Quality improvement in sand casting processes: A hybrid approach incorporating machine learning and metasezgisel optimization

Tez Türü: Doktora

Tezin Yürütüldüğü Kurum: Erciyes Üniversitesi, Fen Bilimleri Enstitüsü, Türkiye

Tez Danışmanı: Doç. Dr. Feyza Gürbüz

Tezin Onay Tarihi: 2025

Tezin Dili: Türkçe

Özet:

This thesis presents a hybrid process optimization approach that integrates machine learning (ML) with metaheuristic algorithms (MA) to reduce quality defects in manufacturing systems. Minimizing defective products often requires the effective optimization of complex process parameters. Although ML models, increasingly adopted in quality management, are effective in defect prediction, their limited interpretability restricts the identification of optimal input settings. To address this limitation, a novel ML-MA integration framework was developed, learning from routine production data without additional experimental costs. In this framework, ML models predict defect rates, while MA techniques identify the optimal process conditions that minimize these predictions. Unlike previous studies focusing primarily on feature-level insights, the proposed framework directly generates actionable parameter recommendations, contributing to tangible process improvements. Its dynamic and bidirectional architecture extends beyond conventional ML-MA integrations, offering an innovative approach to the field. The methodology was applied to the sand casting process in the foundry industry, targeting the reduction of sintering defects. Five ML algorithms were evaluated for defect prediction, and three MA methods were employed to optimize the most accurate model's output. Experimental validation in this study was limited to ten production trials due to operational constraints; however, the framework reduced sinter defects by 78.9%, confirming its practical viability.

Keywords: Quality improvement, Machine learning, Optimization, Sand casting, Metaheuristic algorithms