Scholarlink

<p>The acquisition of a three-dimensional temperature field remains a challenge in laser directed energy deposition (L-DED). The experimental measurement approach is limited to obtaining an overall temperature distribution, and numerical simulations, such as the finite element method (FEM), are time consuming. In this study, an expanded physics-informed neural network (PINN) was proposed to predict the temperature using multi-parameter, multi-track, and experimental data involved in L-DED. The key process parameters—laser power, scanning speed, and laser spot diameter—were integrated as inputs into the network to fully obviate repetitive calculations and significantly improve the prediction efficiency. The training time was reduced by approximately 90% using transfer learning for the multi-track deposition. The surface temperature data measured using an infrared camera were fused as one of the sources to drive the network with high reliability. All the results of the expanded PINN were validated on accuracy using FEM simulations.</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

A physics-informed neural network on 3D-temperature prediction with multi-track, multi-parameter and measured data in laser deposition process

  • 期刊论文

  • 发表日期：

    2025年10月13日

查看全文
  • Yining Yuan,
  • Gang Wang,
  • Yuelan Di,
  • Wei Shi,
  • Liping Wang

摘要

The acquisition of a three-dimensional temperature field remains a challenge in laser directed energy deposition (L-DED). The experimental measurement approach is limited to obtaining an overall temperature distribution, and numerical simulations, such as the finite element method (FEM), are time consuming. In this study, an expanded physics-informed neural network (PINN) was proposed to predict the temperature using multi-parameter, multi-track, and experimental data involved in L-DED. The key process parameters—laser power, scanning speed, and laser spot diameter—were integrated as inputs into the network to fully obviate repetitive calculations and significantly improve the prediction efficiency. The training time was reduced by approximately 90% using transfer learning for the multi-track deposition. The surface temperature data measured using an infrared camera were fused as one of the sources to drive the network with high reliability. All the results of the expanded PINN were validated on accuracy using FEM simulations.