[Numericer]

My computational domain is an annular axisymmetric cylinder with an angled jet like inlet and an open outlet. I've used wedge type boundary with an angle of one degree. Please find attached below the mesh.

The domain represents a bluff-body burner. The flow, a mixture of ethylene and air, enters the domain at about 20 degrees about the axis. I am running the simulation with reactingFoam with combustion turned off, to validate the cold flow before moving on to hot flow.

Initially, I used the k-epsilon model for the cold and the hot flow. The simulation ran fine but the results weren't satisfactory as the model doesn't capture vortex shedding. Thus I switched to Reynold Stress Model LRR.

I've since been experiencing quite an unusual issue. The simulation runs well for the base case. To perform the grid independent study, I refined the grid 1.3x-2x and right when the flow looks like it is about to reach steady state, out of nowhere the epsilon values blow up at a wall below inlet (named bluffBody). After searching through the OpenFOAM forum, I've done the following to resolve this issue:

1. use upwind schemes for turbulent quantities
2. ensure no issue with Courant number; I've set it to a maximum value of 0.8; I've tried values of 0.5, 0.6 as well
3. use different blending function for epsilon: binomial
4. change inlet epsilon/R values: sometimes the case crashes earlier and sometimes later

As a side note, I've also observed that in order to continue from the latestTime, if I freshly decompose the domain for parallel run, the case crashes. While continuing from previously decomposed files, the case runs fine. These issues appear to be related to LRR model. I haven't tested SSG.

I've attached below the case directories for both the base case and the 1.3x refined case. Please help me to resolve these issues.

[dlahaye]

Lower relaxation factors?

See also: simpleFoam tutorial PitzDaily using Reynolds stress tensor (LRR RASModel)

[Numericer]

The issue was with Courant number. I’m having to set maxCo to about 0.4~0.5 for bounded solution.

Incidentally, this also fixed problems I was having with decomposePar.