Digital technologies for self-learning optimization of laser based additive manufacturing processes
Laser metal deposition, an established cladding technology, contains the capability to print three dimensional metal parts. Layer-wise production characteristic with metal powder enables new parts that consist of one or more elements and which cannot be produced with conventional technologies. However, with increasing complexity (geometrical variety) the planning effort increases in such a way that economical applications are not possible anymore. New part geometries need lots of process understanding, experience and test phases. In fact, complex geometries and delicate structures cannot be realized today due to missing process control. The goal of the project is the integration of artificial intelligence into additive manufacturing. The artificial intelligence will take over process planning, -execution and -monitoring. It will be provided with fundamental "experience" (databases and models) from current research and it will start its own learning process. With every produced part its experience will grow. Process parameters will be optimized by monitoring and analyzing errors and deviations. Finally, complex geometries will be produced with the help of machine learning.
This high degree of automation provides essential improvements: Process development periods of new parts will be drastically shortened due to autonomous test phases. In head-to-head comparison to conventional, subtractive techniques (e.g. milling, drilling) laser metal deposition already stands out because of its resource efficiency. This advantage will be amplified by canceling the necessity of producing numerous sample structures. Eventually, the artificial intelligence enables unmatched complexities and new construction options in geometry and material selection.
Optical sensors - The eyes of artificial intelligence
An important pillar of this self-learning production technology is a comprehensive acquisition of sensor data that can be used to monitor and analyze all relevant process states, primarily 3D geometry and temperature profiles. Some information will be given to the process control system, which stabilizes the welding process in real-time. All sensor data will be fused into the self-learning system and produce a feedback for the following process planning. In this way geometrical deviations during the welding process will be corrected online and considered for further production processes. To ensure that, the artificial intelligence will use given, physical models and its own "experiences", automatically obtained from feedback loops and sensor data acquisition. The application of unknown materials and geometries will be optimized by fully-automated trial and error. Time consuming parameter studies by hand will become unnecessary.