Orthotropic behavior characterization in sheet metal forming: parameter identification of Yld2011-27p model using deep learning and genetic algorithm

TY - JOUR

T1 - Orthotropic behavior characterization in sheet metal forming

T2 - parameter identification of Yld2011-27p model using deep learning and genetic algorithm

AU - Shafaie, Majid

AU - Keneshlou, Mehrdad

AU - Askarinejad, Sina

N1 - Publisher Copyright: Copyright © 2025. Published by Elsevier Ltd.

PY - 2026/1

Y1 - 2026/1

N2 - Yld2011-27p is a highly accurate criterion for modeling anisotropic material behavior, though its anisotropic parameter identification traditionally requires extensive experimentation. The present study proposes two data-driven approaches, direct and indirect, based on the earing geometry of a single deep drawing test for efficiently determining these parameters. A deep neural network (DNN) trained by preliminary finite element data is used in the direct method while a combination of deep neural network and genetic algorithm (GA) is used in the indirect method to calibrate the Yld2011-27p anisotropic parameters. These models are iteratively updated through finite element simulations via a Python script and an Abaqus VUMAT subroutine, until the simulated results align with experimental observations. The entire process is automated, requiring only the experimental output and parameter bounds from the user. The approach significantly reduces experimental effort while achieving high prediction accuracy. The direct and indirect frameworks reached final contour prediction errors of 0.94 mm and 0.88 mm, respectively, which are lower than the error of the experimentally calibrated parameters (0.97 mm).

AB - Yld2011-27p is a highly accurate criterion for modeling anisotropic material behavior, though its anisotropic parameter identification traditionally requires extensive experimentation. The present study proposes two data-driven approaches, direct and indirect, based on the earing geometry of a single deep drawing test for efficiently determining these parameters. A deep neural network (DNN) trained by preliminary finite element data is used in the direct method while a combination of deep neural network and genetic algorithm (GA) is used in the indirect method to calibrate the Yld2011-27p anisotropic parameters. These models are iteratively updated through finite element simulations via a Python script and an Abaqus VUMAT subroutine, until the simulated results align with experimental observations. The entire process is automated, requiring only the experimental output and parameter bounds from the user. The approach significantly reduces experimental effort while achieving high prediction accuracy. The direct and indirect frameworks reached final contour prediction errors of 0.94 mm and 0.88 mm, respectively, which are lower than the error of the experimentally calibrated parameters (0.97 mm).

KW - Data-driven system

KW - Deep neural network

KW - Genetic algorithm

KW - Sheet metal forming

KW - Yld2011-27p

UR - https://www.scopus.com/pages/publications/105026068493

U2 - 10.1016/j.matdes.2025.115389

DO - 10.1016/j.matdes.2025.115389

M3 - Article

AN - SCOPUS:105026068493

SN - 0264-1275

VL - 261

JO - Materials and Design

JF - Materials and Design

M1 - 115389

ER -