis. Different classification techniques such as Support Vector Machines-SVM, k- nearest Neighbor-KNN, Ensemble, Decision Tree, Linear discriminant, and Nave Bayes are applied on data of stator current to find which algorithm is giving the highest percentage of efficiency. KNN classification model was providing the highest efficiency. The proposed model can detect whether the motor is healthy or if some percentage of fault has occurred in it so as to plan maintenance in the near future depending upon the percentage of fault.