Thermal analysis of electronic system 3

[elogesh]

Hi,

This is regarding the thermal analysis of electronic system.

Now a days,the application of electronic components in all industry is steadily increasing.But they should be robust enough to withstand the vibration and thermal loads.

We are trying to carryout the CFD analysis of electronic system. Earlier, we outsourced this project to an external agency. But the results were deviating considerably from the test results.They executed the project using Flowtherm. They represented all cylindrical components (Cylindrical capacitor, cylindrical transfomer) as rectangular, due to limitations in the software. I hope this resulted in different flow characteristics, then observed in reality.
Now, We are trying to carryout the analysis using industry standard CFD-Fluent software.But the software requires material properties for PCB board (FR4+ Copper), capacitor (Electrolytic), transformer, resistor and inductor needs to be specified. But these are not made of single material and the properties are not readily avialble. can we suggest anything rational about the consideration of material properties for the thermal analysis.
I look forward for your help.

Regards,
Logesh.E

 

[Transient1]

Are you looking at just a single board with components or an entire system? For board level work, I really like ANSYS TasPCB. Block components compared to cylinders will change the heat transfer a bit, but I suspect not by much. I would hope that the models the consulting company used at least had equivalent surface area. What kind of deviations are you getting?

The properties for most components can be found at

http://www.alldatasheet.com

 

[ko99]

elogesh,

Unfortunatly, modeling electronics in CFD is not an exact science -- it's still an engineering task. The right amount of detail to add to a particular component depends on the concern you have that it may overheat or cause ther parts to overheat.

1. Some components that do not generate heat, like capacitors, can be modeled with a bulk conductivity of about 5 W/mK. You are correct that the actual construction is much more complicated but the cap temperature will not be sensitive to those details. Cap temp will be primarily dictated by the local air temp and secondarily by the PCB temp.

2. You can also greatly simplify the material of most resistors. Yes, they generate heat but typically you want to know how that heat effects other components, the sensitive ones. You rarely need to know the temperature of the resistor element.

3. I agree that sometimes round components, like capacitors and resistors, should not be modeled as rectangular blocks. This is important if there is forced air and nearby heat-generating components. No CFD mesh perfectly fits a curved shape, but I know Flotherm is capable of modeling capacitors as near-round objects (about as round as any CFD program can handle) and I'm a little surprised your consultant didn't model them as Cylinder Objects.

4. For PCBs, use FR4 conductivity of 0.3 W/mK and Cu of 360 W/mK. There are short-cuts for simplifying the layers and vias so you're not bogging down your model with 0.04mm details. Let me know if you want some tips.

5. I have modeled transformers and inductors in great detail, but it's not a trivial task no matter what software you use. You have to get detailed construction info from the manufacturer and/or take the component apart.



ko (

http://www.ecooling.biz)

 

[IRstuff]

Unless the component is making a substantial impact on the air flow, whether it's rectangular or cylinder is probably irrelevant, so long as it's drag and turbulence effects are the same

TTFN

Eng-Tips Policies FAQ731-376

 

[tkordyban]

Getting the shapes of the components correct and using accurate material properties can be important (especially thermal conductivity of the board, which is a complex topic), but in my experience the most important contributor to differences between CFD and test results is the power dissipation. Component power is generally not known to within plus or minus 50%, and yet you expect to get component temperature predictions within plus or minus 5%. In heat tranfer, the temperature rise is directly proportional to the power dissipation. If you guess the power 50% off, the temperature rise above ambient will be 50% off, too.

 

[ko99]

Yes, that's exactly what I see. You model everything very carefully but when you measure the actual components, they are often cooler than expected. Typically this is due to an overestimate of power or a lack of the proper software to exersize the components. Revise the model to reflect measured power and the deltaT's are often within 5-10%.

The next largest error seems to occur when case-air convection is poor. Conduction from die to leads to pcb to air becomes very important, as well as radiation. So your model should reflect this (easier said than done...)

In models with good case-air, which usually means high power density, the interface losses can become more dominant and should be modeled more carefully. Don't just use bulk conductivity and an average TIM thickness. Contact force, surface finish, TIM wetting factor, etc, play a larger role in the accuracy of the model.


ko (

http://www.ecooling.biz)

 

[elogesh]

Dear KO,

Thanks for your comments.

It has taken a quite bit of time to respond.

Regarding your first response

1) We observe similar kind of behaviour in the test.

2) Regarding the resistors, we observe slightly higher temperatures compared to other components in the forced convection conditions.

3) Yes.We understtod that Flowtherm can indeed include rounded objects.

4)There are short-cuts for simplifying the layers and vias so you're not bogging down your model with 0.04mm details. Let me know if you want some tips.

We will be definitely enriched with your tips. Advanced thanks for the tips.

Your tips based on the expertise and experience will be definite help practising young engineers like me...

5) Do you think that using fluent, even I can capture intricate details, which is already inbuilt in softwares like Flowtherm...?

I was under the impression that using Fluent may lead to greater insight into the details compared to the Flowtherm but at the cost of turnaround time. can We have your comments.

I am also equally thankful to tkordbyan, we are not quite sure with the power dissipation in most of the cases.
As we start acquiring data from simulation and test, hope I will be in better position to feed proper inputs.

Thanks once again transient1, K0,IR stuff,tkordbyan.

We have the similar uncertainities in the structural analysis of these electronic components.

Regards,
elogesh