New AI Framework Improves Fault Diagnosis in Manufacturing

A new framework named deep hierarchical knowledge loss (DHK) has been developed to improve fault intensity diagnosis (FID) within intelligent manufacturing. This method tackles the challenges posed by neglecting the interdependencies among target classes, which has limited its practical application. Researchers introduced a hierarchical tree loss to establish a comprehensive mapping for classes with shared characteristics, utilizing tree-based positive and negative hierarchical knowledge constraints. To enhance adaptability, a focal hierarchical tree loss was created alongside two adaptive weighting methods based on tree height. Furthermore, a group tree triplet loss featuring a hierarchical dynamic margin was proposed, integrating hierarchical group concepts and tree distance to represent boundary structural knowledge across various classes. The combination of these loss functions significantly enhances the detection of subtle faults. This framework was detailed in a paper released on arXiv, identified as arXiv:2604.16459v1, under the category of cross announcement. Extensive experiments were carried out to confirm the approach's effectiveness.