Convolutional Neural Network based Diagnosis of Electric Rotating Machines using Field Sensor Signals

The safe operation of power substation equipment is fundamental for guaranteeing the performance and reliability of the power systems. However, the electric equipment and devices are confirmed prone to component failure and breakdown that may directly lead to the power outage. In this paper, a cost-effective solution based on the convolutional neural network is presented for the analysis of faults of electric rotating machines. To reinforce the robustness under a noisy environment, a linear discriminant criterion based metric learning technique is also employed to improve the loss function during the training process. The developed approach can automatically extract self-learned fault features and conduct fault diagnosis. The developed solution has been carefully evaluated through simulation and experiments to quantify the performance. The experimental results demonstrated the effectiveness of the proposed solution for fault diagnosis.