CAREER: Understanding the Integrated Cyber-Physical Resilience of Continuous Critical Manufacturing

工业物联网(IIoT)技术引发了人们对制造业安全和弹性的日益浓厚的兴趣。然而，目前的解决方案缺乏对复杂系统中网络物理弹性的整体理解，无法将IIoT网络漏洞与动态制造流程联系起来，以进行有效的检测和控制。为了解决这些差距，该学院早期职业发展(CALEAR)项目旨在开发集成建模、检测和控制措施的新方法，以了解连续关键制造系统的网络物理弹性。这项研究将致力于消除制定制造业IIoT应用的新政策、法规和标准的障碍。通过与业界利益攸关方合作，该项目承诺将其方法和工具扩展到其他关键基础设施，从而提高国家的网络物理复原力，从而带来长期好处。同时，该项目的教育和推广活动在未来和现有的制造业专业人员中培养了对网络物理复原力的持续认识。引入新的课程和培训材料，增加学生接触先进制造技术的机会，提高他们的数据科学和网络安全技能。K-12外展计划增进了对IIoT和网络物理恢复能力的了解，促进了制造业职业生涯。专门设计的培训软件满足了以工程语言进行直观网络安全培训的需求。这些努力与国家先进制造战略保持一致，有助于确保国家安全的目标。这项研究解决了连续制造系统的网络物理弹性方面的关键挑战。研究目标包括(1)通用工具的开发：该项目旨在建立用于网络物理复原力量化的通用工具。通过创建整合网络连通性和异类组件系统动力学的随机模型，将建立一种新的量化指标。该指标通过随机优化考虑了IIoT网络特征和制造系统动态，揭示了IIoT连接引发的系统级风险。(2)重新思考异常检测：该项目将通过将系统范围的弹性量化纳入基于流程的异常检测算法来重新思考网络物理弹性驱动的异常检测。这涉及到设计新的半监督学习算法，其中包括弹性，重点是理解高维数据驱动的异常检测中支配检测精度和弹性增强的理论。(3)基于协作学习的弹性控制策略：本研究旨在构建基于协作学习的弹性控制策略。利用强化学习和系统连接，这些策略增强了系统对网络攻击的适应性。这涉及探索垂直联合强化学习的未被探索的领域，并产生关于单个机器的控制性能与系统对对手的适应性之间的权衡的新知识。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。

Industrial internet-of-things (IIoT) technologies spark growing interest in manufacturing security and resilience. However, current solutions lack a holistic understanding of cyber-physical resilience in complex systems, failing to connect IIoT network vulnerabilities with dynamic manufacturing processes for effective detection and control. To address these gaps, this Faculty Early Career Development (CAREER) project aims to develop novel methodologies that integrate modeling, detection, and control measures for understanding the cyber-physical resilience of continuous critical manufacturing systems. This study will work to eliminate barriers to the development of new policies, regulations, and standards for IIoT applications in manufacturing. In collaboration with industry stakeholders, this project promises long-term benefits by extending its methods and tools to other critical infrastructures, thereby enhancing national cyber-physical resilience. Meanwhile, the education and outreach activities in this project foster sustained awareness of cyber-physical resilience among both future and current manufacturing professionals. Introducing new courses and training materials enhances students' exposure to advanced manufacturing technologies and improves their data science and cybersecurity skills. K-12 outreach initiatives boost understanding of IIoT and cyber-physical resilience, promoting manufacturing careers. A specially designed training software addresses the need for intuitive cybersecurity training in engineering language. These endeavors align with the National Strategy for Advanced Manufacturing by contributing to the goal of ensuring national security.This study addresses critical challenges in continuous manufacturing systems' cyber-physical resilience. The research objectives include (1) Development of Generalizable Tools: The project aims to build generalizable tools for cyber-physical resilience quantification. By creating stochastic models that integrate cyber connectivity and system dynamics of heterogeneous components, a novel quantification metric will be established. This metric considers both IIoT network features and manufacturing system dynamics through stochastic optimization, revealing system-level risks induced by IIoT connectivity. (2) Rethinking Anomaly Detection: The project will rethink cyber-physical resilience-driven anomaly detection by incorporating system-wide resilience quantification into process-based anomaly detection algorithms. This involves designing novel semi-supervised learning algorithms that incorporate resilience, with a focus on understanding the theories governing detection accuracy and resilience enhancement in high-dimensional data-driven anomaly detection. (3) Collaborative Learning-Based Resilient Control Strategies: The study aims to create collaborative learning-based resilient control strategies. Leveraging reinforcement learning and system connectivity, these strategies enhance a system's adaptability to cyberattacks. This involves exploring the under-explored area of vertical federated reinforcement learning and generating new knowledge regarding the trade-off between the control performance of individual machines and the system's adaptability to adversaries.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.