SBIR Phase I: Agent-Based Identification of Constitutive Relationships from Large Manufacturing Datasets

小型企业创新研究(SBIR)第一阶段项目的更广泛影响是提高数据资源的利用率。虽然数据是当代业务不可或缺的一部分-用于为战略、技术和财务决策提供信息-但数据收集在包括制造业在内的许多领域仍然是联合的，因为物流、实用和战略障碍阻碍了集中化。因此，这些数据资源很快就会变得孤立。不再公平(可查找、可访问、可互操作、可重复使用)，收集如此昂贵的数据的价值丢失了。拟议的技术解决了有效利用联合数据所面临的两个主要问题。首先，它开发了一个统一的界面来分析和探索联邦数据，而不牺牲对数据访问的控制。其次，它将机器学习与对物理系统的理解结合在一起。提出的技术是神经网络和描述潜在物理的本构关系之间的严格数学转换。这两种方法将利用对工艺环境的测量，包括时间、温度和压力，以及机械强度或化学反应活性。神经网络是通用的，易于训练，它通过统计相关性来估计系统行为，这对于重复的、复杂的系统来说是理想的，比如制造过程；但它们需要越来越大和多样化的数据集，以扩大它们可靠的条件。相比之下，通常需要数年时间才能形成的本构关系可以用来预测一个系统在新条件下的表现。这一系统将整合这两种方法。这一奖项反映了NSF的法定使命，并通过使用基金会的智力优势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 improve utilization of data resources. While data is an integral part of contemporary business—used to inform strategic, technical, and financial decisions—data collection remains federated in many fields, including manufacturing, because logistical, practical, and strategic hurdles prevent centralization. Consequently, these data resources quickly become isolated. No longer FAIR (Findable, Accessible, Interoperable, Reusable), the value of data so expensive to collect is lost. The proposed technology addresses two major concerns facing effective utilization of federated data. First, it develops a unified interface to analyze and explore federated data, without sacrificing control over data access. Second, it integrates machine learning with an understanding of the physical system. The proposed technology is a mathematically rigorous translation between neural networks and the constitutive relationships describing the underlying physics. The two approaches will leverage measurements of a process environment, including time, temperature, and pressure, as well as mechanical strength or chemical reactivity. Neural networks, which are general and easy to train, estimate system behavior through statistical correlations, which is ideal for repetitive, complex systems, such as manufacturing processes; but they require increasingly large and diverse datasets to expand the conditions under which they are reliable. In contrast, constitutive relationships, which often take years to develop, can be used to predict how a system will behave under new conditions. This system will integrate both approaches.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.