Refining Discovered Petri Nets by Sequencing Repetitive Components

Abstract. The problem of refining a Petri net (PN) discovered from a single sequence S of events T generated by discrete event processes is addressed. The refinement aims to reduce a possible exceeding language in the discovered model. A technique that extends a t-invariant based discovery method is presented. Given a discovered PN and its set of minimal t-invariants Y, the technique analyses the execution of the t-components in S and determines a sequencing pattern SY that schedules the execution of t-components in the initial PN. Then, SY is used to discover a PN’ that uses transitions in T and new places; PN’ schedules representative transitions of each t-invariant in SY. The refined model is obtained by merging the representative transitions of both PN and PN’ if the pattern is discovered. A first result coping with a subclass of safe PN in which each t-component has at least a transition not shared with any other component is reported.