Tool failure prediction model based on vibration signal processing for lathe machines

Tool failure is a critical issue in lathe machining operations, often leading to unplanned downtime, compromised surface finish, and increased production costs. This study aims to develop an efficient and accurate tool failure prediction model based on vibration signal processing techniques, addressing the growing need for real-time predictive maintenance in modern manufacturing. The primary objective is to analyze the correlation between vibration signal features and different stages of tool wear, and to utilize this information for machine learning-based classification of tool health conditions. Vibration signals were captured from a CNC lathe using a high-frequency accelerometer under varying machining conditions. These signals were preprocessed to remove noise and segmented for analysis. Feature extraction focused on both time-domain (RMS, Kurtosis) and frequency-domain (Spectral Energy) parameters. Statistical validation through one-way ANOVA confirmed significant differences across tool states-Healthy, Worn, and Failed-for all three features, with p-values well below 0.0001. Visualizations using boxplots further demonstrated the non-overlapping distribution of these features among different tool conditions, reinforcing their discriminative power. Supervised learning models including Random Forest, SVM, ANN, and KNN were trained on the extracted features. Random Forest achieved the highest classification accuracy of 95.8%, followed by SVM and ANN. These results validate the practical applicability of vibration-based signal features in detecting tool degradation. The study also compared the outcomes with existing literature, finding strong alignment with previous works on tool wear monitoring. The conclusion emphasizes the model’s relevance for predictive maintenance, recommending its integration into industrial CNC systems to reduce downtime and optimize tool utilization. The proposed model not only contributes to smart manufacturing practices but also offers a scalable approach to condition monitoring for precision machining environments.