Scholarlink

<p>The compression molding of fiber reinforced plastics (FRP) is crucial method for producing lightweight automotive components. However, maintaining dimensional stability is challenging due to the complex interactions between material behavior, process parameters and fiber orientation effects. Predicting post process deformations (warpage) can enhance processing parameters and control the quality of molded parts. Traditional finite element models, such as Moldex3D software are costly and can sometimes lack accuracy due to variability in the production process. This paper introduces an AI-assisted framework that employs software simulations to train a Physics-Informed Neural Network (PINN) to predict warpage in Polyamide 6 (PA6) components reinforced with 45% short glass fibers. The input variables include key process parameters, while the output is the predicted 3D warpage field. The model integrates physical loss terms to ensure consistency with elasticity and thermal strain equations. Compared with Moldex3D predictions, the PINN increased by an average of 23% the proportion of nodes whose warpage predictions fell within the industrial tolerance of 1 mm when validated against 3D-scanned manufactured parts. The aim is to improve the prediction of warpage by combining physics-based machine learning, numerical simulations and experimental measurements obtained from 3D-scanned manufactured components.</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

Development of ANN to predict warpage in FRP components manufactured by compression molding

  • 期刊论文

  • 发表日期：

    2026年6月17日

查看全文
  • Manuel Coca-Gonzalez,
  • Alvaro Frutos,
  • Haile Atsbha,
  • Mariel Alfaro-Ponce,
  • Moises Jimenez-Martinez

摘要

The compression molding of fiber reinforced plastics (FRP) is crucial method for producing lightweight automotive components. However, maintaining dimensional stability is challenging due to the complex interactions between material behavior, process parameters and fiber orientation effects. Predicting post process deformations (warpage) can enhance processing parameters and control the quality of molded parts. Traditional finite element models, such as Moldex3D software are costly and can sometimes lack accuracy due to variability in the production process. This paper introduces an AI-assisted framework that employs software simulations to train a Physics-Informed Neural Network (PINN) to predict warpage in Polyamide 6 (PA6) components reinforced with 45% short glass fibers. The input variables include key process parameters, while the output is the predicted 3D warpage field. The model integrates physical loss terms to ensure consistency with elasticity and thermal strain equations. Compared with Moldex3D predictions, the PINN increased by an average of 23% the proportion of nodes whose warpage predictions fell within the industrial tolerance of 1 mm when validated against 3D-scanned manufactured parts. The aim is to improve the prediction of warpage by combining physics-based machine learning, numerical simulations and experimental measurements obtained from 3D-scanned manufactured components.