| Abstract Scope |
The transition to a circular economy requires next-generation superalloys with exceptional impurity tolerance; however, their lifespan is frequently compromised by detrimental Topologically Close-Packed (TCP) phase precipitation. While data-driven methods can predict these instabilities, standard "black box" neural networks lack the transparency required for critical material certification. This study introduces a Neuro-Symbolic framework to discover interpretable, "white box" kinetic laws governing TCP precipitation.
Our workflow integrates multi-scale physical simulations—Density Functional Theory (EMTO-CPA) for chemically complex energetics and CALPHAD (MatCalc) for precipitation timelines—to generate high-fidelity training data. Unlike traditional deep learning, we utilize Symbolic Regression (HeuristicLab) to extract closed-form analytical equations linking local composition to nucleation rates. [Results indicate the extracted equations reproduce CALPHAD predictions with >90% accuracy while reducing computational cost by orders of magnitude.] By moving from numerical approximation to physical explanation, this framework accelerates the discovery of sustainable, defect-tolerant alloys suitable for recycled feedstock. |