Title: A unified method to decentralized state inference and fault diagnosis/prediction of discrete-event systems

Abstract: The state inference problem and fault diagnosis/prediction problem are fundamental topics in many areas. In this paper, we consider discrete-event systems (DESs) modeled by finite-state automata (FSAs). There exist results for decentralized versions of the latter problem but there is almost no result for a decentralized version of the former problem. We propose a decentralized version of strong detectability called co-detectability which implies that once a system satisfies this property, for each generated infinite-length event sequence, at least one local observer can determine the current and subsequent states after a common observation time delay. We prove that the problem of verifying co-detectability of FSAs is NP-hard. Moreover, we use a unified concurrent-composition method to give PSPACE verification algorithms for co-detectability, co-diagnosability, and co-predictability of FSAs, without any assumption or modifying the FSAs under consideration, where co-diagnosability is firstly studied by [Wang, Yoo, \& Lafortune 2007] under a liveness assumption, while co-predictability is firstly studied by [Kumar \& Takai 2010] by adding at each deadlock state an unbounded unobservable trace. This modification to FSAs used in [Kumar & Takai 2010] does not preserve co-diagnosability or co-predictability. Moreover, by our proposed unified method, one can see that in order to verify co-detectability, more technical difficulties will be met compared to verifying the other two properties, because in co-detectability, generated outputs are counted, but in the latter two properties, only occurrences of events are counted. For example, when one output was generated, any number of unobservable events could have occurred. The PSPACE-hardness of verifying co-diagnosability is already known. In this paper, we prove the PSPACE-hardness of verifying co-predictability.