BI-Sandwich: a cross-domain model for fault diagnosis of aircraft engine main shaft bearings

Abstract

As a core component of the aeroengine, the main shaft bearing faces significant challenges in practical fault diagnosis, including the difficulty of simultaneously capturing both local and global features during feature extraction, and the scarcity of high-quality labeled data under varying operating conditions, which limit the generalization ability of diagnostic models. To address these issues, this article proposes a lightweight transfer learning-based fault diagnosis method built upon a Bi-Sandwich architecture and maximum mean square difference (MMSD). A novel Bi-Sandwich model is designed by integrating multiscale separable convolutions and a broadcast attention mechanism, enabling effective extraction of both local and global features while reducing computational complexity. To mitigate domain discrepancies under different working conditions, an MMSD-based domain adaptation metric using second-order statistics is introduced to align feature distributions between the source and target domains. Experimental results on two aero-engine main shaft bearing datasets demonstrate that the proposed method achieves excellent fault identification accuracy and cross-domain generalization across different loads and rotational speeds.