Thursday, March 11, 2010

Qualitative Science

Here’s one perspective on the history of science: Ug and Grug found that if they took two stones and banged them together, they got sparks and could make fire. They tried to recreate it with other stones, but found that only certain colored stones worked. Their descendents eventually founded atomic theory, created chemicals reactions to purify materials, measured their purity, and even made predictions (successfully) about what other materials would work too.

Short and sweet, science has gone from a very qualitative state to an extremely quantitative and predictive state. This is repeated anytime a new field of science is opened. First people find qualitative observations in which they can then begin to observe patterns. This then leads to models that make predictions. Failures in the predictions lead to a refinement of the model, allowing for better predictions. Repeat ad infinitum.

Every once in a while we get clients that want to go in the opposite direction. Many years ago I was charged with a project involving heat transfer in an electronics case. Because of a number of design elements, we could treat it as a 2-D case. We pulled out our standard heat transfer equations, did some manipulations, calculated a few cases in detail, and made suggestions of what could be done to improve the situation. We thought we had nailed it, and certainly in a technical sense, we had.

But it wasn’t what they wanted. They wanted a bunch of 3D FEA images. The images, being full colored, certainly looked better than equations and spreadsheets, and non-technically-trained upper management liked looking at pretty pictures more than equations, but this was a complete reversal of 15000 years of scientific development taking us right back to Ug and Grug. I am not surprised that the company has been struggling financially in the intervening years.

While I’m on the subject of FEA, let me add a few thoughts. What I don’t like about FEA modeling is that it is difficult to take a fundamental learning from the output in both cases of success and failure.

Because of the expense in completing the computations, the outputs are usually limited to only a few images. As such, if the model is successful, there are most often few if any patterns to allow for a generalization to be noted so that future predictions can be made without the use of FEA.

And on the other hands, as I noted above, failure in the predictions from an equation means that the equation needs more work to understand what is truly occurring, what subtlety has been missed, what nuance needs to be tickled out. Failure of an FEA model prediction could mean any of a number of things – that the software package is bad, that the inputs were bad, that the mesh generated was appropriate… all sorts of possibilities, but not that the underlying science is in need of correction.

Either way, I usually don’t take much away from the output of an FEA modeling effort.


Eric F. Brown said...

As Terry Gilliam said in Monty Python and the Holy Grail-- "It's only a model."

John said...

Apparently that was the only true line in the movie.

I was watching a bio of Terry Gilliam (a favorite director of mine) and he was talking about during the making of the movie, they originally had permission from the British government to shoot scenes in actual castles. The government had second thoughts about such a disrespectful movie being made in their distinguished castles and cancelled the deal, forcing them to actually use plywood models for their castles.

Gilliam noted cynically that considering all the tortures and other gastly things that had occured in the castles, he couldn't see their point.