Scholarlink

<p>Additive manufacturing (AM) has gained significant traction across various engineering sectors due to its flexibility in design and manufacturing. Nevertheless, manufacturing defects with highly scattered features in AM metals dominate the scatter of their fatigue behavior. Among the traditional methodologies for predicting the fatigue life of AM metals, semiempirical mechanical models are often limited by incomplete prior assumptions for describing the mechanism of fatigue failure controlled by manufacturing defects. Meanwhile, machine learning (ML) models supported by limited volumes of experimental fatigue data are often considered unreliable and lack interpretability. To address these challenges, an interpretable physics-informed ML framework, integrated Fourier expansion-Kolmogorov Arnold networks (FE-KAN), has been developed. By integrating the physical constraints and the Fourier layer, FE-KAN can effectively assess and correct potential systematic errors in the fatigue life prediction results of the physics-driven models. The effectiveness of FE-KAN in improving the prediction performance has been examined on multiple datasets of AM metals, along with the validity of its interpretation. In contrast to other state-of-the-art methods, the superiority of FE-KAN is demonstrated in both prediction accuracy and interpretability. Ultimately, a unified fatigue life prediction framework for AM metals is established.</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

Fatigue life prediction for additive manufactured metals using an interpretable physics-informed machine learning framework

  • 期刊论文

  • 发表日期：

    2026年7月28日

查看全文
  • Borui Wu,
  • Shun-Peng Zhu,
  • Lanyi Wang,
  • Zijian Xu,
  • Changqi Luo,
  • Qingyuan Wang

摘要

Additive manufacturing (AM) has gained significant traction across various engineering sectors due to its flexibility in design and manufacturing. Nevertheless, manufacturing defects with highly scattered features in AM metals dominate the scatter of their fatigue behavior. Among the traditional methodologies for predicting the fatigue life of AM metals, semiempirical mechanical models are often limited by incomplete prior assumptions for describing the mechanism of fatigue failure controlled by manufacturing defects. Meanwhile, machine learning (ML) models supported by limited volumes of experimental fatigue data are often considered unreliable and lack interpretability. To address these challenges, an interpretable physics-informed ML framework, integrated Fourier expansion-Kolmogorov Arnold networks (FE-KAN), has been developed. By integrating the physical constraints and the Fourier layer, FE-KAN can effectively assess and correct potential systematic errors in the fatigue life prediction results of the physics-driven models. The effectiveness of FE-KAN in improving the prediction performance has been examined on multiple datasets of AM metals, along with the validity of its interpretation. In contrast to other state-of-the-art methods, the superiority of FE-KAN is demonstrated in both prediction accuracy and interpretability. Ultimately, a unified fatigue life prediction framework for AM metals is established.