Beyond Soundness: On the Semantic Consistency of Executable Process Models

Executable business process models build on the specification of process activities, their implemented business functions (e.g., Web services) and the control flow between these activities. Before deploying such a model, it is important to verify control-flow correctness. A process is sound if its control-flow guarantees proper completion and there are no deadlocks. However, a sound control flow is not sufficient to ensure that an executable process model indeed behaves as expected. This is due to business functions requiring certain preconditions to be fulfilled for execution and having an effect on the process (postconditions). Semantic annotations provide a means for taking such further aspects into account. Inspired by OWL-S and WSMO, we consider process models in which the individual activities are annotated with logical preconditions and postconditions specified relative to an ontology that axiomatizes the underlying business domain. Verification then means to determine whether the interaction of control flow and logical states of the process is correct. To this end, we formalize the semantics of annotated processes and point out which kinds of flaws may arise. We then identify a class of processes with restricted semantic annotations where correctness can be verified in polynomial time; and we prove that the semantic annotations cannot be generalized without losing computational efficiency. The paper is written at a semi-formal level using an illustrative example, details can be looked up in a longer technical report.