Advanced Ensemble Machine Learning Approach for Wear Rate Prediction in Coated Materials: A Comparative Analysis of Tree-Based Regression Models
Sahil Sonawane, Aditya Sangle, Lalit Sope, Vijay Rajput
Abstract—This research explores a comprehensive machine learning approach to predict wear rates in coated materials under varying conditions. Five tree-based ensemble regression models Gradient Boosting, XGBoost, Stacking Regressor, Extra Trees, and Random Forest were evaluated for their predictive accuracy. The Gradient Boosting Regressor demonstrated superior perfor- mance with 98.20% test accuracy, followed by XGBoost (98.05%) and Stacking Regressor (97.88%). The models effectively cap- tured complex relationships between material properties, coating characteristics, speed, ash type, concentration, and time. A Voting Regressor ensemble was developed to enhance stability and leverage the strengths of individual models. Residual analysis revealed minimal bias across predictions, with performance metrics including mean squared error below 0.03 and R² values exceeding 0.99 for top models. This study shifts the approach to wear prediction from a reactive maintenance strategy to a proactive optimization method, which holds considerable promise for lowering equipment replacement expenses and minimizing downtime in industrial settings. The economic evaluation suggests potential reductions of 28-34% in maintenance expenditures through the adoption of predictive wear models for mineral processing machinery. The proposed framework serves as a basis for real-time monitoring systems with established confidence intervals to aid maintenance decisions. Subsequent efforts will aim to enhance the model by accounting for combined wear mechanisms and integrating online learning features to adapt to changing operational conditions.
Index Terms—Keywords — Wear Rate Prediction, Machine Learning, Performance Metrics, Random Forest, Gradient Boost- ing, XGBoost, Stacking Regressor, Extra Trees, Mean Squared Error (MSE), Root Mean Squared Error (RMSE), R² Score, Predictive Modeling, Error Rates, Material Coatings.
