Unified Approach to Detection of Induction Motor Faults from Vibration Signals using Hilbert Transform based Autoencoder Features and Hybrid Efficient Machine Learning Algorithms

Since vibration signals are sensitive to changes, they are regularly used to assess the condition of rotating electrical machines. These signals include vital data that makes it possible to find mechanical problems and identify unusual behaviors. Revealing important patterns from information in vibration data is now greatly assisted by machine learning. Once faults have been prepared and processed properly, ML systems are able to identify both the type and severity of each fault. Three-phase induction motor vibration signals are very important for fault diagnosis to ensure reliable operation and minimize downtime. In this study, vibration signals are collected and pre-processed by the Hilbert Transform to eliminate noise and outliers. The pre-processed signals are fed to an Autoencoder for feature extraction and subsequent classification by effective learning algorithms. Among the models under study, the proposed Hybrid RF-GMM model provided the highest classification accuracy of 95.91%, outperforming RF (92.28%) and GMM (88.18%), reflecting its excellence in fault detection.