From Deep to Physics-Informed Learning of Turbulence: Diagnostics

  • 2018-12-05 18:10:19
  • Ryan King, Oliver Hennigh, Arvind Mohan, Michael Chertkov
  • 0

Abstract

We describe tests validating progress made toward acceleration and automationof hydrodynamic codes in the regime of developed turbulence by three DeepLearning (DL) Neural Network (NN) schemes trained on Direct NumericalSimulations of turbulence. Even the bare DL solutions, which do not take intoaccount any physics of turbulence explicitly, are impressively good overallwhen it comes to qualitative description of important features of turbulence.However, the early tests have also uncovered some caveats of the DL approaches.We observe that the static DL scheme, implementing Convolutional GAN andtrained on spatial snapshots of turbulence, fails to reproduce intermittency ofturbulent fluctuations at small scales and details of the turbulence geometryat large scales. We show that the dynamic NN schemes, namely LAT-NET andCompressed Convolutional LSTM, trained on a temporal sequence of turbulencesnapshots are capable to correct for the caveats of the static NN. We suggest apath forward towards improving reproducibility of the large-scale geometry ofturbulence with NN.

 

Introduction (beta)

None

 

Conclusion (beta)

None