Research map / Multi-agent systems and games / LLM agents
From Text to Simulation: A Multi-Agent LLM Workflow for Automated Chemical Process Design
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes a four-agent LangGraph workflow with Enhanced MCTS that turns natural-language chemical process specs into simulator-executable configurations, evaluated on the self-built Simona dataset.
Problem
Translating textual chemical process requirements into executable, convergent simulation flowsheets is manual, slow and parameter-intensive; existing automation stops at flowsheet or graph representations.
Approach
A multi-agent LLM workflow (task understanding, topology generation, parameter configuration, evaluation analysis) coupled with direct simulator feedback and an Enhanced MCTS that values failed configurations via a dual-layer immediate/potential value and revisits a candidate pool.
Key results
- Ours*-4o reaches 80.3% simulation convergence rate versus 45.4% for the best multi-agent baseline (Swarm) and 23.4% for GPT-4o.
- Design time is reduced by 89.0% relative to expert manual design (913 s vs 8302 s as reported in the table).
- Overall score of 73.88 for Ours*-4o versus 86.16 for expert manual design, which the authors note ranks second behind experts.
Surprising findings
- Expert manual design scores higher overall (86.16) than the proposed method, which the authors acknowledge.
- Removing E-MCTS lowers SCR but cuts iterations from 15 to 7, showing a trade-off between search depth and convergence.
Limitations
- Evaluated on a self-constructed dataset (Simona) with a self-developed simulator, limiting external validity.
- Expert manual design still outperforms the method on scores and economic feasibility.
- Scoring is by an LLM-based evaluation agent with conservative scoring, so dimension gains are small.
Experimental setup
Datasets: Simona
Benchmarks: Simona
Metrics: simulation convergence rate weighted overall score across five dimensions design time tokens iterations
Baselines: GPT-4o Claude Sonnet 4 Swarm AutoGen CrewAI MetaGPT Expert manual design
Builds on: AFlow LangGraph chain-of-thought
Closest papers by meaning
- Can Agents Price a Reaction? Evaluating LLMs on Chemical Cost Reasoning (NeurIPS 2026 · similarity 0.85)
- Beyond Chemical QA: Evaluating LLM's Chemical Reasoning with Modular Chemical Operations (NeurIPS 2025 · similarity 0.84)
- From What to Why: A Multi-Agent System for Evidence-based Chemical Reaction Condition Reasoning (ICLR 2026 · similarity 0.83)
- EngiAgent: Fully Connected Coordination of LLM Agents for Solving Open-ended Engineering Problems with Feasible Solutions (ICML 2026 · similarity 0.83)
- LLM-Augmented Chemical Synthesis and Design Decision Programs (ICML 2025 · similarity 0.82)
- Flow: Modularized Agentic Workflow Automation (ICLR 2025 · similarity 0.82)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.