Authors:
Markus Bauer
1
;
Benjamin Uhrich
2
;
Martin Schäfer
3
;
Oliver Theile
3
;
Christoph Augenstein
2
and
Erhard Rahm
2
Affiliations:
1
Institute for Applied Informatics, Goerdelerring 9, Leipzig, Germany
;
2
Center for Scalable Data Analytics and Artificial Intelligence, Humboldtstraße 25, Leipzig, Germany
;
3
Siemens AG, Siemensdamm 50, Berlin, Germany
Keyword(s):
Artificial Intelligence, Additive Manufacturing, Physics-Informed Neural Networks, Multi-Modal Imaging, Quality Monitoring.
Abstract:
With emerging technologies such as high-precision Laser Powder Bed Fusion (LPBF), rapid prototyping has gained remarkable importance in metal manufacturing. Furthermore, cloud computing and easy-to-integrate sensors have boosted the development of digital twins. Such digital twins use data from sensors on physical objects, to improve the understanding of manufacturing processes as a whole or of certain production parameters. That way, digital twins can demonstrate the impact of design changes, usage scenarios, environmental conditions or similar variables. One important application of such digital twins lies in early detection of manufacturing faults, such that real prototypes need to be used less. This reduces development times and allows products to be individually, affordable, powerful, robust and environmentally friendly. While typically simple USB-camera setups or melt-pool imaging are used for this task, most solutions are difficult to integrate into existing processes and hard
to calibrate and evaluate. We propose a digital-twin-based solution, that leverages information from camera-images in a self-supervised fashion, and creates a heat transfer based AI quality monitoring. For that purpose, artificially generated labels and physics simulation were combined with a multisensor setup and supervised learning. Our model detects printing issues at more than 91% accuracy.
(More)