The ReAddi project, funded under the Förderkennzeichen 13N15112, ran from 1 October 2019 to 30 September 2023 and aimed to create an intelligent, data‑driven additive manufacturing process chain. US U Software AG was a key partner, providing data‑science expertise, deep‑learning development, and a distributed digital backbone for raw data, processed data, algorithms, and deep‑learning models. The consortium also included Bosch, Daimler, and other industrial partners, with Bosch hosting the final Git LFS‑based collaboration platform and Daimler offering a Databricks environment that was ultimately not adopted.
US U’s core technical contribution was a residual autoencoder (AE) designed for anomaly detection in additive manufacturing. The AE was trained on process‑monitoring signals from “good” test parts produced under optimal conditions. Training data comprised signals from multiple hatch lines, collected by three microphones and two pyrometers that monitored the build chamber. The network’s encoder compresses the high‑dimensional sensor data into a latent vector that is 16 times shorter than the original signal; the architecture allows further compression to 32‑ or 64‑fold reductions. During training, the AE learns to reconstruct the input signals, forcing it to capture the most salient features in the latent space. These compressed representations can be extracted for downstream analysis, enabling automated anomaly detection and quality assessment of the manufacturing process.
In addition to the AE, US U developed a Git LFS‑based collaboration platform in partnership with Bosch Renningen. This platform facilitated secure, versioned exchange of data sets, algorithms, and deep‑learning models across the consortium, bypassing delays that would have arisen from adopting Daimler’s Databricks solution. The platform was hosted by Bosch and administered by US U, ensuring continuous collaboration without interruption. US U also demonstrated its Industrial Analytics Platform (IAP) for data integration and curation, but the consortium ultimately chose the Git LFS solution for the digital backbone due to faster deployment timelines.
The project’s organizational structure included eight milestone meetings with the full consortium, fourteen use‑case and working‑group sessions, and three technology meetings. US U’s AI Services department coordinated the AI work, built GPU‑enabled training infrastructure, and integrated the deep‑learning models into the product development pipeline. The “Monitoring” working group, in which US U was actively involved, focused on designing process‑monitoring schemes suitable for machine‑learning applications and on developing algorithms for automated data evaluation.
Overall, the ReAddi project delivered a robust deep‑learning framework for real‑time anomaly detection in additive manufacturing, a scalable data‑sharing platform, and a foundation for future rule‑based control strategies. These outcomes were achieved through close collaboration among industrial partners, leveraging US U’s data‑science capabilities and the consortium’s shared commitment to a digital, data‑driven manufacturing ecosystem.