Machine Learning-based Explainable Stator Fault Diagnosis in Induction Motor using Vibration Signal

The early detection of stator faults in three-phase induction motors is of great importance for modern smart industries' safety, reliability, and performance. The existing stator fault detection techniques are based on voltage-current parameters collected from the motor control system, making the process invasive and complex. To mitigate these drawbacks, a novel non-invasive data-driven-based technique using vibration signature has been proposed. The statistical data analysis technique is first used to optimize the best accelerometer mounting position for vibration data acquisition. Subsequently, the proposed method extracts several time-domain features from the collected raw data, reducing the training data by 82%. Random forest is used for stator fault identification and fault severity prediction, with a high accuracy of 99.86%. The model can also predict the faulty conditions accurately from noisy data with 99.04% accuracy, thereby proving its robustness. The Explainable AI (XAI) is used to interpret the prediction outcome and enhance the model transparency. Further, the XAI establishes the appropriateness of using vibration data for stator fault detection in the motor instead of current signals so that the entire process becomes less time-consuming and easily deployable, leading to a fundamental contribution towards Industry 4.0.