Multi-fidelity Modeling for Uncertainty Quantification in Laser Powder Bed Fusion Additive Manufacturing

Computer simulation of the additive manufacturing (AM) process involves multi-physics, multi-scale models. These sophisticated higher fidelity (HF) AM models, though more accurate, are computationally very expensive. On the other hand, AM process simulation using lower fidelity (LF) analytical models with simplified physics is fast but has significant prediction error. This paper presents a multi-fidelity (MF) modeling approach for constructing a prediction model for an AM process by fusing information from physics-based models of different fidelity and experimental data, thus maximizing the accuracy within the available computational resources. The LF model is corrected in two stages: first using the HF model simulation results and then the experimental data. A Bayesian calibration approach is used to estimate the correction factors and the MF model parameters to account for both process variability as well as model uncertainty. The proposed methodology is demonstrated by constructing a multi-fidelity model to predict the lack-of-fusion porosity in the laser powder bed fusion AM process, by combining an HF multi-physics computational model and an LF Rosenthal equation-based analytical solution. Further, an approach is developed to measure the effectiveness of the method by validating the prediction against experimental data.