Secure Machine Learning Framework for Defect Detection and Quality Enhancement in Injection Molding Processes

Injection molding processes generate large volumes of heterogeneous process data that reflect complex and dynamic manufacturing conditions, directly influencing product quality. Conventional quality control approaches based on fixed statistical thresholds and operator experience often fail to ensure stable and secure operation under such variability, leading to increased defect rates and reduced process reliability. Moreover, insufficient consideration of data integrity and process stability can compromise the robustness of data-driven manufacturing systems. This study proposes a secure machine learning framework for defect detection and quality enhancement in injection molding processes. The framework integrates systematic data preprocessing, correlation analysis, and an unsupervised learning approach based on a Gaussian Mixture Model (GMM) to achieve reliable defect classification. Key process parameters contributing to quality deviations are identified through statistical correlation analysis, and process data are clustered into one normal group and three abnormal groups corresponding to different defect types without requiring labeled datasets. By emphasizing data integrity, stable clustering behavior, and resilience to process variability, the proposed framework enhances manufacturing security and reduces the risk of misclassification. Experimental results using real injection molding process data demonstrate effective defect reduction, improved process parameter optimization, and enhanced operational efficiency. By enhancing defect detectability and reducing misclassification risks, the proposed framework supports safer and more human-centric manufacturing environments by improving operator trust, decision confidence, and preventive intervention capability.