Vortrag von Dr. Lars Bittrich zum Thema Maschinelles Lernen in der Produktionsoptimierung am Campus Ost (KIT):

Am Campus Ost des KIT fand ein Vortrag von Dr. Lars Bittrich statt, der sich an die Forschergruppe 5339 und deren Gäste richtete.

Dr. Bittrich präsentierte seine aktuellen Erkenntnisse zu dem Thema „Using Machine Learning to Increase Production Accuracy in Tailored Fiber Placement“.

Abstract:

This presentation explores the integration of machine learning (ML) to enhance production accuracy in Tailored Fiber Placement (TFP), an advanced embroidery process for manufacturing high-performance composite preforms. Drawing on results from the research project SIQ4TFP, we demonstrate how TFP achieves accurate stitch placement precision but faces challenges due to systematic (controllable) and random (uncontrollable) deviations in fiber bundle alignment. Leveraging a dataset of over 400 samples - analyzed via optical image recognition with ~2000 data points per sample - we identify correlations to productions parameters in these deviations. Simple neural networks are shown to effectively predict systematic errors, enabling pre-production compensation. Furthermore, we present a calibration strategy to address select random deviations by retraining models immediately prior to manufacturing, ensuring adaptability to dynamic process conditions. This approach bridges engineering-driven process optimization with ML techniques, offering a practical framework to improve TFP reliability for aerospace, automotive, and structural applications. The findings underscore the synergy between data-driven m