Safety-Preserving Filters Against Stealthy Sensor and Actuator Attacks

This article proposes a novel strategy based on control input filtering for mitigating the effects of deception attacks on control and sensor measurement signals. We assume an adversary who can tamper data transmitted from a communication network in order to degrade the plant performance. The proposed strategy consists on adding multiple-input multiple-output (MIMO) filters to the control loop, between the received control actions and the plant actuators. The filter's goal is to dynamically steer the reachable set induced by the attack signals to a safe region of the state space. The article provides a filter synthesis method under the form of a semidefinite programming problem, yielding such filters in a way that attack-free control signals are distorted as little as possible, and plant trajectories are contained in the safe set. At the end of the paper, a set of simulations demonstrate the effectiveness of the approach.