Scholarlink

<p>Grinding technology is widely applied in the manufacturing and mechanical processing sectors. Different from conventional three-dimensional rough surface friction models, ground metals exhibit a striated surface morphology, which can be simplified as a two-dimensional plane strain friction issue. Due to surface morphology diversity and loading condition complexity, numerical modeling and experimental approaches have difficulty achieving rapid prediction of line-contact surface friction behavior. Therefore, this study innovatively proposes a hybrid physics-data-driven model integrating finite element analysis (FEA) with machine learning (ML), enabling efficient and accurate prediction of line-contact friction behavior on two-dimensional rough surfaces. An extensive friction behavior database was generated through finite element simulations. Based on this dataset, the random forest (RF) algorithm was used to achieve high-precision prediction of the friction coefficient. Furthermore, a comprehensive analysis was performed on the effects of surface roughness, normal load, yield strength, and local friction coefficient on friction behavior. The RF model exhibits excellent performance in predicting friction coefficients and also accurately identifies the most influential features governing friction behavior. Residual analysis further verifies our model’s reliability, as the RF predictions agree with the FEA results, demonstrating remarkable adaptability and accuracy. Feature importance analysis results reveal that the local friction coefficient and normal load are the main factors influencing friction behavior, but the surface roughness and yield strength exhibit a relatively minor influence. The study innovatively identifies the coupling effects of key parameters through contour maps. Namely, the influence of local friction coefficient decreases with increasing normal load but becomes significantly more pronounced with elevated material yield strength. By integrating ML, our proposed model maintains the high accuracy of FEA while capturing the complexity of interfacial responses through data-driven approaches. Our study advances traditional tribological research from “experience-driven” to “data-intelligence-driven,” thus providing novel insights for understanding and predicting complex friction behaviors, as well as for optimizing frictional design in engineering applications.</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

Friction Behaviors of Rough Line Contact Using Machine Learning-Assisted Finite Element Analysis

  • 期刊论文

  • 发表日期：

    2025年5月28日

查看全文
  • Shun Zhu,
  • Yanwei Liu,
  • Biao Zhao,
  • Siyuan Zhang

摘要

Grinding technology is widely applied in the manufacturing and mechanical processing sectors. Different from conventional three-dimensional rough surface friction models, ground metals exhibit a striated surface morphology, which can be simplified as a two-dimensional plane strain friction issue. Due to surface morphology diversity and loading condition complexity, numerical modeling and experimental approaches have difficulty achieving rapid prediction of line-contact surface friction behavior. Therefore, this study innovatively proposes a hybrid physics-data-driven model integrating finite element analysis (FEA) with machine learning (ML), enabling efficient and accurate prediction of line-contact friction behavior on two-dimensional rough surfaces. An extensive friction behavior database was generated through finite element simulations. Based on this dataset, the random forest (RF) algorithm was used to achieve high-precision prediction of the friction coefficient. Furthermore, a comprehensive analysis was performed on the effects of surface roughness, normal load, yield strength, and local friction coefficient on friction behavior. The RF model exhibits excellent performance in predicting friction coefficients and also accurately identifies the most influential features governing friction behavior. Residual analysis further verifies our model’s reliability, as the RF predictions agree with the FEA results, demonstrating remarkable adaptability and accuracy. Feature importance analysis results reveal that the local friction coefficient and normal load are the main factors influencing friction behavior, but the surface roughness and yield strength exhibit a relatively minor influence. The study innovatively identifies the coupling effects of key parameters through contour maps. Namely, the influence of local friction coefficient decreases with increasing normal load but becomes significantly more pronounced with elevated material yield strength. By integrating ML, our proposed model maintains the high accuracy of FEA while capturing the complexity of interfacial responses through data-driven approaches. Our study advances traditional tribological research from “experience-driven” to “data-intelligence-driven,” thus providing novel insights for understanding and predicting complex friction behaviors, as well as for optimizing frictional design in engineering applications.