Recovery after attacks of deception on Networked Control Systems

This paper follows on the authors' recent work that developed a network-wide attack detector and isolator. Having determined that the sensor of a regulated process is compromised, the problem shifts to maintaining the stability of the subject process and the integrity of the greater physical plant. First, measurements gathered at other processes, interconnected at the physical layer, provide estimates of the output of the process under attack through a bank of state observers. Next, a weighted consensus algorithm fuses all of these estimates into information that is actionable, being independent of the deceptive sensory data. The estimation error is shown to be sector-bounded. To address the resulting latency time in regulation, the authors employ a self-triggered control policy whose aim is to stabilize the attacked process. Furthermore, they test the proposed concept by simulation and demonstrate the safe operation of the physical plant.