Scholarlink

<p>In metal additive manufacturing (AM), accurately predicting bead geometry in Directed Energy Deposition (DED) is crucial for ensuring the quality and precision of the final part. However, this task is often complicated by the dynamic and nonlinear nature of the DED process. This study introduces a novel spatio-temporal neural network (STNN) model designed to predict bead geometry using real-time melt pool images captured during the DED process. Multiple datasets are developed to capture dynamic process responses. The proposed STNN, which considers critical time intervals and overlap rates of the melt pools, demonstrates superior prediction accuracy compared to existing methods. The results also identify the optimal time interval for real-time control and highlight that future time instances have a greater impact than past instances, as evidenced by the asymmetrical melt pool observed in thermal images. Furthermore, the predictive bead geometry enables real-time feedback control in DED by dynamically compensating for deposition height variations. Experimental validation shows a 56.82% reduction in mean absolute error (MAE) and a 57.05% reduction in root mean square error (RMSE) in final component height accuracy compared to conventional open-loop control. These advancements offer a promising tool for real-time monitoring and optimization of the DED process.</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

Bead geometry prediction of directed energy deposition based on spatio-temporal neural network model

  • 期刊论文

  • 发表日期：

    2026年2月10日

查看全文
  • Yuanzhi Chen,
  • Wenze Zhang,
  • Yichen Wang,
  • Pengcheng Hu,
  • Molong Duan,
  • Zichuan Fu,
  • Xiaoke Deng

摘要

In metal additive manufacturing (AM), accurately predicting bead geometry in Directed Energy Deposition (DED) is crucial for ensuring the quality and precision of the final part. However, this task is often complicated by the dynamic and nonlinear nature of the DED process. This study introduces a novel spatio-temporal neural network (STNN) model designed to predict bead geometry using real-time melt pool images captured during the DED process. Multiple datasets are developed to capture dynamic process responses. The proposed STNN, which considers critical time intervals and overlap rates of the melt pools, demonstrates superior prediction accuracy compared to existing methods. The results also identify the optimal time interval for real-time control and highlight that future time instances have a greater impact than past instances, as evidenced by the asymmetrical melt pool observed in thermal images. Furthermore, the predictive bead geometry enables real-time feedback control in DED by dynamically compensating for deposition height variations. Experimental validation shows a 56.82% reduction in mean absolute error (MAE) and a 57.05% reduction in root mean square error (RMSE) in final component height accuracy compared to conventional open-loop control. These advancements offer a promising tool for real-time monitoring and optimization of the DED process.