Scholarlink

<p>Uncontrolled vibrations and chatter bring about acute problems in milling processes, which all worsen the surface finish, decrease the tool life, and diminish the process stability. This paper suggests a smart chatter detection system applying low-cost dual-sensor fusion (vibration and acoustic measurements) and computationally efficient feature extraction and machine learning classification. Features are extracted in frequency domain based on Fast Fourier Transform (FFT) and optimized based on Recursive Feature Elimination (RFE) to achieve minimal computation resourcing interestingly with a high accuracy. Random Forest classifier shows 98% accuracy to classify labelling the milling stability and instability based on a 1,028 samples dataset fixed via segmentations. This work is novel in: (1) sensor fusion of vibration and sound to get better robustness, (2) industry ready low-cost accelerometer and microphone are used, (3) FFT based feature extraction along with RFE is used to achieve best accuracy-efficiency trade-off, and (4) high classification accuracy is achieved with a minimum number of experimental runs. The offered approach can be scaled down, affordable, and it can be deployed in real time in intelligent manufacturing processes.</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

Intelligent Chatter Detection in Milling Using Vibration-Acoustic Fusion and Machine Learning

  • 期刊论文

  • 发表日期：

    2025年12月3日

查看全文
  • Sunil M. Pondkule,
  • Sachin M. Bhosle,
  • S. C. Mahadik

摘要

Uncontrolled vibrations and chatter bring about acute problems in milling processes, which all worsen the surface finish, decrease the tool life, and diminish the process stability. This paper suggests a smart chatter detection system applying low-cost dual-sensor fusion (vibration and acoustic measurements) and computationally efficient feature extraction and machine learning classification. Features are extracted in frequency domain based on Fast Fourier Transform (FFT) and optimized based on Recursive Feature Elimination (RFE) to achieve minimal computation resourcing interestingly with a high accuracy. Random Forest classifier shows 98% accuracy to classify labelling the milling stability and instability based on a 1,028 samples dataset fixed via segmentations. This work is novel in: (1) sensor fusion of vibration and sound to get better robustness, (2) industry ready low-cost accelerometer and microphone are used, (3) FFT based feature extraction along with RFE is used to achieve best accuracy-efficiency trade-off, and (4) high classification accuracy is achieved with a minimum number of experimental runs. The offered approach can be scaled down, affordable, and it can be deployed in real time in intelligent manufacturing processes.