Scholarlink

Concrete-filled steel tubular (CFST) joints have been widely applied to arch bridges in China. Recently, the fatigue cracks seriously damaging the structural safety were found at the chord-brace intersection in this type bridges. Against such a situation, stress concentration factors (SCFs) have been formulated for CFST joints based on the results of detailed finite element (FE) analyses using the least-squares method. However, their accuracy depends on the form of the function to be set in advance, and it may be difficult to set an appropriate form of the function. In addition, the creation of many FE models requires a great deal of time and effort. To solve these problems, this study attempted to use neural networks (NNs), which can perform regression analysis even with a small number of data and does not require to set the functional form, for SCF calculation of CFST T-joints under axial force in the brace. Furthermore, a method for generating the minimum training data required for the NN to have sufficient accuracy was proposed.

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

Application of Machine Learning to the Development of Calculation Model for Stress Concentration Factor of Concrete-Filled Steel Tubular T-Joints Under Axial Force in the Brace

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Kanji Nogita,
  • Shozo Nakamura,
  • Toshihiro Okumatsu,
  • Takafumi Nishikawa

摘要

Concrete-filled steel tubular (CFST) joints have been widely applied to arch bridges in China. Recently, the fatigue cracks seriously damaging the structural safety were found at the chord-brace intersection in this type bridges. Against such a situation, stress concentration factors (SCFs) have been formulated for CFST joints based on the results of detailed finite element (FE) analyses using the least-squares method. However, their accuracy depends on the form of the function to be set in advance, and it may be difficult to set an appropriate form of the function. In addition, the creation of many FE models requires a great deal of time and effort. To solve these problems, this study attempted to use neural networks (NNs), which can perform regression analysis even with a small number of data and does not require to set the functional form, for SCF calculation of CFST T-joints under axial force in the brace. Furthermore, a method for generating the minimum training data required for the NN to have sufficient accuracy was proposed.