Scholarlink

<p>In aerospace manufacturing, thin-walled components are characterized by complex structures and high specific strength, making them highly susceptible to chatter during milling, which significantly degrades machining quality. Most existing methods for chatter signal acquisition and processing rely on single-signal analysis and are thus vulnerable to environmental interference, leading to signal distortion and reduced data reliability. Furthermore, current research still lacks sufficiently robust theoretical foundations and standardized feature extraction procedures. To address these issues, this study proposes a multi-modal feature coupling method that fuses acoustic emission (AE) and acceleration signals to establish an efficient chatter recognition model. First, Fourier transform analysis is performed to preliminarily verify that the chatter frequency approximates the natural frequency of the workpiece. Second, the coefficient of variation of the AE signals and the wavelet packet energy entropy of the acceleration signals are selected as characteristic parameters to identify chatter states through feature coupling; meanwhile, the theoretical basis for employing energy entropy as a chatter indicator is elaborated. On this basis, a chatter monitoring model integrating K-means clustering and Support Vector Machine (KM-SVM) is established. Threshold criteria derived from the coefficient of variation and wavelet packet energy entropy are employed to implement label initialization for unsupervised clustering. The Support Vector Machine learns the joint decision boundary of multi-modal features to achieve fine-grained identification of stable cutting, slight chatter, and severe chatter, thereby enhancing the model’s robustness to signal noise and non-chatter disturbances. Evaluated through five-fold stratified cross-validation, the KM-SVM model achieves an average classification accuracy of 98.17% ± 0.37%, demonstrating excellent operational stability and significantly outperforming traditional single-signal analysis methods under actual machining conditions. This method provides reliable support for chatter monitoring and suppression in thin-walled component milling, offering a new theoretical basis and engineering reference for aerospace manufacturing.</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

Multi-modal coupled feature fusion for dynamic identification of milling chatter

  • 期刊论文

  • 发表日期：

    2026年5月2日

查看全文
  • Lei Hu,
  • Chen Zhang,
  • Renyi Cao,
  • Yujie Shang,
  • Qinyu Zhang

摘要

In aerospace manufacturing, thin-walled components are characterized by complex structures and high specific strength, making them highly susceptible to chatter during milling, which significantly degrades machining quality. Most existing methods for chatter signal acquisition and processing rely on single-signal analysis and are thus vulnerable to environmental interference, leading to signal distortion and reduced data reliability. Furthermore, current research still lacks sufficiently robust theoretical foundations and standardized feature extraction procedures. To address these issues, this study proposes a multi-modal feature coupling method that fuses acoustic emission (AE) and acceleration signals to establish an efficient chatter recognition model. First, Fourier transform analysis is performed to preliminarily verify that the chatter frequency approximates the natural frequency of the workpiece. Second, the coefficient of variation of the AE signals and the wavelet packet energy entropy of the acceleration signals are selected as characteristic parameters to identify chatter states through feature coupling; meanwhile, the theoretical basis for employing energy entropy as a chatter indicator is elaborated. On this basis, a chatter monitoring model integrating K-means clustering and Support Vector Machine (KM-SVM) is established. Threshold criteria derived from the coefficient of variation and wavelet packet energy entropy are employed to implement label initialization for unsupervised clustering. The Support Vector Machine learns the joint decision boundary of multi-modal features to achieve fine-grained identification of stable cutting, slight chatter, and severe chatter, thereby enhancing the model’s robustness to signal noise and non-chatter disturbances. Evaluated through five-fold stratified cross-validation, the KM-SVM model achieves an average classification accuracy of 98.17% ± 0.37%, demonstrating excellent operational stability and significantly outperforming traditional single-signal analysis methods under actual machining conditions. This method provides reliable support for chatter monitoring and suppression in thin-walled component milling, offering a new theoretical basis and engineering reference for aerospace manufacturing.