Scholarlink

<p>In laser powder bed fusion (LPBF), melt pool (MP) characteristics significantly influence process-induced defects, necessitating efficient virtual prediction of MP geometry for improved build quality. Many physics-driven and data-driven modeling approaches are used to attain ‘First Time Right’ parts. Experimental data used to develop data-driven models often contain noise from diverse LPBF variables. Numerical modeling can aid data-driven models but is constrained by high computational costs and limited datasets. This study addresses the challenge of accurately linking LPBF process parameters to MP geometry and the major process-induced defects, such as lack of fusion, balling, and keyholing. A high-fidelity numerical model previously developed by the authors is employed to generate a reliable MP dataset. Machine learning algorithms, including decision tree, support vector regression, and Gaussian process regression, have been used to train the model and predict MP geometry. In addition, a deep neural network (DNN) has been implemented to leverage deep learning capabilities while mitigating the need for large-scale experimental data. The results demonstrate that the DNN outperforms other models in predicting MP geometry and identifying defects. The developed framework offers an efficient and scalable approach for virtual process optimization, with validation against experimental and numerical studies confirming its robustness.</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

Prediction of Melt Pool Geometry and Defects in Laser Additive Manufacturing of Inconel 718 Alloy: A Machine and Deep Learning Approach

  • 期刊论文

  • 发表日期：

    2025年10月30日

查看全文
  • Anuj Kumar,
  • Mukul Shukla

摘要

In laser powder bed fusion (LPBF), melt pool (MP) characteristics significantly influence process-induced defects, necessitating efficient virtual prediction of MP geometry for improved build quality. Many physics-driven and data-driven modeling approaches are used to attain ‘First Time Right’ parts. Experimental data used to develop data-driven models often contain noise from diverse LPBF variables. Numerical modeling can aid data-driven models but is constrained by high computational costs and limited datasets. This study addresses the challenge of accurately linking LPBF process parameters to MP geometry and the major process-induced defects, such as lack of fusion, balling, and keyholing. A high-fidelity numerical model previously developed by the authors is employed to generate a reliable MP dataset. Machine learning algorithms, including decision tree, support vector regression, and Gaussian process regression, have been used to train the model and predict MP geometry. In addition, a deep neural network (DNN) has been implemented to leverage deep learning capabilities while mitigating the need for large-scale experimental data. The results demonstrate that the DNN outperforms other models in predicting MP geometry and identifying defects. The developed framework offers an efficient and scalable approach for virtual process optimization, with validation against experimental and numerical studies confirming its robustness.