SBIR Phase I: Automatic Generation of Physics-Informed AI Models

这个小企业创新研究（SBIR）第一阶段项目的广泛影响是通过以下方式使多种工业应用的模型构建过程民主化：(1)使构建模型变得更容易，而不是几周；(2)将模型构建成本降低10倍或更多；(3)通过提供受底层物理原理约束的更准确和可解释的模型来显著降低风险。这项技术将开启新一代的建模工作流，它更具可扩展性，具有更少的不确定性，并改善部分被理解的复杂过程的结构可见性和互操作性。因此，公司将有切实可行的方法来降低建模成本，简化优化和操作控制，并最终以经济上可行的速度实现更好的决策。更广泛地说，提议的技术应该激发各种创新，并增强强烈依赖物理建模的产品和服务。预计这将对农业、环境科学、土木工程、制造、航空航天、建设、物流、医药等多个领域产生深远的影响。这项小型企业创新研究（SBIR）第一阶段项目将为自动化和加速从数据中构建简化物理模型所需的技术和业务基础奠定基础，同时减少人为干预。该项目的目标是实现快速可靠的机制：(1)将物理世界的时空数据映射为结构网络表示；(2)利用来自这些网络表示的数据生成一组合适的方程来解释潜在的动态；(3)利用先进的优化技术找到一组模型，这些模型能够捕获与观测数据匹配的网络和方程的最佳组合；(4)将所有这些部分编排到一个人工智能平台中，该平台可以自动整合数据和人类交互，以进行模型构建和可视化。第一阶段的工作特别侧重于证明所得到的模型在物理上是健全的，并且在描述基于不同数量的可用数据和条件的流体输送和扩散动力学方面足够稳健。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 impact of this Small Business Innovation Research (SBIR) Phase I project is to democratize the model building process for multiple industrial applications by: (1) making it easy to build models within hours instead of weeks, (2) cut model building costs by 10x or more, and (3) significantly mitigate risks by providing more accurate and interpretable models that are constrained by underlying physical principles. This technology would unlock a new generation of modeling workflows that are more scalable, have less uncertainty, and improve the structural visibility and interoperability of complex processes that are partially understood. As a result, companies would have tangible ways to reduce modeling costs, streamline optimization and control of operations and ultimately, achieve better decisions at economically viable rates. More broadly, the proposed technology should inspire various innovations and enhancements to products and services that strongly rely on physics modeling. It is expected that tangible results would induce profound effects across multiple sectors, including agriculture, environmental science, civil engineering, manufacturing, aerospace, construction, logistics, and medicine. This Small Business Innovation Research (SBIR) Phase I project will set the technical and business foundations required to automate and expedite the construction of reduced physics models from data with minimal human intervention. The goal for this project is to implement fast and reliable mechanisms for: (1) mapping spatio-temporal data of the physical world into structural network representations; (2) leverage the data from these network representations to generate a suitable set of equations that explain the underlying dynamics; (3) use advanced optimization techniques to find a list of models that capture the best combination of network and equations matching the observed data; and (4) orchestrate all these pieces into a single artificial intelligence platform that automatically consolidates data and human interactions for model building and visualization. Efforts in Phase I are particularly focused on demonstrating that the resulting models are physically sound and sufficiently robust for describing the dynamics of fluid transport and diffusion based on different amounts of available data and conditions.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.