Scholarlink

<p>Accurate and stable milling force prediction is the basis for the process parameter optimization in computer numerical control (CNC) milling machining, which has a significant influence on extending tool life, improving machining accuracy, and machining efficiency. Traditional milling force theoretical models have difficulties in predicting the complex changes of milling force during machining due to their oversimplification of the machining conditions. While data-driven models can achieve high prediction accuracy, this is based on obtaining a large and comprehensive experimental dataset. This paper proposes a modeling method for three-axis milling force transfer learning based on theoretical model priors that achieves excellent prediction accuracy. A mechanistic force model with run-out is established, and extensive simulation samples are generated using this model with geometric simulation. Then an integrated transfer learning model is trained using the TrAdaBoost algorithm with MF-CNN as the base learner, employing a mixed dataset composed of a large number of simulation samples and a small amount of experimental data. Finally, the effectiveness and advantages of the proposed method are validated through physical experiments. The results demonstrate that the proposed model accurately predicts milling forces, outperforming mechanistic force models, MF-CNN, and Fine-tuned models.</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

A transfer learning modeling approach for three-axis instantaneous milling forces based on theoretical model priors

  • 期刊论文

  • 发表日期：

    2026年3月3日

查看全文
  • Kai Liu,
  • Jianzhong Yang,
  • Ronghua Wang,
  • Shihao Li,
  • Jiejun Xie

摘要

Accurate and stable milling force prediction is the basis for the process parameter optimization in computer numerical control (CNC) milling machining, which has a significant influence on extending tool life, improving machining accuracy, and machining efficiency. Traditional milling force theoretical models have difficulties in predicting the complex changes of milling force during machining due to their oversimplification of the machining conditions. While data-driven models can achieve high prediction accuracy, this is based on obtaining a large and comprehensive experimental dataset. This paper proposes a modeling method for three-axis milling force transfer learning based on theoretical model priors that achieves excellent prediction accuracy. A mechanistic force model with run-out is established, and extensive simulation samples are generated using this model with geometric simulation. Then an integrated transfer learning model is trained using the TrAdaBoost algorithm with MF-CNN as the base learner, employing a mixed dataset composed of a large number of simulation samples and a small amount of experimental data. Finally, the effectiveness and advantages of the proposed method are validated through physical experiments. The results demonstrate that the proposed model accurately predicts milling forces, outperforming mechanistic force models, MF-CNN, and Fine-tuned models.