Multisource-Refined Transfer Network for Industrial Fault Diagnosis Under Domain and Category Inconsistencies

Abstract:

Unsupervised cross-domain fault diagnosis has been studied recently. It learns transferable features that reduce distribution inconsistency between source and target domains without target supervision. Most cross-domain fault diagnosis methods assume source and target fault category set consistency. Different working conditions have different fault category sets, which challenges this assumption. This article proposes a multisource-refined transfer network for fault diagnosis under domain and category inconsistencies. First, a multisource-domain-refined adversarial adaptation strategy reduces refined categorywise distribution inconsistency within source–target domain pairs. It avoids global-domainwise-forced alignments’ negative transfer trap. To leverage diagnostic knowledge from multiple sources, a multiple classifier complementation module is created by complementing and transferring source classifiers to the target domain. The adaptation module’s similarity scores complement different classifiers, and smooth predictions guide refined adaptation. Thus, in the training stage, refined adversarial adaptation and classifier complementation can produce target-faults-discriminative and domain-refined-indistinguishable feature representations. The proposed method outperforms domain and category inconsistencies in two cases.

Note: Please discuss with our team before submitting this abstract to the college. This Abstract or Synopsis varies based on student project requirements.