Mode-switching approach to detect stealthy sensor attacks for cyber-physical systems

This paper reports on a detection technique for stealthy sensor attacks on cyber-physical systems (CPS). The considered attack is a pole-dynamics attack, in which false data are injected to cancel the unstable trajectory of a plant captured on the feedback sensor output. In this attack, the systems appear to operate normally in the steady-state while the plant is destabilized. The target system is a linear plant, including unstable pole-dynamics, and we assume that the adversary knows the model to carry out the attack. The proposed detection method employs a switching mechanism of two control modes that play different roles. The first mode is the normal mode, which consists of a linear controller and an anomaly detector, having the same structure as those used in conventional networked control systems. The second mode is the attack detection mode, which utilizes an internal feedback controller. It is assumed that this internal controller is not known to the adversary. When the system is in the attack detection mode, owing to the use of the internal controller, the overall system dynamics differ from those adopted by the adversary, thereby the effect of the attack is revealed. The CPS periodically switches between the attack detection mode and normal mode. The timing of the switch is determined to ensure that any attack is revealed before the physical system becomes unstable. The results are validated via simulations of quadrotor control.