Scholarlink

<p>Static Pressure Support-Single Point Incremental Forming (SPS-SPIF) technology, as an advanced plastic forming method for complex-shaped shell parts, requires precise modeling of the multiparameter-coupled nonlinear mapping relationships between process parameters and forming accuracy to enhance product precision. This study proposes a synergistic modeling framework integrating Conditional Tabular Generative Adversarial Network (CTGAN), Deep Belief Network (DBN), and Non-dominated Sorting Genetic Algorithm-II (NSGA-II) multi-objective optimization. First, CTGAN reconstructs multidimensional probability distributions from original experimental data obtained through response surface methodology, generating high-fidelity augmented datasets to address the limited modeling accuracy inherent in small-sample scenarios. Subsequently, DBN establishes a nonlinear predictive model correlating hydrostatic pressure, sheet thickness, layer spacing, and tool-head rotational speed with axial and radial dimensional accuracy. Finally, NSGA-II algorithm achieves multi-objective parameter optimization. Experimental results demonstrate that the CTGAN-DBN-NSGA-II hybrid model achieves superior predictive performance with goodness-of-fit (<i>R</i><sup>2</sup> = 0.979), mean square error (<i>MSE</i> = 0.0032), and root mean square error (<i>RMSE</i> = 0.056), providing scientific guidance for precision control and dynamic process optimization in SPS-SPIF applications.</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 forming accuracy and process parameter optimization for SPS-SPIF

  • 期刊论文

  • 发表日期：

    2025年9月29日

查看全文
  • ZhangShuai Jing,
  • Jianming Zheng,
  • Mingshun Yang,
  • Haoze Zhang,
  • Hongyan Liu,
  • Xi Chen

摘要

Static Pressure Support-Single Point Incremental Forming (SPS-SPIF) technology, as an advanced plastic forming method for complex-shaped shell parts, requires precise modeling of the multiparameter-coupled nonlinear mapping relationships between process parameters and forming accuracy to enhance product precision. This study proposes a synergistic modeling framework integrating Conditional Tabular Generative Adversarial Network (CTGAN), Deep Belief Network (DBN), and Non-dominated Sorting Genetic Algorithm-II (NSGA-II) multi-objective optimization. First, CTGAN reconstructs multidimensional probability distributions from original experimental data obtained through response surface methodology, generating high-fidelity augmented datasets to address the limited modeling accuracy inherent in small-sample scenarios. Subsequently, DBN establishes a nonlinear predictive model correlating hydrostatic pressure, sheet thickness, layer spacing, and tool-head rotational speed with axial and radial dimensional accuracy. Finally, NSGA-II algorithm achieves multi-objective parameter optimization. Experimental results demonstrate that the CTGAN-DBN-NSGA-II hybrid model achieves superior predictive performance with goodness-of-fit (R2 = 0.979), mean square error (MSE = 0.0032), and root mean square error (RMSE = 0.056), providing scientific guidance for precision control and dynamic process optimization in SPS-SPIF applications.