| Abstract Scope |
Artificial Intelligence (AI) platform was used for iron-based alloy simulation and design, by incorporating materials science-based knowledge into computational graph architecture and learning procedures (DNN layers pre-trained on fuzzy physics, microstructure and process representation, and multi-objective optimization). This causal platform can provide novel design ideas and their interpretability via physics and engineering concepts. The graph-based approach facilitated development of the microstructure evolution models with reduced overfitting to limited datasets. Preliminary results demonstrated that domain science-based design of materials for high-temperature environments can be accelerated by the fusion of data from simulation and experiment, and by advanced AI-based workflows. Unprecedented acceleration of the AI-based workflow can be achieved using wafer scale engines, which was demonstrated on a toy problem. |