Scholarlink

<p>In nuclear industry, producing high-quality welds with precise weld bead dimensions is essential to ensure weld integrity and prevent radioactive leakage. The present study applies artificial neural networks to predict weld bead geometry during the fabrication of zircaloy-2 cladded fuel pins for boiling water reactors, using tungsten inert gas welding. Welding experiments were conducted with varied input parameters to develop a robust training and testing dataset. An artificial neural network model was designed with an optimized architecture, exploring different configurations of hidden layers and neurons. The model’s performance was evaluated across a range of activation functions, batch sizes, and learning rates. Once the optimal network configuration and training strategy were identified, the model was trained using a training dataset. The predictive accuracy of the trained model was assessed using an independent test dataset, and it was found that predicted values were within ± 10% of experimentally obtained values.</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 Weld Bead Geometry in TIG Welding Process of Zircaloy Fuel Pins Using Artificial Neural Network

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年6月3日

查看全文
  • Shibashankar Das,
  • K. V. Mani Krishna,
  • U. K. Arora,
  • Raghvendra Tewari

摘要

In nuclear industry, producing high-quality welds with precise weld bead dimensions is essential to ensure weld integrity and prevent radioactive leakage. The present study applies artificial neural networks to predict weld bead geometry during the fabrication of zircaloy-2 cladded fuel pins for boiling water reactors, using tungsten inert gas welding. Welding experiments were conducted with varied input parameters to develop a robust training and testing dataset. An artificial neural network model was designed with an optimized architecture, exploring different configurations of hidden layers and neurons. The model’s performance was evaluated across a range of activation functions, batch sizes, and learning rates. Once the optimal network configuration and training strategy were identified, the model was trained using a training dataset. The predictive accuracy of the trained model was assessed using an independent test dataset, and it was found that predicted values were within ± 10% of experimentally obtained values.