Scholarlink

<p>Accurately predicting viscosity is essential for the development of high-quality welding fluxes. However, there are limited experimental data available, rendering corresponding predictions rather challenging. To counter such technical adversity, a new stacked ensemble model based on a deep neural network (DNN) and data augmentation has been successfully established, aiming to generate viscosity datasets with high fidelity. Moreover, generative adversarial networks (GANs) have been employed to yield augmented data toward reliable data warehouses. Then, the stacked ensemble viscosity prediction model was architected. Validation trials on the testing sets have demonstrated a coefficient of determination (R<sup>2</sup>) of 0.9671 and a mean absolute error (MAE) of 0.1195 Pa·s. Finally, cross-validation with four sets of experimental data has confirmed that the GANs-DNN model exhibits excellent performance, with MAE values being 0.0354, 0.0925, 0.1218, and 0.0623 Pa·s, respectively. This study not only enables a capable model for viscosity prediction about welding flux but also potentially offers a viable tool for rational welding flux design.</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

Predicting Viscosity of Submerged Arc Welding Fluxes: A Deep Neural Network Approach Facilitated by Data Augmentation

  • 期刊论文

  • 发表日期：

    2025年6月2日

查看全文
  • Hongyu Liu,
  • Yanyun Zhang,
  • Shuai Shi,
  • Hang Yuan,
  • Zushu Li,
  • Cong Wang

摘要

Accurately predicting viscosity is essential for the development of high-quality welding fluxes. However, there are limited experimental data available, rendering corresponding predictions rather challenging. To counter such technical adversity, a new stacked ensemble model based on a deep neural network (DNN) and data augmentation has been successfully established, aiming to generate viscosity datasets with high fidelity. Moreover, generative adversarial networks (GANs) have been employed to yield augmented data toward reliable data warehouses. Then, the stacked ensemble viscosity prediction model was architected. Validation trials on the testing sets have demonstrated a coefficient of determination (R2) of 0.9671 and a mean absolute error (MAE) of 0.1195 Pa·s. Finally, cross-validation with four sets of experimental data has confirmed that the GANs-DNN model exhibits excellent performance, with MAE values being 0.0354, 0.0925, 0.1218, and 0.0623 Pa·s, respectively. This study not only enables a capable model for viscosity prediction about welding flux but also potentially offers a viable tool for rational welding flux design.