Semi-MCDA-Net: A novel class-specific cross-domain fault diagnosis technique under time-varying speed conditions with limited labeled data

Abstract

Time-varying rotational speed conditions cause domain shifts in rotating machines, resulting in discrepancies between training and testing datasets, preventing transfer learning models operating at constant speed conditions from detecting invariant features and reducing their generalization efficacy. Moreover, the current transfer learning methods for varying speed scenarios mainly focus on aligning the marginal data distribution while neglecting the influence of class-specific feature alignment on the diagnostic accuracy. To address these limitations, this research develops a novel end-to-end semi-supervised marginal and conditional distribution alignment network (Semi-MCDA-Net) to deal with the issues of fault diagnosis under variable speed working conditions, particularly in the context of insufficient labeled data in the target domain. Our method integrates multi-kernel Maximum Mean Discrepancy (MMD) and Wasserstein distance (WD) to develop a unified domain alignment module, systematically applied across multiple convolutional layers of a shared 1D-CNN. In contrast to conventional transfer learning approaches, the proposed methodology effectively acquires features that are both domain-invariant and class-discriminative, explicitly aligns both marginal and conditional distributions, and thereby generalizes to data in the target domain while improving accuracy near the class decision boundaries. Two case studies are carried out to verify the efficacy and generalizability of Semi-MCDA-Net method.