Scholarlink

<p>This research conducted the hot compression test at the temperature range of 700-850&#xa0;°C and strain rate range of 1-0.001&#xa0;s<sup>−1</sup> to study elevated temperature deformation behavior. Traditional Arrhenius constitutive model, RNN and LSTM models were developed to analyze the flow behavior. PredRNN-V2, MotionRNN and MotionRNN-SSIM models were constructed to predict the hot processing map. The maximum flow stress decreased with the increase in temperature and decrease in strain rate. Metric <i>R</i><sup>2</sup> of tradition Arrhenius is 0.9575 and the data points, suggesting the moderate prediction accuracy. The <i>R</i><sup>2</sup> of LSTM and RNN were 0.99918 and 0.99784, respectively, the data points located between the 5% deviation line, and no overfitting was observed, suggesting the excellent prediction and generalization performance. Metrics of three models were greater than 0.981, suggesting the excellent prediction accuracy of models. Compared with the MotionRNN model, the introduction of SSIM loss function address the issue of image blurring and improve the prediction accuracy of both maps. This work presented an effective method for intelligent evaluated temperature forming for titanium alloys.</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

Research on the Deformation Behavior at Elevated Temperature of Ti-6Al-4 V Titanium Based on the Recurrent Neural Network

  • 期刊论文

  • 发表日期：

    2025年11月11日

查看全文
  • Taowen Wu,
  • Minghe Chen,
  • Hongrui Dong,
  • Xiangwei Wen,
  • Qianlong Sui

摘要

This research conducted the hot compression test at the temperature range of 700-850 °C and strain rate range of 1-0.001 s−1 to study elevated temperature deformation behavior. Traditional Arrhenius constitutive model, RNN and LSTM models were developed to analyze the flow behavior. PredRNN-V2, MotionRNN and MotionRNN-SSIM models were constructed to predict the hot processing map. The maximum flow stress decreased with the increase in temperature and decrease in strain rate. Metric R2 of tradition Arrhenius is 0.9575 and the data points, suggesting the moderate prediction accuracy. The R2 of LSTM and RNN were 0.99918 and 0.99784, respectively, the data points located between the 5% deviation line, and no overfitting was observed, suggesting the excellent prediction and generalization performance. Metrics of three models were greater than 0.981, suggesting the excellent prediction accuracy of models. Compared with the MotionRNN model, the introduction of SSIM loss function address the issue of image blurring and improve the prediction accuracy of both maps. This work presented an effective method for intelligent evaluated temperature forming for titanium alloys.