Predicting Sand Casting Defects using a Data-Driven Supervised Machine Learning Approach: A Case Study of Akaki Basic Metals Industry
| dc.contributor.advisor | Mesfin Gizaw (PhD) | |
| dc.contributor.advisor | Henok Zewdu (Mr.) Co-Advisor | |
| dc.contributor.author | Demewez Demeke | |
| dc.date.accessioned | 2024-07-31T08:32:22Z | |
| dc.date.available | 2024-07-31T08:32:22Z | |
| dc.date.issued | 2024-06 | |
| dc.description.abstract | This research investigates supervised machine learning to predict sand casting defects and its severity, aiming to enhance product quality and reduce costs in metal casting. Effective quality control is essential for maintaining structural integrity, energy efficiency, and environmental sustainability. Defects such as porosities, inclusions, shrinkages, cracks, and blowholes increase energy consumption and environmental impact. Rework, scrap, and product rejection due to defects gain significant production costs and reduce profitability. The study identifies and mitigates defects in bronze, steel, and cast iron products weighing from 15 kg to 16,800 kg. Using a dataset of 1001 samples with 37 features, it evaluates machine learning algorithms: Decision Tree, K-Nearest Neighbors, Gradient Boosting, Random Forest, XGBoost, SVC, Ensemble methods, and NN. XGBoost is most effective, with 87% accuracy in defect type prediction and 94% in severity classification. Specifically, the ensemble XGBoost model achieves 93.07% accuracy in defect severity and 86.67% in defect types. The Neural Network also performs well but shows signs of overfitting due to the small dataset. Severity is classified into severe, minor, and moderate; defect types include non-defect, porosity, shrinkage, and others (misrun, blowhole inclusion, crack, and metal penetration). User-friendly tools based on these models are accessible via URLs (https://scdp-dt.streamlit.app/ and https://scdp-severity.streamlit.app/), aiding defect assessment and decision-making in sand casting. In conclusion, machine learning enhances operational efficiency and product quality while promoting sustainability. It also reduces energy use, minimizes rework costs, and enhances quality control, aligning with global environmental goals. | |
| dc.identifier.uri | https://etd.aau.edu.et/handle/123456789/3363 | |
| dc.language.iso | en_US | |
| dc.publisher | Addis Ababa University | |
| dc.subject | Machine learning | |
| dc.subject | Quality control | |
| dc.subject | Sand casting defects | |
| dc.subject | Supervised algorithms | |
| dc.subject | Sustainability | |
| dc.subject | XGBoost | |
| dc.title | Predicting Sand Casting Defects using a Data-Driven Supervised Machine Learning Approach: A Case Study of Akaki Basic Metals Industry | |
| dc.type | Thesis |