Novel deep network-based transfer learning approach for fault detection of three-phase induction motor

As induction motors in the industry operate in difficult and confined environments, condition monitoring is a critical topic to verify industry safety. This paper proposes a new hybrid application that uses a pre-trained model with machine learning classifiers for fault diagnosis using the induction motor thermal images. A deep learning system based on the EfficientNet-B0 model is developed as a feature extractor. The use of pre-trained model architecture does not need a manual attribute extraction, which can improve classification performance without a noticeable increase in computational complexity. The methodology of this work includes two steps: The first step is to develop a deep learning model as a feature extractor tool with imageNet based weight, while the second step considers the classification assessment. Thermal images have been applied to investigate the task of fault diagnosis. The achieved results have accurately proved the superiority and robustness of the proposed model. The combination of the proposed pre-trained network with a Random Forest classifier (RF) has achieved the highest accuracy of 97%. Moreover, this model might be used for other applications to detect the related class