An open set domain adaptive based generic fault diagnosis framework for marine power unis

To address the limitations of traditional fault diagnosis models for marine power units, which focus on single working conditions, lack generality, and cannot identify unknown faults, this paper proposes a general fault diagnosis model based on open domain adaptation-the Consensus Separation Subdomain Adaptive Network (CSSAN). The model first applies the Consistent Consensus Separation strategy to perform consistency discrimination and domain consensus clustering, adaptively isolating unknown fault samples under varying working conditions without human intervention. Next, the Local Subdomain Adaptive strategy is introduced, leveraging the separated unknown samples to define the decision boundary for unknown faults. This strategy also learns fine-grained subdomain information to improve the matching of public class samples across different conditions and facilitate cross-domain knowledge transfer. Ultimately, this model enables the diagnosis of known fault classes and the identification of unknown fault classes under unknown working conditions. In the experimental section, the effectiveness and generalizability of CSSAN are validated using the PU bearing dataset and a marine engine dataset, representing two distinct types of marine power units. A series of open-set, cross-condition fault diagnosis tasks were designed to assess the model's performance. The experimental results demonstrate that the proposed CSSAN model exhibits superior diagnostic performance across different types of marine power units and varying degrees of openness in cross-working condition fault diagnosis tasks. It successfully diagnoses known faults and detects unknown faults under unknown working conditions, offering a novel solution for cross-working condition fault diagnosis in marine power units.