Automatic discovery of data-centric and artifact-centric processes

Process discovery is a technique that allows for automatically discovering a process model from recorded executions of a process as it happens in reality. This technique has successfully been applied for classical processes where one process execution is constituted by a single case with a unique case identifier. Data-centric and artifact-centric systems such as ERP systems violate this assumption. Here a process execution is driven by process data having various notions of interrelated identifiers that distinguish the various interrelated data objects of the process. Classical process mining techniques fail in this setting. This paper presents a fully automatic technique for discovering for each notion of data object in the process a separate process model that describes the evolution of this object, also known as artifact life-cycle model. Given a relational database that stores process execution information of a data-centric system, the technique extracts event information, case identifiers and their interrelations, discovers the central process data objects and their associated events, and decomposes the data source into multiple logs, each describing the cases of a separate data object. Then classical process discovery techniques can be applied to obtain a process model for each object. The technique is implemented and has been evaluated on the production ERP system of a large retailer.