Scholarlink

<p>Surface polishing by laser remelting (SP-LRM) is a rapid and versatile manufacturing technique capable of producing high-quality surface finishes. However, due to the complex interplay of laser parameters, the process exhibits three distinct regimes termed as shallow, intermediate, and deep, each associated with different surface outcomes. Currently, regime identification and surface quality assessment rely on tedious offline analysis. As an alternative to this approach, the current study proposes an in-situ, data-driven approach using k-means clustering applied to near-infrared (NIR) thermographic emission images to automatically classify process stability and detect anomalies. Along these lines, silhouette analysis revealed that image clustering is influenced by positional dependency at lower laser power, an aspect that was mitigated by centering the melt pool in each frame. Stable processes, characterized by a single dominant image cluster, correlated with improved surface roughness (Sa reduction of 35.3%), while unstable processes exhibited multiple clusters and resulted in surface degradation (Sa increase of 26%). Taken together, these findings demonstrate the novelty and feasibility of unsupervised learning for real-time monitoring and regime classification in SP-LRM, paving the way for intelligent process control.</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

Detection of stability and positional dependency of surface polishing by laser remelting via K-means clustering of co-axial thermographic emission images

  • 期刊论文

  • 发表日期：

    2026年1月27日

查看全文
  • Srdjan J. Cvijanovic,
  • Evgueni V. Bordatchev,
  • O. Remus Tutunea-Fatan

摘要

Surface polishing by laser remelting (SP-LRM) is a rapid and versatile manufacturing technique capable of producing high-quality surface finishes. However, due to the complex interplay of laser parameters, the process exhibits three distinct regimes termed as shallow, intermediate, and deep, each associated with different surface outcomes. Currently, regime identification and surface quality assessment rely on tedious offline analysis. As an alternative to this approach, the current study proposes an in-situ, data-driven approach using k-means clustering applied to near-infrared (NIR) thermographic emission images to automatically classify process stability and detect anomalies. Along these lines, silhouette analysis revealed that image clustering is influenced by positional dependency at lower laser power, an aspect that was mitigated by centering the melt pool in each frame. Stable processes, characterized by a single dominant image cluster, correlated with improved surface roughness (Sa reduction of 35.3%), while unstable processes exhibited multiple clusters and resulted in surface degradation (Sa increase of 26%). Taken together, these findings demonstrate the novelty and feasibility of unsupervised learning for real-time monitoring and regime classification in SP-LRM, paving the way for intelligent process control.