Scholarlink

<p>Tool-condition monitoring (TCM) plays a critical role in improving machining efficiency by reducing operational costs, minimizing downtime, and preventing scrap formation. This study presents a sound-based ensemble learning (EL) framework for classifying tool wear during the turning of AISI 1050 steel using only microphone measurements—offering a practical, low-cost, and non-contact alternative to vibration- or force-sensor-based systems. A comprehensive set of sound features—including amplitude, spectral descriptors, wavelet energy, and Mel-frequency cepstral coefficients (MFCCs)—was extracted to capture essential temporal and spectral characteristics of the cutting process. To enhance generalization, a stratified tenfold cross-validation (CV) strategy was employed. Three high-performing classifiers —GBM, LightGBM, and XGBoost—were integrated in a soft-voting ensemble, which outperformed individual models in both accuracy and robustness. The proposed model achieved 95% accuracy and classified severe wear and unworn conditions with 100% accuracy. Misclassifications were primarily observed between slight wear and worn classes due to overlapping sound signatures. The results demonstrate that the proposed sound-based EL framework is effective, scalable, and suitable for near real-time TCM applications in modern manufacturing 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

Tool-Wear Classification Using Sound-Based Ensemble Learning in the Turning of AISI 1050 Steels

  • 期刊论文

  • 发表日期：

    2026年2月25日

查看全文
  • Savaş Koç,
  • Ramazan İlenç

摘要

Tool-condition monitoring (TCM) plays a critical role in improving machining efficiency by reducing operational costs, minimizing downtime, and preventing scrap formation. This study presents a sound-based ensemble learning (EL) framework for classifying tool wear during the turning of AISI 1050 steel using only microphone measurements—offering a practical, low-cost, and non-contact alternative to vibration- or force-sensor-based systems. A comprehensive set of sound features—including amplitude, spectral descriptors, wavelet energy, and Mel-frequency cepstral coefficients (MFCCs)—was extracted to capture essential temporal and spectral characteristics of the cutting process. To enhance generalization, a stratified tenfold cross-validation (CV) strategy was employed. Three high-performing classifiers —GBM, LightGBM, and XGBoost—were integrated in a soft-voting ensemble, which outperformed individual models in both accuracy and robustness. The proposed model achieved 95% accuracy and classified severe wear and unworn conditions with 100% accuracy. Misclassifications were primarily observed between slight wear and worn classes due to overlapping sound signatures. The results demonstrate that the proposed sound-based EL framework is effective, scalable, and suitable for near real-time TCM applications in modern manufacturing environments.