中国塑料 ›› 2025, Vol. 39 ›› Issue (7): 72-79.DOI: 10.19491/j.issn.1001-9278.2025.07.012

• 加工与应用 • 上一篇    下一篇

基于机器学习方法预测3D打印零件的性能

洪学银1(), 高尚2   

  1. 1.扬州职业技术大学信息工程学院,江苏 扬州 225009
    2.江苏科技大学计算机学院,江苏 镇江 212003
  • 收稿日期:2024-07-28 出版日期:2025-07-26 发布日期:2025-07-22
  • 作者简介:洪学银(1976-),教授,研究方向为人工智能,382593054@qq.com

Performance prediction of 3D printed parts based on machine learning methods

HONG Xueyin1(), GAO Shang2   

  1. 1.School of Information Engineering,Yangzhou Polytechnic University,Yangzhou 225009,China
    2.College of Computer Science,Jiangsu University of Science and Technology,Zhenjiang 212003,China
  • Received:2024-07-28 Online:2025-07-26 Published:2025-07-22

摘要:

采用拉丁超立方实验设计,研究了层高、壁厚、顶底厚、顶底线条方向、填充密度、填充线条方向、打印速度、挤出温度、床温、工作空间温度10种熔融沉积建模(FDM)工艺参数对丙烯腈⁃丁二烯⁃苯乙烯共聚物(ABS)零件拉伸性能的影响,对比了人工神经元网络(ANN)、随机森林(RF)和梯度提升算法(GB)3种机器学习方法预测拉伸性能的准确性。结果表明,ANN预测拉伸强度和断裂伸长率的相关系数R仅为0.883 5和0.892 4,在训练和测试数据集上,预测的均方误差(MSE)在5~10和20~24之间;RF预测的R值为0.913 6和0.924 0,MSE在3~8和15~20之间;GB预测准确性最高,R值为0.975 9和0.981 2,MSE最低,在1~4和8~10之间。在10种工艺参数中,在采用RF模型时,拉伸性能的显著影响因素为填充密度、壁厚、填充线条方向和顶底厚,在采用GB模型时,拉伸性能的显著影响因素为填充密度、壁厚、层高和填充线条方向。填充密度是影响拉伸性能最显著的因素,对GB预测结果的影响显著性达到80 %左右,远大于RF模型中的40 %。

关键词: 熔融沉积建模, 人工神经元网络, 随机森林, Gradient Boosting, 拉伸性能

Abstract:

This study systematically investigated the influence of ten fused deposition modeling process parameters on the tensile properties of ABS parts using a Latin hypercube experimental design. The examined parameters include layer height, wall thickness, top/bottom thickness, top/bottom line direction, infill density, infill line direction, printing speed, extrusion temperature, bed temperature, and workspace temperature. Three machine learning approaches, artificial neural network (ANN), random forest (RF), and gradient boosting (GB), were compared for their predictive accuracy of tensile strength and elongation at break. Results demonstrated that GB outperformed other methods, achieving superior correlation coefficients (R=0.975 9~0.981 2) and the lowest mean square errors (MSE=1~10). RF showed intermediate performance (R=0.913 6~0.924 0, MSE=3~20), while ANN yielded the lowest accuracy (R=0.883 5~0.892 4, MSE=5~24). Feature importance analysis revealed infill density as the most influential parameter, contributing approximately 80 % to GB predictions compared to 40 % in RF. Other significant factors included wall thickness, infill line direction, and layer height. These findings provide valuable insights for optimizing FDM process parameters and selecting appropriate machine learning techniques for mechanical property prediction.

Key words: fused deposition modeling, artificial neural network, random forest, gradient boosting, tensile performance

中图分类号: