A Deep Transfer Nonnegativity-Constraint Sparse Autoencoder for Rolling Bearing Fault Diagnosis With Few Labeled Data

Abstract

Rolling bearing fault diagnosis can greatly improve the safety of rotating machinery. In some cases, plenty of labeled data are unavailable, which may lead to low diagnosis accuracy. To deal with this problem, a deep transfer nonnegativity-constraint sparse autoencoder (DTNSAE) is proposed, which takes advantage of deep learning and transfer learning. First, a novel NSAE is adopted to enhance sparsity. Then, a base deep NSAE (DNSAE) is established to automatically capture the latent features from raw vibration signals. Next, a parameter transfer learning strategy is used to build the DTNSAE to tackle the diagnosis problems with a few labeled data. Finally, two datasets from different domains are used to verify the effectiveness of the proposed method. The testing results suggest that the proposed method is able to remove manual feature extraction and is more effective than the existing intelligent methods when only a few labeled data are available.

Keywords

• Rolling bearing
• fault diagnosis
• deep transfer nonnegativity-constraint sparse autoencoder
• parameter transfer learning
• few labeled data