Session: 16-01-01: Government Agency Student Poster Competition

Paper Number: 145593

145593 - Data-Driven Inverse Design of Material Symmetry-Induced Architected Cellular Materials. 

Imagine a world where revolutionary cellular materials possessing exceptional properties are swiftly produced through instant 3D printing and tailored to meet specific requirements in real-time. This vision highlights the need for a real-time inverse material design system, which is now more attainable with the rapid advancement of machine learning and data-driven approaches. However, the efficacy of state-of-the-art data-driven design approaches is mainly limited to periodic cellular materials. The constraints within the design space, ambiguity in structure-property mapping, and challenges in manufacturing the generated intricate structures make the inverse design challenging. Despite extensive research, much of the design space for cellular materials remains unexplored due to the completeness of the design representations. Data-driven methods speed up property-to-structure mapping but also bring an issue: inversion ambiguity--multiple structures having identical properties with distinct architectures. Furthermore, the manufacturability and compatibility of generated structures hinder the multi-scale design. This research aims to establish a data-driven inverse design framework, trained on a broad spectrum of architected cellular materials, that effectively handles inversion ambiguity and produces manufacturable functional designs. The work is supported by the NSF project Data-Driven Inverse Design of Additively Manufacturable Aperiodic Architected Cellular Materials.

Firstly, we are proposing a novel strategy for generating a seamless cellular dataset and shaping novel structures unbound by a singular repository, which is inspired by the symmetrical properties of crystallographic materials. Unlike conventional methods of design, which initially construct cellular materials by different design methods followed by endeavors to understand their mechanical attributes through material symmetries, this approach charts a different course. Here, material symmetry plays a key role in the design framework. This leads to a continuous dataset filled with various structures, each representing a potential subset category. Such diversity results in a wide range of mechanical properties, allowing for adaptability and versatility to meet different design needs. Then, topology constraints are enforced within the dataset to ensure the manufacturability of the designed structures. The material symmetry generation method facilitates the connectivity between different unit-cells when applying some constraints, which means the establishment of aperiodic structures and, therefore, the touch of extensive design space without the need for complicated designs of the unit-cells. Lastly, these proposed datasets are intended to be trained on models that naturally include uncertainty estimation, such as diffusion geometric data-driven approaches. These models prove highly effective in handling inversion ambiguity with remarkable prediction precision. In addition, the nature of the obtained cellular dataset and the simple constraints enforced within the diffusion model training ensure the manufacturability of the predicted structures. 

In summary, the proposed method for inverse design in cellular materials aligns with NSF project objectives by harnessing computation and data analysis to advance knowledge and accelerate the discovery of novel architected material while maintaining high accuracy and fidelity. It also ensures the simultaneous generation of multiple manufacturable design alternatives. We plan to involve designing 3D cellular materials that could have curved elements exhibiting nonlinear behavior under various load cases in our following research. The proposed design formwork will not only benefit the mechanical design community but also extend to other fields such as robotics, biomechanics…etc.

Presenting Author: Mohammad Abu-Mualla University of Illinois at Chicago

Presenting Author Biography: I am currently a Ph.D. student in mechanical engineering at the University of Illinois at Chicago. I earned my bachelor's degree in mechanical engineering from the University of Jordan in 2019, receiving the Victorian Award for academic excellence. Following this, I gained experience as a land systems design engineer at JODDB. Motivated by my passion for design, I pursued a Fulbright scholarship for a master's degree in mechanical engineering, focusing on design, and was placed at UIC's Dream Lab in 2021. Now, in my first year of a Ph.D. program, I am actively engaged in research on mechanical metamaterial design and the automation of the design process with the help of machine learning and topology optimization

Authors:

Mohammad Abu-Mualla University of Illinois at Chicago
Jida Huang University of Illinois at Chicago