Artificial Neural Networks for the application to reactive flow simulations

Abstract

Objectives:

  • Documentation of current applications of ANNs in CFD
  • Development/Implementation of a neuronal net for the prediction of chemical kinetics within flows

Results:

  • Validated ANN for the application to reactive flow simulations
  • Model for the usage of ANN for OpenFOAM simulations

Description:

Although computational resources are still increasing and more and more simulations are enabled, the computational time can still be prohibitive for reactive flow simulations using complex kinetic mechanisms. There are many different strategies to overcome this problem. Tabulation can be used or the mechanisms can be reduced to skeletal, i.e. simpler, mechanisms. Another way is to use ANN trained by kinetic data and use it to predict the concentration changes in combustion simulations.

Research on ANNs currently used in CFD, especially in reactive flow simulations shoul be studied. An ANN shall be trained and tested for a chemical mechanism for methane combustion. The application of the ANN for CFD simulations of flames should be possible in the end.

Requirements:

  • CFD-exercise beneficial (f.ex.: 166.050)
  • Matlab/Python and C/C++ skills beneficial
  • English or German possible

Start:

As from now

Contact:
DI Eva-Maria Wartha
Ass. Prof. DI Dr. Michael Harasek

Financial award for excellent work possible.