CFDAgent with a general kinematics module: driving biological flapping flight simulation

Artikel i vetenskaplig tidskrift, 2026

Automating moving-boundary computational fluid dynamics (CFD), particularly for simulations of biological flapping flight, requires complex coordination across geometry, meshing, kinematics, and postprocessing. We present an enhanced CFDAgent, a large language model (LLM)-orchestrated platform that executes the entire workflow end-to-end through natural-language dialogue, spanning geometry intake, solver execution, and iterative figure generation. The agent features a general kinematics module that directly converts natural-language motion specifications or measured trajectories into the periodic body motion incorporated into the solver. We validate the platform’s accuracy and versatility against three distinct benchmarks: a rotating sphere, a flapping plate, and a slow-flying Pallas’s long-tongued bat (Glossophaga soricina) with deforming wings. Furthermore, we apply the workflow to investigate the aerodynamic effects of hindwing damage in migrating painted lady butterflies (Vanessa cardui). The simulations show that increasing hindwing trailing-edge erosion alters the wake vortex structures, resulting in progressively greater aerodynamic lift loss.