CPS: Medium: Real-Time Learning and Control of Stochastic Nanostructure Growth Processes Through in situ Dynamic Imaging

这笔网络物理系统(CPS)赠款将支持研究，这些研究将贡献与纳米材料生长的新兴监测和控制技术相关的新知识，这些技术对新型电池和光伏设备等应用至关重要，因为精确的物质结构对于实现所需的电荷、质量和能量流动模式至关重要，这些模式是能量转换和存储的基础。随着动态纳米级成像产生的海量数据的快速到来，国家纳米技术倡议将缺乏过程中的监测和控制确定为阻碍新材料设计和发现的一个巨大挑战，因为“现有方法耗时、昂贵，需要高科技基础设施和高技能水平才能执行。”这笔赠款支持一个由数据科学、控制、电路设计和材料科学的专家组成的多学科团队，旨在通过设计一种能够可靠地将动态成像数据转换为机器可理解的信息以进行过程监控的网络物理系统来应对这一挑战。这项研究的结果将有助于纳米材料的发现，并为规模化生产铺平道路。多学科方法将有助于扩大代表性不足的群体对研究的参与，并对科学和工程教育产生积极影响。这项研究的智能核心是解决实时学习和控制反映纳米对象演化的集体随机行为的多变量、非参数概率密度函数模型的基本问题。需要克服几个科学和工程障碍，包括非参数学习的有效方法，纳米物体演化经验分布的自适应随机控制，以及提高计算效率和满足实时要求的硬件/软件联合设计。研究团队将测试随机浅层体系结构，这是一种解决实时非参数学习问题的新方法，探索支持结构化浅层网络的过程控制，这将是神经网络在实时动态环境中的早期应用之一，并为选定的算法设计硬件加速器，使高精度可扩展纳米制造的即时解决方案成为可能。该奖项反映了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。

This Cyber-Physical Systems (CPS) grant will support research that will contribute new knowledge related to emerging monitoring and control techniques of the growth of nanomaterials, which are crucial for applications such as new types of batteries and photovoltaic devices, because precise structuring of matter is essential to realize the desired charge, mass, and energy flow patterns that underpin energy conversion and storage. With the fast arrival of tremendous amount of data produced by dynamic nanoscale imaging, the National Nanotechnology Initiative has identified the lack of in-process monitoring and control as a grand challenge impeding the design and discovery of new materials, because "existing methods are time-consuming, expensive, and require high-tech infrastructure and high skill levels to perform." This grant supports a multidisciplinary team, comprising experts from data science, control, circuit design, and material sciences, aiming to tackle this challenge by designing a cyber-physical system that can reliably convert dynamic imaging data to machine intelligible information for process monitoring and control. The results from this research will benefit nanomaterial discovery and pave a path to scalable production. The multidisciplinary approach will help broaden participation of underrepresented groups in research and positively impact science and engineering education. The intellectual core of this research is to address the foundational problem of real-time learning and control of a multivariate, nonparametric model of probability density functions that reflect the collective stochastic behavior of evolving nano objects. Several scientific and engineering barriers are to be overcome, including efficient methods for nonparametric learning, adaptive stochastic control of evolving empirical distributions of nano objects, and hardware/software co-designs for boosting computational efficiency and meeting real-time requirements. The research team will test randomized shallow architectures, which are a novel approach for addressing real-time nonparametric learning issues, explore structured-shallow-networks-enabled process control, which will be one of the early applications of neural networks in real-time dynamic settings, and design hardware accelerators for select algorithms that enables on-the-fly solution for high-precision scalable nanomanufacturing.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.