DRAGen - A deep learning supported RVE generator framework for complex microstructure models.

Area of Science:

• Computational Materials Science
• Materials Informatics
• Multiphysics Simulation

Background:

• Representative Volume Elements (RVEs) are crucial for accurate material modeling.
• Generating RVEs with realistic microstructural features is computationally challenging.
• Existing frameworks may lack the flexibility to incorporate diverse microstructural complexities and solver compatibility.

Purpose of the Study:

• To present an enhanced Discrete RVE Automation and Generation Framework (DRAGen).
• To enable the generation of RVEs with complex and realistic microstructural features.
• To expand DRAGen's compatibility with various computational materials engineering (ICME) solvers.

Main Methods:

• Incorporated a generator for RVEs with complex microstructure features extracted from real materials.
• Enabled input data processing from trained neural networks and .csv files.
• Added modules for generating input files for Finite Element (FE) and spectral solvers.

Main Results:

• DRAGen can now reconstruct diverse features including pores, inclusions, martensite bands, hierarchical substructures, and crystallographic textures.
• The framework generates input files for DAMASK, Abaqus, and MOOSE multiphysics frameworks.
• Evaluations and simulations confirm DRAGen's power and flexibility for ICME research.

Conclusions:

• The enhanced DRAGen is a versatile tool for generating realistic RVEs with complex microstructures.
• Its modular design allows for easy expansion and adaptation.
• DRAGen provides researchers with a broad spectrum of microstructures for studies and design developments.