Scholarlink

<p>This paper presents a novel data-driven framework for chatter detection and classification in CNC turning using microphone-based acoustic signal acquisition. Chatter is an unwanted self-excited vibration that is a major source of poor surface finish and reduces tool life. The framework integrates a structured preprocessing pipeline (Butterworth filtering, clipping, smoothing, and min–max scaling), Local Mean Decomposition (LMD) for signal decomposition, multi-algorithm feature ranking (Chi-square, ReliefF, ANOVA, Kruskal–Wallis), Principal Component Analysis (PCA) for dimensionality reduction, and probabilistic Naive Bayes classification; a combination not previously reported for CNC turning chatter detection. The study analytically discusses these methods, which entail signal preprocessing, feature extraction, dimensionality reduction, and classification. Experiments were conducted on a CNC lathe where a microphone-based acquisition system was used to collect chatter signals. Preprocessing of the recorded vibration signals was performed in a series of filters, clipping, smoothing, and scaling to reduce environment noise (transient perturbation) effects. Local Mean Decomposition (LMD) technique was used to break down the preprocessed signals into a set of Product Functions (PFs). Based on the chatter signal (reconstructed) which is made up of the most dominant PFs, several statistical features were obtained. As it was observed that some of the extracted features were redundant, the dimensionality reduction was conducted with the help of Principal Component Analysis (PCA), and the ranking of features was carried out with the help of Chi-square (Chi<sup>2</sup>), Relief, ANOVA and Kruskal Wallis algorithms. The most notable characteristic was identified as the Absolute Mean Amplitude (AMA) and was used to classify the Chatter Indicator (CI). In the case of chatter classification, Gaussian Naive Bayes and Kernel Naive Bayes models were carried out and tested. Lastly, further experimental studies were carried out to ascertain the strength and forecasting of the suggested methodology.</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

Optimizing CNC machining: naive bayes classifier for achieving chatter-free stable turning

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年5月25日

查看全文
  • Pankaj Gupta,
  • Ali B. M. Ali,
  • Sunil Kumar,
  • Ahmed A. Alamiery,
  • Aseel Smerat,
  • Wahaj Ahmad Khan

摘要

This paper presents a novel data-driven framework for chatter detection and classification in CNC turning using microphone-based acoustic signal acquisition. Chatter is an unwanted self-excited vibration that is a major source of poor surface finish and reduces tool life. The framework integrates a structured preprocessing pipeline (Butterworth filtering, clipping, smoothing, and min–max scaling), Local Mean Decomposition (LMD) for signal decomposition, multi-algorithm feature ranking (Chi-square, ReliefF, ANOVA, Kruskal–Wallis), Principal Component Analysis (PCA) for dimensionality reduction, and probabilistic Naive Bayes classification; a combination not previously reported for CNC turning chatter detection. The study analytically discusses these methods, which entail signal preprocessing, feature extraction, dimensionality reduction, and classification. Experiments were conducted on a CNC lathe where a microphone-based acquisition system was used to collect chatter signals. Preprocessing of the recorded vibration signals was performed in a series of filters, clipping, smoothing, and scaling to reduce environment noise (transient perturbation) effects. Local Mean Decomposition (LMD) technique was used to break down the preprocessed signals into a set of Product Functions (PFs). Based on the chatter signal (reconstructed) which is made up of the most dominant PFs, several statistical features were obtained. As it was observed that some of the extracted features were redundant, the dimensionality reduction was conducted with the help of Principal Component Analysis (PCA), and the ranking of features was carried out with the help of Chi-square (Chi2), Relief, ANOVA and Kruskal Wallis algorithms. The most notable characteristic was identified as the Absolute Mean Amplitude (AMA) and was used to classify the Chatter Indicator (CI). In the case of chatter classification, Gaussian Naive Bayes and Kernel Naive Bayes models were carried out and tested. Lastly, further experimental studies were carried out to ascertain the strength and forecasting of the suggested methodology.