Abstract
Multi-scale, multi-physics, computational models are a promising tool to provide detailed insights to understand the process-structure-property-performance relationships in additive manufacturing (AM) processes. To take advantage of the strengths of both physics-based and data-driven models, we propose a novel, hybrid modeling framework for laser powder bed fusion (L-PBF) processes. Our unbiased, model integration method combines physics-based data and measurement data for approaching more accurate prediction of melt-pool width. Both a high-fidelity computational fluid dynamics (CFD) model and experiments utilizing optical images are used to generate a combined dataset of melt-pool widths. From this aggregated dataset, a hybrid model is developed using data-driven modeling techniques, including polynomial regression and Kriging methods. The performance of the hybrid model is evaluated by computing the average relative error and comparing it with the results of the simulations and surrogate models constructed from the original CFD model and experimental measurements. It is found that the proposed hybrid model performs better in terms of prediction accuracy and computational time. Future work includes a conceptual introduction on the use of an AM ontology to support improved model and data selection when constructing hybrid models. This study can be viewed as a significant step towards the use of hybrid models as predictive models with improved accuracy and without the sacrifice of speed.
In-situ characterization and quantification of melt pool variation under constant input energy density in laser powder bed fusion additive manufacturing process
