[hebeldebel]

Hi,
I have a simulation which shows deviations from the measured temperature depending on the ambient temperature, in run it for.

e.g.
ambient temperature: 10 °C => deviation 11 K
ambient temperature: 20°C => deviation 10 K
ambient temperature: 40°C => deviation 6 K
ambient temperature: 60°C => deviation 0.7 K

I tried increasing the number of cells from 3 Mio. to 6 Mio.
=> does not converge with k-epsilon, with sst- menter only

I tried to set the heat transfer coefficients (on the outer boundaries of the simulation), depending upon the temperatur.
=> helps a little but not much

I tried to increase and decrease two contact resistances I use to simulate thermal adhesive.
=> does not change the fact, that the deviation changes with the ambient temperature

Switching from k-epsilon to sst-menter with 3 Mio. cells and ally+ option decreases the deviation a little, but has no effect with 6 Mio. cells.

Does this make sense to anyone? Any ideas or hints?

[ggulgulia]

It will be more helpful if you could tell what exactly you are trying to simulate, how many regions you have, have you considered radiation as a possible mode of heat transfer, where did you get the convection coefficient from, etc.

[hebeldebel]

Of course:

I am simulating an LED on a heatsink within a Lexan-box, which is kept inside a climate cabinet with a constant temperature. The simulation does not include the climate cabinet, nor the lexan box, but only the fluid and solid regions inside it. I measured the flow velocity outside the lexan box and calculated an average alpha value for each side using the equation for a flat plane (as approximation). I use the surface-to-surface radiation model for radiation simulation. Each boundary representing the lexan walls in reality has got a transmission value of 0.9 (IR transmission from lexan datasheets) and the heat transfer coefficient value I calculated from the velocity measurement. The electrical power of the led has been measured and the light output has been calculated from the datasheet and with SPEOS data. The difference was used as basis for the volumetric heat source (phosphor of the LED Module). There is a contact resistance between the LED-Modul and the heat sink representing the thermal compound. All surfaces of the LED and heat sink show wall y+ values between 0.0008 to 0.04 . k-omega sst menter with ally+ treatment was used and gamma re theta with this field function for the free stream edge: "$WallDistance > 0.005?1:0" (maybe there is something wrong with this?).

[ggulgulia]

I do not know much about LEXAN box. so i assume you are using Multiband Readiative Heat Transfer physics model? If you do, you also need transmissivity of the surface of LED.

You can also refer to a similar article published in Dynamics (CD-adapco's quarterly magzine) on their website. As far as I remember, the problem was solved by coupling STAR CCM+ with spice.

[ggulgulia]

In case you are using radiation physics model, you need to select multiband radiation model.

[hebeldebel]

I am using gray thermal radiation, because the power of the visible light output has already been removed from my equation to set the heat source power (the data from speos) and the lexan/makrolon/polycarbonate box has 90% transmission for visible light.

[hebeldebel]

Some more information:
The temperature readings of my measurements show, that for every 10 K I increase the ambient temperature, the LED temperature increases by 10 K as well. when I look at my simulation, I can see, that the temperature increases by 8 K for every ambient temperature increase of 10 K.

The simulation is about 10 K hotter than the measurement for 10 °C ambient temperature and 2 K colder for 60°C ambient temperature.

[hebeldebel]

I looked at the gray thermal radiation temperature, which was set to 333K and set it to the ambient temperature for each simulation, that seems to have decreased the deviation a lot, it went down to 0,4K increase in deviation for each 10K increase of ambient temperature.

I looked at the y+ values of the heat sink varying from 1 to 30, which seems to be just in the wrong range (using all y+ treatment).

I also retrieved the specified y+ heat transfer coefficient on the fluid interface to the heat sink and saw the average increase with the ambient temperature decrease.

Can inaccurate meshing lead to an increase in heat transfer coefficient, the lower the ambient temperature is?