Scholarlink

<p>In the dynamic forming process of profile during roll bending, the downward pressure parameters at different times exert a nonlinear coupled effect on the final curvature radius, making it difficult to predict the ultimate curvature radius accurately. This has become a challenging issue in the field of industrial precision forming. To address this problem, a CNN-TCN-TPA neural network model is proposed to model the complex coupled relationships during the dynamic roll bending forming process. Firstly, a multi-scale CNN is employed to extract the implicit features of roll bending at different time scales, enabling the model to understand the inherent patterns of roll bending data comprehensively. Subsequently, TCN is utilized to learn the influence relationships before and after roll bending forming. Finally, a temporal attention mechanism is adopted to learn the impact of different historical moments on the final outcome, thereby establishing the CNN-TCN-TPA roll bending forming curvature radius prediction model and achieving accurate prediction of the roll bending forming curvature radius. The prediction performance of the CNN-TCN-TPA model is compared with traditional neural network models, TCN models, and TCN-TPA models. The results indicate that the CNN-TCN-TPA model exhibits higher prediction performance compared to other neural network models, with mean square error and mean absolute error of 5971.65 and 24.42, respectively.</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

Roll bending forming curvature radius prediction based on the CNN-TCN-TPA neural network model

  • 期刊论文

  • 发表日期：

    2025年4月7日

查看全文
  • Guoyan Huang,
  • Yafeng Zhang,
  • Tong Wu,
  • Peng Shi,
  • Menghang Wan

摘要

In the dynamic forming process of profile during roll bending, the downward pressure parameters at different times exert a nonlinear coupled effect on the final curvature radius, making it difficult to predict the ultimate curvature radius accurately. This has become a challenging issue in the field of industrial precision forming. To address this problem, a CNN-TCN-TPA neural network model is proposed to model the complex coupled relationships during the dynamic roll bending forming process. Firstly, a multi-scale CNN is employed to extract the implicit features of roll bending at different time scales, enabling the model to understand the inherent patterns of roll bending data comprehensively. Subsequently, TCN is utilized to learn the influence relationships before and after roll bending forming. Finally, a temporal attention mechanism is adopted to learn the impact of different historical moments on the final outcome, thereby establishing the CNN-TCN-TPA roll bending forming curvature radius prediction model and achieving accurate prediction of the roll bending forming curvature radius. The prediction performance of the CNN-TCN-TPA model is compared with traditional neural network models, TCN models, and TCN-TPA models. The results indicate that the CNN-TCN-TPA model exhibits higher prediction performance compared to other neural network models, with mean square error and mean absolute error of 5971.65 and 24.42, respectively.