Scholarlink

<p>The peak weld temperature serves as a critical indicator of joint quality in rotational friction welding. Its thermal history directly governs material plastic flow behavior and the resulting mechanical properties of the joint. Conventional multiphysics simulations reconstruct thermal behavior during the welding process. However, high-fidelity simulations require substantial computational resources and long simulation times, which limit their application in rapid process parameter evaluation and optimal design. To overcome these limitations, this study introduces an artificial intelligence–based data-driven surrogate model. The proposed model learns the nonlinear relationships between rotational friction welding process parameters and weld temperature responses. It enables rapid prediction of the weld temperature history during the welding process. The surrogate model significantly reduces computational cost while preserving the key thermal characteristics of the process. Furthermore, physics-informed feature selection strategies enhance prediction reliability and physical interpretability. The results demonstrate that the proposed approach supports process design and welding quality evaluation in rotational friction welding. Experimental results show that the temperature evolution is divided into three distinct stages, including the friction stage from 0 to 5&#xa0;s, the forging stage from 5 to 12.5&#xa0;s, and the cooling stage after 12.5&#xa0;s. When the rotational speed increases from 1000&#xa0;rpm to 4000&#xa0;rpm, the peak interfacial temperature rises from approximately 215&#xa0;°C to 260&#xa0;°C, indicating a strong positive correlation between rotational speed and heat input. Comparative model evaluation demonstrates that XGBoost provides the best generalization performance, achieving a testing coefficient of determination of 0.9324 and a root mean square error of 11.95&#xa0;°C, which outperforms Random Forest and Support Vector Regression models. Feature importance analysis reveals that process time contributes 75.22%, while rotational speed and friction time contribute 13.29% and 11.48%, respectively. Validation under unseen operating conditions shows low prediction errors ranging from 1.94% to 3.85%. In addition, the proposed artificial intelligence-assisted parameter decision system reduces electrical energy consumption from 4.66 kWh to 0.47 kWh and decreases carbon emissions by approximately 90%, demonstrating strong potential for intelligent and sustainable manufacturing applications. These outcomes directly support the United Nations Sustainable Development Goals related to industry innovation and infrastructure, responsible consumption and production, and climate action, demonstrating the contribution of intelligent manufacturing technologies to sustainable industrial development.</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

Artificial intelligence-driven prediction of weld-zone thermal behavior in rotary friction welding of polymers

  • 期刊论文

  • 发表日期：

    2026年3月9日

查看全文
  • Chil-Chyuan Kuo,
  • Chen Fang,
  • Armaan Farooqui,
  • Song-Hua Huang

摘要

The peak weld temperature serves as a critical indicator of joint quality in rotational friction welding. Its thermal history directly governs material plastic flow behavior and the resulting mechanical properties of the joint. Conventional multiphysics simulations reconstruct thermal behavior during the welding process. However, high-fidelity simulations require substantial computational resources and long simulation times, which limit their application in rapid process parameter evaluation and optimal design. To overcome these limitations, this study introduces an artificial intelligence–based data-driven surrogate model. The proposed model learns the nonlinear relationships between rotational friction welding process parameters and weld temperature responses. It enables rapid prediction of the weld temperature history during the welding process. The surrogate model significantly reduces computational cost while preserving the key thermal characteristics of the process. Furthermore, physics-informed feature selection strategies enhance prediction reliability and physical interpretability. The results demonstrate that the proposed approach supports process design and welding quality evaluation in rotational friction welding. Experimental results show that the temperature evolution is divided into three distinct stages, including the friction stage from 0 to 5 s, the forging stage from 5 to 12.5 s, and the cooling stage after 12.5 s. When the rotational speed increases from 1000 rpm to 4000 rpm, the peak interfacial temperature rises from approximately 215 °C to 260 °C, indicating a strong positive correlation between rotational speed and heat input. Comparative model evaluation demonstrates that XGBoost provides the best generalization performance, achieving a testing coefficient of determination of 0.9324 and a root mean square error of 11.95 °C, which outperforms Random Forest and Support Vector Regression models. Feature importance analysis reveals that process time contributes 75.22%, while rotational speed and friction time contribute 13.29% and 11.48%, respectively. Validation under unseen operating conditions shows low prediction errors ranging from 1.94% to 3.85%. In addition, the proposed artificial intelligence-assisted parameter decision system reduces electrical energy consumption from 4.66 kWh to 0.47 kWh and decreases carbon emissions by approximately 90%, demonstrating strong potential for intelligent and sustainable manufacturing applications. These outcomes directly support the United Nations Sustainable Development Goals related to industry innovation and infrastructure, responsible consumption and production, and climate action, demonstrating the contribution of intelligent manufacturing technologies to sustainable industrial development.