On 19 February 2025, Prof. Dr. Volker Lohweg's working group at Institute Industrial IT (inIT) of Technische Hochschule Ostwestfalen-Lippe (TH OWL) organised a customer clinic together with the consortium of the AI4ScaDa project. The innovative AI workflow, which was developed as part of the project for test planning and analysis, was tested intensively.
Insights into the AI-supported workflow
The newly developed demonstrator covers the entire process from test planning to the continuous improvement of the models. The workflow includes the following core functions:
- Experimental planning and design: Supports systematic experimental planning for efficient data collection.
- Data transformation into tabular data records: Automated structuring and preparation of data for further analysis.
- Optimisation of the data records: Space-filling optimisation of the initial data in order to capture as much process knowledge as possible with as scarce data as possible.
- Labelling of the data sets: Recording and labelling of the output variables relevant for the modelling.
- Analysing data with a GUIDE tree: Use of a decision tree for transparent and interpretable data analysis.
- Evaluation of new data points: Application of existing models to new data for sound forecasting or process optimisation.
- Iterative model improvement through active learning: Identification of areas with high uncertainty and targeted data generation to improve model accuracy.
Methodological principles of the demonstrator
At the beginning of the event, the methodological basics of the AI workflow were explained, including
- Sliced Latin Hypercube Designs (SLHD): This method is applied in experimental design to efficiently and uniformly cover the parameter space with both numerical and categorical parameters.
- GUIDE decision tree: An interpretable AI technique that enables robust and transparent prediction models.
- Active learning for decision trees: This approach focuses on the targeted generation of new data points in areas of high uncertainty in order to gradually improve the accuracy of the models.
The use cases in the project have shown that new approaches are required for these methods in order to support and improve targeted data collection with limited data. Further details and information on the methods can be found in the publications: https://plattform.its-owl.de/projects/ai-for-scarce-data-maschinelles-lernen-und-informationsfusion-zur-nachhaltigen-nutzung-von-labor-und-kundendatenName/AXYP8egrEQ
Interactive workshop and practical test
After the theoretical introduction, the participants were able to test the demonstrator in an interactive workshop. The individual components of the workflow were tested using a practical use case. The tasks took them step by step through the entire process - from test planning to data-driven decision-making. This practical test provided valuable insights into the functionality and user-friendliness of the graphical user interface (GUI). At the same time, numerous suggestions were made for further improving the interface, particularly with regard to ease of use, visualisation options and interactivity.
Productive event with a positive outlook
Christoph-Alexander Holst, research group leader of Prof. Dr Volker Lohweg's working group, emphasises: "The customer clinic confirmed that the GUI developed is intuitive to use and effectively supports the entire workflow." Research associate Julian Bültemeier, who was responsible for conducting the Customer Clinic on 19 February 2025, agrees: "The feedback from the test phase will now be taken to further optimise the tool before the workflow will be published on the it's OWL innovation platform at the end of March. Many thanks to everyone involved for the valuable feedback and dedicated collaboration!"