Collaborative Research: Calibrating Digital Twins in the Era of Big Data with Stochastic Optimization

该项目将通过提供新的校准方法，为能源、医疗保健和制造业等许多应用中的数字双胞胎创造价值机会，为国家繁荣做出贡献。数字孪生是复杂物理系统的数字表示，可用于在虚拟世界中监控，预测和测试系统。利用观测数据对数字孪生进行参数校准是使它们能够密切复制物理系统的最重要步骤之一。 如今，先进的数据传感和收集技术可以从复杂系统的许多组件中提供大量数据点。该项目的成功将通过从这些大型数据集进行有效采样，提供一种可靠的估计方法，从而大大减少校准的计算负担。该项目的外联活动将通过与行业从业人员接触来改善劳动力准备，通过让代表性不足的学生参与研究来扩大参与，并为K提供机会-12名学生了解数据科学领域。在这个项目中建立的数字孪生校准的定量方法将充分利用大数据的力量，同时解决数据集的大小和复杂性。具体研究任务包括：开发与统计理论相协调的随机优化方法，通过确定最佳的（最小的信息量最大）的计算效率的数据子集，扩展的综合优化框架，适用于广泛的校准问题，包括多维，功能，和时变校准，与理论和实践的影响;以及将输入不确定性与优化无缝结合，以显著增强解决方案的鲁棒性，同时保持计算的易处理性。该奖项反映了NSF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。

This project will contribute to the national prosperity by providing new calibration methods to generate value-producing opportunities for digital twins in many applications, including energy, healthcare, and manufacturing. A digital twin is a digital representation of a complex physical system that can be useful for monitoring, forecasting, and testing the system in a virtual world. Parameter calibration of digital twins with observational data is one of the most important steps in enabling them to closely replicate a physical system. Today, advanced data sensing and collection technologies provide massive data points from many components of a complex system. The success of this project will provide a means of robust estimation by efficient sampling from these large datasets, thereby significantly reducing the computational burden of calibration. The outreach activities of the project will improve workforce preparation through engagement with industrial practitioners, broaden participation through involvement of underrepresented students in research, and provide opportunities for K-12 students to learn about the field of data science.Quantitative methods established during this project for digital twin calibration will fully leverage the power of Big Data while addressing the research challenges brought forth by the size and complexity of the datasets. Specific research tasks include: development of stochastic optimization approaches reconciled with statistical theories that will optimally guide simulation experiments by identifying the best (smallest most informative) subsets of data for computational efficiency; extending the integrative optimization framework to be applicable for a wide range of calibration problems, including multi-dimensional, functional, and time-variant calibrations, with theoretical and practical implications; and seamless incorporation of input uncertainty with optimization to dramatically enhance the solution's robustness while maintaining computational tractability. The approach will be validated through real-word case studies in building energy systems and wind power systems.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.