EAGER:Real-Time:Automated Control-Assisted Data-Based Model Development for Real-Time Feedback Control

美国过程工业未来的经济竞争力需要成功实施下一代控制器，以实时优化过程经济性能，并考虑安全因素和传统上被忽视的动态，这些动态会显著影响控制回路性能。提出的探索性研究旨在开发一种新颖的、自动化的、控制辅助的方法，用于从在线数据中构建基于物理的模型，并为实时应用制定下一代控制器。如果成功，这项研究可能会对广泛的化学加工工业产生影响。从数据中开发模型的技术目前受到提供给模型识别程序的状态测量信息和假设表示系统动力学的结构的限制。因此，从过程状态/输入测量中导出控制设计的基于物理的模型的能力需要获得适当数据的方法，从中可以推导出正确的模型结构。拟议的研究将：a)开发一个控制辅助框架，用于从有利于开发基于物理的模型的过程中获取在线操作数据；B)开发将数学函数结构与操作数据趋势联系起来的方法；C)探索基于优化控制的偏微分方程模型的潜在类比或基于机器学习的计算时间减少技术；D)开发控制器更新技术，以促进控制律设计中使用的功能或参数的自动化开发；e)展示利用高保真过程模型的化学过程模拟的开发方法，以研究开发的技术在促进基于物理的模型开发自动化方面的潜力，用于开发实时应用的基于模型的控制器。建议与工业界进行互动，以验证所提出的方法，并计划通过网络广泛传播结果。除了对研究生和本科生进行研究培训外，还建议将课程开发和面向初高中学生以及普通公众的推广活动的重点放在吸引更广泛的受众和鼓励代表性不足的少数民族从事工程职业上。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 future economic competitiveness of the US process industries requires the successful implementation of next-generation controllers to optimize process economic performance in real time and account for both safety considerations and traditionally-neglected dynamics that can significantly impact control loop performance. The proposed exploratory research aims to develop a novel, automated, control-assisted methodology for building physics-based models from on-line data and formulating next-generation controllers for real-time applications. If successful, the proposed research can potentially have an impact in a broad spectrum of chemical process industries.Techniques for developing models from data are currently limited by the state measurement information provided to a model identification procedure and by the structure assumed to represent the system dynamics. Therefore, the ability to derive physics-based models for control design from process state/input measurements requires methods for obtaining appropriate data from which the correct model structures can be deduced. The proposed research will : a) develop a control-assisted framework for obtaining on-line operating data from a process that is conducive to developing a physics-based model; b) develop methods for relating mathematical function structure to operating data trends; c) explore potential analogy or machine learning-based computation time reduction techniques for partial differential equation models in optimization-based control; d) develop controller update techniques which facilitate automated development of functions or parameters used in the control law design; and e) demonstrate the developed methods utilizing chemical process simulations of high fidelity process models to investigate the potential of the developed techniques to promote automated physics-based model development for use in developing model-based controllers for real-time applications. Interactions with industry are proposed to validate the proposed methodologies and broad dissemination of results via the web is planned. In addition to training graduate and undergraduate students in research, curriculum development and outreach activities to middle- and high-school students as well as the general public are proposed focused on engaging a broader audience and encouraging underrepresented minorities to pursue careers in engineering.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.