面 向 民 机 跨 域 故 障 诊 断 的 多 源 域 层 级交 互 图 融 合 原 型 网 络

Extracting diverse and correlated transferable knowledge from multiple domains for cross-domain diagnosis in unlabeled civil aircraft target scenarios is a critical and pressing need. Graph-based multi-source domain adaptation is an effective solution to this problem. However， existing methods are inaccurate in extracting multi-domain features and have insufficient depth of feature interaction， which hinders the performance of cross-domain fault diagnosis. To address these limitations， a multi-source domain hierarchical-interactive graph fusion prototype network (MDHGFPN) is proposed for civil aircraft cross-domain fault diagnosis. A de-redundancy global prototype network is built to iteratively refine the global prototype features from both spatial and channel dimensions， ensuring accurate multi-domain prototype embedding to guarantee the construction of high-quality knowledge graphs. A hierarchical-interactive graph fusion strategy is constructed to model a wider range of neighborhoods without restriction. This strategy fully exploits multi-domain knowledge interactions，adaptively fusing neighborhood information most beneficial to the target node， thereby achieving effective cross-domain fault diagnosis. Experimental cases show that MDHGFPN is effective and superior in cross-domain fault diagnosis tasks for civil aircraft in unlabeled scenarios.