Scholarlink

<p>This study introduces a predictive surrogate model for real-time control in cold upsetting processes, incorporating both material and machine behaviors. Traditional approaches often simplify machine behavior as rigid or with constant stiffness; however, the proposed method dynamically couples material and machine responses, accounting for efficiency changes across different upsetting operations. This is achieved through the integration of a data-driven billet-related surrogate model with a machine-related analytical blow efficiency prediction, accurately capturing elastic energy losses. For the construction of the surrogate model in this use case, a multilayer perceptron artificial neural network (MLP ANN) was employed, demonstrating high predictive accuracy with a dataset comprising 2000 entries generated using Latin Hypercube Sampling (LHS) and numerical simulations. The model provides precise predictions for key outputs like forging load and plastic energy. Experimental validation shows prediction errors below 5% for energy setpoints, reduced to under 1% with blow efficiency correction. The general methodology of surrogate model creation can be adapted for various metal-forming processes, providing a versatile framework for real-time simulation and control.</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

Real-time forging process control: integrating billet-related surrogate and machine behavior models

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年4月9日

查看全文
  • David Uribe,
  • Camille Durand,
  • Cyrille Baudouin,
  • Régis Bigot

摘要

This study introduces a predictive surrogate model for real-time control in cold upsetting processes, incorporating both material and machine behaviors. Traditional approaches often simplify machine behavior as rigid or with constant stiffness; however, the proposed method dynamically couples material and machine responses, accounting for efficiency changes across different upsetting operations. This is achieved through the integration of a data-driven billet-related surrogate model with a machine-related analytical blow efficiency prediction, accurately capturing elastic energy losses. For the construction of the surrogate model in this use case, a multilayer perceptron artificial neural network (MLP ANN) was employed, demonstrating high predictive accuracy with a dataset comprising 2000 entries generated using Latin Hypercube Sampling (LHS) and numerical simulations. The model provides precise predictions for key outputs like forging load and plastic energy. Experimental validation shows prediction errors below 5% for energy setpoints, reduced to under 1% with blow efficiency correction. The general methodology of surrogate model creation can be adapted for various metal-forming processes, providing a versatile framework for real-time simulation and control.