ENHANCING RESILIENT OPERATIONS VIA PREDICTIVE MAINTENANCE: A MACHINE LEARNING APPROACH FOR BEARING FAULTS PREDICTION IN INDUCTION MOTORS

This thesis presents a machine learning based approach for bearing fault prediction in induction motors to enhance the resilience of operations in industrial facilities. Squirrel cage induction motors are commonly used in industrial facilities to supply critical loads. Therefore, the failure in these induction motors leads to unplanned downtime and production losses. The study investigates bearing fault prediction through utilizing a Feed-forward Neural Network (FFNN) to predict current and speed signals that are then fed into a Random Forest classifier to identify bearing faults in the predicted signals. Both models are implemented, integrated, and tested in MATLAB and Simulink using a simulated induction motor. The results demonstrate the system's ability to predict faults accurately, and the addition of a set-reset logic to minimize redundant alarms supports its practical applicability in industrial facilities. Future work can investigate the use of different machine learning techniques to forecast signals and detect faults, and the system can be tested in a laboratory or industrial settings to validate its performance and reliability.