Using SU2's Adjoint Optimization For Inverse Problem

[Gaurav11ME]

I am exploring the use of SU2's adjoint capability for solving an inverse problem. I wish to know if it is possible to use this to optimize a spatially varying parameter (e.g., spatially varying viscosity). I have measured values of the velocity field in the flow domain, and below is the process I wish to follow for the inference.

1. Guess viscosity values at grid points.
2. Run a RANS solution
3. Compare the RANS velocity field with the measured field
4. Use loss function = (u_rans-u_measure)^2
5. Use adjoint optimization to get the updated viscosity field
6. Repeat 2-5 until convergence (Loss<set_threshold)

[pcarruscag]

Yes, this is possible, but you would need to write code for:
Reading the viscosity distribution from somewhere
Using it in the solver
Computing the new objective function
Extracting derivatives wrt new variables

You can have a look at CElasticitySolver, it does similar actions to what I described.

[Gaurav11ME]

Thank you for the quick response. I have figured out how to add a configuration option for reading data from a file. I will check CElasticitySolver.

[Gaurav11ME]

I could not locate CElasticitySolver file. I found CFEASolver. Are they the same?

[bigfooted]

I recommend working from Visual Studio Code , and then search for "Elasticity" to see files associated with this solver, for instance the files in the directory src/numerics/elasticity

[Gaurav11ME]

Ok. I will try that.