Scholarlink

<p>The present research aims to estimate the necking and fracture limits of various sheet metals using machine learning (ML) models&#xa0;and to&#xa0;predict the formability of different steel and aluminum sheet metals. Experiments of stretch forming (SF) and single-point incremental forming (SPIF) were performed under uniaxial, plane strain, and biaxial strain paths to estimate failure limits. Further, using different ML algorithms, a supervised ML methodology was proposed to predict the forming limit diagram (FLD) and fracture forming limit diagram (FFLD) of various sheet materials. Subsequently, the ML-predicted FLDs and FFLDs were validated with the experimental data. It was observed that the random forest regressor (RFR), decision tree regressor, and extreme gradient boosting ML models showed high accuracy in prediction, with the RFR model outperforming all other ML models. The accuracy in the prediction of FLD and FFLD for the RFR model was 92% and 96%, respectively. Furthermore, the ML-predicted FLDs and FFLDs were incorporated as failure initiation models into the finite element simulations coupled with the&#xa0;advanced anisotropic Yld2000 material model to perform the post-forming analyses during SF and SPIF processes. It was found that the mean absolute percentage error values in the&#xa0;dome height prediction of all the SF and SPIF samples&#xa0;using the&#xa0; best-predicted RFR FLD and RFR FFLD were within 10% error values. Moreover, the surface strains and thickness distributions for the SF and the SPIF samples were efficiently predicted using the RFR model-based FLD and FFLD.</p> Graphical Abstract <p></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

Machine Learning Enabled Estimation of Formability for Anisotropic Sheet Metals

  • 期刊论文

  • 发表日期：

    2025年6月19日

查看全文
  • Abdul Samad,
  • Ankit Kumar Thakur,
  • Shamik Basak,
  • Kaushik Pal

摘要

The present research aims to estimate the necking and fracture limits of various sheet metals using machine learning (ML) models and to predict the formability of different steel and aluminum sheet metals. Experiments of stretch forming (SF) and single-point incremental forming (SPIF) were performed under uniaxial, plane strain, and biaxial strain paths to estimate failure limits. Further, using different ML algorithms, a supervised ML methodology was proposed to predict the forming limit diagram (FLD) and fracture forming limit diagram (FFLD) of various sheet materials. Subsequently, the ML-predicted FLDs and FFLDs were validated with the experimental data. It was observed that the random forest regressor (RFR), decision tree regressor, and extreme gradient boosting ML models showed high accuracy in prediction, with the RFR model outperforming all other ML models. The accuracy in the prediction of FLD and FFLD for the RFR model was 92% and 96%, respectively. Furthermore, the ML-predicted FLDs and FFLDs were incorporated as failure initiation models into the finite element simulations coupled with the advanced anisotropic Yld2000 material model to perform the post-forming analyses during SF and SPIF processes. It was found that the mean absolute percentage error values in the dome height prediction of all the SF and SPIF samples using the  best-predicted RFR FLD and RFR FFLD were within 10% error values. Moreover, the surface strains and thickness distributions for the SF and the SPIF samples were efficiently predicted using the RFR model-based FLD and FFLD.

Graphical Abstract