Roller Bearing Fault Diagnosis Using Deep Transfer Learning and Adaptive Weighting

Abstract A fault diagnosis approach for roller bearings utilizing deep transfer learning and adaptive weighting is suggested to address the issue that extra state samples in target domain data of impair diagnostic accuracy. CNN-LSTM a network model proposed by Lecun et al., which has good performance image processing processing. It can effectively apply predictive local perception time series weight sharing CNN, greatly reduce number networks improve efficiency learning. The method first establishes feature extraction module, uses convolutional neural map bearing high-dimensional space. Secondly, concept design weighted discriminator, adaptively weights samples; finally, through confrontation space, are classified. Training increase similarity healthy shared source domain. Then measuring between these based on sample size, setting threshold label additional domains as unknown. technique validated using gearbox data, from Case Western Reserve University, locomotive wheel data. accuracy less than 80%, suggesting successfully overcome effects diagnose faults.