Scholarlink

<p>This research proposes a coupled machine learning (ML) and finite element (FE) approach to predict the formability of steel and aluminum metallic sheets under the influence of pre-strain and temperature. Two ML algorithms, viz. support vector regression (SVR) and random forest regression (RFR), were selected to predict the forming limit diagrams (FLDs) under two scenarios: various pre-strained conditions during a two-stage forming process and at elevated temperature forming process. A novel shift variable and shift direction were introduced in the ML models to account for the dynamic nature of FLDs. The RFR outclassed the SVR in predicting FLDs with a higher coefficient of determination (<i>R</i><sup>2</sup>) value. Further, RFR and SVR could predict the shifting nature of the FLDs depending on the magnitude of pre-strain and temperature. The dynamic nature of the pre-strained RFR&#xa0;FLD was successfully restricted in the path-independent polar effective plastic strain (PEPS)-based locus during two-stage forming. Limiting&#xa0;dome heights were predicted with the Barlat Yld2000 anisotropic material model and ML&#xa0;FLD into the FE simulations, with an error of 5.92% for pre-strained steel sheet and 6.61% for aluminum sheet&#xa0;at 200&#xa0;°C. The ML-FE approach closely captured the surface strain distributions for both the forming processes.</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

Formability Prediction of Anisotropic Thin Sheets Using Machine Learning Framework: Influence of Pre-Strain and Temperature

  • 期刊论文

  • 发表日期：

    2025年4月10日

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

摘要

This research proposes a coupled machine learning (ML) and finite element (FE) approach to predict the formability of steel and aluminum metallic sheets under the influence of pre-strain and temperature. Two ML algorithms, viz. support vector regression (SVR) and random forest regression (RFR), were selected to predict the forming limit diagrams (FLDs) under two scenarios: various pre-strained conditions during a two-stage forming process and at elevated temperature forming process. A novel shift variable and shift direction were introduced in the ML models to account for the dynamic nature of FLDs. The RFR outclassed the SVR in predicting FLDs with a higher coefficient of determination (R2) value. Further, RFR and SVR could predict the shifting nature of the FLDs depending on the magnitude of pre-strain and temperature. The dynamic nature of the pre-strained RFR FLD was successfully restricted in the path-independent polar effective plastic strain (PEPS)-based locus during two-stage forming. Limiting dome heights were predicted with the Barlat Yld2000 anisotropic material model and ML FLD into the FE simulations, with an error of 5.92% for pre-strained steel sheet and 6.61% for aluminum sheet at 200 °C. The ML-FE approach closely captured the surface strain distributions for both the forming processes.