UniMate: A Unified Model for Mechanical Metamaterial Generation, Property Prediction, and Condition Confirmation

Wangzhi Zhan, Jianpeng Chen, Dongqi Fu, Dawei Zhou

TL;DR: We propose a unified model for mechanical metamaterial generation, prediction and conditioning.

Abstract: Metamaterials are artificial materials that are designed to meet unseen properties in nature, such as ultra-stiffness and negative materials indices. In mechanical metamaterial design, three key modalities are typically involved, i.e., 3D topology, density condition, and mechanical property. Real-world complex application scenarios place the demanding requirements on machine learning models to consider all three modalities together. However, a comprehensive literature review indicates that most existing works only consider two modalities, e.g., predicting mechanical properties given the 3D topology or generating 3D topology given the required properties. Therefore, there is still a significant gap for the state-of-the-art machine learning models capturing the whole. Hence, we propose a unified model named UniMate, which consists of a modality alignment module and a synergetic diffusion generation module. Experiments indicate that UniMate outperforms the other baseline models in topology generation task, property prediction task, and condition confirmation task by up to 80.2%, 5.1%, and 50.2%, respectively. We open-source our proposed UniMate model and corresponding results at https://github.com/wzhan24/UniMate.

Lay Summary:

Metamaterials are specially engineered materials that can achieve mechanical behaviors not commonly found in nature, such as being extremely stiff while staying lightweight. Designing these materials is complex because it involves three important parts: the 3D structure (topology), the material density, and the mechanical performance. Most existing methods can only handle two of these parts at a time.
Our work introduces UniMate, a new AI model that, for the first time, can consider and connect all three parts together in a single system. UniMate can generate material structures, predict how they will perform, and determine the right conditions they need to meet specific goals.
We tested UniMate on all three tasks and found that it performs better than previous methods, creating higher-quality structures, predicting properties more accurately, and confirming conditions more effectively. To support this, we also built a new dataset and evaluation tools that cover a wider range of material design problems.
UniMate is a step toward making the design of mechanical metamaterials more comprehensive and effective.

Primary Area: Applications->Chemistry, Physics, and Earth Sciences

Keywords: metamaterial, topology generation, property prediction

Submission Number: 12617