As part of the KiRo3D research project, a 3D printing robotic cell has been digitalized. This involved integrating a pyrometer and a CMOS camera for real-time melt pool monitoring via a beam splitter in the optical path. Coupled with the RSI interface of the KUKA robot, the process data is efficiently captured and stored in a database.
Additionally, a comprehensive process monitoring tool has been developed, empowering machine operators to oversee the process in real-time. For subsequent analysis, a web-based analytical tool has been created, enabling visualization and evaluation of process data post-build. This tool not only facilitates location-specific representation of sensor data (geomapping) but also supports dataset generation and training of machine learning models.
These models are utilized to train reinforcement learning agents, which incrementally learn the process characteristics from historical data. Ultimately, these agents can be deployed within the process to automate the control of process parameters. This approach stabilizes the process, enhances product quality, minimizes scrap, and reduces production costs.
