Scholarlink

<p>The melt rate is a key factor influencing the quality of ingots produced in electroslag remelting (ESR) process. However, existing research on the mechanism of ESR melting rates suffers from limitations in terms of model complexity and computational efficiency. To overcome this challenge, this study integrates principles of thermodynamic equilibrium and slag bath dynamics to develop a physical model for predicting electrode melting rates. Based on this model, a comprehensive database incorporating multiple influencing factors was established. Leveraging this database, four distinct neural network models were proposed, and the Bayesian optimization algorithm was employed to determine the optimal architecture for each model. Through a systematic comparison of the predictive performance of the different models on both the database and a generalization dataset, coupled with a comprehensive analysis of prediction errors and correlation coefficients, the results demonstrate that the long short-term memory physics-informed neural networks (LSTM-PINNs) model exhibits superior predictive accuracy and generalization capability. Consequently, the LSTM-PINNs model proves capable of effectively predicting the electrode melting rate during the ESR process, thereby providing a reliable theoretical foundation for real-time optimization, control, and decision-making&#xa0;of this process.</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

Prediction of Melt Rate in Electroslag Remelting Process Based on LSTM-PINNs

  • 期刊论文

  • 发表日期：

    2025年10月29日

查看全文
  • Xin Hu,
  • Xuechi Huang,
  • Zhongqiu Liu,
  • ZiJing Zhang,
  • Baokuan Li,
  • Fang Wang

摘要

The melt rate is a key factor influencing the quality of ingots produced in electroslag remelting (ESR) process. However, existing research on the mechanism of ESR melting rates suffers from limitations in terms of model complexity and computational efficiency. To overcome this challenge, this study integrates principles of thermodynamic equilibrium and slag bath dynamics to develop a physical model for predicting electrode melting rates. Based on this model, a comprehensive database incorporating multiple influencing factors was established. Leveraging this database, four distinct neural network models were proposed, and the Bayesian optimization algorithm was employed to determine the optimal architecture for each model. Through a systematic comparison of the predictive performance of the different models on both the database and a generalization dataset, coupled with a comprehensive analysis of prediction errors and correlation coefficients, the results demonstrate that the long short-term memory physics-informed neural networks (LSTM-PINNs) model exhibits superior predictive accuracy and generalization capability. Consequently, the LSTM-PINNs model proves capable of effectively predicting the electrode melting rate during the ESR process, thereby providing a reliable theoretical foundation for real-time optimization, control, and decision-making of this process.