A machine learning method for predicting the elongation to failure of Al–Si alloy in high pressure die casting combining experiment and modeling

Abstract

High-pressure die casting (HPDC) of Al–Si alloys faces challenges in improving mechanical properties and early failure due to process-induced defects. Traditional quality assessment experiments and modeling methods are costly and lack predictive capability. This study proposed a machine learning (ML) method integrating numerical simulations and experimental data to predict elongation. The data were extracted from experiments and well validated mathematical models of the HPDC process and underwent preprocessing such as standardization and feature selection. The performance of twelve common ML models was evaluated, among which the eXtreme Gradient Boosting (XGBoost) and Gradient Boosting (GB) algorithms performed the best. By using the Crested Porcupine Optimizer (CPO) and Bayesian Optimization for hyperparameter optimization, the accuracy of the models was further improved. The R2 value of the CPO-XGBoost model reached the optimal value of 0.882. The SHapley Additive exPlanations (SHAP) analysis revealed that total shrinkage volume dominantly reduced elongation (El), while die temperature and pouring temperature negatively affected El. Validation with independent experiments demonstrated that the difference between the predicted values and the measured values was within 5 %. This work establishes a novel methodology combining physics-based simulations and experiments to comprehensively predict and analyze the failures of HPDC Al–Si alloys from multiple perspectives. It can be further extended to the prediction of various types of alloys or multiple mechanical properties.