Neural Networks Predict Fluid Dynamics Solutions from Tiny Datasets

Cristina White, Daniela Ushizima, Charbel Farhat
January 26, 2019

In collaboration with the Stanford professor Charbel Farhat, the Vivian Church Hoff Professor of Aircraft Structures in the School of Engineering and his DOE CGF awarded student Cristina White, we created a new approach to classify AMReX-based simulation of fluid dynamic solutions with potential to advance aerodynamic designs such as jets, spacecraft, or gas turbine engines. AMReX is an open-source tool designed by the Center of Computational Sciences and Engineering at LBNL, led by Ann Almgren, also a co-author of this work.

Abstract: In computational fluid dynamics, it often takes days or weeks to simulate the aerodynamic behavior of designs such as jets, spacecraft, or gas turbine engines. One of the biggest open problems in the field is how to simulate such systems much more quickly with sufficient accuracy. Many approaches have been tried; some involve models of the underlying physics, while others are model-free and make predictions based only on existing simulation data. However, all previous approaches have severe shortcomings or limitations. We present a novel approach: we reformulate the prediction problem to effectively increase the size of the otherwise tiny datasets, and we introduce a new neural network architecture (called a cluster network) with an inductive bias well-suited to fluid dynamics problems. Compared to state-of-the-art model-based approximations, we show that our approach is nearly as accurate, an order of magnitude faster and vastly easier to apply. Moreover, our method outperforms previous model-free approaches.

Featured Fellows

Daniela Ushizima

Computational Research Division, CAMERA, LBNL
Research Affiliate