Scholarlink

Background <p>Cutting tool wear directly affects product quality, machining cost, and productivity in CNC milling processes. Predictive maintenance strategies therefore rely heavily on accurate prediction of the Remaining Useful Life (RUL) of cutting tools.</p> Purpose <p>This study aims to develop a hybrid machine learning framework for accurate RUL prediction of milling tools using vibration-based signal features.</p> Methods <p>Time-domain vibration features, including root mean square (RMS), skewness, kurtosis, and crest factor, were extracted. Savitzky–Golay smoothing and Robust Scaler normalization were applied as preprocessing techniques. Five machine learning models were benchmarked across four wear modes (flank, nose, notch, and crater): Random Forest, XGBoost, LightGBM, Support Vector Regression, and Multilayer Perceptron.</p> Results <p>Preprocessing improved prediction accuracy by approximately 39% and reduced RMSE from 3.42 to 2.08 cycles. Ensemble tree-based models outperformed kernel-based and neural network models. Random Forest achieved the best overall performance, providing a strong balance between prediction quality (R² = 0.997) and robustness across all wear modes. Model reliability was further validated using predicted-versus-actual plots, 3D scatter plots, and heatmap visualizations.</p> Conclusion <p>The proposed hybrid framework effectively predicts tool RUL and demonstrates the superiority of ensemble learning methods for tool condition monitoring. Random Forest serves as a robust and explainable baseline model suitable for industrial applications. The framework is computationally efficient and well-suited for real-time smart manufacturing and Industry 4.0 environments.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

RUL Prediction of Milling Tools Using Hybrid Machine Learning and Signal Processing: A Novel Framework for Tool Condition Monitoring

  • 期刊论文

  • 发表日期：

    2026年2月18日

查看全文
  • Sunil Mahadeo Pondkule,
  • Sachin Madhavrao Bhosle

摘要

Background

Cutting tool wear directly affects product quality, machining cost, and productivity in CNC milling processes. Predictive maintenance strategies therefore rely heavily on accurate prediction of the Remaining Useful Life (RUL) of cutting tools.

Purpose

This study aims to develop a hybrid machine learning framework for accurate RUL prediction of milling tools using vibration-based signal features.

Methods

Time-domain vibration features, including root mean square (RMS), skewness, kurtosis, and crest factor, were extracted. Savitzky–Golay smoothing and Robust Scaler normalization were applied as preprocessing techniques. Five machine learning models were benchmarked across four wear modes (flank, nose, notch, and crater): Random Forest, XGBoost, LightGBM, Support Vector Regression, and Multilayer Perceptron.

Results

Preprocessing improved prediction accuracy by approximately 39% and reduced RMSE from 3.42 to 2.08 cycles. Ensemble tree-based models outperformed kernel-based and neural network models. Random Forest achieved the best overall performance, providing a strong balance between prediction quality (R² = 0.997) and robustness across all wear modes. Model reliability was further validated using predicted-versus-actual plots, 3D scatter plots, and heatmap visualizations.

Conclusion

The proposed hybrid framework effectively predicts tool RUL and demonstrates the superiority of ensemble learning methods for tool condition monitoring. Random Forest serves as a robust and explainable baseline model suitable for industrial applications. The framework is computationally efficient and well-suited for real-time smart manufacturing and Industry 4.0 environments.