ECLIPSE: Adaptable Model Predictive Control on a Chip for Personalized and Point-of-Care Plasma Medicine

生物医学设备在医疗诊断和治疗中变得越来越普遍，其中即时护理设备在个性化医疗方面显示出巨大的前景。尽管如此，医疗治疗和设备的日益复杂，需要发展先进的决策和自动控制系统，以确保生物医学设备的安全、可预测和治疗有效的操作。为此，该项目侧重于利用低温等离子体生物医学设备的医疗潜力。一些最有前途的血浆医学应用包括治疗生物膜相关感染、治疗伤口和皮肤病、协助癌症治疗、治疗病毒感染和生物植入物的表面修饰。该项目旨在利用机器学习和基于优化的控制方面的进步，开发智能决策系统，用于使用即时等离子体生物医学设备为个体受试者提供个性化等离子体剂量。例如，在资源有限的社区，智能决策系统可以创造前所未有的机会，使等离子装置在使用点等离子医学和生物技术应用中实现安全和（治疗上）有效的控制。这项研究的结果将用于一门新的基于学习的控制研究生课程，PI将与伯克利工程师和导师项目合作，指导本科生进行高中水平的科学课程。尽管最近在等离子体生物医学设备的模型预测控制（MPC）方面取得了进展，但在个性化和护理点等离子体生物医学应用方面仍存在重要挑战。该项目旨在开发一个系统理论框架，用于优化设计和数据高效适应mpc片上控制器，用于使用点护理生物医学设备的复杂接口的安全和个性化等离子体治疗。这一目标将通过以下方式实现：(1)开发一个多目标优化框架，用于快速动态（即kHz采样率）的不确定非线性系统的mpc片上软件和硬件协同设计；(2)开发安全、最优性能自适应的贝叶斯优化方法，特别适合于控制策略自适应性能数据匮乏的应用；(3)实验展示了具有典型生物医学应用的冷大气等离子体射流的自适应MPC-on-a-chip，用于细菌消毒。该项目系统理论发展的影响将超越等离子体，并可以转化为其他技术系统，其中自适应mpc -片上控制器可以发挥关键作用。该项目还提供了计算基准和开源代码，以提供在最先进的技术上取得的改进的可测量的演示。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。

Biomedical devices are becoming increasingly ubiquitous for medical diagnosis and therapies, wherein point-of-care devices have shown great promise for personalized medicine. Nonetheless, the increasing complexity of medical therapeutics and devices necessitates the development of advanced decision-making and automatic control systems to ensure safe, predictable, and therapeutically effective operation of biomedical devices. To this end, this project focuses on harnessing the medical potential of low-temperature plasma biomedical devices. Some of the most promising plasma medicine applications include treatment of biofilm-related infections, treatment of wounds and skin diseases, assistance in cancer treatment, treatment of virus infections, and surface modification of bioimplants. This project aims to leverage advances in machine learning and optimization-based control to develop intelligent decision-making systems for personalizing plasma dose delivery for individual subjects using point-of-care plasma biomedical devices. Intelligent decision-making systems can create unprecedented opportunities by enabling safe and (therapeutically) effective control of plasma devices for point-of-use plasma medicine and biotechnology applications, for example, in resource-limited communities. The results of this research will be used in a new graduate course on learning-based control, and the PI will partner with the Berkeley Engineers and Mentors program to mentor undergraduate students in conducting high school level science lessons.Despite recent advances in model predictive control (MPC) of plasma biomedical devices, there yet remains important challenges towards personalized and point-of-care plasma biomedical applications. This project aims to develop a systems-theoretic framework for optimal design and data-efficient adaptation of MPC-on-a-chip controllers for safe and personalized plasma treatment of complex interfaces using point-of-care biomedical devices. This objective will be realized through: (1) developing a multi-objective optimization framework for co-design of software and hardware for MPC-on-a-chip for uncertain nonlinear systems with fast dynamics (i.e., kHz sampling rates); (2) developing Bayesian optimization methods for safe and optimal performance-oriented adaptation of MPC-on-a-chip that is especially suited for applications in which performance data for control policy adaptation are scarce; and (3) experimentally demonstrating adaptive MPC-on-a-chip of a cold atmospheric plasma jet with prototypical biomedical applications for bacterial disinfection. The impact of the systems-theoretic developments of this project will be beyond plasmas and can translate to other technical systems in which adaptive MPC-on-a-chip controllers can play a critical role. The project also delivers computational benchmarks and open-source codes to provide a measurable demonstration of the improvements achieved over the state-of-the-art.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.