FCDG: A Central Dogma-Inspired Approach for Cross-Domain Fault Diagnosis

Although deep learning-based detection technology has significantly improved the efficiency and accuracy of fault diagnosis, its development is limited by the differences in sample distribution caused by operating condition changes. To address the challenge, a feature codon intelligent bearing fault diagnosis method feature codon domain generalization (FCDG) with wide-area generalization capability is proposed. FCDG introduces the concepts of feature codons and feature anticodons and proposes a new fault diagnosis method based on the nonprobabilistic modeling of feature codons. That is, feature anticodons and feature codons are extracted from the source domain and target domain, respectively, and the feature anticodons and feature codons are matched and translated into the corresponding fault category. The paradigm effectively integrates the distribution of feature codons in the source domain to infer the class affiliation of samples in the target domain and expands the theoretical framework of domain generalization (DG) fault diagnosis by introducing the matching of feature codons and feature anticodons. The classification method of feature codon matching replaces the traditional classification, avoiding the influence of classification boundary failure caused by domain shift. A large number of experimental results demonstrate the effectiveness of the proposed FCDG under various variable operating conditions.