Automated CNC Tool Wear Detection System Using Multi-Domain Signal Processing and Machine Learning | IJORET – Volume 11- Issue 2 | IJORET-V11I3P1

Remaining Useful Life (RUL) prediction and machine accuracy monitoring are critical components of predictive maintenance in CNC machining systems. However, many existing approaches rely heavily on labeled datasets, computationally expensive models, and complex feature engineering, which restrict their deployment in real industrial environments, particularly on resource-constrained IoT and edge devices. This paper presents an efficient multi-domain signal processing and machine learning framework for CNC machine accuracy prediction. The proposed system follows a five-stage methodology. First, multi-sensor data are acquired from vibration, current, temperature, and cutting force sensors. Second, data preprocessing is performed through cleaning, noise reduction using Mean, Median, Gaussian, Wiener, Wavelet, and Kalman filters, and normalization using Min-Max, Z-Score, and L2 methods. Third, multi-domain signal analysis is conducted in the time domain, frequency domain using Fast Fourier Transform (FFT), and time-frequency domain using Maximal Overlap Discrete Wavelet Transform (MODWT). Fourth, a total of 49 discriminative features are extracted, including statistical features such as mean, variance, skewness, kurtosis, and wavelet-based descriptors. Finally, multiple machine learning models, including Logistic Regression, Support Vector Machine (SVM), Random Forest, XGBoost, and Artificial Neural Network (ANN), are employed for classification. A significant contribution of this work is the development of a node-based visual interface using React and ReactFlow, enabling users to construct analytical pipelines through drag-and-drop operations without programming expertise. The platform supports both manual and automatic modes to enhance usability and flexibility. Experimental evaluation using the PHM 2010 dataset demonstrates strong predictive performance, where the Random Forest model achieved 94.8% classification accuracy. Results further indicate that sensor fusion and multi-domain feature extraction significantly improve prediction performance, with wavelet-based features providing the highest contribution. The proposed framework offers a scalable, cost-effective, and interpretable solution for Industry 4.0 predictive maintenance applications. Future work will focus on real-time deployment, edge-device integration, and cloud-based monitoring systems.

This study presented an intelligent and efficient framework for CNC machine tool wear prediction using multi-domain signal processing and machine learning techniques. The system combines preprocessing, feature extraction, sensor fusion, and classification into a unified workflow supported by a node-based visual interface. Experimental evaluation using the PHM 2010 dataset demonstrated strong predictive performance, with the Random Forest model achieving the highest accuracy of 94.8%. The obtained results confirm that multi-sensor data and hybrid feature extraction significantly improve tool condition monitoring performance. The proposed system provides a practical, cost-effective, and interpretable solution for predictive maintenance in modern manufacturing environments. With further development toward real-time deployment and cloud integration, the framework has strong potential for future smart factory applications.

1.Widodo, A., and Yang, B. S., “Support Vector Machine in Machine Condition Monitoring and Fault Diagnosis,” Mechanical Systems and Signal Processing, vol. 21, no. 6, pp. 2560–2574. 2.Teti, R., Jemielniak, K., O’Donnell, G., and Dornfeld, D., “Advanced Monitoring of Machining Operations,” CIRP Annals, vol. 59, no. 2, pp. 717–739. 3.Jemielniak, K., “Commercial Tool Condition Monitoring Systems,” International Journal of Advanced Manufacturing Technology, vol. 15, pp. 711–721. 4.Breiman, L., “Random Forests,” Machine Learning, vol. 45, pp. 5–32. 5.Chen, T., and Guestrin, C., “XGBoost: A Scalable Tree Boosting System,” Proceedings of the ACM SIGKDD Conference, pp. 785–794. 6.PHM Society, “PHM 2010 Data Challenge Dataset for High-Speed Milling Tool Wear Monitoring.” 7.Bishop, C. M., Pattern Recognition and Machine Learning, Springer, 2006. Goodfellow, I., Bengio, Y., and Courville, A., Deep Learning, MIT Press, 2016. [Citation Format]

Submit Your Paper Call for Papers About IJORET Journal Scope Review Process