Inclined Eccentricity Fault Diagnosis in Induction Motors Using Deep Learning

Deep learning recently has proven effective for fault diagnosis in induction motors. Compared to traditional methods, it offers better accuracy and provides an automated process for fault detection. However, a key challenge with using deep learning model is the collection of data, which can be difficult and expensive to obtain on real-world motors. To address this issue a precise simulation model can be used, replicating the induction motors under various fault conditions and thus, data can be generated for various motor configuration and fault conditions. Therefore, in this paper an analytical model based on the modified winding function method is employed to simulate inclined eccentricity faults in a wound rotor induction motor. These faults are characterized by nonuniform air gap and axial misalignment. The data generated from this analytical model is then used to train a five-layer feedforward neural network. To validate the model's performance, predictions are made using data acquired from a scaled-down 2.5 MW wound rotor induction motor, where controlled eccentricity is introduced via a dual screw mechanism. Results demonstrate that the deployed model achieves 97 % accuracy in predicting fault occurrence within this experimental setup.