In industrial environments, the collection of vast amounts of operational and instrumentation data serves critical purposes such as monitoring, control, preventative maintenance, fault detection, and troubleshooting. Networked control systems have revolutionized traditional methodologies, offering seamless data transfer capabilities while minimizing wiring and maintenance issues. Their ease of implementation and scalability make them applicable across a wide spectrum of operations, from small-scale setups to large industrial complexes. However, the efficient functioning of industrial process control systems in real-time heavily relies on the accuracy of recorded data and the dependability of networked communication channels. Any compromise in the integrity or confidentiality of this data due to unauthorized access or manipulation by malicious entities can result in severe consequences, impacting operational safety and economic performance. As intelligent cyber-attacks have the potential to access system information, it is necessary to develop networked control systems that maintain the confidentiality of industrial data, and have cyber-attack detection and resilient operation strategies to address cybersecurity issues beyond fault diagnosis, and is the focus of this thesis.
Large-scale industrial processes encounter numerous control and operational challenges, such as nonlinearity, high dimensionality, complex interacting process dynamics, inherent state and input delays, and limited sensor measurements. To address these challenges effectively, a comprehensive mathematical model representing plant dynamics is essential, with appropriate integrations to tackle specific challenges. For example, model predictive control systems can handle multivariable interactions and input/state constraints, state predictors can address input delays, time-lag models are employed to account for state delays, and observers are integrated to estimate unavailable data accurately. Additionally, distributed and decentralized control structures offer improved computational efficiency compared to centralized frameworks, particularly advantageous for large-scale processes. Real-time adaptation to fluctuating economics is another crucial aspect for maintaining competitiveness in the market. Balancing these challenges with the need to enhance cybersecurity and ensure the confidentiality of system data necessitates the development of innovative control frameworks.
Motivated by the above, this thesis introduces novel control architectures featuring encrypted communication tailored for various nonlinear processes. Encryption techniques are integrated into centralized, decentralized, and distributed model predictive control (MPC) systems. Additionally, this thesis introduces two-layer and two-tier encrypted control frameworks that combine linear and nonlinear control strategies. Within these frameworks, linear control systems perform control input computations in an encrypted space, eliminating the need for decryption and ensuring the preservation of data confidentiality. Incorporating machine-learning-based and logic-based cyberattack detectors with reconfiguration mechanisms further fortifies these encrypted control frameworks for cyber-resilient operation. System-specific integrations in the control system address complexities like limited feedback, input and state delays, and the dynamic nature of process economics. Numerical simulations of nonlinear chemical process examples and Aspen Plus simulations of large-scale chemical process networks demonstrate the effectiveness of the proposed frameworks. The results highlight their ability to improve operational safety, cyber-security, computational efficiency, and overall closed-loop and economic performance in nonlinear processes.
