VFEAgent: AI framework automates finite element analysis from images

Finite Element Analysis (FEA) is essential for modern engineering but notoriously complex, requiring deep domain expertise for modeling, meshing, and simulation setup. A new paper from Peking University and China Agricultural University introduces VFEAgent, an end-to-end multi-agent system that automates the entire FEA workflow directly from multimodal inputs—images (e.g., CAD drawings, hand sketches) and natural language problem descriptions. The system combines two core innovations: (1) a multimodal vision-language multi-agent pipeline using ReAct (Reasoning + Acting) driven reasoning to parse heterogeneous inputs and generate structured FEA specifications, and (2) a verification-first code synthesis framework that includes robust self-debugging and fallback mechanisms to ensure the generated simulation code is both executable and physically valid. This eliminates the need for manual geometry preparation, mesh generation, and solver configuration.

Evaluated across various engineering mechanics scenarios, VFEAgent achieved high success rates in producing complete, physically valid simulations, significantly outperforming existing LLM-based baseline methods in reliability and correctness. The framework demonstrates that automating the full FEA workflow is feasible, potentially reducing the time engineers spend on repetitive tasks. While still a preprint (9 pages, 3 figures), the results highlight how large language models combined with multimodal reasoning can bridge the gap between design intent and numerical simulation. The authors note that VFEAgent could ultimately liberate engineers from tedious manual analysis, allowing them to focus on higher-level design decisions. The code and data have not yet been released, but the approach signals a major step toward AI-driven engineering simulation.