Decentralized failure diagnosis in discrete event systems

In this paper, we develop and analyze decentralized failure diagnosis schemes for systems that can be modeled as finite state machines (FSMs). We consider a scenario where we are given multiple local diagnosers (with partial access to the events occurring in the system under diagnosis) and focus on designing a global coordinator which gathers the estimates of the local diagnosers at irregularly spaced time intervals and combines them to reach the final diagnosis decision. The functionality of the system and the local diagnosers is known, and each local diagnoser does not necessarily observe the same events as other diagnosers. We require the global coordinator to perform diagnosis using only simple processing of the results it receives from the local diagnosers. More specifically, we consider two possible strategies for the coordinator to reach a final diagnosis decision: (i) intersection of local estimates or (ii) consultation of a stored table that is constructed offline. In both cases, the global coordinator can operate in real time without knowledge of the functionality of the system or the local diagnosers. For each scheme, we derive conditions for single-request diagnosability and multiple-request diagnosability. We also explore the trade-offs between the two schemes in terms of processing and memory requirements on the global coordinator