Scholarlink

<p>Laser Powder Bed Fusion (L-PBF) is distinguished by its precision in fabricating intricate structures, yet accurately predicting thermal profiles remains challenging due to the computational complexity of traditional physics-based models. This study explores the application of the Pix2Pix Generative Adversarial Network (GAN) framework for predicting thermal maps in L-PBF processes across multiple materials, including 316 stainless steel alloy (SS316L), aluminum alloy (AlSi10Mg), pure copper (Cu), and pure tungsten (W). Unlike conventional image-to-image translation tasks, where input and output images exhibit pixel-to-pixel spatial correlation, this work maps additive manufacturing process parameters encoded as images lacking spatial correlation to thermal profiles representing temperature distributions. The proposed approach demonstrates high accuracy (Mean Squared Error (MSE) <InlineEquation ID="IEq1"> <EquationSource Format="TEX">\(&lt;\mathbf {0.018\%}\)</EquationSource> </InlineEquation> and Structural Similarity Index Measure (SSIM) <InlineEquation ID="IEq2"> <EquationSource Format="TEX">\(&gt;\mathbf {92.5\%}\)</EquationSource> </InlineEquation>) and rapid prediction speeds (100 images per second) using data derived from finite element simulations, which were validated with experimental results. This study provides a computationally efficient methodology for thermal profile prediction, showcasing the utility of GANs in capturing the relationship between process parameters and thermal behavior in additive manufacturing.</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

Machine learning-based prediction of thermal profiles during laser-based additive manufacturing

  • 期刊论文

  • 发表日期：

    2026年2月25日

查看全文
  • Aishwarya Manjunath,
  • Venkata Mani Krishna Karri,
  • Amrutha Anantatamukala,
  • Selvamurugan Palaniappan,
  • Shashank Sharma,
  • Song Fu,
  • Narendra B. Dahotre

摘要

Laser Powder Bed Fusion (L-PBF) is distinguished by its precision in fabricating intricate structures, yet accurately predicting thermal profiles remains challenging due to the computational complexity of traditional physics-based models. This study explores the application of the Pix2Pix Generative Adversarial Network (GAN) framework for predicting thermal maps in L-PBF processes across multiple materials, including 316 stainless steel alloy (SS316L), aluminum alloy (AlSi10Mg), pure copper (Cu), and pure tungsten (W). Unlike conventional image-to-image translation tasks, where input and output images exhibit pixel-to-pixel spatial correlation, this work maps additive manufacturing process parameters encoded as images lacking spatial correlation to thermal profiles representing temperature distributions. The proposed approach demonstrates high accuracy (Mean Squared Error (MSE) \(<\mathbf {0.018\%}\) and Structural Similarity Index Measure (SSIM) \(>\mathbf {92.5\%}\) ) and rapid prediction speeds (100 images per second) using data derived from finite element simulations, which were validated with experimental results. This study provides a computationally efficient methodology for thermal profile prediction, showcasing the utility of GANs in capturing the relationship between process parameters and thermal behavior in additive manufacturing.