Context is all you need: Towards autonomous model-based process design using agentic AI in flowsheet simulations

Pascal Schäfer; Lukas J. Krinke; Martin Wlotzka; Norbert Asprion

cs.AIMar 13, 2026

Context is all you need: Towards autonomous model-based process design using agentic AI in flowsheet simulations

Authors: Pascal Schäfer, Lukas J. Krinke, Martin Wlotzka, Norbert Asprion

Organizations: BASF SE, Ludwigshafen, Germany · RWTH Aachen University, Aachen, Germany

Abstract

Agentic AI systems integrating large language models (LLMs) with reasoning and tooluse capabilities are transforming various domains - in particular, software development. In contrast, their application in chemical process flowsheet modelling remains largely unexplored. In this work, we present an agentic AI framework that delivers assistance in an industrial flowsheet simulation environment. To this end, we show the capabilities of GitHub Copilot (GitHub, Inc., 2026), when using state-of-the-art LLMs, such as Claude Opus 4.6 (Anthropic, PBC, 2026), to generate valid syntax for our in-house process modelling tool Chemasim using the technical documentation and a few commented examples as context. Based on this, we develop a multi-agent system that decomposes process development tasks with one agent solving the abstract problem using engineering knowledge and another agent implementing the solution as Chemasim code. We demonstrate the effectiveness of our framework for typical flowsheet modelling examples, including (i) a reaction/separation process, (ii) a pressure-swing distillation, and (iii) a heteroazeotropic distillation including entrainer selection. Along these lines, we discuss current limitations of the framework and outline future research directions to further enhance its capabilities.

Figures & tables

Figure 1: Multi-agent system architecture.

Case	A	B	C	D	E
1	Ethylene	Benzene	Ethyl	Diethyl	–
			benzene	benzene
2a	n-Propanol	Benzene	–	–	–
2b	Acetone	Chloroform	–	–	–
3	Water	1,4-dioxane	Benzene	Dimethyl	–
				carbonate

Table 1: Component mapping for the masked case studies in the Results and discussion section.

Reaction	Stoichiometry	$T$	$p$	Conditions
		[°C]	[bar]
Main	A + B $\rightarrow$ C	180	20	90% conversion w.r.t. A
				2:1 molar B:A ratio
Side	A + C $\rightarrow$ D	180	20	10% conversion w.r.t. A
Transalkylation	D + B $\rightarrow$ 2 C	240	20	80% conversion w.r.t. D
				5:1 molar B:D ratio, pure D required

Table 2: Reaction network and operating conditions for case study 1. All reactions occur in the liquid phase.

$p$	$T_{\mathrm{b}}$ (A)	$T_{\mathrm{b}}$ (B)	$T_{\mathrm{b}}$ (C)	$T_{\mathrm{b}}$ (D)
bar	°C	°C	°C	°C
0.12	$-133.0$	23.9	71.4	112.9
0.15	$-130.5$	28.7	77.0	119.1
0.25	$-124.3$	40.7	90.8	134.1
0.50	$-114.9$	58.8	111.7	157.0
1.00	$-104.0$	79.7	135.7	183.3

Table 3: Boiling points of pure components A–D at selected pressures for case study 1.

Figure 2: Final process flow diagram for case study 1 - reaction-separation process. Blue numbers show the molar balance as pre-computed by the process development agent. Green numbers show the results from a rigorous simulation of the flowsheet built by the Chemasim modelling agent. Dashed lines indicate purge streams planned by the Chemasim modelling agent to prevent the accumulation of impurities in the recycle. If no composition is given, the process development agent assumes a pure component stream.

Panel of Figure 3 — Figure 3: Pressure-dependent xy-diagrams for case study 2 at selected pressures. Colors and line-styles indicate different pressures as given in the legend.

Case study	$p$ [bar]	$T_{\mathrm{b}}$ (A) [°C]	$T_{\mathrm{b}}$ (B) [°C]	$T_{\mathrm{azeo}}$ [°C]
2a (MI)	0.5	79.82	58.80	56.79
	1.0	97.07	79.66	76.13
	2.0	116.65	103.91	98.05
	5.0	147.08	142.70	131.90
2b (MA)	0.25	20.26	23.86	27.03
	1.0	55.68	60.57	63.97

Table 4: Pure-component boiling points and azeotrope temperatures for case study 2 at selected pressures. MI and MA refer to minimum and maximum azeotrope, respectively.

Figure 4: Final process flow diagram for case study 2a - binary pressure-swing distillation for a minimum azeotrope. Blue numbers show the mass balance as pre-computed by the process development agent. Green numbers show the results from a rigorous simulation of the flowsheet built by the Chemasim modelling agent. Dashed lines indicate purge streams planned by the Chemasim modelling agent to prevent the accumulation of impurities in the recycle. If no composition is given, the process development agent assumes a pure component stream.

Panel of Figure 5 — Figure 5: Ternary diagram for the heteroazeotropic distillation systems for separating A from B in case study 3. Diagrams include the azeotropes and (ternary) miscibility gaps. Squares mark azeotropes, with MI denoting homogeneous minimum azeotropes and MIH heterogeneous minimum azeotropes. Thin lines inside the miscibility gap show tie lines connecting liquid phases in equilibrium. The critical point of the liquid-liquid equilibrium is marked with a star.

