Comparing the Effectiveness of Two Convolutional Neural Networks Methods on Fault Diagnosis

Fault diagnosis is an important branch of modern control system and plays an important role in industrial production. Researchers keep pursuing more convenient and practical methods for fault diagnosis. As data collection becoming more convenient, data-driven methods develops rapidly for their excellent performance, especially the deep learning methods have been a popularity way to fault diagnosis. This paper tries to use two data-driven methods ( ResNet-50 and SE-ResNet-50 ) on fault diagnosis without transfer learning. More uniquely, training data and testing data are collected under different conditions. SE-ResNet-50 achieves the highest accuracy 99.1%, which is better than ResNet-50 obviously. This experiment shows that SE-ResNet-50 achieves good performance without large training samples in fault diagnosis.