基于时间序列深度学习的金属板材本构模型建立

To accurately predict the mechanical response of pure aluminum sheets under arbitrary complex loading conditions, the elastoplastic constitutive relationship was established by combining the visco-plastic self-consistent (VPSC) model with a gated recurrent unit (GRU) neural network structure. First, the parameters of the VPSC model were calibrated using experimental data. Then, 12500 groups of complex strain paths were generated through the piecewise cubic Hermite interpolation polynomial method. Subsequently, the VPSC model was employed to obtain stress and strain sequence data under various complex loading conditions, which were then converted into strain increment-stress sequence datasets (the input was the strain increment sequence and the output was the stress sequence). Finally, the datasets were divided into training and validation sets for training and prediction of the neural network model. The results show that, without adopting the traditional constitutive formulations or assuming any deformation mechanisms, the GRU-based sequence deep learning model can effectively capture the stress and strain response of metal sheets under complex nonlinear deformation. The prediction accuracy of normal stress reaches 98. 885%, and that of shear stress reaches 97. 42%, demonstrating that the data-driven plastic constitutive modeling of metallic sheets, based on the combination of deep learning models and microstructural models, is both accurate and efficient.