<![CDATA[Hydraulic systems are widely used in industry due to high power density and precision control; however, failures in components such as coolers, valves, pump leakage, and hydraulic accumulators can cause downtime and high maintenance costs. Conventional corrective and preventive maintenance is limited due to reliance on fixed schedules and manual inspection. Therefore, this study applies machine learning-based predictive maintenance to predict hydraulic component conditions. This study aims to predict the condition of four components (Cooler, Valve, Internal Pump Leakage, and Hydraulic Accumulator), evaluate Histogram Gradient Boosting (HGB) using Accuracy and macro F1-score, and implement the model in a Streamlit application. The Condition Monitoring of Hydraulic Systems dataset (2,205 cycles) is used. Sensor data from .txt files is transformed into tabular features per cycle, standardized using StandardScaler, and split into 70:30 training and testing sets. Four independent HGB models are trained and evaluated using confusion matrix, Accuracy, and macro F1-score. Results show strong performance: Cooler_condition (Accuracy 0.998, F1 0.998), Valve_condition (0.834, 0.786), Internal_pump_leakage (0.968, 0.963), and Hydraulic_accumulator/bar (0.989, 0.987). The trained models are saved as .pkl files and integrated into a Streamlit application for interactive prediction and CSV export.]]>
Copyrights © 2026
