From Text to Simulation: A Multi-Agent LLM Workflow for Automated Chemical Process Design

TL;DR

The paper presents an end-to-end, AI-assisted workflow that converts natural-language chemical process descriptions into executable simulation configurations by coordinating four specialized LLM-driven agents (Task Understanding, Topology Generation, Parameter Configuration, Evaluation Analysis) within a LangGraph-based framework and guided by Enhanced Monte Carlo Tree Search (E-MCTS). It integrates direct, bidirectional validation with industrial process simulation software and includes an auxiliary thermodynamic analysis step to ensure feasible designs, achieving substantially faster design times and improved convergence on the Simona dataset while approaching expert performance on multiple criteria. Key contributions include direct generation of executable simulations from text, a modular multi-agent architecture that fuses semantic understanding with domain knowledge, and an E-MCTS scheme that values and revisits failed configurations to refine designs. The results demonstrate meaningful gains in design speed and convergence, offering a practical path toward AI-assisted chemical process design with broad industrial relevance.

Abstract

Process simulation is a critical cornerstone of chemical engineering design. Current automated chemical design methodologies focus mainly on various representations of process flow diagrams. However, transforming these diagrams into executable simulation flowsheets remains a time-consuming and labor-intensive endeavor, requiring extensive manual parameter configuration within simulation software. In this work, we propose a novel multi-agent workflow that leverages the semantic understanding capabilities of large language models(LLMs) and enables iterative interactions with chemical process simulation software, achieving end-to-end automated simulation from textual process specifications to computationally validated software configurations for design enhancement. Our approach integrates four specialized agents responsible for task understanding, topology generation, parameter configuration, and evaluation analysis, respectively, coupled with Enhanced Monte Carlo Tree Search to accurately interpret semantics and robustly generate configurations. Evaluated on Simona, a large-scale process description dataset, our method achieves a 31.1% improvement in the simulation convergence rate compared to state-of-the-art baselines and reduces the design time by 89. 0% compared to the expert manual design. This work demonstrates the potential of AI-assisted chemical process design, which bridges the gap between conceptual design and practical implementation. Our workflow is applicable to diverse process-oriented industries, including pharmaceuticals, petrochemicals, food processing, and manufacturing, offering a generalizable solution for automated process design.

Figures (3)

• Figure 1: Representation of different stages of chemical process design workflow. "Diagram" refers to a general process flow diagramdouglas1988conceptual, while "Hypergraph" represents a process hypergraph with parameter annotations mann2024esfiles. The entire stage requires a lot of manual operation design, while existing automated methods only focus on various representations of graphs. Our method achieves cross-dimensional and fully automated implementation
• Figure 2: The overview of our comprehensive workflow, consisting of four key agents: (i) the Task Understanding Agent, which parses user inputs to extract key parameters, constraints, and design objectives. (ii) the Topology Generation Agent, which determines appropriate chemical components and thermodynamic methods, selects suitable unit operations and establishes their connections. (iii) the Parameter Configuration Agent, which assigns operating conditions to each unit within the generated topology, such as temperature and pressure. (iv) the Evaluation Analysis Agent, which assesses simulation results to determine whether to iterate or terminate.
• Figure 3: Illustration of the design generated by different methods, where different colors represent different cases.