| Abstract Scope |
Alloy design often requires in-depth knowledge of materials science, expertise in applying modeling/experimental toolboxes and cognitive skills of processing information and making decisions, thus mainly reserved for human experts. Recent advances in manufacturing capabilities (e.g., additive manufacturing) have opened novel possibilities and raised the bar for the next generation of alloy designs. To address those opportunities and challenges, automation of reliable alloy design can have unique potential for high efficiency and reduced costs.
In this presentation, we will discuss our recent exploration on automating alloy design by leveraging large language model (LLM) powered multi-agent conversional frameworks. Specially, we will discuss how individual agents based on general purposed LLM (e.g., GPT4) can be profiled to handle specific subtasks (e.g., planning and modelling) and how their conversations can lead to continuous improvements and effective alloy design. Our framework provides important insights and may lay the foundation for achieving robust automatic alloy designs. |