Computational Generation of Virtual Concrete Mesostructures.

Area of Science:

• Materials Science
• Computational Mechanics
• Civil Engineering

Background:

• Concrete's disordered morphology significantly impacts its behavior.
• Accurate simulation of concrete requires realistic mesoscale models.

Purpose of the Study:

• To introduce the Concrete Mesostructure Generator (CMG) for creating ultra-realistic virtual concrete.
• To enable advanced mesoscale and multiscale computational modeling of concrete.

Main Methods:

• Generating polyhedral aggregates with concave depressions from cuboids.
• Assembling aggregates using a hierarchical random sequential adsorption algorithm.
• Calibrating virtual mesostructures with laboratory aggregate distributions.

Main Results:

• Validated virtual concrete mesostructures against laboratory measurements.
• Compared elastic properties from computational homogenization with experimental data.
• Trained a 3D-convolutional neural network to predict elastic properties from voxel data.

Conclusions:

• The CMG tool generates validated, realistic virtual concrete mesostructures.
• Computational homogenization of virtual models accurately predicts concrete's elastic properties.
• Machine learning offers a direct route to predict material properties from morphology.