Dynamic Distribution Adaptation Based Transfer Network for Cross Domain Bearing Fault Diagnosis

Abstract In machinery fault diagnosis, labeled data are always difficult or even impossible to obtain. Transfer learning can leverage related diagnosis knowledge from fully source domain enhance the performance in sparsely unlabeled target domain, which has been widely used for cross diagnosis. However, existing methods focus on either marginal distribution adaptation (MDA) conditional (CDA). practice, and distributions discrepancies both have significant but different influences divergence. this paper, a dynamic based transfer network (DDATN) is proposed bearing DDATN utilizes instance-weighted maximum mean discrepancy (IDMMD) (DDA), dynamically estimate of adapt with domain. The experimental evaluation demonstrates that outperformance state-of-the-art methods.