Merging alignments for decomposed replay

In the area of process mining, conformance checking aims to find an optimal alignment between an event log (which captures the activities that actually have happened) and a Petri net (which describes expected or normative behavior). Optimal alignments highlight discrepancies between observed and modeled behavior. To find an optimal alignment, a potentially challenging optimization problem needs to be solved based on a predefined cost function for misalignments. Unfortunately, this may be very time consuming for larger logs and models and often intractable. A solution is to decompose the problem of finding an optimal alignment in many smaller problems that are easier to solve. Decomposition can be used to detect conformance problems in less time and provides a lower bound for the costs of an optimal alignment. Although the existing approach is able to decide whether a trace fits or not, it does not provide an overall alignment. In this paper, we provide an algorithm that is able provide such an optimal alignment from the decomposed alignments if this is possible. Otherwise, the algorithm produces a so-called pseudo-alignment that can still be used to pinpoint non-conforming parts of log and model. The approach has been implemented in ProM and tested on various real-life event logs.