pod-surrogate's Introduction

Implementation of a POD based surrogate model for aerodynamic shape optimization

Please refer to the thesis for a in depth description of the methods involved. Some data has not been included into this git repository due to the 100MB file size limit of GitHub

Make sure the imports of SU2, Armadillo, PyDOE and RoDeO within the source files work for your setup.

TL;DR

Requirements: SU2, RoDeO, Armadillo, PyDOE a SU2 compatible config file and mesh for your model.
Compile model_order_reduction.hpp, model_training.cpp, model_evaluation.cpp and model_order_decompression
Surrogate model creation:

python acquire_samples.py -f [CONFIG.cfg] -p [SAMPLE_DIR] -s [SAMPLENR]
mkdir [REDUCED_MODEL_DIR]
model_order_reduction -i[SAMPLE_DIR] -o[REDUCED_MODEL_DIR] -n[MODECOUNT]
python prepare_training_data.py -f [REDUCED_MODEL_DIR]/PODCoefficients.txt -p [SAMPLE_DIR] -o [TRAINING_DIR]
mkdir [KRIGING_DIR]
model_training -i[TRAINING_DIR] -o[KRIGING_DIR] -n[MODECOUNT] -i[ITERATIONS] -r[REGPARAM]

Surrogate model evaluation (INPUT.txt containing line separated DV_VALUES)

model_evaluation -i[INPUT.txt] -m[KRIGIN_DIR] -p[REDUCED_MODEL_DIR] -o[FLOW_VECTOR.txt]
python  -i [FLOW_VECTOR.txt] -t [SAMPLE_DIR]/flow_file_template.vtk -o [OUTPUT_FLOWFILE.vtk]