[baazar]

To All,

I'm going to post this on Solidworks forum as well but I figured I should try here too.
We recently purchased Floworks and are using it to thermally model an electronics project. Before we start using it for any projections on this design, we're testing its accuracy on small experiments.

Right now I've got a 3 Ω resistor burning 3 Watts attached to a heatsink. I've entered heat generation, thermal resistance (measured and calculated), ambient air conditions etc.
I'm quite sure I have all of the initial conditions as close to reality as possible.
Unfortunately I cannot get the results to get any closer than 10°C below what I am physically measuring (Calc = 40°C, Meas = 30°C).

Have any of you guys gotten floworks to produce highly accurate results (i.e. ±2.5°C) or is it always off or is there some major variable that is easy to miss that would influence the results?

Thanks in advance - Bram

[Boris_M]

Hi Bram,
actually the software is used quite extensively by a lot of companies such as Bosch, Miele, Renault and Valeo etc.
There are many factors you have to consider and it is important as you said to have the measurement reproduced accurately.
For example:
Did you use radiation?
The right radiation properties?
Was the mesh fine enough in the heatsink fins?
Any contact resistance like TIM material used with the correct value (they can deviate to datasheets a lot depending on the vendor, thickness and pressure applied)?

It is hard to tell what your settings are and what your measurement model was like.
I did a paper recently with LEDs on a heat sink for ISAL 2013 where we used an IR camera and thermal characterization to get measurement data and compare them to the simulation. In one case we found also 11.5°C deviation between two LEDs in the IR camera and in the thermal characterization we saw then that the TIM was not good, probably an air bubble beneath it (it was a two-sided adhesive TIM). The result of the other cases were up to +/-3°C with the simulation. And when we adjusted the TIM thermal resistance for the one LED we got the same values. So sometimes it is not easy to get things right. The smallest inconsitancy such as this air bubble would have caused a 10°C between the measurements and the simulation.

I recently had a customer calling us where they also had issues with the temperature and I said I would have to see the model if everything was defined correctly but the model was confidential so an NDA would have taken some time. In the mean time they went back to check if the measurement and the simulation model were really equally designed and found that there was a small deviation in the measurement setup and after they corrected it the results matched perfectly.

Now in your case, go back and check if you really are able to model the resistor correctly in CAD with all the materials it consists of. For example there are some of the high power resistors that are made of the metal core and a kind of ceramic hull and then the metal enclosure with some of the ribs and mounting holes to mount it onto a heatsink. Do you have all the material properties of the resistor? A deviation in the heat conductivity can change the temperature already by 5-7°C and that would be then your too high deviation.

I hope this helps,
Boris

[baazar]

So I've been doing some experimentation and I've gotten much closer to real results.

Playing around with contact resistance (as measured), air-flow (as assumed), and turning on things like gravity and radiation calculations have all contributed to more accurate results.

Unfortunately the two most influential attributes are the contact resistance and air flow. Contact resistance must be measured first since nearly all real-life junctions are not found in tables, and air flow which is assumed because it is difficult to accurately measure CFM at the inlet.

I can play with these numbers and get drastically different results which isn't boasting well for the software providing a "max-temp."

It does however, tell me where the hot spots are and which configurations are relatively better which is a strong plus.

Boris, your information does seem to help for real life comparison but for building prototypes it would seem I'm still left using "best judgement."

- Bram

[Boris_M]

Hi Bram,
Of course there are parameters that influence the result accuracy a lot if they were just assumed but that is the case with every simulation software.
See the image below on differences of TIM properties between measurements and datasheets.
And for the air flow rate it is of course sensitive to the amount of CFM.

And again yes to the comparison of real life models. As long as there are too many variables since the model is a very early prototype, the measurement and simulation results can be far off. Often at this stage some parameters are rather vague and estimated can be off when comparing to measurements but as you say, they can give a good idea of the direction to go with the design decisions.
As soon as the design comes closer to the end and the variables are reduced and therefore less uncertainties are left in the setup, the results to match more and more to the measurements.
This is also the reason why fundamental design decisions at the initial draft stage are done by hand calculations or some spread sheets. The error is in the same order as simulations with only very limited information and rough geometry.

So it is always important to model the reality as accuratly as possible. Every deviation from that will increase the deviation between measurements and reality.

Boris