This research introduces a foundation model approach for predicting composite material properties, addressing the challenge of scarce materials datasets through transfer learning techniques.

• Enables microstructure reconstruction and accurate prediction of mechanical properties
• Leverages pre-trained models to extract latent features from limited data
• Demonstrates significant improvements in stiffness prediction and nonlinear behavior analysis
• Provides a scalable framework for accelerating materials engineering with fewer training samples

For engineering applications, this breakthrough means faster material development cycles, reduced testing costs, and more efficient design of composites with specific mechanical properties.

Foundation Model for Composite Microstructures: Reconstruction, Stiffness, and Nonlinear Behavior Prediction