Scholarlink

<p>This paper presents the results of a study on the effectiveness of machine learning methods for predicting the mechanical properties of steels based on data from 40 key chemical, structural, and deformation parameters of their microstructure. These parameters have a determining influence on the formation of its mechanical characteristics and can be obtained directly from the surface of structures using non-destructive testing techniques and special mobile equipment. To establish a quantitative relationship between various combinations of the steel’s microstructural parameters and its mechanical properties, a neural network based on a multi-layer perceptron was developed. The neural network was trained on 355 experimental datasets obtained from carbon and low-alloy steels of various strength classes, subjected to different types of thermo-mechanical processing. Each dataset included values for the 40 structural parameters and 6 mechanical properties. The neural network testing results demonstrated the feasibility of predicting mechanical properties with high accuracy. The mean absolute percentage error (MAPE) for property prediction was as follows: ultimate tensile strength—1.27%; yield strength—2.54%; elongation—4.17%; reduction of area—2.84%; impact toughness—11.03%; fracture toughness—6.74%.</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

Using Machine Learning for Non-destructive Evaluation of Mechanical Properties of Steel Structures

  • 期刊论文

  • 发表日期：

    2025年11月15日

查看全文
  • Alexander Zorin,
  • Konstantin Rochev,
  • Polina Kozhevnikova,
  • Vitaly Kuntsev,
  • Anatoly Pavlov

摘要

This paper presents the results of a study on the effectiveness of machine learning methods for predicting the mechanical properties of steels based on data from 40 key chemical, structural, and deformation parameters of their microstructure. These parameters have a determining influence on the formation of its mechanical characteristics and can be obtained directly from the surface of structures using non-destructive testing techniques and special mobile equipment. To establish a quantitative relationship between various combinations of the steel’s microstructural parameters and its mechanical properties, a neural network based on a multi-layer perceptron was developed. The neural network was trained on 355 experimental datasets obtained from carbon and low-alloy steels of various strength classes, subjected to different types of thermo-mechanical processing. Each dataset included values for the 40 structural parameters and 6 mechanical properties. The neural network testing results demonstrated the feasibility of predicting mechanical properties with high accuracy. The mean absolute percentage error (MAPE) for property prediction was as follows: ultimate tensile strength—1.27%; yield strength—2.54%; elongation—4.17%; reduction of area—2.84%; impact toughness—11.03%; fracture toughness—6.74%.