Hi,
Anyone here ever looked into fitting a curve to transient thermal simulation data to extract an R-C network?
I’m wondering if much of it can be simplified by forcing the point extraction at the minimum needed resolution so filtering is not required before curve fitting. Just curious if this is something that has been looked into already.
Thanks,
Do you mean simulating some electronic components and calculating their thermal resistances and capacitances by thermal impedance curve fitting in order to obtain a 1D electrical circuite-like model ? For that, do you need temperature variation in time for selected locations ? This could be done with a history output request. Resolution can be controlled with proper incrementation settings, output frequency (1 = every increment) or time points (can be added via keywords as explained in your previous thread).
Yes, similar. I don’t think they typically do data extraction at specific points mainly because a real world system can’t measure temperature in between layers of the stack.
Typically you get temperature vs time in the 1us to 100s range. Divide by power in to get impedance. Convert to log domain and filter data for noise/fitting simplification. Do some math to find potential inflection points. Fit a curve of multiple step functions to extract Foster RC nodes and then covert to Cauer if needed.
You can make a Cauer RC based solely on material properties but I’ve found the thermal spread used in the calculation changes in the results a lot. So, I wanted to try simulation based Foster RC and compared to Cauer ideal. Just hoping someone had already tried this and if it was something that could be built-in or something like a script/macro.
Maybe history outputs at each interface and separate curve fits for each section would get pretty close?
Yeah, history outputs in several locations (you will get a table/curve for each node). Then in the newest version of PrePoMax you can do some basic math operations on those results or you can export them to ParaView, spreadsheet or some numerical computing package like Scilab for more complex calculations. Such advanced operations on the results are normally done outside of FEA software (even when working with Abaqus).