Scholarlink

<p>The objective of this research is to investigate the feasibility of utilizing tooling vibration signals and cutting parameters to predict workpiece surface profiles in the milling process, enabling a comprehensive evaluation of surface quality. Information-contained (IC) signals were synthesized from milling vibration measurements using principal component analysis (PCA) and empirical mode decomposition (EMD). Statistical features and sweeping-frequency features were then extracted from individual intrinsic mode functions (IMFs) of the IC signals in both time and frequency domains. Subsequently, independent long short-term memory (LSTM) networks were used to estimate the individual IMFs of the surface profiles, incorporating the milling parameters and selected signal features as inputs. Results demonstrated that mid-low frequency components of surface profiles could be estimated accurately with an average mean absolute percentage error (MAPE) of 14.72%, while low-frequency components achieved an average MAPE of 7.79%.</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

Surface profile estimation in milling through vibration analysis and long short-term memory networks

  • 期刊论文

  • 发表日期：

    2025年2月25日

查看全文
  • Tian-Yau Wu,
  • Cheng-Yi Lin

摘要

The objective of this research is to investigate the feasibility of utilizing tooling vibration signals and cutting parameters to predict workpiece surface profiles in the milling process, enabling a comprehensive evaluation of surface quality. Information-contained (IC) signals were synthesized from milling vibration measurements using principal component analysis (PCA) and empirical mode decomposition (EMD). Statistical features and sweeping-frequency features were then extracted from individual intrinsic mode functions (IMFs) of the IC signals in both time and frequency domains. Subsequently, independent long short-term memory (LSTM) networks were used to estimate the individual IMFs of the surface profiles, incorporating the milling parameters and selected signal features as inputs. Results demonstrated that mid-low frequency components of surface profiles could be estimated accurately with an average mean absolute percentage error (MAPE) of 14.72%, while low-frequency components achieved an average MAPE of 7.79%.