Comparative Study of Ensemble Learning for Predictive Maintenance in HDD Manufacturing

Thailand is a major hub for hard disk drive (HDD) manufacturing, where maintaining high equipment reliability is essential for product quality and export readiness. Critical testing machines, such as the Guzik Signal Analyzer (GSA 6000 Series) and the Read-Write Analyzer (4000 TDMR Series), can experience performance issues over time due to environmental changes and prolonged use. This paper presents a predictive maintenance (PdM) framework using ensemble machine learning models to predict equipment failures based on sensor data, particularly temperature readings. There are 6 models such as XGBoost, LightGBM, AdaBoost, CatBoost, Gradient Boosting Tree, and Bagging were trained on a real dataset. The Synthetic Minority Oversampling Technique (SMOTE) was used to handle class imbalance. Among all models, XGBoost achieved the best performance with 98.91% accuracy, 98.91% precision, 98.91% recall, F1-score 98.91% and an AUC of 0.9995. The proposed PdM system enables early fault detection, reduces unplanned downtime, and improves overall production efficiency.