<p>This study is devoted to the development of models for predicting the microhardness of bulk composite materials by machine learning (ML) techniques. The microhardness prediction is based on standard mechanical tests, specifically employing the Vickers indentation method. To quantitatively assess the influence of material composition on the microhardness of the composites, we have developed a&#xa0;novel methodology. This approach involves constructing a&#xa0;surrogate model based on an ML Random Forest Method (RFM) and an analytical model for calculating the material hardness. The RFM simulates the probability distribution of the indenter imprint diagonal after microhardness tests which are the input data for the analytical model to compute the material hardness. The results of application of the RFM showed significantly greater accuracy (MSE is 7.42·10<sup>−4</sup>%) on the test data. Our findings underscore the synergistic potential of combination of experimental and computational simulation techniques, including machine learning, to predict the mechanical properties of the materials.</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 driven models for microhardness estimation of composite materials

  • A. E. Rezvanova,
  • M. I. Kochergin,
  • N. A. Luginin,
  • V. V. Chebodaeva

摘要

This study is devoted to the development of models for predicting the microhardness of bulk composite materials by machine learning (ML) techniques. The microhardness prediction is based on standard mechanical tests, specifically employing the Vickers indentation method. To quantitatively assess the influence of material composition on the microhardness of the composites, we have developed a novel methodology. This approach involves constructing a surrogate model based on an ML Random Forest Method (RFM) and an analytical model for calculating the material hardness. The RFM simulates the probability distribution of the indenter imprint diagonal after microhardness tests which are the input data for the analytical model to compute the material hardness. The results of application of the RFM showed significantly greater accuracy (MSE is 7.42·10−4%) on the test data. Our findings underscore the synergistic potential of combination of experimental and computational simulation techniques, including machine learning, to predict the mechanical properties of the materials.