Diagnosability and attack detection for discrete event systems under sensor attacks

This paper extends the theory of diagnosability by investigating fault diagnosis in discrete event systems under sensor attacks using finite-state automata as models. It assumes that an attacker has compromised the communication channel between the system’s sensors and the diagnostic engine. While the general attack model utilized by the attacker has been previously studied in the context of supervisory control, its application to fault diagnosis remains unexplored. The attacker possesses the capability to substitute each compromised observable event with a string from an attack language. The attack model incorporates event insertion and deletion, as well as static and dynamic attacks. To formally capture the diagnostic engine’s ability to identify faults in the presence of the attacker, a novel concept called CA-diagnosability is introduced. This extends the existing notions of CA-controllability and CA-observability. A testing procedure for CA-diagnosability is developed, and its correctness is proven. Some sufficient conditions for CA-diagnosability that can be easily checked are also proposed and proved. The paper then investigates conditions under which the role of an attacker can be reverted from malicious to benevolent, that is, to help the diagnoser to diagnose faults. The paper further applies diagnosability theory to investigate conditions under which the presence of the attacker can be detected.