Intelligent sensor-based fault diagnosis and tool wear prediction for CNC milling using hybrid machine learning

Abstract Computer Numerical Control (CNC) machining plays a crucial role in modern manufacturing, where the condition of cutting tools directly affects product quality, production efficiency, and maintenance costs. Accurate monitoring of tool wear and fault conditions is therefore essential for enabling predictive maintenance and preventing unexpected machine downtime. However, conventional Tool Condition Monitoring (TCM) approaches often exhibit limitations in feature extraction, classification accuracy, and decision-making capabilities. To address these challenges, this study proposes an intelligent sensor-based fault diagnosis framework that integrates advanced feature selection and hybrid machine learning techniques for CNC cutting TCM. The proposed approach utilizes the FeatureAndMetadata_Milling.csv dataset derived from CNC milling operations, which contains statistical features extracted from vibration and current sensor signals collected throughout the tool life cycle. Data preprocessing is performed using mean imputation for missing values, Min–Max normalization for feature scaling, and Z-score-based outlier removal to improve data quality. Subsequently, Boruta-Random Forest (BRF) feature selection is employed to identify the most informative sensor features and reduce data dimensionality. A hybrid classification model combining Light Gradient Boosting Machine (LightGBM) and Extreme Gradient Boosting (XGBoost) is then developed, with hyperparameters optimized through the Whale Optimization Algorithm (WOA) to enhance predictive performance. Experimental evaluation demonstrates that the proposed framework achieves an accuracy of 0.991, precision of 0.989, recall of 0.986, F1-score of 0.987, and ROC–AUC of 0.992. These results indicate excellent fault detection capability and robust multi-class classification performance across different tool wear conditions.