Mechanical response of additively manufactured foam: A machine learning approach

Abstract

This paper uses ensemble and automated machine learning algorithms to predict the mechanical properties (tensile and flexural strength) of a three-dimensionally printed (3DP) foamed structure. The closed cell foams were made from the most commonly used thermoplastic, High-Density Polyethylene (HDPE). The hollow glass microspheres are infused in HDPE at varying volume %. The available data on these foams' mechanical properties are used by the chosen machine learning (ML) algorithms to propose the best suited algorithm for such a three-phased microstructure as these closed cell foams exhibit. Finally, the strength predictions from the models were validated using experimental data. The models were trained with nozzle temperature, bed temperature, and force values as input parameters. The output parameters predicted were the tensile and flexural strength. LightGBM outperforms all other models in terms of performance among ensemble-based models, while H2OAutoML outperforms all other models. All the ML algorithms produced models with greater than 95% accuracy. Finally, memory and time consumption for each model are presented.