Framework for Integrating the Finite Element Method and a Statistical Learning Approach for Characterizing the Effect of Process Parameters on Mechanical Response

Reliable analysis and insights are needed to guide processing parameters for additive manufacturing (AM) based material designs to reduce the cost associated with a trial-and-error approach. This work demonstrates an integrated framework that combines thermomechanical simulation with a statistical framework as a tool for providing interpretable insights into the interplay between AM process parameters on metallic alloy printing. Specifically, a large number of simulations (192) were performed using processing parameters (laser power, laser scan speed, layer height, and cooling assumptions) to predict the stress and deformation outcomes for a nickel alloy, Inconel-625. Instead of traditional plots for post-processing, two unsupervised machine learning approaches were applied to the outcomes, namely the finite mixture models (FMM) and the finite mixture of regression models (MoRM). The FMM used a mixture of Gaussians and regression to cluster the regions based on von Mises stress and deformation outcomes to identify locations (within and across simulations) that behaved similarly. The MoRM approach used a mixture regression model with Gaussian error, which directly allowed for the effect of each manufacturing process parameter to be considered. The MoRM performed well at capturing the variance present within the simulations. This approach also identified areas of geometry that were sensitive to changes in the manufacturing parameters. For example, the stress clusters at the lower layers were robust to change, while those at the higher layers were sensitive to changes. For the deformation, the unsupported regions were the most sensitive to change, while the rounded areas tended to be robust. The underlying thermomechanistic interpretation supported these outcomes. The findings collectively demonstrated how an independent statistical approach can be used to provide insights across a large number of process parameters and simulation outcomes for identifying optimum process and printing outcomes.