| Abstract Scope |
The guiding principles of materials science are based on fundamental relationships between elemental composition, process parameters, structure, and properties of materials. Explainable artificial intelligence (AI) and machine learning (ML) can speed up the materials discovery process if they incorporate the scientific and engineering knowledge. Advanced AI tools can support the fusion of observed data and engineering information and, ultimately, the explainable design of new materials and processes. This presentation illustrates it on the example of iron-based alloy development. Graph Neural Networks’ (GNN) designer was used as a flexible framework, in combination with other advanced computational tools. The designer models proved to be effective not only in capturing the fundamental relationships hidden in the data but also in designing new alloys and processes. It was observed that the adversarial models were aligned with the known brittleness vs ductility trends, after the variables’ transformation. The proposed compositions favored well-known alloy strengthening mechanisms. |