Scholarlink

<p>The quality assurance of the Laser Powder Bed Fusion Process (LPBF) has been extensively investigated over the last decade for in-situ monitoring of metal additive manufacturing. The process inherently generates voids within the bulk of the part, which can detrimentally affect the quality of the printed part. The characterization of these voids by estimating their size and identifying their geometrical features remains a challenge. This study introduces a Machine Learning (ML) based framework for estimating void sizes of varying geometries using layer-wise one-dimensional (1D) average light intensity signal obtained from the optical tomography system during the 3D printing of metallic components. The framework integrates a synthetic data generation module to enhance the training of the size estimation model. The proposed approach enables rapid defect detection and characterization, which are essential for achieving a defect-free LPBF process, and facilitates real-time parameter correction during builds. The inclusion of synthetic data improves the ML model’s performance by reducing the Root Mean Square Error (RMSE) from 67.65 to 58.2 and improving the correlation coefficient between the observed and the estimated pore size from 0.67 to 0.77. The cylindrical-shaped defects had the best results with a reduction in the RMSE from 51.6 to 40.88 and an improvement in the correlation coefficient from 0.81 to 0.89. Overall, the integration of synthetic data significantly improved defect detection performance by reducing the RMSE and improving the correlation coefficients across all the defect types.</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

Data driven estimation of pore size using 1D light emissions for laser powder bed fusion additive manufacturing

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年5月21日

查看全文
  • Jose Galarza,
  • Jorge Barron,
  • Luis Jimenez,
  • Tamer Oraby,
  • Jianzhi Li,
  • Farid Ahmed

摘要

The quality assurance of the Laser Powder Bed Fusion Process (LPBF) has been extensively investigated over the last decade for in-situ monitoring of metal additive manufacturing. The process inherently generates voids within the bulk of the part, which can detrimentally affect the quality of the printed part. The characterization of these voids by estimating their size and identifying their geometrical features remains a challenge. This study introduces a Machine Learning (ML) based framework for estimating void sizes of varying geometries using layer-wise one-dimensional (1D) average light intensity signal obtained from the optical tomography system during the 3D printing of metallic components. The framework integrates a synthetic data generation module to enhance the training of the size estimation model. The proposed approach enables rapid defect detection and characterization, which are essential for achieving a defect-free LPBF process, and facilitates real-time parameter correction during builds. The inclusion of synthetic data improves the ML model’s performance by reducing the Root Mean Square Error (RMSE) from 67.65 to 58.2 and improving the correlation coefficient between the observed and the estimated pore size from 0.67 to 0.77. The cylindrical-shaped defects had the best results with a reduction in the RMSE from 51.6 to 40.88 and an improvement in the correlation coefficient from 0.81 to 0.89. Overall, the integration of synthetic data significantly improved defect detection performance by reducing the RMSE and improving the correlation coefficients across all the defect types.