Scholarlink

<p>This study investigates the application of machine learning techniques to predict the deformation behavior of the AA 5052-H32 aluminum alloy over a wide range of processing conditions. Tensile tests were conducted at different temperatures (100–450&#xa0;°C) and strain rates (0.01 and 1&#xa0;s⁻<sup>1</sup>), enabling the acquisition of the material’s flow curves. The experimental data were fitted to the Hensel–Spittel constitutive equation and subsequently employed in the development of Artificial Neural Network (ANN) and eXtreme Gradient Boosting (XGBoost) models. The AI-based models exhibited superior predictive performance compared to the phenomenological approach. The optimized ANN, with a dense architecture consisting of two hidden layers with 256 and 128 neurons, achieved a mean absolute error (MAE) of 3.61&#xa0;MPa and a mean squared error (MSE) of 19.09 MPa<sup>2</sup>. The XGBoost model, configured with 100 decision trees and a maximum depth of 4, delivered even more accurate results, with MAE of 0.74&#xa0;MPa, MSE of 2.46 MPa<sup>2</sup>, and a coefficient of determination (R<sup>2</sup>) of 0.9988, showing an almost perfect overlap between predicted and experimental curves. These findings confirm the high accuracy and robustness of machine learning techniques, highlighting their potential as a superior alternative to traditional phenomenological models for predicting flow behavior in forming processes.</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

Artificial intelligence-based modeling of hot deformation behavior in AA 5052-H32 alloy

  • 期刊论文

  • 发表日期：

    2025年10月7日

查看全文
  • Rafael Pandolfo da Rocha,
  • Matheus Henrique Riffel,
  • André Rosiak,
  • Luis Fernando Folle,
  • Tiago Nunes Lima,
  • Lirio Schaeffer

摘要

This study investigates the application of machine learning techniques to predict the deformation behavior of the AA 5052-H32 aluminum alloy over a wide range of processing conditions. Tensile tests were conducted at different temperatures (100–450 °C) and strain rates (0.01 and 1 s⁻1), enabling the acquisition of the material’s flow curves. The experimental data were fitted to the Hensel–Spittel constitutive equation and subsequently employed in the development of Artificial Neural Network (ANN) and eXtreme Gradient Boosting (XGBoost) models. The AI-based models exhibited superior predictive performance compared to the phenomenological approach. The optimized ANN, with a dense architecture consisting of two hidden layers with 256 and 128 neurons, achieved a mean absolute error (MAE) of 3.61 MPa and a mean squared error (MSE) of 19.09 MPa2. The XGBoost model, configured with 100 decision trees and a maximum depth of 4, delivered even more accurate results, with MAE of 0.74 MPa, MSE of 2.46 MPa2, and a coefficient of determination (R2) of 0.9988, showing an almost perfect overlap between predicted and experimental curves. These findings confirm the high accuracy and robustness of machine learning techniques, highlighting their potential as a superior alternative to traditional phenomenological models for predicting flow behavior in forming processes.