I've been within 0.5 deg in some sims. But we spent a lot of time making sure that the settings exactly matched the physical.
Have you included radiation?
Have you done mesh convergence?
Have you considered that there will be some discrepancy between physical and numerical such that you have looked at trends to see that trends are correct and that the discrepancy may just be a constant to keep track of as an assumption between designs?
Brian, Getting the fidelity between simulation and what you have on the bench is a mission. Is the problem with the test or simulation or both?
Bench or Heat chamber consideration that you may think make no difference but do:
- Solar from windows
- The building HVAC moving air. Look a JEDEC specifications testing boards.
- The heat chamber fan
- Radiation from ancillary equipment
- The UUT is on a table or bench
- Proper ambient air temperature measurement
- I the component temperature method verifiable.
- Add radiation as mentioned - but what is the proper emissitivity and absorption?
- Add the bench or support features
- Add and air velocity match bench or heat chamber
- Add heat chamber walls and fan...
How dose the temperature compare with calculations using basic heat transfer?
You need to experiment with the lab test setup and the simulation until satisfied that both are robust. Tweak simulation until it matches the bench test. Then for the next equipment build you will have a process and confidence.
EIA/JESD51-2 or newer..
More or less it's a very rudimentary test. Resistor measured exactly 3.0 Ohm, Voltage set to exactly 3.0 V, Geometry approximately 98% identical. I seem to be getting closer results when using radiation but I'll have to figure out how to properly implement it.
I'm not sure what mesh convergence is but I have "Use for convergence" checked wherever possible.
It's also likely that there's some airflow that is influencing it as well. I'm happy to know that others have gotten accurate results which means I can keep fiddling with the settings to make it more accurate and I haven't hit a brick wall of computational limitations.
Thanks guys, I think I can try to figure it out from here. I'll start a new post if I need specific information for the test I'm running.