Depthwise Separable Convolutional Neural Network for Fault Diagnosis in Motor of CRF Pump-Unit Using Infrared Thermal Imaging

In this study, we aim to develop an efficient lightweight deep learning approach for the classification task of infrared thermal imaging of electric motor faults. To simulate real-world conditions, we focus on motors installed in a circulating water pump unit (CRF pump-unit). With the advancement of industrial automation, accurate and rapid fault diagnosis becomes increasingly crucial for the reliable operation of equipment. We introduce a lightweight network structure based on depthwise separable convolution, which significantly reduces the number of model parameters and computational requirements, offering advantages in realtime and embedded applications. Our experimental results highlight that the proposed method has achieved significant results in the classification of electric motor faults. The model's superior performance is demonstrated on real datasets using t-SNE clustering and confusion matrices. Additionally, we analyze the model's number of parameters and computational load, emphasizing its advantages in lightweight design. In conclusion, this research presents a novel perspective in the field of electric motor fault diagnosis and provides a strong foundation for further studies and applications.