The extended subloading surface (ESS) model is the materials model in which the subloading surface and the elastic-core are introduced in stress space. One of the challenges of this model is that there is no established method for determining material parameters rationally. In practice, parameter tuning seems to be done on a trial-and-error process or by using optimization functions included in the commercial FEM analysis software packages at present. The difficulty can be attributed to the fact that the internal state variables, such as elastic-core, cannot be measured directly. To address this problem, we attempted to utilize LSTM, a deep learning method, which is effective for processing time series data and is expected to be highly effective in predicting internal state variables that depend on strain history. The estimation of material parameters would also be possible by predicting the changes of the internal state variables. The learning-model trained by LSTM was constructed with the training data which are numerical simulation results. The simulation results contain strain, stress and internal state variables, such as back stress and elastic-core, from the ESS model with various material parameters. The learning-model showed good predictive performance on some of the training data. The remaining challenges are predicting the unknown data and increasing the prediction accuracy of the learning-model.
Ishizu et al. (Wed,) studied this question.