Scaling Usability of ML Analytics with Knowledge Graphs: Exemplified with A Bosch Welding Case

Baifan Zhou, Dongzhuoran Zhou, Jieying Chen, Yulia Svetashova, Gong Cheng, Evgeny Kharlamov

Abstract: Automated welding is heavily used in the automotive industry to produce car bodies by connecting metal parts with welding spots. Modern welding solutions and manufacturing environments produce a high volume of heterogeneous data. Analytics of these data with machine learning (ML) can help to ensure high quality of welding operations. However, due to heterogeneity of data and application scenarios, it is challenging to scale usability of such ML analytics is challenging, namely saving time in developing new solutions and reusing already developed solutions. We address this challenge by relying on knowledge graphs (KG) that not only conveniently allow to integrate welding data, but also to serve as a basis for layering ML-based analytical applications, thus enabling quality monitoring of welding operations. In this work we focus on the construction of a KG for welding that is tailored towards further use for ML applications. Furthermore, we demonstrate how selected ML analytical tasks are supported by this KG.