Scholarlink

<p>In this study, an Artificial Neural Network (ANN) approach is employed to predict natural frequencies in multi-wall carbon nanotube (MWCNT) reinforced GFRP hybrid composite plates with double delaminations considering without overlapping two delaminations. This methodology is implemented to overcome the computational challenges associated with modeling complex double delamination scenarios implemented in developed finite formulation using higher order shear deformation theory. It was understood from the study that influence of double delamination is significantly more than single delaminations in the plates. It was demonstrated that by utilizing four ANN models that machine learning approaches, particularly Random Forest Regression, can effectively predict properties of MWCNT interfaces with high accuracy (R2 of 0.995). The significant performance gap between linear regression (MSE of 17.20) and ensemble methods like RFR (MSE of 0.26, representing a 98.5% reduction in error) highlights the complex, non-linear nature of the relationships in MWCNT interface systems. The findings conclude that the Random Forest Regression (RFR) model offers the most accurate predictions, closely aligning with the results obtained from the Finite Element Model (FEM) developed for composite plates with double delamination. The complete computational effort using FEM involved analyzing 71,280 delamination scenarios, which required approximately 61.875&#xa0;days to complete on an HP Workstation Z8 G4. This effort was undertaken to determine the natural frequencies corresponding to various combinations of double delamination positions, weight fractions, and interface characteristics under CCCC boundary conditions. In contrast, by employing Artificial Neural Network (ANN) prediction techniques, the same predictive coverage can be achieved using only 14,256 FEM cases, completing the task in just 12.375&#xa0;days with significantly reduced error thereby offering a more efficient and reliable alternative to exhaustive FEM simulations.</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

ANN-based prediction of natural frequencies in MWCNT-reinforced GFRP hybrid composite plates with double delaminations

  • 期刊论文

  • 发表日期：

    2025年6月24日

查看全文
  • Dhivya Elumalai,
  • Mohit Gupta,
  • Anuj Kumar Sharma

摘要

In this study, an Artificial Neural Network (ANN) approach is employed to predict natural frequencies in multi-wall carbon nanotube (MWCNT) reinforced GFRP hybrid composite plates with double delaminations considering without overlapping two delaminations. This methodology is implemented to overcome the computational challenges associated with modeling complex double delamination scenarios implemented in developed finite formulation using higher order shear deformation theory. It was understood from the study that influence of double delamination is significantly more than single delaminations in the plates. It was demonstrated that by utilizing four ANN models that machine learning approaches, particularly Random Forest Regression, can effectively predict properties of MWCNT interfaces with high accuracy (R2 of 0.995). The significant performance gap between linear regression (MSE of 17.20) and ensemble methods like RFR (MSE of 0.26, representing a 98.5% reduction in error) highlights the complex, non-linear nature of the relationships in MWCNT interface systems. The findings conclude that the Random Forest Regression (RFR) model offers the most accurate predictions, closely aligning with the results obtained from the Finite Element Model (FEM) developed for composite plates with double delamination. The complete computational effort using FEM involved analyzing 71,280 delamination scenarios, which required approximately 61.875 days to complete on an HP Workstation Z8 G4. This effort was undertaken to determine the natural frequencies corresponding to various combinations of double delamination positions, weight fractions, and interface characteristics under CCCC boundary conditions. In contrast, by employing Artificial Neural Network (ANN) prediction techniques, the same predictive coverage can be achieved using only 14,256 FEM cases, completing the task in just 12.375 days with significantly reduced error thereby offering a more efficient and reliable alternative to exhaustive FEM simulations.