| Abstract Scope |
The rapid progress in additive manufacturing of alloys brings opportunities in controlling microstructures and geometry, thus unlocking unprecedent performances. However, to fully access such potential, efficient models for navigating the tremendous design space of alloy compositions and processing conditions are of great research interest. Here, we propose AlloyGPT, an autoregressive alloy language model, to learn the composition-structure-property relationship and generate novel design for additively manufacturable alloys.
Specifically, we develop efficient grammar to convert knowledge-rich alloy datasets into readable language records for both forward prediction and inverse design tasks. Then, we construct a generative pre-trained transformer (GPT) model to master this alloy language through auto regression. After training, our AlloyGPT can handle both prediction and design tasks simultaneously, and continuously learn about other alloy systems. Our alloy language model presents a novel way of integrating comprehensive knowledge of materials science and is expected to be valuable for accelerating alloy design. |