Session: 01-05-02: Applications of Artificial Intelligence/Machine Learning for Aerospace Structures 2

Paper Number: 161652

161652 - Machine Learning-Assisted Multiscale Modeling for Exploring the Structure-Property Relationships of I-Beam Lattice Metamaterials 

Beam lattice metamaterials are made of beam networks that can be tailored to achieve customizable properties, which have many potential applications in aerospace, mechanical, and biomedical engineering. Recently, beam lattices using beams with I-shape cross-sections were developed that have been demonstrated to have remarkably high strength/stiffness-to-weight ratios and excellent energy absorption capabilities compared with conventional beam lattices with circular beam cross-sections. Additionally, joint reinforcements can be added to the I-beam lattices to further improve the stiffness, strength, and energy absorption. In addition to greatly enhanced mechanical properties, these I-beam lattices also introduced highly tailorable, anisotropic material properties determined by several cross-sectional geometries. Compared to conventional beam lattices, the effective material properties are not only determined by the relative density of the lattices but also by the beam cross-sectional geometries. This poses a computational challenge to explore the complex structure-property (SP) relationships of the I-beam lattices.

In this work, we employed the computational homogenization method to perform multiscale modeling for I-beam and reinforced I-beam lattices with different cross-sectional geometries, such as the thickness of web and flanges, the width and height of the beam cross-section, and the thickness of the reinforcements. The unit cells of these beam lattices were created in finite element software Abaqus and the multiscale modeling was performed using Abaqus Micromechanics Plug-in. The simulation results, such as the stiffness and yield strength, were experimentally validated from quasi-static compression tests. The samples were fabricated using a Stratasys J35 Pro 3D printer. The validated multiscale models were employed to generate over 100 simulation data, containing effective engineering material constants and yield strength in the major loading direction. These data points were sampled using Sobol sampling, covering relative volume densities from 10% to 60%.  To efficiently explore the SP relationships that connect cross-sectional designs to the material properties, a machine learning model was developed using the Gaussian Process Regression (GPR) model.

The results showed that the proposed multiscale models can accurately capture the experimentally measured stiffness and strength while offering improved computational efficiency compared to direct numerical simulations using high-fidelity finite element models. The proposed GPR model was trained with high prediction accuracy. For a specific relative density of the conventional circular beam lattices, the lattices showed a single value for stiffness and strength as expected. However, multiple values of stiffness and strength are discovered for I-beam and reinforced I-beam lattices, which suggests enhanced design flexibility when customizing the lattice properties for specific applications. The developed GPR model will be connected to Explainable Artificial Intelligence (XAI) to investigate how the cross-sectional geometries affect the final material properties and strength of the I-beam and reinforced I-beam lattices.

The presented work enables a better understanding of the complex SP relationships of a recently developed beam lattice made of I-beam and reinforced I-beam. The results provide critical insight into the vital cross-sectional geometries of the final material properties of the lattices, which will help guide the material design to support the development of lightweight, high-performance materials for customized aerospace applications.

Presenting Author: Twinkle Kothari The University of Texas at Arlington

Presenting Author Biography: Twinkle Kothari is a PhD candidate in the Mechanical and Aerospace Engineering department at the University of Texas at Arlington. Her research focuses on multiscale modeling and experiments of I-beam lattice metamaterials.