Scholarlink

<p>Laser Powder Bed Fusion (L-PBF) is an additive manufacturing technology that involves complex multi-scale and multi-physics interactions, often resulting in defects such as residual stresses and deformation. Optimizing scanning paths to control temperature distribution is crucial. Traditional geometry-based methods like the Least Heat Influence (LHI) are efficient but lack comprehensive physical principles. Recent advances in machine learning (ML) provide alternatives, yet their application in L-PBF path optimization is limited by high training cost issues. This study presents a novel ML-based approach, termed SmallDataBoosted U-Net (SDBU), which enhances model performance using small datasets. By incorporating a tailored loss function that aligns with the optimization objectives and physics-based feature engineering enriching the prediction criteria, SDBU achieves significant improvements in prediction accuracy and stability. Additionally, an integrated scanning path optimization framework tailored to specific printing patterns and utilizing genetic algorithms further reduces dependence on data and ensures practical applicability. Experimental results demonstrate the superior performance of SDBU, achieving a 45.09% improvement in strip mode and a 10.83% improvement in island mode compared to the LHI algorithm. This approach represents a significant advancement in machine learning applications for laser powder bed fusion (L-PBF), offering scalable and precise path optimization solutions. The source code is available on GitHub: <a href="https://github.com/henyoujingshen/LPBF-ScanningPath-Optimization-with-SDBU">https://github.com/henyoujingshen/LPBF-ScanningPath-Optimization-with-SDBU</a>.</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

Optimization of scanning path in laser powder bed fusion: training and application of deep learning model using small dataset

  • 期刊论文

  • 发表日期：

    2025年6月21日

查看全文
  • Hubocheng Tang,
  • Huilin Ren,
  • Yi Xiong

摘要

Laser Powder Bed Fusion (L-PBF) is an additive manufacturing technology that involves complex multi-scale and multi-physics interactions, often resulting in defects such as residual stresses and deformation. Optimizing scanning paths to control temperature distribution is crucial. Traditional geometry-based methods like the Least Heat Influence (LHI) are efficient but lack comprehensive physical principles. Recent advances in machine learning (ML) provide alternatives, yet their application in L-PBF path optimization is limited by high training cost issues. This study presents a novel ML-based approach, termed SmallDataBoosted U-Net (SDBU), which enhances model performance using small datasets. By incorporating a tailored loss function that aligns with the optimization objectives and physics-based feature engineering enriching the prediction criteria, SDBU achieves significant improvements in prediction accuracy and stability. Additionally, an integrated scanning path optimization framework tailored to specific printing patterns and utilizing genetic algorithms further reduces dependence on data and ensures practical applicability. Experimental results demonstrate the superior performance of SDBU, achieving a 45.09% improvement in strip mode and a 10.83% improvement in island mode compared to the LHI algorithm. This approach represents a significant advancement in machine learning applications for laser powder bed fusion (L-PBF), offering scalable and precise path optimization solutions. The source code is available on GitHub: https://github.com/henyoujingshen/LPBF-ScanningPath-Optimization-with-SDBU.