Bearing Fault Diagnosis: An Efficient Architecture Based on CNN and Transfer Learning

This paper presents an effective architecture for bearing fault diagnosis based on convolutional neural networks (CNN) and transfer learning. Unlike traditional methods that rely on manual feature extraction, the proposed approach integrates automatic feature extraction and pattern classification within a single model. The approach utilizes transfer learning techniques by incorporating SqueezeNet v1.1, a pre-trained deep learning model trained on ImageNet, to classify wavelet scalogram images of bearing faults. Transfer learning facilitates faster and simpler model training by enabling the transfer of learned features using a reduced number of training images.