CATVis: A Collaborative Multi-Agent Workflow for Turbomachinery Simulation Data Visualization
Authors
Zhe Wang (Computer Network Information Center, Chinese Academy of Sciences), Zehao Lou (Computer Network and Information Center, University of Chinese Academy of Sciences), Guanghui Zhao (Computer Network Information Center (CNIC)), Yu Dong (Computer Network Information Center, Chinese Academy of Sciences), Guan Li (Computer Network Information Center, Chinese Academy of Sciences), Pengyi Xu (Michigan Technological University), GAORONG LIANG (University of Chinese Academy of Sciences), Jun Liu (Computer Network Information Center, Chinese Academy of Sciences), Guihua Shan (Computer Network Information Center, Chinese Academy of Sciences)
Presentation
- Session
- Big Data, Bigger Physics
- Time
- Friday, Nov 13, 08:54 – 09:03 (US/Eastern) · session 08:00 – 09:30
- Location
- Hall America south
Keywords
Scientific workflow, LLM-based agent, CFD visualization
Abstract
Recent advances in AI for Science have enabled natural language (NL) interfaces for scientific data analysis. In turbomachinery CFD post-processing, translating ambiguous high-level analytical goals (e.g., vortex identification) into precise visualization procedures supporting complex domain-specific analysis is challenging. We present CATVis, Collaborative multi-agent workflow system that bridges this gap by transforming NL intents into structured middle representation for visualization. Our approach reformulates domain-specific visualization procedures as composable workflow representations, and use multi agent to generate workflow representations via intent planning, template generation, and error-aware refinement, where each stage incrementally updates a shared structured representation. We evaluate the impact of external knowledge and workflow structuring on generation accuracy, demonstrating that the proposed approach significantly improves complex workflow generation correctness while reducing prompt complexity.
For Practitioners
This work targets practitioners in computational fluid dynamics (CFD), turbomachinery design, and scientific visualization who routinely analyze complex simulation outputs but rely on manual, script-heavy post-processing pipelines. CATVis provides a workflow-centric and LLM-assisted alternative that translates high-level analytical intents (e.g., vortex identification, blade-to-blade flow inspection, or spanwise analysis) into structured, executable visualization workflows. By abstracting domain-specific operations into composable DAG-based tasks, the system reduces the need for expertise in low-level scripting environments such as Tecplot macro languages or ParaView Python pipelines, while preserving domain control and interpretability. In practice, CATVis can be integrated into existing CFD post-processing workflows as an intermediate “visualization workflow orchestration layer,” enabling engineers to iteratively generate, inspect, and refine visualization pipelines through natural language and interactive adjustment. The multi-agent design improves robustness in complex, multi-step visualization scenarios common in turbomachinery simulation data analysis, while the structured representation supports debugging, reuse, and collaboration across teams.