[aerosayan]

I've looking at block structured AMR capable codes like AMRex, CHOMBO, SAMRAI.


Example : https://amrex-codes.github.io/amrex/gallery.html



Based on my rudimentary understanding of these codes, it looks like they don't generate structured grids that conform to the body (aerofoil, rocket etc).


They don't have to. That's one of the benefits of using these codes. Since some of them were designed with the intention of solving exascale simulations of stuff like supernova explosions, black hole formation etc etc, it makes sense why the generated grid doesn't need to conform to any body.


It's not to say that the codes won't be able to solve for flows over bodies. They can. As per my rudimentary understanding, they use the Immersed Boundary Method to basically represented the boundary conditions near the body.


However, in the world of viscous flow with boundary layer interactions, it would be beneficial to generate a structured grid which conforms to the body, in order to better resolve the boundary layer flow.

At the same time, it would be really beneficial to have the same local block structured refinement capabilities of the previously mentioned codes.


The expected result is shown in the attached picture.



The thing is, I have not been able to find any CFD code that does both. (Most probably because I don't know where to look.)



I only found one paper by E. Steinthorsson, David Modiano and Phillip Colella. "Computations of Unsteady Viscous Compressible Flows Using Adaptive Mesh Refinement in Curvilinear Body-Fitted Grid Systems" - NASA.


The attached picture was taken from the paper mentioned above. And we can see that a body conforming structured AMR code would be very helpful in transient simulations of aerofoils, turbine blade cascades, CD nozzles, and other applications of structured grids.


Anyone know of any paper and/or codes developed using such a body conforming block structured AMR algorithm?

Thanks
sayan

[sbaffini]

Maybe I misunderstand your question but, basically, any serious unstructured CFD code will do it. The structured starting grid is just a special case.

[praveen]

Have you seen http://www.overtureframework.org ?

[aerosayan]

Originally Posted by sbaffini

Maybe I misunderstand your question but, basically, any serious unstructured CFD code will do it. The structured starting grid is just a special case.

I haven't been able to find such a software.

The big name softwares using block structured AMR techniques are based on Dr. Phillip's work in LBNL. Well most of them are from LBNL.

I looked at Dr. Philip's website, and most of his work from early 1990s to early 2000s, is based on the block structured AMR techniques with Immersed/Embedded Boundary Method (I/EBM)


https://crd.lbl.gov/departments/comp...hillip-colella


I could only find one paper in this period (the one I included above) which was about using a body-fitted grid as the base mesh and not using EBM.


EBM is quite powerful for simulations which have moving boundaries. However the performance cost (I don't know enough to be able to optimize it correctly) and development cost (time, in my case) would be quite high if I was to try and use EBM to create a small 2D Unsteady Euler/NS CFD solver.



The body-fitted structured grid version has the benefit of not requiring high refinement boxes to capture the boundary. As shown in the attached gif, the EBM based method of AMReX has to use a lot of refinement boxes to capture the body. All of those refinement boxes and computational resources could have been spent in resolving the shocks and turbulence instead.

[aerosayan]

Originally Posted by praveen

Have you seen http://www.overtureframework.org ?

I will take a look. Thanks!

[sbaffini]

Ok, if you really want a block structured code, then I don't know (besides overture mentioned above). But any decent unstructured finite volume code has AMR.

[aerosayan]

Originally Posted by praveen

Have you seen http://www.overtureframework.org ?

I checked the solver. It's really good for representing the boundary layers. They seem to be using overlapping/overset/chimera grids to create their domain, and their AMR method is somewhat similar to that used by AMReX.


NASA's CFL3D also allows creating multiple refiened grid regions and to embed them into regions of high gradients. It is static though.


I like CFL3D for its simplicity, but the mathematics used in overture will be useful.


Thanks