Deep learning for in situ data compression of large turbulent flow simulations

Abstract

Large turbulent flow simulations can lead to information bottlenecks as data is generated faster than it can be processed and saved. Innovative in-situ data compression techniques are needed. Here we examine a deep learning approach to in-situ compression using a novel autoencoder architecture customized for three-dimensional turbulent flows. We compare it to a randomized single-pass singular value decomposition method and demonstrate improved compression and reconstruction, particularly with respect to important statistical quantities such as turbulent kinetic energy, enstrophy, and Reynolds stresses, at lower computational cost.