Fixing Nondeterminism in Large Discrete-Event Knowledge

Abstract

Discrete-event knowledge (DEK) consists in a variety of automaton-based data structures which are generated by knowledge-based and engineering systems, including intelligent diagnosis systems. Model-based diagnosis methods for discrete-event systems (DESs) are typically supported by large automata that allow for the efficient generation of the candidate diagnoses when the DES is being operated. However, the generation of the DEK may involve a determinization step, where a nondeterministic finite automaton (NFA) is transformed into an equivalent deterministic finite automaton (DFA) by means of the classical Subset Construction algorithm. When this determinization is performed online and the NFA is large, the diagnosis engine may slow down to such an extent that the entire diagnosis task is jeopardized. This is why a novel determinization algorithm is proposed, called Quick Subset Construction, which, instead of generating from scratch the equivalent DFA, removes the nondeterminism by operating directly on the NFA. This way, the larger the NFA and the smaller the portion of nondeterminism involved, the faster Quick Subset Construction will be over Subset Construction. Massive experimentation on large automata has confirmed this result empirically.