Die-casting defect prediction and diagnosis system using process condition data

This study aims to construct a system for predicting and diagnosing defects in casting products and their causes to improve the productivity of the casting process in the die casting industry. Three data analysis algorithms are proposed to predict defects and diagnose the causes of the defects. First, diagnosing the pre-heating state, which appears during the casting process, is important, as casting products produced during the pre-heating state are treated as defects and managed separately. Thus, Random Forest is applied to classify the normal and pre-heating states of the casting process. Second, before training the defect prediction algorithm, the Synthetic Minority Over-sampling Technique (SMOTE) method is executed to address a data imbalance issue between normal and defect products. This paper applies the Random Forest algorithm to predict whether the product is defective and which defect type is expected. Third, the rules of defect causes are configured based on the conditions extracted from the decision trees, which are generated by Random Forest. All three algorithms presented in this study have over 89% accuracy. Finally, this study can support various engineering process tasks and upgrade the quality of the advanced die-casting process beyond the primary level of ICT (Information and Communications Technologies) in the existing die-casting industry.