Hi,

I`m working on a spray simulation of a quench-hardening process. Water ist atomized by a twophase spray nozzle and hits a hot steel plate. Interest is at the heat transfer that should be compared to experimental data.

Steady state simulation is working so far, but I`ve some problems with the unsteady approach. The calculations runs fine for a long time and then suddenly within a timestep it diverges or crashes. Maybe someone has some experience with sprays and can spot a important error in my setup.

I`m using a 3d-hex-mesh, all faces are map-type and only a small number of cells has a equiskew of 0.6. The mesh is build as a wide cylinder, with a small rectangle as core. The injector (DPM model, air assisted atomizer, water) is placed 0.5 mm in front of a mass-flow inlet (in line with the "axis" of the cylinder), 100mm above the bottem of the cylinder, which is a 3mm high zone of solid and represents the steel plate. The top of the cylinder is a mass-flow inlet (low speed) and the side is defined as pressure outlet. Backflow is occuring, but I guess it`s not relevant because it relatively far away from the spray.

Flow is compressible, all (but pressure) is solved in 2nd order, unsteady second implicit is used, spray breakup/collision and dynamic drag is activated, viscious model is ke-realizable with enhanced wall treatment. Air speed at the inlet behind the injector is ~ 180m/s, the plate has a temperature of 870K. For experimental data a twophase-spray-nozzle with inner-mixture was used. With fluent you are modeling a outer-mixture nozzle if I got it right. I haven`t found any common values for a air speed in these nozzles, any ideas what usually used?

I was unsure about modeling the grid in the boundary layer at the plate. I think I used 30 cells within 3 mm, the first node at 0.075 mm. Any guidelines for BL in wall-spray-interaction setups?

I think my mesh(es) are fine enough, one has 900K cells, another 1.5M cells and finally I build a third mesh with 950K cells, which represents only a quarter of the cylinder and uses periodic boundaries.

The spray model "wall-jet" was used most times. The "wall-film" model caused regular programm crashes ... these with messages point 1-4 . But a restart doesnt help Wink.

Timestep size was another question. With 1e-04 sec it`s working, 1e-05 may work better, but it crashes/diverges too ... as with 1e-06. I would like to simulate at least 2 seconds Sad.

The most frequent error is "Floating point error: invalid number" ... calculating in double precision didn`t make a noticable difference. Roughly, in the iteration before the crash one variable "explodes". Temperature, pressure or viscosity ratio ... the number of limited cells jumps up to ~60.000 cells. Often the number of limited cells is growing very slowly in a few iterations before a crash, but there is alway this huge jump

I usually load significant older data and re-runs it with altered URF-settings and this often helps ... but that`s not useful.

Sometimes I thinks it`s primarily a problem in the DPM-fluid-wall interaction in the boundary layer, but I`ve no proof.

Thanks for reading this rather long story Wink ... if someone has any comments/tips/ideas ... feel free to reply. Thanks!

Mathias

I will be simulating spray in a day or so...

I know DPM is a great model, but I would rather start with VOF model, because literature says that if your droplet size is smaller than the spacing between droplet i.e your spray is dense, which might be your case. but it is still suggested to start with VOF and then go to DPM.

DPM is more a higher resolution thing.

Enhanced wall treatment tend to complicate things a little, did you try switching that off.

For spray drying, WAVE model is more suitable with injections ( give some parcel number).

Maybe I need some help i future from you, I dont think there are many people actively do sprays using CFD.

Thanks