This conference paper was submitted for presentation at the NAFEMS World Congress 2025, held in Salzburg, Austria from May 19–22, 2025.

Abstract

The paper presents an initial investigation into using fast-running surrogate models to approximate the numerical simulation of a Material Extrusion (MEX) additive manufacturing process. This study'™s goal was to explore several off-the-shelf surrogate model solutions, within either a commercial code or open source, that could mimic a long-running MEX process simulation of thermoplastic parts, with a reasonable accuracy at a fraction of the required time. The Additive Manufacturing (AM) process was simulated as a sequentially coupled thermo-mechanical analysis within a Finite Element (FE) environment. Several parts were considered in this study, ranging from simple benchmark shapes (purposely designed to showcase specific deformation behavior after cooling) to a more complex geometry. The underlying assumption here is that the part displacement field during the whole AM process is contiguous, meaning no cracks or other discontinuities would occur throughout. Temperature-dependent material properties were considered where appropriate for the neat Polyethylene Terephthalate Glycol (PETG) material used in this study. Several process and material parameters (e.g. convection coefficient, deposition temperature, thermal conductivity, coefficient of thermal expansion, elastic modulus) were considered as input variables and a Latin Hypercube Sampling (LHS) of the design space was selected for generating training data for the surrogate models. The output variables of interest were part deformations and stresses at various locations, at the end of the cooldown phase. Several popular techniques (Response Surface Models, Radial Basis Functions, Kriging and Artificial Neural Networks) were used for building the surrogate models, and statistical tools were used to compare their results with the corresponding FEA models. Although experimental validation is not a focus of the current study, some of the shapes analyzed here were physically manufactured to provide basic confidence in the resultant deformations, and this aspect is presented in the paper as well. The paper ends with conclusions, recommendations and ideas for expanding such surrogate models further, to mimic the MEX process simulation of large-scale fiber-filled thermoplastic parts.