Predicting and Evaluating Decoring Behavior of Inorganically Bound Sand Cores, Using XGBoost and Artificial Neural Networks

Abstract

Complex casting parts rely on sand cores that are both high-strength and can be easily decored after casting. Previous works have shown the need to understand the influences on the decoring behavior of inorganically bound sand cores. This work uses black box and explainable machine learning methods to determine the significant influences on the decoring behavior of inorganically bound sand cores based on experimental data. The methods comprise artificial neural networks (ANN), extreme gradient boosting (XGBoost), and SHapley Additive exPlanations (SHAP). The work formulates five hypotheses, for which the available data were split and preprocessed accordingly. The hypotheses were evaluated by comparing the model scores of the various sub-datasets and the overall model performance. One sand-binder system was chosen as a validation system, which was not included in the training. Robust models were successfully trained to predict the decoring behavior for the given sand-binder systems of the test system but only partially for the validation system. Conclusions on which parameters are the main influences on the model behavior were drawn and compared to phenomenological–heuristical models of previous works.

Keywords

• casting technology
• inorganically bound sand cores
• decoring behavior
• artificial neural networks
• XGBoost
• SHAP