Feature-Optimised Machine Learning Framework for Intelligent CNC Grinding Tool Monitoring

Abstract

Tool Condition Monitoring (TCM) is vital in CNC grinding to ensure dimensional precision, surface quality, and process efficiency. This study introduces a machine learning-based TCM framework that integrates vibration signal analysis, feature extraction, and metaheuristic optimization. Vibration data from industrial CNC grinding were processed to extract time- and frequency-domain features, which were then applied to tool wear prediction using Support Vector Regression (SVR). To improve accuracy and robustness, the Marine Predators Algorithm (MPA) was employed for hyperparameter tuning. The MPA-optimized SVR achieved the lowest mean squared error (0.0028), outperforming autoregressive (0.0088) and power-law (0.0084) models. These findings demonstrate the potential of combining vibration analysis with metaheuristic-optimized machine learning for intelligent, scalable, and real-time TCM. The proposed framework aligns with Industry 4.0 objectives by reducing downtime, improving reliability, and enabling adaptive monitoring in precision manufacturing.