Discovering Instance Spanning Exceptions from Process Execution Logs

Exceptions in process execution occur frequently and require appropriate handling strategies in order to avoid undesired consequences. For quality control in manufacturing processes, for example, when a blade gets missing, the set of affected instances is put on hold until the blade is found. As it can be seen from this example, the exception affects multiple instances and is hence denoted as instance spanning exception. Exceptions leave footprints in the process execution logs of the affected instances. Hence, process execution logs provide a valuable data source for discovering and analyzing exceptions. However, the discovery of instance spanning exceptions is still an open challenge. Thus, this paper proposes i) a classification of instance spanning exceptions based on literature and a set of real-world examples, followed by ii) a description of how instance spanning exceptions manifest in process execution logs along with an elicitation of minimal requirements for enabling their discovery, and iii) five instance spanning exception discovery algorithms, one for each class. The discovery algorithms are implemented and evaluated on a set of synthetic process execution logs reflecting real-world instance spanning exceptions and on a real-world process execution log from the public transport domain demonstrating the feasibility as well as applicability of the presented algorithms.