Scholarlink

<p>The formation of oxide scale on steel surfaces during high-temperature processing affects surface quality, mechanical properties, and subsequent manufacturing steps. Traditional characterization methods are often destructive and lack real-time feedback. This work proposes a non-contact method for estimating oxide-layer thickness in AISI 1045 steel using thermal imaging and machine learning. Five Joule-heated oxidation experiments were conducted, and infrared thermograms were recorded throughout each thermal cycle. Each image was flattened and processed using principal component analysis (PCA), and the resulting features were aligned with micrometer-scale ground-truth thickness values. Regression models—Linear, SVR, Random Forest (RF), and XGBoost—were trained and validated, with cross-validation used to assess robustness. Tree-based models, particularly RF, outperformed linear and kernel-based approaches, achieving an <Emphasis Type="BoldItalic">R</Emphasis><sup><Emphasis Type="BoldItalic">2</Emphasis></sup> of 0.89 and a mean absolute error of &lt; 7&#xa0;µm in cross-validation. Visual inspections confirmed that the predicted thickness trends closely followed real measurements, even across independent test sets and full experiment sequences. The proposed framework enables accurate, real-time estimation of thermally induced dimensional changes in steel specimens from infrared thermograms alone. Although the reference measurement captures total surface expansion—including oxide growth and thermal dilation—the close alignment with micrometer readings highlights the potential of this method for non-contact thermal process tracking in steels.</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

Non-Contact Thickness Estimation of High-Temperature Oxide Scales on AISI 1045 Steel Using Infrared Thermography and Tree-Based Machine Learning

  • 期刊论文

  • 发表日期：

    2026年3月19日

查看全文
  • Antony Morales-Cervantes,
  • Gerardo Marx Chávez-Campos,
  • Héctor Javier Vergara-Hernández,
  • Maritza Fabiola León-Bejarano,
  • Edgar Guevara,
  • Jorge Sergio Téllez-Martínez

摘要

The formation of oxide scale on steel surfaces during high-temperature processing affects surface quality, mechanical properties, and subsequent manufacturing steps. Traditional characterization methods are often destructive and lack real-time feedback. This work proposes a non-contact method for estimating oxide-layer thickness in AISI 1045 steel using thermal imaging and machine learning. Five Joule-heated oxidation experiments were conducted, and infrared thermograms were recorded throughout each thermal cycle. Each image was flattened and processed using principal component analysis (PCA), and the resulting features were aligned with micrometer-scale ground-truth thickness values. Regression models—Linear, SVR, Random Forest (RF), and XGBoost—were trained and validated, with cross-validation used to assess robustness. Tree-based models, particularly RF, outperformed linear and kernel-based approaches, achieving an R2 of 0.89 and a mean absolute error of < 7 µm in cross-validation. Visual inspections confirmed that the predicted thickness trends closely followed real measurements, even across independent test sets and full experiment sequences. The proposed framework enables accurate, real-time estimation of thermally induced dimensional changes in steel specimens from infrared thermograms alone. Although the reference measurement captures total surface expansion—including oxide growth and thermal dilation—the close alignment with micrometer readings highlights the potential of this method for non-contact thermal process tracking in steels.