Scholarlink

<p>This study addresses the challenge of predicting and controlling melt pool behavior in Hot-Wire Laser Metal Deposition (HW-LMD) technology by proposing a transfer learning strategy based on simulation datasets for melt pool information prediction. First, a large amount of simulated data was generated using a numerical model to pre-train a deep neural network (DNN). Then, transfer learning was applied by incorporating actual experimental data to enhance the model’s accuracy in predicting melt pool size information. The experimental results demonstrate that this method significantly reduces the demand for experimental data and lowers prediction errors. The model trained with traditional methods exhibited an error rate of 21.16%, whereas the error was significantly reduced to 2.03% after optimization using the transfer learning strategy based on the simulation dataset. The findings offer a novel approach to process optimization and quality control in the field of 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

Optimization of process prediction models for hot-wire laser metal deposition using transfer learning strategies based on simulation datasets

  • 期刊论文

  • 发表日期：

    2025年1月4日

查看全文
  • Chunkai Li,
  • Yu Pan,
  • Yu Shi,
  • Wenkai Wang

摘要

This study addresses the challenge of predicting and controlling melt pool behavior in Hot-Wire Laser Metal Deposition (HW-LMD) technology by proposing a transfer learning strategy based on simulation datasets for melt pool information prediction. First, a large amount of simulated data was generated using a numerical model to pre-train a deep neural network (DNN). Then, transfer learning was applied by incorporating actual experimental data to enhance the model’s accuracy in predicting melt pool size information. The experimental results demonstrate that this method significantly reduces the demand for experimental data and lowers prediction errors. The model trained with traditional methods exhibited an error rate of 21.16%, whereas the error was significantly reduced to 2.03% after optimization using the transfer learning strategy based on the simulation dataset. The findings offer a novel approach to process optimization and quality control in the field of additive manufacturing.