Vibration-Based Tool Wear Prediction via Ensemble Learning and AutoML-Guided VMD Mode Selection

Purpose: This work proposes an advanced Tool Wear Condition Monitoring (TWCM) approach that integrates machine learning with variational mode decomposition to predict tool wear progression during turning. Methods: AISI 1045 steel was machined using a TNMG carbide insert. The generated vibration signals were acquired and analyzed using variational mode decomposition to extract correlations between tool wear behavior and machining dynamics. An AutoML approach was applied to identify VMD modes with strong correlations with flank wear (VB). Power spectral density (PSD) analysis was then performed on the selected modes to capture frequency variations induced by tool wear. Several machine learning models, including an ensemble model, were trained using the extracted features. Results: AutoML qualified two VMD modes as highly correlated with flank wear. The ensemble model achieved an R² value of 0.98, demonstrating the predictive capability. Conclusion: The integrated approach accurately predicts flank wear from vibration signals, confirming its effectiveness for tool wear monitoring. The findings also highlight the benefits of ensemble learning for achieving accurate predictions.