Research Information System
JOURNAL OF INTELLIGENT MANUFACTURING, 2025 (SCI-Expanded, Scopus)
This paper presents a hybrid process optimization approach that combines machine learning (ML) and metaheuristic algorithms (MA) to reduce quality defects in production systems. Defect reduction in such systems often requires the optimization of complex process parameters. While ML models are widely used for defect prediction, they may be insufficient for identifying optimal input settings due to their limited interpretability. To address these challenges, a novel ML-MA integration framework has been developed, enabling learning from routine process measurements without incurring additional experimental costs. Within this framework, ML models are used to estimate defect rates, while MA techniques determine the optimal process parameters that minimize these rates. Unlike previous studies focused on feature-level insights, the proposed framework directly generates the optimal set of process parameters, offering a more actionable and data-driven solution for defect minimization. As a case study, the framework was applied to reduce sintering defects in the sand mold casting process within the cast iron industry. Three ML algorithms, Gradient Boosting Regression (GB), Random Forest Regression (RF), and Extra Trees Regression (ET), were evaluated for defect rate prediction. Subsequently, three MA techniques, Genetic Algorithm (GA), Particle Swarm Optimization (PSO), and Simulated Annealing (SA), were employed to minimize the defect rates predicted by the best-performing ML model. Experimental validation in this study was limited to ten production trials due to operational constraints; however, the framework achieved a 78.9% reduction in the sinter defect rate, clearly demonstrating its practical viability for identifying optimal process settings in real-world conditions.