To ensure successful process executions, process models must contain all important information for reflecting reality appropriately. This includes relevant process details (RPDs) which describe specifications or configurations of tasks affecting process success. However, RPDs are not always known in advance since they are hard to detect, even by process experts. Furthermore, RPDs are often not directly process-related but more context-related. Image data that are handled in a process have great potential to contain such hidden but crucial process information. Approaches that aim at identifying and extracting RPDs, e.g. from image data, mostly come with demanding prerequisites like the availability of large amounts of execution data. Consequently, these techniques prove impractical for the implementation in small enterprises, as such entities typically lack access to a sufficiently extensive dataset. In this paper, we demonstrate how RPDs can be extracted from images recorded during process execution by using Association Rule Mining (ARM) without the demand for huge input data. In an experimental setup, different ARM algorithms are evaluated in two use cases addressing pick-and-place scenarios from a real manufacturing process. The results confirm the effectiveness of the developed approach, demonstrating its suitability for smaller companies.
Loading....