SVR-MTS prediction approach for shrinkage defects

This study presents a quantitative prediction model with an optimal dimensional reduction technique for the first time in the context of shrinkage defects in a compact graphite iron (CGI) engine cylinder head. The model is built on the support vector regression (SVR) machine learning algorithm and the Mahalanobis-Taguchi system (MTS) is incorporated for dimensional reduction purposes. An in-depth analysis of 41 process variables has been conducted to identify the crucial factors influencing the shrinkage defect. The chosen variables were then used to feed the intelligent prediction model. Optimal tuning of SVR hyperparameters was achieved by evaluating adjustments over the root mean square error (RMSE) for training and testing datasets while simultaneously minimizing RMSE when predicting the defect size. Results were experimentally validated through scanning electron microscopy with energy-dispersive X-ray detection (SEM-EDX).