Enhanced resilient sensor attack detection using fusion interval and measurement history

Abstract

For the Cyber-Physical Systems which are equipped with multiple sensors that measure the same physical variables, some of the sensors may be subject to malicious attacks and provide the wrong measurements to the controller, resulting in an unsafe state of the system. Although there are several methods for detecting sensor attacks, they may not work well when an attacker manipulates the sensor output very slightly or infrequently over a period of time, e.g., stealth attacks. To address this problem, we take the system dynamics model into account, and propose a novel sensor attack detection algorithm based on historical measurements and fusion intervals. The core idea is to build different fault models for different sensors, add a virtual sensor to use pairwise inconsistencies between two sensors to identify attacks. Finally, we validate the performance of the algorithm on real measurements obtained from the LEGO EV3 ground vehicle, and the results show that the proposed method outperforms state-of-the-art algorithm, especially for stealth attacks, with about 92% promotion in attack detection and identification rate.