Scholarlink

<p>Aiming at the problem of multivariable strong coupling and complex relationship structure in the prediction of mechanical properties of pipeline steel, the existing methods usually only consider the pairwise interaction between chemical compositions and process parameters, so it is difficult to effectively characterize the multivariable strong coupling relationship between them, and ignoring the potential correlation between input features and different mechanical properties. Therefore, this paper proposed a prediction model of mechanical properties of pipeline steel based on hypergraph convolution network. The model considers the correlation between features and the correlation between features and target property by introducing two types of hypergraphs. One type of hypergraph is used to learn the multivariate coupling relationship between chemical compositions and process parameters to capture the deep coupling relationship of input features. The other type of hypergraph further introduces mechanical property nodes to extract the correlation between input features and target property in the training phase, thereby providing structural guidance for the prediction process. Aiming at the problem of insufficient local sensitivity and limited global dependence in the feature aggregation process of the dual-hypergraph structure, the model introduces a stochastic attention mechanism (SAM) to construct attention weights at both the node and hyperedge levels, enabling adaptive weighting of feature aggregation across different hierarchical levels. At the same time, the hypergraph attention mechanism (HAM) is introduced to strengthen the global coupling relationship between different features and enhance the overall perception ability of the model. The experimental results show that the prediction accuracy of the model for yield strength, tensile strength and elongation of pipeline steel reaches 99.2%, 98.49%, and 98.08%, respectively, which fully proves that it can effectively capture the complex nonlinear interaction between composition and process and the superiority of overall performance prediction.</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

A dual-attention hypergraph convolutional network for predicting mechanical properties of pipeline steel

  • 期刊论文

  • 发表日期：

    2026年2月19日

查看全文
  • Qiwen Zhang,
  • Minghui Gao

摘要

Aiming at the problem of multivariable strong coupling and complex relationship structure in the prediction of mechanical properties of pipeline steel, the existing methods usually only consider the pairwise interaction between chemical compositions and process parameters, so it is difficult to effectively characterize the multivariable strong coupling relationship between them, and ignoring the potential correlation between input features and different mechanical properties. Therefore, this paper proposed a prediction model of mechanical properties of pipeline steel based on hypergraph convolution network. The model considers the correlation between features and the correlation between features and target property by introducing two types of hypergraphs. One type of hypergraph is used to learn the multivariate coupling relationship between chemical compositions and process parameters to capture the deep coupling relationship of input features. The other type of hypergraph further introduces mechanical property nodes to extract the correlation between input features and target property in the training phase, thereby providing structural guidance for the prediction process. Aiming at the problem of insufficient local sensitivity and limited global dependence in the feature aggregation process of the dual-hypergraph structure, the model introduces a stochastic attention mechanism (SAM) to construct attention weights at both the node and hyperedge levels, enabling adaptive weighting of feature aggregation across different hierarchical levels. At the same time, the hypergraph attention mechanism (HAM) is introduced to strengthen the global coupling relationship between different features and enhance the overall perception ability of the model. The experimental results show that the prediction accuracy of the model for yield strength, tensile strength and elongation of pipeline steel reaches 99.2%, 98.49%, and 98.08%, respectively, which fully proves that it can effectively capture the complex nonlinear interaction between composition and process and the superiority of overall performance prediction.