ARTFEED — Contemporary Art Intelligence

LLMs Guide Experimental Phase Diagram Construction for Alloys

ai-technology · 2026-04-24

A study has shown that large language models (LLMs) can facilitate the experimental planning needed to create phase diagrams for multicomponent alloys, a process that typically requires significant time. Within a closed-loop system, a versatile LLM serves as an experimental planner, recommending measurement compositions for each cycle, alongside high-throughput synthesis and X-ray diffraction for phase identification. This method successfully generated the ternary phase diagram for the Co-Al-Ge system at 900°C through repeated synthesis and analysis. Two approaches were evaluated: one utilized initial compositions suggested by a domain-specific LLM (aLLoyM) trained on phase diagram information, while the other depended solely on the general-purpose LLM. Both methods exhibited complementary advantages, with aLLoyM guiding the initial exploration phase.

Key facts

  • LLMs can guide experimental planning for phase diagram construction.
  • Framework uses LLM as planner in closed loop with high-throughput synthesis and X-ray diffraction.
  • Ternary phase diagram of Co-Al-Ge system at 900°C was constructed.
  • Two strategies compared: aLLoyM (domain-specific) vs. general-purpose LLM.
  • aLLoyM directed initial exploration; strategies showed complementary strengths.
  • Research published on arXiv (2604.20304).
  • High-throughput experimentation reduces time for phase diagram construction.
  • Domain-specific LLM trained on phase diagram data.

Entities

Institutions

  • arXiv

Sources