Scholarlink

<p>Within the scope of linear-elastic fracture mechanics (LEFM), several dimensionless parameters, known as geometry factors, exist. Calculating geometry factors is a time-consuming and computationally resource-intensive process. This study evaluates the predictive abilities of various machine learning models for dimensionless fracture parameters (<i>Y</i><sub><i>I</i></sub>, <i>Y</i><sub><i>II</i></sub>, <i>T</i>*). In this regard, three fracture specimens were considered, including single-edge notched bend (SENB), semi-circular bend (SCB), and edge-notched disc bend (ENDB). The evaluations were conducted under a full range of Mode I/II brittle fracture conditions. As the first step, the datasets were statistically described, including assessment of the distributions, correlation of parameters, and examination of the dataset’s sensitivity to input variables. Using a diverse dataset, the study tests linear, Lasso, Ridge, partial least squares, decision tree, artificial neural networks, random forest, gradient boosting machine (GBM), boosted tree, support vector regression, and Gaussian process regression models. Different train/test ratios were assumed to evaluate the dependency of models on the train ratio.</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

Machine learning based predicting the dimensionless fracture parameters from different bend-type specimens

  • 期刊论文

  • 发表日期：

    2025年12月6日

查看全文
  • Xinmin Hong,
  • Haytham F. Isleem

摘要

Within the scope of linear-elastic fracture mechanics (LEFM), several dimensionless parameters, known as geometry factors, exist. Calculating geometry factors is a time-consuming and computationally resource-intensive process. This study evaluates the predictive abilities of various machine learning models for dimensionless fracture parameters (YI, YII, T*). In this regard, three fracture specimens were considered, including single-edge notched bend (SENB), semi-circular bend (SCB), and edge-notched disc bend (ENDB). The evaluations were conducted under a full range of Mode I/II brittle fracture conditions. As the first step, the datasets were statistically described, including assessment of the distributions, correlation of parameters, and examination of the dataset’s sensitivity to input variables. Using a diverse dataset, the study tests linear, Lasso, Ridge, partial least squares, decision tree, artificial neural networks, random forest, gradient boosting machine (GBM), boosted tree, support vector regression, and Gaussian process regression models. Different train/test ratios were assumed to evaluate the dependency of models on the train ratio.