Developing a Predictive Maintenance Model for Addressing Underfill in Heineken Brewery’s Share Company
| dc.contributor.advisor | Ameha Mulugeta (PhD) | |
| dc.contributor.advisor | Natnael Mekonnen (Mr.) Co-Advisor | |
| dc.contributor.author | Ruth Wondosen | |
| dc.date.accessioned | 2025-04-25T08:26:09Z | |
| dc.date.available | 2025-04-25T08:26:09Z | |
| dc.date.issued | 2024-10 | |
| dc.description.abstract | Predictive maintenance (PdM) is crucial for enhancing operational efficiency and reducing downtime in industrial processes. This study introduces a novel approach to developing a predictive maintenance model for addressing underfilling issues in the bottling processes of Heineken Company, which significantly impact revenue and product quality. Unlike previous works that rely solely on historical data, this research incorporates synthetic data generated using Conditional Generative Adversarial Networks (cGAN) to overcome data limitations and enhance model robustness. The methodology involved comprehensive data preprocessing, including imputation and feature engineering, to prepare the dataset for training a Random Forest classifier. The model development was refined through hyperparameter tuning via Grid Search and validated using cross-validation. The results demonstrated a strong predictive capability, with a training accuracy of 90.67%, test accuracy of 90.21%, and cross-validation accuracy of 90.89%, indicating reliable generalization. The use of cGAN contributed to increased data variability, mitigating overfitting and ensuring realistic model training scenarios. This study advances the field of predictive maintenance by demonstrating how synthetic data can augment limited datasets to improve model accuracy and resilience. Integrating this model into the production line enables proactive maintenance scheduling, reducing disruptions and enhancing product consistency. Future work will focus on expanding the model's scope to incorporate real-time sensor integration for adaptive learning and exploring ensemble models, such as hybrid Random Forest and LSTM architectures, to handle temporal patterns and further optimize predictive performance. | |
| dc.identifier.uri | https://etd.aau.edu.et/handle/123456789/5369 | |
| dc.language.iso | en_US | |
| dc.publisher | Addis Ababa University | |
| dc.subject | Predictive Maintenance (PdM) | |
| dc.subject | Machine Learning | |
| dc.subject | Underfill | |
| dc.subject | Random Forest | |
| dc.subject | conditional Generative Adversarial Network(cGAN) | |
| dc.title | Developing a Predictive Maintenance Model for Addressing Underfill in Heineken Brewery’s Share Company | |
| dc.type | Thesis |