Cybersecurity in process control: Machine-learning detection and encrypted control

由于网络控制系统容易受到网络攻击，化工过程控制系统的安全性变得至关重要。未能确保网络安全可能会导致化工过程操作中的不安全和潜在的灾难性后果，导致环境破坏、资本损失和人员伤亡。近年来，网络攻击是由经验丰富的对手设计的，目的是修改执行器、传感器或控制动作，但仍然无法通过经典的检测方法检测到。因此，实时检测网络攻击并缓解其影响是一个重要的研究问题，其解决方案直接影响到化工过程行业的安全。基于这些考虑，本研究项目的目标是基于基于模型的控制方法框架内的机器学习技术和加密工具，开发检测和处理针对大类非线性过程的过程控制系统的智能网络攻击所需的理论和计算方法。将开发用户友好的软件，并将其整合到最广泛使用的化学工艺CAD软件中；将开设短期课程和讲习班来传播这些计算工具。此外，研究成果将被纳入加州大学洛杉矶分校的本科过程控制和高级过程设计与经济学课程，以介绍机器学习技术的应用，以符合部门教唆目标以及加州大学洛杉矶分校关于实施数据科学辅修课程的目标。最后，将通过参加加州大学洛杉矶分校的工程教育和多样性中心，让不同的本科生和研究生群体参与研究，并通过向高素质学生提供暑期实习机会，向社区学院进行外联。该研究计划的重点是设计和实现计算方法，以防止可能危及工业过程控制系统的数据完整性、闭环系统稳定性和过程操作安全的网络攻击。为此，将开发一种基于数据的检测方法，使用机器学习算法来解决使用时间序列测量数据的网络攻击的分类问题。这种方法的核心是设计基于加密的模型预测控制(MPC)方案，该方案可以安全地运行该过程，并且在存在网络攻击的情况下将性能降级降至最低。这将通过开发一种新的网络安全的MPC稳定性-性能控制体系结构来实现，该体系结构可以检测网络攻击并在保持闭环稳定性的同时隔离控制网络中受影响的传感器和执行器。由此产生的网络安全框架的有效性将通过在Aspen模拟环境中结合工业过程可变性数据的高保真大规模化学过程网络的应用来展示。该奖项反映了NSF的法定使命，并已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。

The security of chemical process control systems has become crucially important since networked control systems are vulnerable to cyber-attacks. The failure to ensure cyber-security can lead to unsafe and potentially catastrophic consequences in chemical process operations, causing environmental damage, capital loss, and human injuries. In recent years, cyber-attacks have been designed by sophisticated adversaries to modify the actuator, the sensor, or the control action, yet remain undetectable by classical detection methods. Therefore, real-time detection of cyber-attacks and mitigation of their effects is an important research problem whose solution could directly impact the safety and security of the chemical process industries. Motivated by these considerations, the goal of this research program is to develop the theory and computational methods needed for detecting and handling intelligent cyber-attacks on process control systems for broad classes of nonlinear processes based on machine learning techniques and encryption tools incorporated within the framework of model-based control methods. User-friendly software will be developed and integrated into the most widely used chemical process CAD software; short courses and workshops will be created to disseminate these computational tools. Furthermore, the research results will be incorporated within the undergraduate process control and senior process design and economics course curricula at UCLA to introduce applications of machine learning techniques in accordance with departmental ABET goals as well as UCLA campus goals on the implementation of a Data Science minor. Finally, the involvement of a diverse group of undergraduate and graduate students in the research through participation in the Center for Engineering Education and Diversity at UCLA, and outreach to Community Colleges by offering summer internships to highly qualified students, will be pursued. The focus of this research program is on the design and implementation of computational methods to prevent cyber-attacks that can compromise data integrity, closed-loop stability, and process operational safety of industrial process control systems. This will be carried out by developing a data-based detection approach using machine learning algorithms to solve classification problems of cyber-attacks using time-series measurement data. Central to this approach is the design of encryption-based model predictive control (MPC) schemes that can operate the process safely and with minimal performance degradation in the presence of cyber-attacks. This will be accomplished through the development of a novel cyber-secure MPC stability-performance control architecture that can detect cyber-attacks and isolate the affected sensors and actuators in the control network while maintaining closed-loop stability. The effectiveness of the resulting cyber-security framework will be demonstrated through applications to high-fidelity, large-scale chemical process networks in an ASPEN simulation environment incorporating industrial process variability data.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.