Knowledge Transfer Between Machines in Laser Powder Bed Fusion - Transfer Learning with Small Training Data Sets

Laser Powder Bed Fusion PBF-LB/M is currently one of the most versatile and adopted additive manufacturing technologies for printing metals. To take new PBF-LB machines into service, an thorough characterization and calibration is often necessary to get the desired output. This is often done empirically, however data driven methods become more and more available. This research explores the use of Transfer Learning (TL) with already existing, pretrained, neural network-based process models from a different machine to predict tensile properties of printed AlSi10Mg0.5. The TL models were evaluated regarding their optimal post-learning approach and performance, while using minimal amounts of training data to keep the empirical testing expenses low. Eight single transfer model variants were tested, based on their degree of training freedom. Additionally, a weighted mean model ensemble of all eight single models was derived. The validation trials showed that of both single models and the weighted mean model did offer usable predictions to optimize processing conditions and entailing tensile properties.