Predictive Quality Defects Combining Design of Experiments and Machine Learning: A Case Study on Shrinkage in Injection Molding

Abstract

Injection molding is extensively used across industries due to its cost-efficiency and performance benefits. However, variations within the molding process can significantly affect the quality of manufactured parts, presenting challenges in accurately predicting defects. This research introduces an innovative methodology for modeling volumetric shrinkage defects. The study initiates with a factorial design of experiments to simulate a range of process conditions. Following this, three distinct machine learning algorithms—Random Forest, K-Nearest Neighbor, and Support Vector Machine—are applied to forecast volumetric shrinkage. The outcomes reveal that the Random Forest model exhibits superior predictive capabilities, achieving an $R^{2}$ value of 0.75, a root mean square error (RMSE) of 0.0345, and a mean absolute error (MAE) of 0.099, surpassing the performance of the other models tested. These findings highlight the efficacy of the proposed approach in enhancing the prediction accuracy of quality defects in injection molding processes.