TUNNEL_PRECONDITIONER for multiple meshes with different y/z

[Helllink]

Since the pressure term for tunnels is a function of x, does someone has an idea to adjust the poisson equations to a tunnel that has different cross sectional areas and thus is build with multiple meshes with different size in y/z-direction. Of course the tunnel is still closed on all y/z and abutting surfaces between meshes with different size.

[mcgratta]

The TUNNEL_PRECONDITIONER is not designed for this. The algorithm is based on the assumption that the pressure profile in the axial direction is approximately one-dimensional. If that assumption is violated, the approximation is no longer valid.

Of course, you can try a simple test case, but my guess is that it will fail.

[Helllink]

I have to corret myself. there are differences in stability between the to options when it comes to higher velocities.

This brings me to the first question, if the equations of the tunnel preconditioner can be adjusted to respect the different cross sectional areas. Since the flow is the same for both options and approximately one-dimensional.

[mcgratta]

There are several examples in the FDS Validation Guide where the TUNNEL_PRECONDITIONER is used for a tunnel that has obstructions and a non-rectangular cross section, like Memorial Tunnel. However, I have never tried a case where the y and z coordinates of the meshes are not the same. If we consider FDS 6.10.0, which method runs faster, the one where you have different mesh cross sections or the one where you use OBSTs to block off a volume near the ceiling, floor, or walls?

[Helllink]

I performed some more tests. The option with the same size in y/z-direction and the use of OBST runs much faster in FDS 6.9.1 than 6.10.0. Did the handling of blocked cells changed between the two versions?

[mcgratta]

We found an error in 6.9.1. Even though it runs faster, it can lead to significant errors in the pressure.

[Helllink]

Thank you for the information. This explains a lot to me. For my case the computational time is doubled between 6.9.1 and 6.10.0. That's why I tried to optimize.

[mcgratta]

At some point, maybe between 6.9.1 and 6.10.0 we added an error statement if the meshes did not align. This was done because the method does not converge. You can certainly use non-aligned meshes without TUNNEL_PRECONDITTIONER.