cammyd32 wrote:Also what could be worth considering is forming a design programme that is made entirely from experimental data, have you ever considered such an idea yourself?
It is certainly possible and attractive for the reasons you detail. However, it would be difficult even for simple low-speed pneumatics if we don't test the effects of dynamic friction and the fineness ratio of the pipes. If you non-dimensionalize the parameters which control the energy efficiency (a non-dimensional number itself --- % has no units) with the mentioned assumptions you'll get at least 5 parameters. A few years ago I had 6, but I realized that the right non-dimensionalization eliminated one of them.
So you have 5 parameters to vary. Let's say that you try 3 different values and don't do multiple trials for each value. You'd have to do (3^5 =) 243 tests! 3 values with 2 trials each are ((2 * 3)^5 =) 3125 tests! You can see this is getting unmanageable already. For the tests I did in 2010, as I recall, I stopped at around 100 tests because I was getting really bored. If I did 100 tests every day for a month, I could get 3000 tests, but that'd be a very significant effort.
Springers have significantly more parameters at a minimum. So the situation is much worse there.
This is why computer simulations are attractive. We can't test everything for many reasons. Some tests require way too many experiments (though this is helped with a large team of scientists working independently). Some tests are expensive (full scale fire tests). Some tests are illegal (nuclear testing). Etc.