The paper introduces H-mPFDNN, a hierarchical extension of the Material-Property-Field-based Deep Neural Network (mPFDNN), designed to predict material properties across multiple length scales. It includes a scale-aware feature aggregation mechanism, an emergent property predictor, and cross-scale coupling operators to model interactions between microstructural features and mesoscale fields.
Key findings
H-mPFDNN bridges atomistic representations with mesoscale emergent behavior through a multi-level interaction framework.
Introduces scale-specific interaction fields to capture intra-scale correlations and cross-scale couplings.
Develops an emergent property predictor module to model collective phenomena arising from hierarchical organization.
Proposes a physics-informed attention mechanism for efficient information propagation across hierarchical levels.
Limitations & open questions
The paper is a research proposal and does not include experimental results or validation against real-world data.