Research on gear cross-condition fault diagnosis based on transfer learning

This paper proposes a method for cross-condition fault diagnosis of gears that combines the Convolutional Block Attention Module (CBAM) with subdomain adaptation networks. The channel submodule employs both the maximum pooling output and the average pooling output from the shared network. Similarly, the spatial submodule utilizes analogous outputs, which are pooled across the channel axis and then passed to convolutional layers. The advanced subdomain adaptation network enhances the functionality of the Deep Adaptation Network (DAN) by capturing precise information for each category, synchronizing the overall distributions of source and target domains, and coordinating the distributions of relevant subdomains. Through comparative experiments, our proposed method outperforms Convolutional Neural Networks (CNN), Deep Adaptation Network (DAN), and Deep Subdomain Adaptation Network (DSAN).