(207f) Model Predictive-Based Detection of Cyberattacks on Actuators Controlling Nonlinear Systems with Changing Process Dynamics

The implementation of advanced control algorithms is dependent on regular data updates from sensors in order to guarantee sufficient recursive feasibility and stability of real-world processes. A majority of these processes especially in the field of chemical engineering not only display nonlinear characteristics but also have dynamics that change with time for various reasons ranging from inherent dynamics to faults and disturbances affecting the system. The decision of industries to incorporate standards associated with Industry 4.0 by integrating physical processes with communication and control systems ^[1] are critical to not only increase production efficiency and improve transparency of production lines but also ensure better control of processes using advanced control strategies. However, this integration also opens these processes and associated components to cyberattacks which can affect process safety compromising not only profits but also endangering human lives.

Prior research from our group investigated various strategies to detect cyberattacks on process sensors ^[2][3][4] and actuators ^[5][6] to provide explicit guarantees of stability and feasibility in this environment. Initial detection strategies developed guaranteed stability only for limited time periods in the presence of cyberattacks, or made assumptions on how many sensors could be attacked. These short-comings motivated the detection strategy that will be discussed in this work, called the Directed Randomization Method ^[7], to attempt to gain stability and detect attacks over extended periods of time. In this strategy, one of two control actions is randomly selected at every sampling time such that the state of the process is maintained within a certain region determined a-priori over the subsequent sampling period. The goal is to make it challenging for an attacker to guess the correct next state measurement that will not identify the system as being under attack when this randomness exists in the control action selection, even if both options are known to an attacker. The objective of this work is to extend the Directed Randomization Method to the case that process dynamics can change. When this occurs, if an unexpected state measurement is obtained, it may not be clear whether the process dynamics changed or whether an attack on the sensors occurred. We discuss how to extend the Directed Randomization Method to this case.

References:

[1] Lezzi, M., Lazoi, M., Corallo, A. Cybersecurity for industry 4.0 in the current literature: A reference framework. Computers in Industry 103, 97-110 (2018).

[2] Rangan, K. K., Oyama, H., & Durand, H. Integrated Cyberattack Detection and Handling for Nonlinear Systems with Evolving Process Dynamics under Lyapunov-based Economic Model Predictive Control. Chemical Engineering Research and Design (2021).

[3] Oyama, H. and H. Durand. Integrated Cyberattack Detection and Resilient Control Strategies Using Lyapunov-Based Economic Model Predictive Control, AIChE Journal, 66 ("Futures" issue), ee17084 (2020).

[4] Liu, S., Wei, G., Song, Y., Liu, Y. Extended kalman filtering for stochastic nonlinear systems with randomly occurring cyber attacks. Neurocomputing 207, 708-716 (2016).

[5] Li, Y., Voos, H., Rosich, A., Darouach, M. A stochastic cyber-attack detection scheme for stochastic control systems based on frequency-domain transformation technique, in: International Conference on Network and System Security, Springer, 209-222 (2015).

[6] Huang, K., Zhou, C., Tian, Y.-C., Yang, S., & Qin, Y. Assessing the Physical Impact of Cyberattacks on Industrial Cyber-Physical Systems. IEEE Transactions on Industrial Electronics, 65(10), 8153–8162 (2018).

[7] Oyama, H., Messina, D., Rangan, K.K., Leonard, A.F., Nieman, K., Durand, H., Tyrrell, K., Hinzman, K. and Williamson, M. Development of directed randomization for discussing a minimal security architecture. Digital Chemical Engineering 6, 100065 (2023).