On the Possibility of Replacing Scheil-Gulliver Modeling with Machine Learning and Neural Network Models

Abstract

Resource-efficient manufacturing is a foundation for sustainable and circular manufacturing. Semi-solid processing typically reduces material loss and improves productivity but generally requires a better understanding and control of the solidification of the cast material. Thermal analysis is commonly used in high-pressure die casting (HPDC) processes to determine casting process parameters, such as liquidus and solidus temperatures. However, this method is inadequate for semi-solid casting processes because the eutectic temperature is also a crucial parameter for successful semi-solid casting. This study explores the feasibility of using machine learning and artificial neural networks to predict fundamental values in Al-Si alloy casting. The Thermo-Calc 2022 software Scheil-Gulliver calculation function was used to generate the training and the test datasets, which included features such as melting temperature, alpha aluminium solidification temperature, eutectic temperature, and the solid fraction amounts at eutectic temperature. The results show that both models have a symmetric mean absolute percentage error (SMAPE) of less than 2 % with temperature prediction, with the machine learning model achieving a better accuracy of less than 1 %. A case study comparing practical measurements with prediction results is also discussed, demonstrating the potential of AI methods for predicting semi-solid casting processes.