Scholarlink

<p>In the manufacturing process of mechanical parts, it is crucial to conduct intelligent damage recognition of key components through sound signal analysis. This practice enables proactive defect detection, reduces equipment downtime. Sound signal analysis effectively extracts damage state information to meet industrial detection needs. Thus, this paper proposes an integrated RegNet-random forest (RF) framework based on sound signal processing, so as to identify cracks in mechanical critical structures with different damage extents accurately and stably. First, sound signals of structures with different damage levels are collected via pulse excitation experiments. Second, short-time Fourier transform (STFT) converts these signals into time-frequency images, from which RegNet extracts features. Subsequently, the model is improved in terms of optimizing the RF algorithm and network structure. Finally, the construction of the RegNet-RF model is completed and for recognition. Experiments verify its effectiveness in mechanical structure health monitoring, with comparative analyses showing superiority in damage identification.</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

Crack damage intelligent recognition of mechanical key structures based on RegNet and random forest with sound signal

  • 期刊论文

  • 发表日期：

    2026年5月12日

查看全文
  • Jinxiu Qu,
  • Jiayan Wu,
  • Changquan Shi,
  • Yumei Bai,
  • Fei Ke,
  • Minzhi Qin,
  • Jie Yun

摘要

In the manufacturing process of mechanical parts, it is crucial to conduct intelligent damage recognition of key components through sound signal analysis. This practice enables proactive defect detection, reduces equipment downtime. Sound signal analysis effectively extracts damage state information to meet industrial detection needs. Thus, this paper proposes an integrated RegNet-random forest (RF) framework based on sound signal processing, so as to identify cracks in mechanical critical structures with different damage extents accurately and stably. First, sound signals of structures with different damage levels are collected via pulse excitation experiments. Second, short-time Fourier transform (STFT) converts these signals into time-frequency images, from which RegNet extracts features. Subsequently, the model is improved in terms of optimizing the RF algorithm and network structure. Finally, the construction of the RegNet-RF model is completed and for recognition. Experiments verify its effectiveness in mechanical structure health monitoring, with comparative analyses showing superiority in damage identification.