Hi,

my original incompressible fractional step NS code is used for simulating moving bodies with c-grid.

after converting to overset grid, ie background cartesian + original c-grid, i found 1 problem.

in the cl,cd graph vs time, there are some spikes in the graphs. on investigation, it is found that the pressure on the surface changes quite a lot before and after the spike occurs e.g. -0.4 to 8. usually from 1 time step to another, the change is around 20 to 100%. Since this is an imcompressible code, only the difference in pressure matters. But this sudden change is still not normal.

i've also used tecpot to calculate the cl,cd and the same spike happens. the connection between c-grid and cartesian grid is done using bilinear or biquadratic interpolation.

i've found another strange thing too. if i fixed the time step to 1e-4 and the spike occurs at time=.522, when i change the time step to 0.9e-4, the spike will not occur at time=.522, but at another time.

the velocity vectors all seems normal and the viscous force calculation has no such problem. changing Re from 100 to 1000 to 10000 did not make a difference. moreover, if the body remains stationary, there's no such problem e.g. vortex shedding of a cylinder. the BC used is dp/dn=0 on surface, while at far field, p is fixed at 0 or dp/dn=0. this has not caused any problem in the past.

anyone has experience this before? i'm still wondering what's the cause....

thanks for reading.

Hi Ben,

Ive observed this while doing some 3d simulations on overset grids for a rotating wing. Im also not quite sure why this happens. Similar to your observation, I tried to change the initial position of the wing and rotate. The spike dissapeared. But if i place the wing initially at a position just before the spike occurs (in the earlier case), it comes on instantly.. So i guess its got to do with the condition of the grid at that instant of time..

Did you try to refine the grid. It sometimes works that way. It dissapeared by refining. Does it happen for low amplitude flapping motion as well?

Regards, Dominic

If you have blanked cells that are moving from a hole region to a region where they become active in the flowfield, then you need to make sure those points have reasonable values before they are uncovered. If they are initially set to freestream and then blanked, these points will typically carry the freestream initial conditions into the flowfield as they are uncovered. Could this be your problem?

Hi, I've already done that. I have done an interpolation using the other grid values for those cells which changed status.

I also realised that at Re=100, for the same number of pts, if my nearest pt is 0.01 units away from the boundary, there is no such problem. however, if i reduced the distance to 0.001, spikes start appearing.

the original code with just the c-grid does not 've this problem, even with smaller boundary distance. i've no clue why this is happening. it also seems that if the spikes are removed, the result 'll be the correct graph, although i can't be exactly sure since too many spikes makes visualization difficult.

What is usually done is to set the hole points to a running average of the neighboring cells even before they emerge into the flowfield - this provides a smoother emergence into the flow field. Don't know if that's what you're currently doing. It might help if not.

Hi ag, with regard to the background cartesian hole points, how do u usually deal with it?

For the momentum eqn, I merely force u=v=0. As for the pressure poisson eqn, I either fixed the value, or simply do not impose any condition. There's not much difference in answer though.

I'll try to use your suggestion. btw, do u know of any papers which deals with such problems? I have only managed to get 1 or 2 aiaa papers. it seems that overset grid is not very popular nowadays. moreover, most author rarely touch on such issues.

thanks!

Most of my experience is using compressible solvers - hole points are updated each time step by averaging from the grid points surrounding them. Thus, if a hole point is initialized to freestream, and the points immediately around are also at freestream (say all hole points) then the new value will still be freestream. If the neighbor points are at some other value, then the average will reflect this. Then as the point moves into the flowfield, its value undergoes a smoother change and introduces a smoother change to the flowfield calculation. Overset is actually very popular nowadays - look for papers from the most recent Overset Grid Symposium (google "Overset Grid Symposium") for some examples.