Decision support is essential when humans are responsible for choosing critical courses of action about large, complex, distributed, partially-observable, dynamical systems. When the system behavior looks abnormal, a decision is expected to be taken based on the root cause of the undesired behavior. Still, several alternative root causes, or candidates, can explain the same observed behavior, each of them consisting of a set of faults. Since finding such candidates requires heavy diagnostic reasoning, a vast collection of automated diagnosis tools have been proposed in the literature. When the diagnosis tool is model-based, the knowledge stored in the system model is drawn not only from human experts but also from design and/or operation data. Up to a few years ago, model-based diagnosis was invariably set-oriented: a candidate is a set of faults that explains the observation(s). Recently, a temporal-oriented perspective to diagnosis of dynamical systems was proposed: a candidate is a chronological sequence of faults, and the set of all candidates is a regular language over the alphabet of faults. This new perspective may help a practitioner better understand what has happened in the system, thus supporting the decision-making process more adequately. This chapter deals with the decision support provided by a model-based temporal-oriented approach to diagnosis of partially-observable discrete-event systems. A (distributed) discrete-event system consists of several components that are modeled as communicating automata. Three temporal-oriented diagnosis techniques are investigated: (1) blind diagnosis, with no compiled knowledge, (2) greedy diagnosis, with (offline) total knowledge compilation, and (3) lazy diagnosis, with initial (offline) partial knowledge compilation and subsequent (online) multiple extensions of the compiled knowledge. Experimental results suggest that only lazy diagnosis may be viable in real application domains, as both blind and greedy diagnoses suffer from serious complexity difficulties.
Supporting decision-making in diagnosis of discrete-event systems by model-based temporal techniques
Gian Franco Lamperti
;Stefano Trerotola;Marina Zanella
2024-01-01
Abstract
Decision support is essential when humans are responsible for choosing critical courses of action about large, complex, distributed, partially-observable, dynamical systems. When the system behavior looks abnormal, a decision is expected to be taken based on the root cause of the undesired behavior. Still, several alternative root causes, or candidates, can explain the same observed behavior, each of them consisting of a set of faults. Since finding such candidates requires heavy diagnostic reasoning, a vast collection of automated diagnosis tools have been proposed in the literature. When the diagnosis tool is model-based, the knowledge stored in the system model is drawn not only from human experts but also from design and/or operation data. Up to a few years ago, model-based diagnosis was invariably set-oriented: a candidate is a set of faults that explains the observation(s). Recently, a temporal-oriented perspective to diagnosis of dynamical systems was proposed: a candidate is a chronological sequence of faults, and the set of all candidates is a regular language over the alphabet of faults. This new perspective may help a practitioner better understand what has happened in the system, thus supporting the decision-making process more adequately. This chapter deals with the decision support provided by a model-based temporal-oriented approach to diagnosis of partially-observable discrete-event systems. A (distributed) discrete-event system consists of several components that are modeled as communicating automata. Three temporal-oriented diagnosis techniques are investigated: (1) blind diagnosis, with no compiled knowledge, (2) greedy diagnosis, with (offline) total knowledge compilation, and (3) lazy diagnosis, with initial (offline) partial knowledge compilation and subsequent (online) multiple extensions of the compiled knowledge. Experimental results suggest that only lazy diagnosis may be viable in real application domains, as both blind and greedy diagnoses suffer from serious complexity difficulties.File | Dimensione | Formato | |
---|---|---|---|
dss.pdf
accesso aperto
Descrizione: Capitolo preceduto dalle pagine iniziali del libro
Tipologia:
Full Text
Licenza:
Copyright dell'editore
Dimensione
1.01 MB
Formato
Adobe PDF
|
1.01 MB | Adobe PDF | Visualizza/Apri |
I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.