A novel integrated TDLAVOA-XGBoost model for tool wear prediction in lathe and milling operations

Tool wear in machining operations compromises tool lifespan and performance. Machine learning models, particularly eXtreme Gradient Boosting (XGBoost), demonstrate pattern recognition capabilities for such predictions. However, their effectiveness is highly dependent on hyperparameters, and empirical identification of optimal configurations remains challenging. This study proposes an integrated model for tool wear prediction in CNC machining that combines improved algorithms with XGBoost. Specifically, we developed an enhanced African Vulture Optimization Algorithm (TDLAVOA), incorporating Tent chaotic mapping to improve population diversity during initialization. The algorithm optimized individual fitness by guiding boundary searches toward optimal ranges, thereby increasing elite population generation. A lens imaging reverse learning strategy further refined global search capabilities through iterative optimization of top individuals. TDLAVOA was validated against classical and modern algorithms on benchmark functions. Subsequently, the XGBoost-TDLAVOA model was developed by employing TDLAVOA to auto-tune seven critical XGBoost hyperparameters using MSE loss, which enhances adaptability and robustness. The model's performance was validated with two datasets: flank wear data from CNC lathe tools (sourced from the Kaggle database) and milling simulations (generated using DEFORM-3D). Comparative experiments with seven established models confirmed its superior predictive accuracy. Results demonstrated the model's effectiveness in learning nonlinear relationships between machining variables and tool wear. Specifically, on the test sets, XGBoost-TDLAVOA achieved an RMSE of 0.0851 (R²=0.9843) for CNC turning and an RMSE of 2.85 × 10⁻⁵ (R²=0.9779) for milling operations. This work advances predictive maintenance strategies and presents a novel integrated framework for tool wear modeling.