Scholarlink

<p>IF steel has a complex production process and unclear microstructural evolution, making it hard to optimize processes and control properties accurately. Current studies either overlook microstructure - related factors or fail to combine unstructured data (such as metallographic images with structured data, resulting in insufficient prediction accuracy. To solve these problems, this study proposes a multimoda full - process property prediction model (WPPPM) based on ResNet50-MLP. lt integrates composition, process parameters, and microstructural features (enhanced metallographic images, precipitation behavior, and characteristic texture strength). The novelty of this model is reflected in three aspects: First, using CycleGAN to enhance metallographic images for better extraction of microstructure distribution features. Second, combining PSO and BO - XGBoost to quantify precipitation behavior and the composition - process - texturerelationship, addressing key issues in parameter determination and quantitative modeling. Third, fusing structured and unstructured metallographic data to directly connect composition-process, microstructure, and properties. Comparative experiments show that WPPPM is moreaccurate than traditional models (which only use composition and process parameters) in predicting yield strength, tensile strength, elongation, and r-value. This work offers a new way for precise property control and process optimization of lF steel.</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

Research on key parameters and correlation model of IF steel based on multimodal information fusion: From composition and process to properties

  • 期刊论文

  • 发表日期：

    2026年7月10日

查看全文
  • Jiawen Tian,
  • Yuqiao Wang,
  • Haotian Wu,
  • Zhiyang Li

摘要

IF steel has a complex production process and unclear microstructural evolution, making it hard to optimize processes and control properties accurately. Current studies either overlook microstructure - related factors or fail to combine unstructured data (such as metallographic images with structured data, resulting in insufficient prediction accuracy. To solve these problems, this study proposes a multimoda full - process property prediction model (WPPPM) based on ResNet50-MLP. lt integrates composition, process parameters, and microstructural features (enhanced metallographic images, precipitation behavior, and characteristic texture strength). The novelty of this model is reflected in three aspects: First, using CycleGAN to enhance metallographic images for better extraction of microstructure distribution features. Second, combining PSO and BO - XGBoost to quantify precipitation behavior and the composition - process - texturerelationship, addressing key issues in parameter determination and quantitative modeling. Third, fusing structured and unstructured metallographic data to directly connect composition-process, microstructure, and properties. Comparative experiments show that WPPPM is moreaccurate than traditional models (which only use composition and process parameters) in predicting yield strength, tensile strength, elongation, and r-value. This work offers a new way for precise property control and process optimization of lF steel.