Many everyday products are manufactured using complex production systems that require experienced specialist staff to operate. Due to the shortage of skilled workers and demographic change, it is becoming increasingly difficult to maintain stable and efficient production. In order to still be able to produce competitively, production facilities must also be able to be operated by inexperienced employees. One key to this can be to expand production machines and systems with intelligent assistance systems in order to reduce training times and increase employee satisfaction and overall productivity.
In ADELeS (anomaly correction through extracted expert knowledge and learning systems), we are developing an AI-based assistance system to detect and correct quality deviations and errors during production. The quality assurance process combines learning systems with expert knowledge via neuro-symbolic learning. The latter is extracted based on experience and data.
Anomaly mitigation through extracted expert knowledge and learning systems
In research project ADELeS, the University of Augsburg, the Friedrich-Alexander University Erlangen Nuremberg and the companies REHAU Industries SE & Co. KG and XITASO GmbH have joined forces to tackle these challenges in the context of an extrusion process for the extrusion of edgebanding, which are used, for example, in furniture production.
REHAU uses 40-60m long extrusion lines for the production of edgeband, which consist of a large number of individual machines for separate processes (e.g. extruder, cooling, printing units).
In order to predict the expected product quality on the one hand (predictive quality) and to suggest parameter adjustments to the operators of the extrusion line to mitigate quality anomalies on the other, the research approach provides for the combination of expert knowledge with machine learning methods based on time series data from the extrusion line.
The assistance systems that presents these suggestions is triggered by automatic detection of quality anomalies.
Expert knowledge is captured, formalized and quantified through experience- and data-based knowledge extraction. The result is used in both explainable learning classifier systems and neuro-symbolic approaches.
The focus of XITASO is on the further development of the innovative methodology for data-based knowledge extraction developed in the predecessor project AIPE, the representation of this procedural knowledge in knowledge graphs, their representation in vectors, and the neuro-symbolic combination of expert knowledge and neural networks.
Bavarian Ministry of Economic Affairs, Regional Development and Energy
Bayerisches Verbundförderprogramm (BayVFP)
– Digitisation funding line –
Information and communication technology funding programme
VDI/VDE Innovation + Technik GmbH
KI-Produktionsnetzwerk Augsburg
With research and innovation projects, we explore the potential of tomorrow’s technologies. For this purpose, we are always looking for committed colleagues who would like to continue and shape the XITASO path with us.
Dr. Richard Nordsieck
Tel. +49 821 885882-89
richard.nordsieck@xitaso.com