Cyber-physical security using geometric control theory and model-predictive control

Cyber-Physical Systems (CPS), referred to as those systems that integrate physical processes with computational resources and communication capabilities, are ubiquitous in modern society, which is characterized by networked infrastructures in many different domains.

Transportation networks, power generation and distribution networks, industrial automation systems are examples of such systems, where control of physical plant is mediated by and integrated with a wireless communication network. If this integration can improve efficiency, it nonetheless makes the system more vulnerable to attacks launched in the cyber-domain. Recent real world attacks targeting physical plants raised the problem of cyber-physical security, suggesting that information security mechanisms have to be complemented with specifically designed control systems, possibly resilient against attacks and/or equipped with attack monitors.

Attacks might be designed to result undetectable, i.e. invisible with respect to the system measurements by combining the injection of malicious signals in actuators and sensors. The synthesis of such attacks is based on the implicit weak redundancy of the plant: since the number of inputs is larger than the number of measured outputs, a malicious signal can be designed to be aligned with the null space of the transfer function of the plant with the purpose of corrupting the system behavior while remaining undetected. Based on the framework of geometric control theory, control allocation and model predictive control, the ideas proposed in this master project arise from a “fight fire with fire” principle: designing and implementing an input signal that does not alter the desired system performance and simultaneously keeps the total control effort close to its limits, the occurrence of an undetectable attack can be revealed by the violation of input constraints with the consequent loss of invisibility.

Emneord: Cyber-physical systems, Optimization algoritms, Model-predictive control
Publisert 25. sep. 2018 14:39 - Sist endret 25. sep. 2018 14:39

Omfang (studiepoeng)

60