Simulation doesn't meet reality
Firstly, please allow me to introduce myself, this is my first thread. My name is Hans, located in Austria and I’m working in a small engineering office, I shall be in charge of thermal issues.
I came across to this forum searching the internet for Flotherm issues….
So, here I am and I have a basic problem as it seems.
Yes, I am a newbie. Never worked with CFD so far and I am certainly missing the basics. I hope you guys may direct me the right way :)
I did all the tutorials on Flotherm and my calculation results match their results. In the meantime I would like to add "of course" - as I used same settings (e.g. grids) they recommended.
After that I had the idea to verify (or validate) my simulation results … so I took a resistor (28,5W) mounted it on an Al plate (234x81x2,5) and here my problem already starts. I still keep wondering why I measure different results than Flotherm predicts. I actually expect that this may look very ridiculous to you, a simple plate and a resistor don't match, hehe.
Well, Tutorial 1 is actually the same setup, a plate + a resistor, so I was expecting comparable results of let's say +/- 1°K but Flotherm predicts about 30% too much.
So, I thought that my Flotherm model was wrong. So I started playing with the grids and observed that Flotherm results changed substantially with the number of the grids, which confused me. I was expecting "press the button FINE" and here is a precise prediction, as I was used from SPICE (electronics), but no, Flotherm's results changed even up to 5..6°K by that? That confuses me, as I have no clear guidance on what is a sufficient grid in order to get a +/-1C result?
In the mean time it looks to me that a grid of not more than 1 or 2 mm side length covering the Al plate would give a reliable prediction. My computer takes ages to do simulation with 500.000 grids, ages means in this case up to 90 minutes. And despite the fact that the mesh is so fine I get surprisingly often “the solution did not converge”. Sometimes this happens if I change the power dissipation of the resistor only…?
Q1: why do I get a “converge” problem when the grid is as fine?
Q2: what do you estimate, what calculating time is normal for such simply model and are about 300.000 grids something which sounds OK for you? I am a bit unsure; maybe I have a problem with my windows installation? I run a 2,68 GHz Quad Core I7 with 5GB RAM.
Q3: the only restriction I know about the solving room is that there should be at least two grids between the ambient boundaries and the simulation object. On the other side it is mentioned that a bigger simulation cuboid would be helpful if we do natural convection. So, what would you recommend from practice?
Q4. What regards the residual error: if I have a NOT converging solution with a residual error of 2 (shown in the profiles window graph) what does that mean to me? Is a none converged solution a reliable prediction or could it be totally wrong. The Flotherm help says residual error is 0.5% of fixed heat source”. Well, in my case I use 28,5W, so does that mean that I should consider an uncertainty of 2x0,5% of the max temperature of the resistor (hot spot)? Why is the Y axis going over 200 when 200*0,5% = 100%, what is an error >100%?
I believe some practical guidance from you is what I need at a most.
Overall I spent now more than 10 days playing around with Flotherm … frankly speaking I need help as I am stuck, I did everything what I could.
Due to upload file size restrictions I can't upload any files herein, however I would be happy to send them by email to one of you experts for debugging.
And one final question: is there any book you guys recommend to me? I was searching on Amazon but with low success.
Simulation doesn't meet reality
First of all I'd like to suggest that you contact your local support office who have a lot of experience in getting people up and running with FloTHERM. Also, as with any simulation tool, training is advised.
In terms of obtaining a 'grid independent' solution we always advise that you run a few models, refining the grid until such time as the results you are interested do not change. Experience that you will gain will lead you to appreciate that there is a need to add grid cells where there are high temperature gradients, where there are low gradients you can have a coarse grid. For you application attach a grid constraint to the plate and define an inflation of say 3mm on both air sides with say about 5 cells. Same for the resistor. Localize the assembly containing the plate+resistor to keep those small cells close to the assembly.
For this application I would think that 50,000 cells, well placed, should be enough. A typical solution should take minutes.
For this natural convection application of course make sure that radiation is on and you have radiation attributes attached to the plate and resistor.
Define a monitor point(s) in areas of interest. In the profiles window, when those monitor points do not change with subsequent iterations (usually when the residuals <10) then you model is converged enough.
How is the resistor attached to the plate? Are you modelling any interface resistance between the two correctly?
For background info on the use of FloTHERM feel free to check out my blog (I'm not allowed to post the url here, just google "Robin Bornoff's Blog"). Again, I strongly advise you to contact your local support office, spending 10 days getting stuck is never a good thing!
|All times are GMT -4. The time now is 05:16.|