Mini-symposium F6

F6: Using Artificial Intelligence to Enhance Computational Fluid Dynamics

Chair: 

Claire E. Heaney (Imperial College London)*

Keywords:  Artificial intelligence, scientific machine learning, deep learning, data-driven methods, computational fluid dynamics, numerical methods.

ABSTRACT

Solving complex physics and engineering problems to the required degree of precision places large demands on computing power. For most problems of interest, many degrees of freedom are needed to solve the discretised governing equations accurately enough. The problem of computational resources is exacerbated for multi-query problems, such as data assimilation, uncertainty quantification and inverse problems, which require results from hundreds or more simulations.

Machine learning underlies many recent efforts to reduce computational times. A number of new directions have emerged that combine physics-based methods with data-driven methods. These include physics-informed neural networks (and related methods), which incorporate physical constraints in the loss function of the neural network [1]; operator learning, which learns a map from physical coordinates to a variable of interest, exemplified by neural operators [2, 3] and DeepONet [4]; transformer networks, which offer the promise of improved time series prediction [5]; and discovering unknown physics, such as subgrid-scale models for turbulence [6] or learning equations from data [7].

This mini-symposium welcomes contributions on applying any of the above techniques, or indeed, any methods from Artificial Intelligence, to enhance computational fluid dynamics simulations.

REFERENCES

[1] Raissi, Wang, Triantafyllou and Karniadakis (2019) Deep learning of vortex-induced vibrations, J. Fluid Mech. 861:119–137.

[2] Kurth, Subramanian, Harrington, et al. (2023) FourCastNet: Accelerating Global High-Resolution Weather Forecasting Using Adaptive Fourier Neural Operators, in ACM's PASC '22, Davos, Switzerland.

[3] Li, Liu-Schianchi, Kovachki et al. (2024) Learning dissipative dynamics in chaotic systems, in Proceedings of the 36th International Conference on Neural Information Processing Systems (NIPS '22).

[4] Lu, Jin, Pang, et al. (2021) Learning nonlinear operators via DeepONet based on the universal approximation theorem of operators, Nature Machine Intelligence 3:218–229.

[5] Geneva and Zabaras (2022) Transformers for modeling physical systems, Neural Networks 146:272–289.

[6] Sanderse, Stinis, Maulik and Ahmed (2024) Scientific machine learning for closure models in multiscale problems: a review, arXiv preprint: 2403.02913.

[7] Brunton, Proctor and Kutz (2016) Discovering governing equations from data by sparse identification of nonlinear dynamical systems, Proc. Natl Acad. Sci. 113(15):3932–3937.