Classification of Cyber-Physical System Adversaries
funded through the U.S. National Security Agency SoS Lablet
researchers: Sayan Mitra and Geir Dullerud
Cyber-Physical Systems (CPS) are vulnerable to elusive dynamics-aware attacks that subtly change local behaviors in ways that lead to large deviations in global behavior, and to system instability. The broad agenda for this project is to classify attacks on different classes of CPS based on detectability. In particular, we are identifying attacks that are impossible to detect in a given class of CPS (with reasonable resources), and we are developing detection algorithms for those that are possible. The methods developed will primarily be aimed at scenarios in which attackers have some ability to intermittently disrupt either the timing or the quality-of-service of software or communication processes, even though the processes may not have been breached in the traditional sense. Much of the work will also apply to cases where such limited disruptions are introduced physically. Our approach is based on a set of powerful technical tools that draw from and combine ideas from robust control theory, formal methods, and information theory.