Navigation

Bachelor's Thesis

Unsupervised learning for the classification of process steps in spatial-temporal data

Advisors

Matthias Zürl (M.Sc.) , André Hanak (Frauenhofer IIS)

Duration

05/2020-08/2020

Abstract

Process Mining is an up and coming sub-field of process management. It aims to analyze and improve
processes, based on their digital footprint. This limits the scope of process mining, by excluding
physical tasks which are not being documented digitally. [1]
One approach to digitize physical actions is through IoT based positional tracking. This results in a
semantic gap between the spatial-temporal data and interpretable process steps, which are required for
process mining.
Past work of the Fraunhofer Institute for Integrated Circuits IIS utilized this approach in a
manufacturing context. The industrial production site of a manufacturing company was equipped with
an IoT-Technology to track components during their production cycle. Afterwards, a simple distancebased
algorithm was used to map the spatial-temporal data to manually determined physical process
steps (e.g polishing or drying of a component). [2]
The proposed work attempts to develop an approach, which utilizes clustering and supervised learning
methods, to train a classifier that outperforms the previous analytical approach.
It attempts to lower the cost of application by implementing the automated detection of process steps,
effectively minimizing manual modeling and data-labeling. The labeled data-set is then used to train a
classifier.
The final model will be compared to the analytical approach based on cost, accuracy, and scope of
potential application.
Lack of expressive data will be compensated by developing a framework, which creates synthetic
data-sets, by simulating physical processes based on observational parameters.
This provides a base truth for meaningful validation and ensures that edge cases can be analyzed.

 

References:

[1] van der Aalst, W. (2016): Process Mining. Data Science in Action. Second Edition.
Heidelberg: Springer.
[2] Hanak A., Janda P., Meyer S. (2018). Optimizing Production Processes with Wireless Smart
Sensors and Tracking. Wireless Congress 2018 Munich.