Real added value for companies through AI
On 2 December 2024, the partners of the it's OWL research project Artificial Intelligence for Scarce Data (AI4ScaDa) met for the fifth coordination meeting at Miele to present the results obtained so far and discuss them together. The innovative project combines the expertise of the companies SU BIOTEC, GEA and Miele with the competences of the Center for Applied Data Science (CfADS) at Bielefeld University of Applied Sciences and the Institute Industrial IT (inIT) at Technische Hochschule Ostwestfalen-Lippe (TH OWL).
Process optimisation driven by AI
AI4ScaDa aims to optimise processes that work with limited and cost-intensively generated data - so-called "scarce data" - through the use of Artificial Intelligence (AI). The project is characterised by three different fields of application as well as a wide-ranging partnership and offers new approaches for data-based innovations in industry.
Modelling of a donor plant cultivation
One example of the importance of process-controlling measures is plant breeding, where factors such as the choice of light spectrum or temperature adjustments at different stages of development are decisive for breeding success. AI methods support this process by estimating the most important parameters and settings and providing valuable knowledge for optimising the cultivation process. SU BIOTEC has developed a donor plant cultivation modelling system for this purpose, the generated data of which will be recorded and evaluated beyond the project.
Preserving expert knowledge
GEA, one of the world's largest suppliers of systems for the mechanical clarification and separation of liquids in various industries, faces the challenge of retaining the expert knowledge of laboratory engineers that has been built up over decades. As part of the project, the expert knowledge was collected in a targeted manner through data augmentation and summarised in AI models. Applying the interpretable models, the experts were able to validate the models and check their plausibility. This knowledge will be made available to other employees for support in the future.
"It was impressive to see that the experts were able to validate the plausibility of the models. We have thus shown that we can achieve real added value for companies with the models," agree Julian Bültemeier, research associate in Prof. Dr. Volker Lohweg's Discrete Systems working group, and Christoph-Alexander Holst, research group leader of this working group.
Learning on scarce time series
Miele's use case in the AI4ScaDa project involves the development of intelligent tumble dryers. In particular, time series from endurance tests are available for this, which are numerous but often only record the "good" condition of the machines, which means that the variance in the data is very low. As part of the project, methods for processing such time series were therefore investigated. The aim is to create a blueprint for learning and handling such time series in order to be able to better analyse the data in the future.
Generalisation of the AI workflow
The project is working on merging the implemented solution modules into software to generate the AI workflow. An initial live demonstration of this workflow was presented at the milestone meeting. Work is currently underway on the final components. The consortium is evaluating the usability of this workflow in order to enable a broad transfer and usability in industry in the future.
More about AI4ScaDa
The it's OWL AI4ScaDa innovation project is dedicated to researching special AI techniques that may be inserted profitably in the context of scarce data. In contrast to big data, the term "scarce data" refers to a limited amount or incomplete data that are often collected in laboratories. The key challenge is that AI systems are heavily dependent on the quality and quantity of available data. If only very scarce data are available, AI processes can often barely keep up with human expertise. AI4ScaDa aims to close this gap and support companies in exploiting the benefits of Artificial Intelligence even in data-scarce environments.