Transforming Unstructured Process Models: Translating Implicit and Scattered Parallel Structures into Structured Hierarchies
The usage of process models in healthcare is low. Even though this industry could benefit greatly from an increased usage of process models, since these models help with the understanding of processes taking place. We therefore set out to find a way for increasing the understandability of these models. The goal is to transform existing models into better understandable models. The focus of this research is not on the construction of a good model, but on improving an existing model without changing its behavior. As suggested by literature, hierarchy is used to achieve this improvement in understandability.
For the research PN2HSC was used, a transformation implementation in the GrGen software suite. This transformation tool transforms petri nets to state charts with hierarchy. Though PN2HSC can transform structured process models without a problem, it is not able to transform unstructured models correctly.
Unfortunately, models from healthcare and other industries are often unstructured, thereby severely limiting the usage of PN2HSC. This research aimed to improve PN2HSC by extending it to handle unstructured models. Four patterns of unstructured behavior where found in literature, which served as the basis for developing rules for finding violations of structure in process models.
Edges that cause this unstructuredness are temporary hidden from the model, so a structured model remains, that can be transformed using PN2HSC (or another transformation tool). Afterwards the hidden edges are made visible again.
Applying these rules to models showed that they do not take multi‐level hierarchies into account: parallelism as part of another parallel path is incorrectly marked as a violation. A filter is created to prevent this incorrect marking. Additionally, resource structures and structures where there is a crossover between parallel paths, require additional rules to be handled correctly. Thus the four existing patterns were extended with two additional patterns. Subsequently, additional rules are created as well.
The results were validated using real‐life models from the healthcare field, to see how these patterns perform in real situations. It was noticed that the degree of success varied: the rules created can uncover latent hierarchy in models, but they will not find hierarchy where there is originally none. It was noted that finite models, with a start and end node, performed better in contrast to continuing (infinite) models. Also, black token nets outperformed colored nets. There should not be any conclusions connected to these observations, since the number of models tested was too low to substantiate any claims. Though it seems possible that certain types of models contain a higher degree of hierarchy than other model types.
The validation did show that the developed rules are very successful in uncovering latent hierarchy, and some complicated models benefitted greatly from the additional transformation rules. These models became much easier to understand, due to hierarchical presentation of the model.
Get the PDF from the TU/e library.