Discovering a Class of Workflow Nets with Reduced Exceeding Language

Abstract

This paper deals with the automated discovery of workflow nets (WFN) from logs of event traces that represent the observed behaviour of a discrete event process. The aim is to obtain a WFN N that represents closely the language of the log λ, whose surplus language of the model ℒ(N)\λ is low. The approach pursues to build a subclass of block-structured WFN composed of block subnets with cycles discovered from subsets of λ. This paper focuses on the discovery of blocks from subsets of traces that have the same event alphabet. In the proposed technique, a block is formed by two subnets using the same event labels: a) N <inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">&lt;</inf> , a WFN state-machine derived from the causal relations between events, and b) N <inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">C</inf> , a WFN marked-graph that allows firing once all the different transitions. Then, N is obtained by the synchronous composition of N <inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">&lt;</inf> and N <inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">C</inf> (N<||N <inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">C</inf> ). The algorithms derived from the method, which have polynomial time complexity, build highly precise models.