Resilient Control Systems for Industrial Process Automation

数字控制和监测单元自动调节物理过程，而通信网络用于传输信息。根据需要，人工操作员通过与监控单元交互来手动调节过程。在工业过程自动化的实际设置中，过程动态在连续时间内演变，而控制、监视、通信和操作员动作动态在离散时间内演变，具有周期性、时间驱动和事件触发事件。这些不同的动态和耦合产生了一种新的具有挑战性的系统类别--网络-物理-人类系统--这使得控制系统的开发变得复杂，特别是对于具有多个物理上相互连接的进程以及多个相互连接的控制和监测单元的大规模分布式应用。使问题进一步复杂化的是，何时发生异常事件(例如，故障和网络攻击)，需要控制系统及时响应。拟议研究计划的目标是发现对工业过程自动化中的异常提供弹性的控制系统(即弹性控制系统)的新的设计和分析方法。该计划的重点是发展一个以网络-物理-人类系统为基础的大规模互联系统范式。这一范例将捕捉到所涉及的物理过程、数字系统和信号以及人类操作员行为之间的依赖关系，以连续和离散时间动态的混合方式建模。它将是全球最全面的现代控制系统的代表，并将有效地模拟我们社会中的工业应用，例如在制造工厂、石油和天然气精炼厂以及供水网络中。这项研究将发现故障诊断、网络安全和警报管理的新解决方案，这将提高我们开发完全弹性和自主的工业过程自动化的能力。将开发新的计算方法和工具来检测随机故障和恶意网络攻击，定位它们的确切位置，并确定控制程序以减轻它们的影响。这项研究将建立在控制、优化和图论的基础上，并将涉及状态估计、控制器和观测器设计，以及系统稳定性和性能标准。当检测到故障和网络攻击等异常事件时，它们会作为警报消息在监控单元上发出，以提醒操作员并指导他们的行动。该研究计划的一个关键目标是开发算法程序，以更好地管理警报的显示并使操作员的操作自动化。为了得到这样的基本结果和实用的解决方案，这项研究将开发数据挖掘和机器学习程序。这一新颖的研究计划将促进知识和培训人员，为控制系统社区和行业做出创新贡献。

Digital control and monitoring units automate the regulation of physical processes, while a communication network is used to transfer information. As needed, human operators manually regulate the processes by interacting with the monitoring units. In this practical setup for industrial process automation, process dynamics evolve in continuous-time, whereas control, monitoring, communication, and operator action dynamics evolve in discrete-time, with periodic, time-driven, and event-triggered events. These different dynamics and couplings give rise to a new and challenging class of systems - the cyber-physical-human systems - which complicate the development of control systems, especially for large-scale distributed applications with multiple physically interconnected processes and multiple interconnected control and monitoring units. What further complicates the matter is when abnormal events occur (e.g., faults and cyber attacks), requiring timely response from control systems. The objective of the proposed research program is to discover new design and analytical methods for control systems that offer resiliency to abnormality in industrial process automation (namely, resilient control systems). The program focuses on developing a large-scale interconnected system paradigm with a cyber-physical-human system basis. This paradigm will capture the dependency between the involved physical processes, digital systems and signals, and human operator actions, modelled in a hybrid manner with continuous- and discrete-time dynamics. It will be the most comprehensive representation of modern control systems globally, and will effectively model industrial applications in our society, such as in manufacturing plants, oil & gas refineries, and water distribution networks. The research will uncover novel solutions for fault diagnosis, cyber security, and alarm management, which will advance our ability to develop fully resilient and autonomous industrial process automation. New computational methods and tools will be developed to detect random faults and malicious cyber attacks, locate their exact location, and determine control procedures to mitigate their effects. The research will build on control, optimization, and graph theories, and will address state estimation, controller and observer designs, and system stabilization and performance criteria. When abnormal events like faults and cyber attacks are detected, they are annunciated on monitoring units as alarm messages, to alert human operators and direct their actions. A key aim of the research program is to develop algorithmic procedures to better manage the display of alarms and automate human operator actions. To arrive at such fundamental results yet practical solutions, the research will develop data mining and machine learning procedures. This novel research program will advance knowledge and train personnel to deliver innovative contributions for the control systems community and industry.