DVMTICANET: A Synthetic Multisource Interactive Adaptation Framework for Intelligent Fault Diagnosis Under Large Cross-Domain Discrepancies

Multisource transfer fault diagnosis is currently a hot topic in the field of intelligent fault diagnosis (IFD). However, in practical scenarios, large cross-domain discrepancies usually exist among datasets of different working conditions, which brings more difficulties and challenges to multisource transfer fault diagnosis. To solve the tough issue, a synthetic multisource interactive adaptation framework is proposed in this article. First, in order to acquire satisfying multisource transfer fault diagnostic performances under large cross-domain discrepancies, the entire framework is established based on strategies of domain shared level adaptation, domain specific level adaptation, and personalized diagnostic knowledge learning. Second, a distinctive domain specific transferability metric is designed and utilized to induce these three strategies to better facilitate the positive transfer of the entire framework. Third, the deep convolutional variational autoencoder (DCVAE) is selected as the shared feature extractor, which can help preliminarily narrow the large cross-domain discrepancies due to its strong feature distribution regularization ability. Eventually, the experiments based on two different bearing datasets are conducted to verify the superiority and effectiveness of the proposed method, especially when the cross-domain discrepancies are quite large.