Deep Transfer Learning-Based Broken Rotor Fault Diagnosis For Induction Motors

Abstract

Due to their starting and running torque needs as well as their four-quadrant operation, modern industrial drives utilise induction motors (IM). Failures in the rotor bars of the motor can be found using the voltages and currents of each of the three phases as well as the acceleration and velocity signals. For the diagnosis of the quantity of broken rotor bars for a failed IM, conventional signal processing-based feature extraction techniques and machine learning algorithms have been applied in the past. The number of broken rotor bars is determined in this study by looking into a novel technique. For the aforementioned aims, specifically, the deep learning methodologies are studied. In order to do this, convolutional neural network (CNN) transfer learning algorithms are described. Initially, a bandpass filter is used for denoising, and then the signals are transformed using the continuous wavelet transform to create time-frequency pictures (CWT). The collected images are used for deep feature extraction and classification using the support vector machine (SVM) classifier, as well as for fine-tuning the pre-trained ResNet18 model. Metrics for performance evaluation employ categorization accuracy. Additionally, the results demonstrate that the deep features that are recovered from the mechanical vibration signal and current signal yield the greatest accuracy score of 100%. Nonetheless, a performance comparison with the publicly available techniques is also done. The comparisons also demonstrate that the proposed strategy outperforms the compared methods in terms of accuracy scores.