Building Hierarchical Workflow Nets for Discrete-Event Processes Discovery

This paper presents a method to build hierarchical workflow nets (WFN) from a log of event traces representing the behavior of a discrete event process. The proposed method operates in two phases: event bundling, and model construction. The bundling of events is performed according to traces of the event log. The event alphabet is partitioned into clusters called Ordered Execution Sets (OES), in which all the events in the class always occur together in the same traces of the event log; besides, the cluster contains a unique subset of concurrent events. Afterwards, the two-level hierarchical WFN is constructed. For every OESj, a Petri Net (PN) substructure is built according to the causal and concurrent relations between events in the bund; each substructure is net of level two represented by a macro-transition within the net of level one. Macro-transitions are linked through places following simple rules that consider the boundary events of the OES substructures. The algorithms derived from the method, which have polynomial-time complexity, have been implemented and tested on artificial event logs.