Deep learning for synthetic microstructure generation in a materials-by-design framework for heterogeneous energetic...

Area of Science:

• Materials Science
• Computational Science
• Chemical Engineering

Background:

• The sensitivity of heterogeneous energetic (HE) materials is intrinsically linked to their microstructure.
• Hot spots, formed by energy localization at defects like porosities, initiate chemical reactions in HE materials.
• Meso-scale models are crucial for predicting HE material response, integrating physics from statistically representative microstructural features.

Purpose of the Study:

• To develop a method for generating synthetic microstructures of HE materials.
• To enable the creation of large ensembles of microstructures for robust model development.
• To facilitate the design of novel HE materials with engineered microstructures for specific performance.

Main Methods:

• Utilized generative adversarial networks (GANs) to learn from images of HE microstructures.
• Generated ensembles of synthetic HE material microstructures.
• Demonstrated control over porosity distribution and spatial manipulation within generated microstructures.

Main Results:

• The GAN method produced qualitatively and quantitatively realistic HE microstructures.
• The approach successfully generated new morphologies with controllable porosity.
• The generated microstructures can be used to inform microstructure-dependent energy deposition models.

Conclusions:

• GANs offer a powerful tool for generating diverse and realistic HE material microstructures.
• Controlled manipulation of microstructure via GANs supports a materials-by-design framework for HE materials.
• This work advances the predictive modeling and engineering of energetic materials.