Learning ordinality-aware multimodal representations for composite materials design

Area of Science:

• Materials Science
• Artificial Intelligence
• Machine Learning

Background:

• Composite material design relies on understanding complex microstructures, often requiring AI integration.
• Existing multimodal learning frameworks struggle with the continuous, nonlinear design spaces of composites, especially under data scarcity.
• Current methods are primarily developed for crystalline or polymer systems with discrete structure-property relationships.

Purpose of the Study:

• To develop a novel multimodal learning framework for composite material design that addresses data scarcity and continuous design spaces.
• To introduce ordinality as a core principle for building robust multimodal representations in composite materials.
• To enable accurate property prediction, cross-modal retrieval, and microstructure generation for composites.

Main Methods:

• Proposed ORDinal-aware imagE-tabulaR (ORDER) alignment to integrate microstructural images with tabular material descriptors.
• Utilized physics-based surrogate signals to reduce the reliance on complete property annotations.
• Ensured that similar target properties are mapped to nearby regions in the latent space, preserving continuity.

Main Results:

• ORDER effectively integrates heterogeneous data sources (microstructures and tabular data) for composite design.
• The framework demonstrated superior performance compared to baseline methods in property prediction, cross-modal retrieval, and microstructure generation.
• ORDER successfully handles data scarcity and preserves the continuous nature of composite properties, enabling interpolation.

Conclusions:

• Ordinality is a key principle for effective multimodal representation learning in composite materials.
• The ORDER framework offers a powerful solution for data-scarce composite design challenges.
• This approach advances the application of AI in designing novel composite materials with desired properties.