Better Late than Never: On Epistemic Diagnosability of Discrete Event Systems

We investigate the diagnosability verification problem in the framework of discrete-event systems. Most of the existing works on this topic assume that faults are related to the internal behaviors of the system such as occurrences of particular events. In this work, motivated by information-flow security considerations, we model faults as some critical information leakages of the system to an intruder, which may have different observations from the system user. Specifically, we say that a fault occurs if the intruder knows that the system has passed by a secret state. We present a formal notion called epistemic diagnosability to capture whether or not the system user can always detect, based on its own observation, the critical information leakage to an intruder within a bounded delay. We show that this new notion subsumes the standard notion of event-based diagnosability. Furthermore, an effective algorithm is provided to verify this new notion.