CAREER: A New Sensor-Driven Framework for Real-time Monitoring, Control, and Decision Making in Dynamic Systems

这项学院早期职业发展计划(CALEAR)资助研究，旨在解决复杂、传感器驱动的工程系统的有效操作和维护，促进科学进步和提高经济竞争力和国家繁荣。最近智能传感技术和大规模数据采集平台的进步为可再生能源和智能制造等不同领域的动态工程系统的实时监测和控制带来了新的机遇。为了充分利用这些新的机会，需要研究可扩展的模型和方法来捕获这些系统的动态行为，并有效地分析从不同来源收集的异类数据，以便能够进行实时决策和控制。通过与国家可再生能源实验室、佛罗里达电力和照明以及阿贡国家实验室的合作，该项目将使用在风力涡轮机和电网上收集的大型数据库作为资产健康监测和控制的测试案例。该项目还旨在加强代表不足的少数群体在STEM领域的参与，特别是在数据分析、数据驱动的决策和计算科学职业方面。该项目将开发一个全新的量化框架，该框架可以有效地利用从工程系统中各种来源收集的大规模不同类型的时间序列数据，以产生实时可操作的见解和决策情报。使用动态贝叶斯分层建模和高级递归神经网络，而不强加强烈的参数/分布假设和先验知识，该项目将开发和验证可扩展的基于优化的方法，以使用过去的数据训练状态空间模型。该框架将产生一种受深度强化学习技术启发的实时最优控制和动态决策的新方法。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 Faculty Early Career Development Program (CAREER) grant supports research that addresses the efficient operation and maintenance of complex, sensor-driven engineered systems, promoting both the progress of science and advancing economic competitiveness and national prosperity. Recent advancements in smart sensing technologies and large-scale data acquisition platforms have brought new opportunities for real-time monitoring and control of dynamic engineering systems in such diverse domains as renewable energy and smart manufacturing. To fully benefit from these new opportunities, research is needed on scalable models and methods to capture the dynamic behavior of these systems and efficiently analyze heterogeneous data collected from disparate sources so as to enable real-time decision making and control. Through collaborations with the National Renewable Energy Laboratory, Florida Power and Light, and Argonne National Laboratory, the project will use large scale databases collected on wind turbines and electric power grids as test cases for asset health monitoring and control. The project also aims to enhance the participation of underrepresented minorities in STEM fields, particularly in data analytics, data-driven decision making, and computing science careers. This project will develop a fundamentally new quantitative framework that can efficiently utilize large-scale heterogeneous time-series data collected from various sources in engineered systems to generate real-time actionable insights and decision-making intelligence. Using dynamic Bayesian hierarchical modeling and advanced recurrent neural networks without imposing strong parametric/distributional assumptions and prior knowledge, the project will develop and validate scalable optimization-based methods for training the state-space model using past data. The framework will result in a new approach for real-time optimal control and dynamic decision- making inspired by deep reinforcement learning techniques.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.