Prediction and Optimization Model for Casting Process Defects with Proposed Process Variable Ranges

Purpose: The purpose of this study was to improve casting process quality by constructing an integrated analysis procedure that combines anomaly detection, process variable selection, and process optimization in the die casting process.Methods: Using historical process data, anomalies were detected through the Isolation Forest algorithm, and Tree SHAP was applied to identify key variables that contributed to the anomaly scores. Subsequently, a Genetic Algorithm was employed to search for optimal combinations of process variables that reduce the defective proportion.Results: The results of this study showed that the integrated use of machine learning and optimization techniques effectively identified process variable associated with abnormal product and suggested optimal value and range of process parameters. The findings indicated that critical variables encompassed not only domain- established features but also statistically relevant process factors that had previously been neglected.Conclusion: The proposed framework enhances quality management without exclusive reliance on expertises. It is applicable to diverse metal sectors and facilitates incremental adoption of data-driven autonomy for small and medium-sized enterprises (SMEs) with restricted AI capacity.