Data-driven Uncertainty Quantification Framework in Metal Additive Manufacturing

Abstract

Abstract This paper presents the uncertainty quantification (UQ) framework with a data-driven approach using experimental data in metal additive manufacturing (AM). This framework consists of four steps. First, the experimental data, including process parameters and signatures, are obtained by performing tests in various conditions. Next, the model is constructed by surrogate modeling and a machine learning algorithm using the obtained data. Then, the uncertainties in a quantity of interest (QoI), such as bead geometry, surface roughness, or mechanical properties, are quantified. Lastly, the UQ is verified and validated using the experimental data. The proposed framework is demonstrated with the data-driven UQ of the bead geometry in gas tungsten arc welding (GTAW)-based wire + arc additive manufacturing (WAAM). In this case study, the uncertainty sources are process parameters and signatures, and the QoI is bead geometry. The process parameters are wire feed rate (WFR), travel speed (TS), and current, while the process signatures are voltage-related features. The bead geometry includes the width and height of single-beads. The results of the case study revealed that (1) verifying and validating the data-driven UQ of bead geometry with the normal beads was conducted, and the predicted values were within the 99% confidence intervals, (2) the bead width was negatively correlated with TS, and (3) the bead height had a positive and negative correlation with WFR and TS, respectively.