Discovering Workflow Performance Models from Timed Logs

Contemporary work ow management systems are driven by explicit process models, i.e., a completely speci ed work ow design is required in order to enact a given work ow process. Creating a work ow design is a complicated time-consuming process and typically there are discrepancies between the actual work ow processes and the processes as perceived by the management. Therefore, we have developed techniques for discovering work ow models. Starting point for such techniques are so-called \work ow logs" containing information about the work ow process as it is actually being executed. In this paper, we extend our existing mining technique [4] to incorporate time. We assume that events in work ow logs bear timestamps. This information is used to attribute timing such as queue times to the discovered work ow model. The approach is based on Petri nets and timing information is attached to places. This paper also presents our work ow-mining tool EMiT. This tool translates the work ow log of several commercial systems (e.g., Sta ware) to an independent XML format. Based on this format the tool mines for causal relations and produces a graphical work ow model expressed in terms of Petri nets.