Hybrid machine learning and finite element modeling for accurate prediction of sintering-induced deformation in material extrusion additive manufacturing

Abstract

The current study developed a physics-based, data-driven finite element analysis (FEA) model using commercially available software to predict the shrinkage and deformation of copper specimens fabricated by material extrusion 3D printing (ME3DP) during the pressureless sintering process. Identifying shrinkage and deformation prior to designing the 3D CAD model helps designers optimize and adjust the component geometry, considering the sintering effects. In this regard, two datasets were captured: one from the developed phenomenological model and the other from the experimental outcomes of the sintering process. The artificial neural network (ANN) being a machine learning (ML) technique, was configured using the complete dataset by optimizing and identifying the most suitable network parameters to determine the relative density during the sintering of ME3DP copper specimens. The configured ANN was rebuilt and used as a constitutive equation to predict the shrinkage and deformation of copper specimens in COMSOL Multiphysics by modifying the existing constitutive laws. The results obtained from the experiments, FEA, and ML-FEA models were compared for two different shapes: cubic and I-section geometries. Additionally, the stresses evolved in the cube and I-section copper specimens captured by the FEA and ML-FEA models are presented. Furthermore, Shapley additive explanations (SHAP), an interpretability tool, was incorporated to quantitatively analyze the influential order of the input features and their contribution to the prediction of relative density.