Diagnosability of a Class of Discrete Event Systems Based on Observations

Abstract

Diagnosability of a system is the ability to detect a fault occurring in the system within a limited time delay. Diagnosability verification precedes fault detection and isolation. For the diagnosability verification, traditionally state based sensor outputs are utilized. Taking advantage of technology developments in the sensor area, this thesis proposes an event-based approach to diagnosability verification. The seminal work of Sampath et al.(1995) proposes a regularity-like condition for diagnosability verification of discrete event system. A new O- diagnosability condition for diagnosability verification is proposed in this thesis based on event outputs called observations. Necessary and sufficient condition for O-diagnosability of a system is derived. Based on the O-diagnosability condition, diagnosability verification complexity is shown to be linear in the power set of alphabets including events and observations whereas in the Sampath s case, the diagnosability verification complexity is exponential. Also, it is shown that a system which is not diagnosable in the sense of Sampath s diagnosability condition becomes O-diagnosable when observations are included.In line with the modular diagnosability approach based on Sampath s diagnosability condition, developed by many researchers, this thesis proposes a new condition named Component O-Diagnosability, or, in short, CO-diagnosability. Diagnosability verification of a system from the CO-diagnosability condition simplifies the system O-diagnosability verification process. Theoretical results are derived for O-diagnosability verification of the system using the concept of CO-diagnosability. A system which satisfies the theoretical results will be O-diagnosable at the system level based on the O-diagnosability of the components incrementally added. Hence the need to build a global diagnoser is avoided and component models (subsystem models) and component diagnosers