Learning Forum

[vedama]

I have a simple 2D case that I'm running grid adaption on. The image below is a zoomed in image near the inlet.

This next image is the set of marked cells in that same region. Note how the first "column" of cells are not being marked.

However, when I adapt, cells that aren't being marked are being adapted, as in the image below. Look at the first "column" again and compare the above 2 images with the image below.

Why is this happening? This is causing divergence issues. The solution is blowing up and giving garbage values.

[Rob]

I'm slightly intriqued as to how you've got such a mix of quad & tri cells, but that shouldn't be an issue. 

What are you using for the adaption criterion? 

[vedama]

I'm using gradient and curvature of density across multiple iterations. This is the 6th cycle of adaption. I have max refinement level set to 5

[Rob]

Which may mean you've got holes in the register because that's what the gradients say. If you look at the contours how do they look (before it fails)?

[vedama]

What do you mean by "holes in the register"?

The contours don't look great and that was going to be my next question. Why is there so much variation in the first cell of the inlet?

[vedama]

Just to be clear, it is a high-speed flow case so the contours should look clean and smooth, especially at the inlet. I'm also using second-order upwind scheme and AUSM inviscid flux type so really, everything should be going downstream. I have a really fine un-adapted case that reflects this accurately.

[Rob]

I have no idea re the inlet as I have no idea what is set. Or what you're modelling. 

Registers of gradient will trigger only when that criterion is met. So, if you have a small patch with lower gradient you may have a "hole" in the register. That then causes odd mesh adaption. If you've not set the coasening criterion (assuming a moving shock) the it's anybody's guess how the mesh will finish up. 

[vedama]

I do not have a coarsening criterion currently.

I'm trying to understand your point about holes just so I can get a better understanding of the functionality. My current register looks for density gradients/curvatures in the solution and marks cells for adaption based on a specific set of minimum threshold values. If the cell values are above this threshold value, they should adapt. If the cell values are lower as stated/shown by the marked cells, what exactly is happening to cause the cell to be adapted?

[Rob]

OK, re the gradient. Let's assume a chess board. We'll adapt for greyscale gradient, so only pick the cells either side of the boundary. That should pick up all cells around the black square, but may omit the corners as they're diagonal to another black square. If we repeat, then we'll just pick the smaller cells on the black-white boundary. However, in your case the solution is evolving so the position of the higher gradients moves, and if the gradient is better resolved in the previous steps it's value is now low enough to not be picked. The register is doing exactly what you told it, hence plotting the values to see what's going on. 

[vedama]

Right, but according to my criterion, the first column of cells is low enough that it is not being marked. But it's still being adapted, and you said it's because of a "hole" in the register. How does the hole relate to the chessboard analogy?

[Rob]

The lowest cells aren't being marked - there's a gap to the grey line on the image. 

[vedama]

 

I’m referring to the first column of cells coming off the inlet, highlighted below. Not the first anisotropic layer.

 

[Rob]

Only one cell's been picked up. That's due to the adjacent adaption, you can't have multiple adaption layers next to unadapted mesh. 

[vedama]

Ok yes that makes sense. Thank you!