Active Diagnosis of Time-Interval Automata: Time Perspectives

Abstract

Language diagnosability captures the capability of the system to detect faults based on observations. When a system is not diagnosable, supervisory control can be used to enforce its diagnosability to prevent the faults from occurring silently, known as the active diagnosis problem. Note that in the context of timed discrete-event systems, an observation contains not only the sequence of the events but also their time information. Therefore, two sequences consisting of the same events but with different occurrence time instants can still reveal the occurrence of faults. This fact motivates us to consider enforcing the diagnosability of a timed discrete-event system by regulating the occurrence time instants of certain controllable events. In this paper, we first construct a verifier for a time-interval automaton to verify its diagnosability. Then, based on the verifier, we enforce the diagnosability of a time-interval automaton by restricting the time intervals of certain controllable events and disabling some controllable events. Note to Practitioners—Fault diagnosis and active diagnosis play a critical role in ensuring the reliability, safety, and efficiency of systems across various industries, ranging from automotive and aerospace to healthcare and smart grids. Discrete-event systems, as general models for complex man-made systems, are well-studied for modeling digital computer systems in the above scenarios. Early detection and correction of faults contribute to improved performance, reduced downtime, and enhanced overall system functionality. This work investigates the active diagnosis problem, i.e., design a supervisor to enforce diagnosability, for discrete-event systems modeled by time-interval automata. The control policy combines the time and logical information of the system, thereby allowing the closed-loop systems to retain more of the original system behavior. Time-interval automata is a model that is not complicated but is closer to actual engineering systems than finite automata, providing new insights for control engineers in modeling and control.