A data-driven framework for structure-property correlation in ordered and disordered cellular metamaterials

This repository provides data and associated codes for a data-driven framework that enables prediction of macroscopic properties of 2D cellular metamaterials, and identifies their connection to key morphological characteristics, as identified by the integration of machine-learning models (Random Forest and GAM) and interpretability algorithms (SHAP analysis).

Table of Contents

• Data
• Code Overview
• Prerequisites
• Contributors
• Contact

Data

Microstructure Data

The microstructural data contains 1646 different tessellations, both ordered and disordered. For visualization, each tessellation is represented by the corresponding nodes and connectivity (in the Tessellation Dataset folder), and two demos are provided to display the tessellation and/or microstructure.

Tessellation_Demo.m Display of tessellation for a certain sample in the dataset.

Microstructure_Demo.m Display of microstructure for a certain sample for a given relative density in the dataset.

Structure-Property Data

The structure-property data contains 42 microstructural features, the corresponding effective stiffness for 1646 different tessellations at 5 different relative densities, which in total consists of 8230 microstructures with 43 parameters.

Structure-Property-Data.csv Each column represents a property, as listed in the header. Every five rows correspond to a distinct topology with five relative densities, using the same sample index as in Tessellation Dataset.

Code Overview

1. Virtual microstructure generation Generation of cellular microstructures.
2. Feature and property measurement Extraction of features and stiffness calculation for each cellular metamaterial in the dataset.
3. Data Preparation Generation of accessible data files for machine learning algorithms.
4. Random Forest and SHAP Random forest regression model for stiffness prediction and utilization of SHAP analysis for structure-property correlation.
5. Generalized Additive Model Generalized additive model for stiffness prediction and structure-property correlation.

Prerequisites

• Matlab (R2020a or later, full toolbox installation recommended)
• Simulia Abaqus (2021)
• Python (3.8 or later)
• R (3.6.1 or later)

Contributors

Shengzhi Luan, Enze Chen, Joel John, Stavros Gaitanaros

Contact

For any further information, please feel free to contact Shengzhi Luan ([email protected]) or Stavros Gaitanaros ([email protected]).