Enhancing Hydraulic System Reliability via XGBoost-Based Predictive Maintenance

Predictive maintenance (PdM) can substantially reduce unplanned downtime and maintenance costs in industrial systems. In this study, we develop a proof-of-concept XGBoost-based machine learning pipeline for detecting component faults in a hydraulic test rig using the publicly available UCI Condition Monitoring of Hydraulic Systems dataset. [1] After filtering unstable cycles, 756 stable operating cycles across 17 sensors were retained. Domain-informed feature engineering and principal component analysis (PCA) retaining 95% of variance (12 components) were applied before training a MultiOutput XGBoost classifier within a stratified cross-validation framework. On a held-out test set comprising 189 cycles, the model achieved a macro F1-score of 0.95, demonstrating balanced performance across the cooler, accumulator, and pump leakage components. A lightweight Streamlit prototype illustrates near-real-time operator input and alerting. These results show that feature-engineered ensemble methods can effectively support PdM in industrial settings. Future work will explore temporal models such as LSTMs, expand pump leakage samples, and integrate with computerised maintenance management systems (CMMS) for field validation.