Scholarlink

<p>In this paper, a general deep learning fusion model was proposed, which integrated intelligent optimization algorithm with deep learning technology to predict the deformation behavior of materials under complex loading and a wide range of temperatures with robustness and universal capability. The Particle Swarm Optimization (PSO) algorithm was selected to improve the accuracy and generalization of the model significantly by optimizing the initial deep neural network (DNN) weights and biases. Taking DH460 steel as an example, the PSO–DNN model was compared in detail with traditional phenomenological models in terms of prediction. The results indicated that the PSO–DNN model achieves the highest accuracy, with the average AARE was 1.5 pct. The capability of the PSO–DNN model to make predictions on out-of-domain applications was then analyzed. In interpolation and extrapolation validation, the average AARE were 2.1 and 4.7 pct, respectively, which demonstrated its wonderful generalization capability. Eventually, the PSO–DNN model was applied to the bearing steel GCr15 and ship plate steel to verify its scalability. The results showed that the PSO–DNN model was far superior to other models, and it can effectively learn the deformation mechanism of materials. This paper provided a robust and well-scalable alternative method for traditional phenomenological constitutive models and offered potential for the design and optimization of the continuous casting 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

Constitutive Modeling of the Continuous Casting Steels Based on PSO–DNN Fusion Model

  • 期刊论文

  • 发表日期：

    2025年4月7日

查看全文
  • Bochun Liang,
  • Cheng Ji,
  • Miaoyong Zhu

摘要

In this paper, a general deep learning fusion model was proposed, which integrated intelligent optimization algorithm with deep learning technology to predict the deformation behavior of materials under complex loading and a wide range of temperatures with robustness and universal capability. The Particle Swarm Optimization (PSO) algorithm was selected to improve the accuracy and generalization of the model significantly by optimizing the initial deep neural network (DNN) weights and biases. Taking DH460 steel as an example, the PSO–DNN model was compared in detail with traditional phenomenological models in terms of prediction. The results indicated that the PSO–DNN model achieves the highest accuracy, with the average AARE was 1.5 pct. The capability of the PSO–DNN model to make predictions on out-of-domain applications was then analyzed. In interpolation and extrapolation validation, the average AARE were 2.1 and 4.7 pct, respectively, which demonstrated its wonderful generalization capability. Eventually, the PSO–DNN model was applied to the bearing steel GCr15 and ship plate steel to verify its scalability. The results showed that the PSO–DNN model was far superior to other models, and it can effectively learn the deformation mechanism of materials. This paper provided a robust and well-scalable alternative method for traditional phenomenological constitutive models and offered potential for the design and optimization of the continuous casting process.