A Machine Learning-Based Surrogate Model for Similarity Criterion of Solidification

The present study develops machine learning-based surrogate models for similarity criterion for solidification. The solidification rate R and Niyama criterion value from the simulation results are compiled, pre-processed, and then used to train the models. A regularization approach is used to minimize variance and avoid overfitting, with the base learners as three-layer artificial neural network (ANN). The predictions from the surrogate model are compared to the training data across both the solidification rate R and Niyama criterion value, considering the different factors affecting the solidification of castings. The trained model has a mean percentage error in the solidification rate R and Niyama criterion value of ~9.89% and ~1.90%, respectively, for the entire dataset. The results show that the predicted and training values are consistent with the parameter changes during the solidification of the castings. Factors affecting the solidification process of castings were evaluated. It is found that the casting temperature has the greatest influence on the solidification of castings, and the validity of the surrogate model is verified by pouring.