Efficient Inverse Design Optimization through Multi-fidelity Simulations, Machine Learning, and Boundary Refinement Strategies
Data for the manuscript "Efficient Inverse Design Optimization through Multi-fidelity Simulations, Machine Learning, and Boundary Refinement Strategies" by Luka Grbcic (LBNL), Juliane Müller (NREL), and Wibe Albert de Jong (LBNL), 2024.
LBNL - Lawrence Berkeley National Laboratory
NREL - National Renewable Energy Laboratory
DATA DESCRIPTION -- Airfoil Inverse Design (AID)
AID_Cp_targets -- containts pressure coefficient target values for the RAE2822 and NACA2410 airfoils. First column of both files are the normalized x coordinates of the airfoil geometry, while the second are the Cp values.
AID_ML_model_data -- containts the data needed to train the ML models. Each row in the x.txt file are the B-Spline coefficients that represent an airfoil shape. Cpmin.txt is the file that contains the minimum pressure coefficient for each shape in x.txt. B-spline coefficients are the inputs to train the ML model, while the Cpmin values are the output.
DATA DESCRIPTION -- Scalar Field Reconstruction (SFR)
SFR_s_targets -- containts the scalar target values for the Sinusoidal (Design Case 1) and Linear (Design Case 2) BC cases. The files contain 30 scalar values measured in the domain as described in the manuscript. Specific points are given in the Appendix of the manuscript, and in the OpenFOAM case files.
SFR_ML_model_data -- containts the data needed to train the ML models. Each row in the x_s.txt file are the scalar values at the boundary that map to the maximum scalar values measured in the domain, stored in the file s_max.txt. x_s are the inputs to train the ML model, while the s_max values are the output.
SFR_OpenFOAM_cases -- containts the OpenFOAM 9 cases for the Sinusoidal (target case 1) and Linear (target case 2) BC cases. LaplacianFoam is the used solver (the diffusion partial differential equation is solved). The blockMesh can be used to remake the domain and HF mesh. The probes1 function in the controlDict contains the coordinates in the domain where the scalar values are measured.