Automated modelling of discrete-event processes. Discovering Petri nets including silent transitions by classifying event traces

Abstract

ABSTRACT This paper addresses the problem of automated modelling of discrete-event processes from event traces that capture the process behaviour. A novel method for building Petri nets (PN) that include silent transitions is presented. The method discovers workflow nets (WFN), a class of PN, from logs of event traces λ. The proposed approach is based on a classification of traces such that λ is partitioned into four classes: normal traces (λN), which do not involve the firing of silent transitions; skip traces (λS), redo traces (λR), and switch traces (λSW), which involve the firing of skip, redo, and switch type silent transitions, respectively. Afterwards, traces in λN are used to build a base WFN N, which may include silent transitions of type initialise and finalise. Then, N is refined by adding the silent transitions corresponding to the classes oftraces λS, λR, and λSW. The algorithms of the method have polynomial-time complexity; implementation and tests using artificial event logs are presented.