Robust deep learning framework for constitutive relations modeling

Modeling the full-range deformation behaviors of materials under complex loading and conditions is a significant challenge for constitutive relations (CRs) modeling. We propose general encoder-decoder deep learning framework that can model high-dimensional stress-strain data histories with robustness universal capability. The employs an encoder to project input information (e.g., history, conditions, information) lower-dimensional hidden space decoder map representation stress interest. evaluated various architectures, including gated recurrent unit (GRU), GRU attention, temporal convolutional network (TCN), Transformer encoder, on two datasets were designed include wide range conditions. All architectures achieved excellent test results root-mean-square error (RMSE) below 1 MPa. Additionally, we analyzed capability different make predictions out-of-domain applications, uncertainty estimation based ensembles. proposed approach provides robust alternative empirical/semi-empirical models CRs modeling, offering potential more accurate efficient design optimization.