Skip to main content

Brent Cullimore

Being perturbed isn’t much fun if you are a human.

But models love getting perturbed. We don’t perturb them enough, in fact.

OK, perturb has negative connotations. And perturber just sounds wrong in so many ways!

Now you understand why we called the new perturbation feature in Version 5.8 of Thermal Desktop the Model Kicker instead. It sounds more violent, I know, but it was intended to be reminiscent of “kicking the tires before you buy a car.” You should “kick your math model before accepting its results.”

This all started a year or two ago when a friend and I were pondering the question of how we could use the increasing power of computers to produce a better model and not just a bigger one. How we could use Moore’s Law for good instead of evil, given that we’re way past the days when we struggled to afford enough nodes to get acceptable accuracy.

Full disclosure: my friend and I were sampling flights of craft beer at one of the many brew pubs here in Boulder Colorado. (Why we would be discussing Thermal Desktop in that situation can’t be explained. Would it surprise you to know that we were alone, and that the booths next to us had mysteriously cleared out?)

Surprisingly, we came up with some good ideas.

Miraculously, we remembered them.

I was especially perturbed (!) by this thought: we’re building such big models these days that we don’t always understand what is going on inside them.

Big models hide mistakes, and make it harder to run the model many times in order to evaluate various scenarios. You have to spend a lot of time staring at colored contours and plotted transient results in order to figure out what exactly is happening inside your model. You can easily find out where heat is flowing, but it is much harder to figure out exactly why it flows where it does when it can follow so many parallel routes.

How about diverting some of those fast new multi-core CPUs and screaming solution algorithms into helping build your intuition about how the system behaves?

The purpose of computing is insight, not numbers.” 
                                                        - Richard Hamming

Does this sound hard? It isn’t! It turns out the Model Kicker is really easy to set up, ridiculously fast to run, and very easy to intuit relative strengths of connection.

Here is a snapshot (at one time point) of a box full of batteries and electronics, isothermalized by four heat pipes. Can you tell how well the heat pipes are helping? Can you even tell whether they are hooked up correctly? You can certainly see what is hottest and what is coldest, but for all you know you forgot to put in the thermal contact in one spot, or you put it in twice, or you put in the wrong value. Or you forgot to merge the edges of the side walls. (Relax, we’ve all goofed up. We just don’t want other people to find those goof-ups before we do!)

Now take a look at what happens when you “kick” one heat pipe and measure not just temperature, but the strength of connection between that heat pipe and the rest of the model on a scale of 0.0 to 1.0. In other words, 0.0 means “thermally not connected” and 1.0 means “extremely well-connected thermally.”

If the heat pipe had experienced a connection problem (e.g., missing contact conductance), that issue would have jumped out of the screen at you.

Or you can perturb all four heat pipes at once to ask questions such as “Are the heat pipes doing that big cylinder any good?”

To learn how to use the Kicker, you can read about it in the TD User’s Manual. You can also view this class or ask us for the class notes.

You don’t have to use the Kicker to explore your model. You can also make your own perturbations in any inputs (such as the bond line conductance, or a convection coefficient scaling factor) and use the powerful TD postprocessing option to compare the temperature differences between two answers. You’ll be surprised what jumps out at you!

Now, go take a look at the last model you built and get perturbed!