SURFACE ROUGHNESS SURROGATE MODELING IN METAL 3D PRINTING USING KRIGING AND BATCH EXPERIMENTAL DESIGN

Laser Powder Bed Fusion Additive Manufacturing has revolutionized the production of geometrically complex components, enabling the creation of intricate designs that were previously difficult or impossible to achieve using traditional manufacturing methods. However, one of the key challenges associated with laser powder bed fusion is the substantial surface roughness that often results in parts produced through this process. The increased surface roughness negatively impacts the mechanical performance of the components and increases the costs associated with post-processing, such as additional polishing or finishing treatments. This study seeks to address this challenge by developing a surrogate model for predicting and controlling vertical surface roughness based on process parameters. To achieve this, we propose the Kriging with Iterative Spatial Prediction (KRISP-Uncertainty) algorithm, which combines regression Kriging with an iterative leave-one-out cross-validation method that utilizes Kullback-Leibler Divergence (KLD). This approach refines the surrogate model using select experimental data points, reducing uncertainty with minimal additional experimental data. Our findings demonstrate that the KRISP-Uncertainty algorithm can effectively optimize surface roughness predictions, providing an efficient method for surrogate modeling and controlling surface quality in laser powder bed fusion. By rapidly tailoring stochastic surrogate models to specific material and hardware configurations, this method enhances the overall efficiency and effectiveness of the laser powder bed fusion process, reducing both surface roughness and the associated post-processing requirements.