Efficient Synthesis of Sensor Deception Attacks Using Observation Equivalence-Based Abstraction
Paper i proceeding, 2020

This paper investigates the synthesis of successful sensor deception attack functions in supervisory control using abstraction methods to reduce computational complexity. In sensor deception attacks, an attacker hijacks a subset of the sensors of the plant and feeds incorrect information to the supervisor with the intent on causing damage to the supervised system. The attacker is successful if its attack causes damage to the system and it is not identified by an intrusion detection module. The existence test and the synthesis method of successful sensor deception attack functions are computationally expensive, specifically in partially observed systems. For this reason, we leverage results on abstraction methods to reduce the computational effort in solving these problems. Namely, we introduce an equivalence relation called restricted observation equivalence, that is used to abstract the original system before calculating attack functions. Based on this equivalence relation we prove that the existence of successful attack functions in the abstracted supervised system guarantees the existence of successful attack functions in the unabstracted supervised system and vice versa. Moreover, successful attack functions synthesized from the abstracted system can be exactly mapped to successful attack functions on the unabstracted system, thereby providing a complete solution to the attack synthesis problem.

Deception Attacks

Supervisory Control Theory

Automaton

Abstraction

Författare

Sahar Mohajerani

Chalmers, Elektroteknik, System- och reglerteknik, Automation

Rômulo Meira-Góes

University of Michigan

Stéphane Lafortune

University of Michigan

IFAC-PapersOnLine

24058963 (eISSN)

Vol. 53 4 28-34

15th IFAC Workshop on Discrete Event Systems, WODES 2020
Rio de Janeiro, Brazil,

Ämneskategorier

Inbäddad systemteknik

Datavetenskap (datalogi)

Datorsystem

DOI

10.1016/j.ifacol.2021.04.069

Mer information

Senast uppdaterat

2021-06-14