Intake manifold convergence

[oagreda]

LINK OF THE PROJECT: https://www.simscale.com/workbench/?pid=8941575546584074889&mi=spec%3A28d13863-0adc-43ff-a753-6099ef70b4ff%2Cservice%3ASIMULATION%2Cstrategy%3A331

Hi, I’m doing a transient analysis for an Intake manifold, I’m studying air performance in the intake for the cylinder heads in the fire order of 1 - 4 - 2 - 3. In the compressible 9 section, I did some transient analysis from ‘transient test 1 to 6’ and failed, maybe my mesh was to fine so I read that reducing the number of cells not to coarse will help, so It started to work for me. From ‘transient test 7 to 10’, each analysis stopped because the runtime over (community plan), but I achieved the pressure pulse of each cylinder and the velocity profile. I will show you my doubts.

Transient test 7:

1. I added 4 non-orthogonal correctors ( <75 degrees).
2. I changed the relaxation factor of the density field to 0.01 (I don’t know why some public projects for transient simulation add this value but I just copied ).
3. I changed the scheme of gradient to Celllimited Gauss linear 1) because I read in a OpenFoam tips that for standard mesh the LeastSquares generate oscilatory data).
4. The time step it’s low because I’m analyzing the engine performance at 5800 RPM and the valve opens and closes it’s fast.

Pressure pulse in the cylinder and inlet:

Note: The overshoot in the inlet pressure, I used Flow rate because I’m using that design parameter.

Velocity profile:

Note: Ends because the runtime it’s not the total end time (0.1289)

Turbulent Kinetic energy:

Note: Appeared to be increasing in time, not periodic.

Residuals:

Transient test 8:

CHANGES IN THE NUMERICS:

1. I changed the relaxation factors to increase convergence, p=0.4, U=0.8, ρ=0.01, h=0.8, e=0.8, k=0.8 and ω=0.8.
2. I write the initial condition for pressure: 142 kPa It’s the pressure for my pressure pulse.

Pressure pulse in the cylinder and inlet:

Note: The pressure in the inlet overshoot at the start but stabilizes faster.

Velocity profile:

Note: Ends in nearly the same time step, and the velocity data per cylinder increases the magnitude.

Turbulent kinetic energy:

Note: Improved the turbulent energy in the start.

Residuals:

Transient test 10:

CHANGES:

1. Initial pressure condition to 125 kPa.
2. I changed the relaxation factors to increase convergence, p=1, U=0.9, ρ=0.01, h=0.9, e=0.9, k=0.9 and ω=0.9.
3. I changed the divergence scheme in div(phi,U) to Gauss linear upwind limited grad.

Pressure pulse:

Note: Pressure inlet overshoot less.

Velocity profile:

Note: Ends in the same range.

Turbulent kinetic energy:

Note: Increases a lot and fluctuates a little beat in the peaks.

Residuals:

RESUME:
So, I don’t figure it out how to speed up the convergence to reach more time steps for mi end time, I tried to change the solver of velocity, enthalpy, internal energy, k and w to PBiCG but made it slower I canceled the run…
If you can help me to set better my numerics in the relaxation factor, the solvers and the schemes.

NOTE: In compressible 8, the steady state runs for Cyl 1 open N12 v7 to Cyl 4 open N12 v7, in average 1 are the solution for the cylinders, y changed a little bit the cad for compressible 9 but not too much.

[Ricardopg]

Hello and thank you for the detailed and thorough explanation.

We actually have a webinar on this exact topic (cost-effective transient analyses). The webinar goes through numerics and other good practices.

PS: please note that the runtime of transient analyses is heavily limited by the fineness of the mesh (due to the Courant number). A very clean CAD model and a good meshing strategy is absolutely necessary for cost-effective analyses. You will find some meshing pointers here.

Cheers

[oagreda]

Hi Ricardo, I followed your tips, and I made my mesh coarse but I don’t now if It’s too coarse (128.3 k cells, 157.6 k nodes), in transient test 14.

Residuals:

Velocity profile:

As an amateur in CFD, I don’t know if my results converged, I read that if you have residuals low to 1e-5 the results maybe converged but also you have to see if you have periodic results even though the residuals are not low.

[Ricardopg]

Well, there are quite some differences between transient and steady-state simulations, when it comes to convergence

For steady-state, most of the iterations (especially towards the beginning of the run), the simulation will not be converged. That’s why it’s important to evaluate if a steady-state simulation is converged when it ends.

For a transient simulation, the simulation should converge for every timestep. That’s why the correctors are used (as discussed in the webinar that I linked). In the webinar, the residual levels are also addressed - if you’d like to bring them down, then tweaking the relaxation factors would be a possibility. But this may also take more correctors to converge.

It may be a good idea to have a second look at the webinar

PS: periodic results are pretty common in transient analyses, especially when you have periodic inlets/outlets (like you do).

Cheers