Automatic generation of optimized process models from declarative specifications

Abstract

Process models often are generic, i. e., describe similar cases or contexts. For instance, a process model for commissioning can cover both vehicles with an automatic and with a manual transmission, by executing alternative tasks. A generic process model is not optimal compared to one tailored to a specific context. Given a declarative specification of the constraints and a specific context, we study how to automatically generate a good process model and propose a novel approach. We focus on the restricted case that there are not any repetitions of a task, as is the case in commissioning and elsewhere, e. g., manufacturing. Our approach uses a probabilistic search to find a good process model according to quality criteria. It can handle complex real-world specifications containing several hundred constraints and more than one hundred tasks. The process models generated with our scheme are superior (nearly twice as fast) to ones designed by professional modelers by hand.