An Adaptive System Identification Approach Using Mobile Sensors

该项目将通过开展研究，提供一种变革性的方法来识别具有少量传感器的大型动力系统，从而使国家利益受益。研究结果将对桥梁、管道、建筑物、风力机塔架和其他结构的损伤诊断和性能评估等应用有帮助。特别是，传感器布置的迭代方法将为系统识别和损伤定位提供前所未有的能力，为指导有效的结构修复提供决策支持框架。这一能力有可能加快目前对民用基础设施的检查进程，并每年减少数亿美元的成本。这个项目的一个组成部分是在系统辨识、反馈控制、机器人学和结构工程领域对本科生和研究生的启发和教育。该项目还将通过塔夫茨的学生教师外展导师计划接触到K-12学生。少数族裔学生的参与是优先事项，将通过直接与塔夫茨大学已有的项目合作来促进这一点。通过这个项目，将开发一个新的系统识别框架，其中使用顺序数据处理进行识别，并反复重新部署移动传感器，以提高识别的系统模型的准确性。对于具有移动传感器的系统辨识，基于贝叶斯推理的辨识和强化学习技术将被用来根据从当前传感器布置获得的反馈来寻找未来的传感器位置。该项目的方法还将包括路径规划部分和数据融合部分。路径规划组件在建模精度与能量和时间约束之间进行权衡，以确定每个迭代步骤中数据采集的最佳候选位置，而数据融合组件将考虑移动传感器处通信和计算的资源限制，确定如何将先前模型与新数据尽可能有效地组合在一起。该项目方法的能力最终将在设想的结构诊断应用的案例研究中进行评估。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 project will benefit national interests by conducting research that provides a transformative approach to enable the identification of large dynamical systems with a small number of sensors. The results will be useful for applications like the damage diagnosis and performance assessment of bridges, pipelines, buildings, wind turbine towers, and other structures. In particular, the iterative approach to sensor placement will provide an unprecedented capability for system identification and localization of damage, feeding into a decision support framework to guide effective structural repair. This capability has the potential to speed current inspection processes for civil infrastructure and reduce costs by hundreds of millions of dollars annually. An integral part of this project is the inspiration and education of undergraduate and graduate students in the areas of system identification, feedback control, robotics, and structural engineering. This project will also reach K-12 students through Tufts' Student Teacher Outreach Mentorship Program. Participation of students from minorities is a priority and will be promoted by working directly with established programs at Tufts University.Through this project a novel system identification framework will be developed, where identification is implemented using sequential data processing, and where mobile sensors are redeployed iteratively to enhance the accuracy of the identified system model. For system identification with mobile sensors, Bayesian inference-based identification and reinforcement learning techniques will be employed to find the future locations of sensors based on the feedback obtained from the current sensor placement. The methodology in this project will also include a path-planning component and a data-fusion component. The path-planning component balances modeling accuracy with energy and time constraints to determine the best candidate locations for data acquisition at each iterative step, while the data-fusion component will determine how to combine prior models with new data as efficiently as possible, considering resource constraints on communication and computation at the mobile sensors. The capabilities of the project's approach will finally be evaluated on a case study for the envisioned structural diagnosis application.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.