SBIR Phase I: A Physics-Informed/Encoded Polymer Informatics Platform for Accelerated Development of Advanced Polymers and Formulations

这个小企业创新研究（SBIR）第一阶段项目的更广泛/商业影响是改变聚合物材料的开发方式。该技术采用了最先进的人工智能（AI）技术，旨在大幅加速对新聚合物配方的探索，高效、准确地发现具有目标性能和应用的聚合物配方，并最终最大限度地减少开发新型高性能功能材料所需的时间和成本。这项技术将能够有针对性地开发用于特定应用的聚合物，如包装或能量存储，同时确保完全可回收性。这种新型聚合物设计可以帮助缓解目前全球塑料垃圾问题。鉴于聚合物是当今使用的最重要的材料之一，预计SBIR一期项目的影响将是重大而深远的。这个小企业创新研究（SBIR）第一阶段项目旨在改变目前用于发现和设计功能聚合物的最先进的基于人工智能的技术。自大约十年前聚合物信息学开始以来，这种基于人工智能的方法已迅速成为设计新型功能聚合物的强大工具。这项技术的核心是机器学习模型，这些模型是根据过去的数据进行训练的，用于评估尚未合成的聚合物材料。目前，这些模型是通过纯粹独立地“学习”可用数据集来开发的，忽略了来自不同来源的不同聚合物类别和性质的数据之间的许多物理控制的相关性。如果没有适当的意识，模型很容易违反相关的物理规则，呈现非物理的结果，特别是在训练数据不够大的情况下。在这个项目中，该公司将开发两个深度学习架构，其中已知的和重要的物理治理相关性是安全的。该架构将成为最先进的深度学习工具，用于解决聚合物信息学中非常常见和重要的小数据和稀疏数据问题。这项新技术有望显著改变功能性聚合物的开发和应用。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

The broader/commercial impacts of this Small Business Innovation Research (SBIR) Phase I project are to transform the way in which polymeric materials are developed. Adopting the most advanced artificial intelligence (AI) techniques, the proposed technology seeks to dramatically accelerate the exploration of new polymer formulations, efficiently and accurately discovering those with targeted performances and applications, and ultimately minimizing the time and the cost needed to develop new and superior functional materials. This technology will enable the targeted development of polymers for specific applications such as packaging or energy storage, while ensuring full recyclability. New polymer designs of this type can help alleviate the current global problem of plastic waste. Given that polymers are one of the most important classes of materials in use today, the impact of this SBIR Phase I project is expected to be significant and far-reaching. This Small Business Innovation Research (SBIR) Phase I project aims at transforming the state-of-the-art AI-based technology currently used to discover and design functional polymers. Since the beginning of polymer informatics about a decade ago, this AI-based approach has quickly become a powerful tool to design new functional polymers. At the center of this technology are the machine-learning models, trained on past data and used to evaluate the polymeric materials yet to be synthesized. Currently, the models are developed by purely “learning” the available datasets independently, ignoring numerous physics-governed correlations across data of different polymer classes and properties that come from different sources. Without proper awareness, the models can easily violate the relevant physic rules and render unphysical results, especially when the training data are not sufficiently large. In this project, the company will develop two deep learning architectures in which known and important physics-governed correlations are secured. The architectures will be the most advanced deep learning tools to combat the small and sparse data problems that are very common in and important for polymer informatics. The new technology is expected to significantly transform the development and deployment of functional polymers.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.