|
Deep Learning and Crystal Plasticity: A Preconditioning Approach for Accurate Orientation Evolution Prediction
P. Saidi, H. Pirgazi, M. Sanjari, S. Tamimi, M. Mohammadi, L.K. Beland, M.R. Daymond, I. Tamblyn Computer Methods in Applied Mechanics and Engineering, 389, 114392 (2022) Predicting how crystal orientations evolve during material deformation is essential for understanding and controlling the mechanical properties of metals, but conventional crystal plasticity simulations are computationally expensive. In this work, we develop a deep learning surrogate model for crystal plasticity that introduces an unsupervised machine learning-based data preconditioning technique to optimize crystal orientation data for neural network training. This preconditioning approach dramatically improves the test score from 0.831 to 0.999 on Taylor model data while reducing the required training iterations by approximately an order of magnitude. By employing both artificial neural networks and recurrent neural networks trained on Taylor model crystal plasticity data, the surrogate model achieves very reasonable agreement with real-world electron backscattered diffraction (EBSD) measurements from rolling experiments. This work is foundational for further data-driven studies, enabling the efficient and precise prediction of texture evolution from experimental and simulated crystal plasticity results. |
