Safety Filters Against Actuator Attacks

This manuscript focuses on mitigating the effect of deception attacks on control signals, that is, in the presence of an adversary that tampers with data coming from the controller to the system actuators in order to degrade the plant performance. We propose adding Multiple-Inputs Multiple-Outputs (MIMO) filters to the loop, between the received control actions (which are potentially corrupted by attacks) and the plant actuators. These filters are designed to dynamically steer the reachable set induced by the attack signals to a known safe region of the state space. We provide a synthesis framework (built in terms of the solution of a collection of semidefinite programs) to design the filters so that attack-free control signals are distorted as little as possible–in terms of $H_{\infty }$ and $H_2$ norms of the difference between original and filtered control signals–and the trajectories are guaranteed to be contained in a predefined safe set. The results are illustrated through a simulation case focusing on the stability augmentation system of an airplane.