Diagnostic Information for Control-Flow Analysis of Workflow Graphs (aka Free-ChoiceWorkflow Nets)

Abstract

A workflow graph is a classical flow graph extended by concurrent fork and join. Workflow graphs can be used to represent the main control-flow of e.g. business process models modeled in languages such as BPMN or UML activity diagrams. They can also be seen as compact representations of free-choice Petri nets with a unique start and a unique end. A workflow graph is said to be sound if it is free of deadlocks and exhibits no lack of synchronization, which correspond to liveness and safeness of a slightly modified version of the corresponding Petri net. We present a new characterization of unsoundness of workflow graphs in terms of three structural, i.e., graphical error patterns. We also present a polynomialtime algorithm that decides unsoundness and returns for each unsound workflow graph, one of the three structural error patterns as diagnostic information. An experimental evaluation on over 1350 workflow graphs derived from industrial business process models suggests that our technique performs well in practice.