Figure 6: Process flow diagram for case study 3 as originally proposed by the agent system using a classic heteroazeotropic distillation approach with one distillation column and a decanter. Blue numbers show the mass balance as pre-computed by the process development agent. Green numbers show the results from a rigorous simulation of the flowsheet built by the Chemasim modelling agent. Dashed lines indicate purge streams planned by the Chemasim modelling agent to prevent the accumulation of impurities in the recycle. If no composition is given, the process development agent assumes a pure component stream.

Figure 7: Process flow diagram for case study 3 enforcing a modified heteroazeotropic distillation approach with a pre-separation step. Blue numbers show the mass balance as pre-computed by the process development agent. Green numbers show the results from a rigorous simulation of the flowsheet built by the Chemasim modelling agent. Dashed lines indicate purge streams planned by the Chemasim modelling agent to prevent the accumulation of impurities in the recycle. If no composition is given, the process development agent assumes a pure component stream.

Panel of Figure 8 — Figure 8: Ternary diagrams for case study 4 (more challenging scenarios for heteroazeotropic distillation and entrainer selection). Diagrams include the azeotropes and (ternary) miscibility gaps. Squares mark azeotropes, with MI denoting homogeneous minimum azeotropes and MIH heterogeneous minimum azeotropes. Thin lines inside the miscibility gap show tie lines connecting liquid phases in equilibrium. The critical point of the liquid-liquid equilibrium is marked with a star.

Panel of Figure 9 — Figure 9: Diagrams for case study 5 (ternary pressure-swing distillation). Diagrams show the pressure-dependence of the binary azeotropes. Squares mark azeotropes, with MI denoting minimum azeotropes.

Agentic Workflows

Agentic workflows coordinate AI agents, tools, and humans across multi-step tasks, emphasizing reliable execution, efficiency, and oversight.

See all 218 papers

Sketch2Simulation: Automating Flowsheet Generation via Multi Agent Large Language Models
Mar 25, 2026Abdullah Bahamdan, Emma Pajak, John D. Hedengren +1Program Synthesis
Sargent Centre for Process Systems Engineering, Imperial College London, London, SW7 2AZ, United Kingdom · Department of Chemical Engineering, Brigham Young University, Provo, Utah 84602, United States
CRAFTS: Collaborative Role-Adaptive Fine-Tuning of LLM Agents for Chemical Process Simulation
Aug 2, 2026Ziyun Zhang, Yuxin Lin, Eldin Wee Chuan Lim +1Fine-Tuning LLM Agent Orchestration
Department of Chemical & Biomolecular Engineering, National University of Singapore · 2Key Laboratory of Multimedia Trusted Perception and Efficient Computing, Ministry of Education of China, Xiamen University, Xiamen, Fujian, China
LLM Agents Perform Controlled Experiments Using Simulation Models
Aug 22, 2026Yuchen Xia, Michael Weyrich, Nasser Jazdi +6Multi-Agent LLM Systems Large Language Model-Guided Scientific Discovery
Institute for Industrial Automation and Software Engineering University of Stuttgart, Stuttgart, Germany · Sustainable Innovation & Transformational Excellence (xSITE), Pharmaceutical Technology & Development, Operations AstraZeneca Macclesfield, UK · Chemical Development, Pharmaceutical Technology & Development, Operations AstraZeneca Macclesfield, UK +1

Mar 25, 2026J/K move · Enter open · S save

Sketch2Simulation: Automating Flowsheet Generation via Multi Agent Large Language Models

Abdullah Bahamdan, Emma Pajak, John D. Hedengren, Antonio del Rio Chanona

Sargent Centre for Process Systems Engineering, Imperial College London, London, SW7 2AZ, United Kingdom · Department of Chemical Engineering, Brigham Young University, Provo, Utah 84602, United States

Program Synthesis

Open paper AskPDF arXiv

Converting process sketches into executable simulation models remains a major bottleneck in process systems engineering, requiring substantial manual effort and simulator-specific expertise. Recent advances in generative AI have improved both engineering-diagram interpretation and LLM-assisted flowsheet generation, but these remain largely disconnected: diagram-understanding methods often stop at extracted graphs, while text-to-simulation workflows assume structured inputs rather than raw visual artifacts. To bridge this gap, we present an end-to-end multi-agent large language model system that converts process diagrams directly into executable Aspen HYSYS flowsheets. The framework decomposes the task into three coordinated layers: diagram parsing and interpretation, simulation model synthesis, and multi-level validation. Specialized agents handle visual interpretation, graph-based intermediate representation construction, code generation for the HYSYS COM interface, execution, and structural verification. We evaluate the framework on four chemical engineering case studies of increasing complexity, from a simple desalting process to an industrial aromatic production flowsheet with multiple recycle loops. The system produces executable HYSYS models in all cases, achieving complete structural fidelity on the two simpler cases and strong performance on the more complex ones, with connection consistency above 0.93 and stream consistency above 0.96. These results demonstrate a viable end-to-end sketch-to-simulation workflow while highlighting remaining challenges in dense recycle structures, implicit diagram semantics, and simulator-interface constraints.

Abstract

Agentic Workflows

Explore similar work

Sketch2Simulation: Automating Flowsheet Generation via Multi Agent Large Language Models

CRAFTS: Collaborative Role-Adaptive Fine-Tuning of LLM Agents for Chemical Process Simulation

LLM Agents Perform Controlled Experiments Using Simulation Models