对偶(语法数字)
计算机科学
断层(地质)
算法
故障检测与隔离
控制理论(社会学)
扩散
噪音(视频)
信号处理
噪声测量
数值模型
人工智能
故障指示器
电子工程
算法设计
矩阵代数
作者
Yiyao An,Ke Zhang,Zhiqin Zhu,Yi Chai,Shen Yin,Simon X. Yang
标识
DOI:10.1109/tii.2026.3712033
摘要
A heat diffusion constrained prototypical dual contrastive network is proposed to address few-shot open-set fault diagnosis. A heat diffusion constrained graph domain adaptation module is proposed to address negative transfer caused by incomplete label space overlap in open-set scenarios with uncovering latent manifold structures. By embedding an energy minimization criterion into the diffusion dynamics, the model is driven to converge toward a low-energy stable manifold, enabling physics-inspired interpretability of the transfer behavior. In parallel, a progressive prototypical dual contrastive module is designed to extract informative features from limited samples and learn well-separated representations for unseen fault categories, enabling estimation of unknown classes and alignment with appropriate cluster hierarchies. Extensive experiments validate the superior generalization ability of the proposed model.
科研通智能强力驱动
Strongly Powered by AbleSci AI