Artificial intelligence (AI), especially large language models (LLMs), are triggering an intelligent transformation in the field of computational fluid dynamics (CFD). LLMs can assist and even partially replace humans in executing complex fluid simulation tasks. Due to such opportunities, we systematically review recent progress and limitations of applying LLMs to the CFD. Specifically, three key applications of LLMs in CFD are surveyed, explored, and analyzed, including: (1) Predictive modeling for prediction of fluid behavior (such as real-time generation of flow field distributions) and automatic discovery of turbulence models (such as improvement of the k-ε model based on direct numerical simulation data, achieving implicit encoding of physical laws); (2) Optimization applications, including CFD model optimization (i.e., optimization of numerical model parameters and their sub-models, such as turbulence model parameters), hyperparameter optimization of machine learning models (such as learning rate, optimizer, number of neurons, and hidden layers involved in neural networks training), geometric structure optimization (such as topological optimization of fluid process equipment structures), and process operating parameters (such as inlet velocity, operating pressure and temperature); (3) Automated execution scheduling for full-process end-to-end automation for efficient CFD simulations, including automatic geometry generation, boundary condition configuration, meshing, and solver invocation. Meanwhile, we summarize and analyze key weaknesses, such as insufficient physical credibility and engineering applicability limitations. Finally, we suggest future directions to possibly address these challenges, including reducing the difficulty of obtaining domain data, continuously updating domain models, and improving model interpretability. The complementary relationship between human intelligence and AI (HI-AI) will promote CFD to better serve academic and engineering applications.
Gai et al. (Thu,) studied this question.