A Comparative Study of Classical Machine LearningAlgorithms for Multi-Modal CMFD of MachiningTools

Accurate prediction of cutting tool wear is a fundamental requirement for effective condition monitoring and fault diagnosis (CMFD) in modern manufacturing environments. While recent research has increasingly used machine learning (ML) approaches for this purpose, existing studies are typically limited to a small number of sensor modalities and focus predominantly on model complexity rather than the scalability of classical algorithms under multi-modal conditions. This paper presents a comparative study of four classical ML algorithms including Random Forest Regression (RFR), Gradient Boosting Regression (GBR), Support Vector Regression (SVR), and K-Nearest Neighbours (KNN), for tool wear prediction using a high-dimensional multi-modal dataset obtained from run-to-failure milling experiments. Eleven sensor modalities including spindle vibration, machine base vibration, cutting force, acoustic emission, and power were simultaneously recorded across 108 machining rounds performed by three independent carbide end mill tools on Ti-6Al-4V workpieces. Fifteen time-domain statistical features were extracted per modality, providing a 165-dimensional feature vector per cutting round. A training methodology incorporating cross-tool validation and 5-fold cross-validation was implemented to ensure rigorous and fair evaluation of generalisation performance across unseen tools. Quantitative evaluation using Mean Absolute Error, Mean Squared Error, and the coefficient of determination (R²) revealed that kernel-based and instance-based approaches outperformed ensemble methods, with SVR and KNN achieving better average predictive accuracy. These findings demonstrate that classical ML techniques maintain strong relevance for multi-modal CMFD applications and that expanding the number of sensor modalities beyond conventional setups improves tool wear prediction performance.