At Boeing, we deal with mathmatical stress analyses in one of two ways:

Use an overly conservative margin of safety, or perform structural (lab) testing to verify computed data.

Almost all formulae used for structural analysis are the results of extensive physical experimentation. The more experimental data you have, the beter you can trust the mathematical equations derived therefrom.

>

> I did indeed contact Gougeon Brothers about a year ago about
> similar data they had published in "On Boat Construction".
> Mr. Watson allowed as how the data was not the best. They previously
> test a few panels of quite different lay ups. The new data is better
> in that they have a larger data set with more closely related samples.
> It is not data which should use for detailed engineering. However,
> as a basis for gross comparison the data is a good start.

> I know George uses mathematical models to predict material properties.
> I would submit that this method is even worse than bad data because
> it has not been correlated back to real data. And if the real data
> is bad the correlation of a model to that data will be even worse.
> Measurements are one step away from reality, models are two steps
> away. Just because models produce more consistent results does not
> mean they are better. Bad data is like a stopped watch, it is correct
> at least twice a day and you know that it is suspect, an un-verified
> mathematical model is like a slow watch, it is never right but it
> looks trustworthy.