Agentic Additive Manufacturing Alloy Discovery

TL;DR

Problem: AM alloy discovery is time-consuming due to multidisciplinary requirements. Approach: proposes an agentic multi-agent system driven by the Model Context Protocol to orchestrate CALPHAD-based property predictions (Thermo-Calc), defect prediction, and printability analysis. Contributions: demonstrates end-to-end automation across known and unknown alloys, with substantial LoF-map results and literature-consistent corrosion guidance; discusses data-structure challenges and a failure mode. Significance: provides a scalable framework to accelerate automated materials discovery in additive manufacturing and to reduce dependence on manual parameter tuning.

Abstract

Agentic systems enable the intelligent use of research tooling, augmenting a researcher's ability to investigate and propose novel solutions to existing problems. Within Additive Manufacturing (AM), alloy discovery remains a complex challenge, often requiring expertise in the various domains of materials science, thermodynamic simulations, and experimental analysis. Large Language Model (LLM) enabled agents can facilitate this endeavor by utilizing their extensive knowledge base to dispatch tool calls via Model Context Protocol (MCP) to perform actions such as Thermo-Calc property diagram calculations and lack of fusion process map generation. In addition, the multi-agent system developed in this work is able to effectively reason through complex user prompts and provide analysis on the printability of proposed alloys. These agents can dynamically adjust their task trajectory to the outcomes of tool call results, effectively enabling autonomous decision-making in practical environments. This work aims to utilize LLM enabled agents to automate and accelerate the task of alloy discovery within the field of additive manufacturing and showcase the benefits of adopting this multi-agent system.

Figures (10)

• Figure 1: (Top Left) An input query for alloy compositions regarding the printability of an additively manufactured part suitable for its intended use case is provided to Claude Sonnet. This Large Language Model (LLM) calls the tools necessary to generate and analyze each potential alloy compositions, providing a response of candidates ranked by their content of their lack of fusion fusion regimes. (Top Right) Thermo-Calc allows for the retrieval of material properties for an arbitrary alloy composition, for instance thermal conductivity, to be used in down stream printability calculations. (Bottom Left) Workspaces provide a way for each of the tools to effectively communicate with one another and handles state management and file organization. (Bottom Right) Tools managed by the Additive Manufacturing subagents then utilize the calculated material properties from the Thermo-Calc subagent to generate a lack of fusion process map to send back to the LLM for analysis and final recommendation.
• Figure 2: (Left) Claude Code provides an interface for integrating agentic tools with Claude Sonnet LLM, allowing for natural language input to execute tasks and response analysis. (Right) Streamlined summary of tool executions and analysis from the prompt given to Claude Code utilizing subagents for Additive Manufacturing, Thermo-Calc, and Workspace.
• Figure 3: A simple tool calling procedure within the Workspace subagent for the task of finding or initializing a workspace. Here an initial tool call is made to list available workspaces and if none are found, a new workspace is created. This newly initialized workspace or the most relevant selected by Claude Sonnet is included in the successful response object.
• Figure 4: Flow diagram outlines the expected tool calling procedure for the Thermo-Calc subagent. In this example the material properties for Stainless Steel are extracted from the calculated property diagram of the alloy's elemental composition. Composition is obtained from a look-up table of known alloys or provided directly to the agentic system and parsed into mass fractions using Claude Sonnet. This process generates a schema file with alloy's material properties recorded for downstream use with other tools.
• Figure 5: Diagram outlines expected tool calling procedure for additive manufacturing subagent for the task for generating and analyzing a lack of fusion process map. Build and material configurations are required to initialize a process map, the latter of which can be obtained from the Thermo-Calc subagent or manually configured by the additive manufacturing subagent. Initializing the process map provides override ranges for power and velocity build parameters for melt pool depth calculations. The tool generates the process map and a response consisting of power and velocity configurations that potentially exhibit lack of fusion defects. Claude Sonnet analyzes this response and provides suggestion for optimal build parameters.