Component level diagnosability of discrete event systems based on observations

Diagnosability of a discrete event system is usually dealt at the system level entailing synchronous composition of component automata leading to state explosion problem. However, many of the systems are modular in nature allowing localized detection and isolation of faults. Moreover, developments in sensor technology allow direct detection of faults based on sensor output which are denoted as observations in this paper. Combining modularity and observations, we propose a new concept of O-diagnosability based on observations at the subsystem or component level to make the diagnosability verification less complex. The concepts of monolithic (system) O-diagnosability and CO-diagnosability (system to component) are introduced and necessary and sufficient conditions for O-diagnosability are derived. Theoretical results on the relation between monolithic O-diagnosability and CO-diagnosability support the system level diagnosability verification through component level analysis in a progressive way. Computational complexity for the proposed diagnosability verification is shown to be of the order of n² where n is the largest number of diagnoser states of a component of the system.