Testing for reliability is very slow, and requires many exemplars for statistical validity.
Repeated detailed analysis of each design iteration should be faster and cheaper; that's pretty much why engineers have jobs.
At least where components like bearings are involved, the hard work has already been done, math models exist and have been validated, so all you have to do is figure out how your design maps to what is known.
If you're going completely 'off label' for everything, you're on your own, and will have to work from the basics. ... and you should be constantly re-evaluating the wisdom of using standard components in odd ways.
On the other hand, test failures do give you signposts that can guide your investigation, by reverse-engineering each failure to its root.
On the other, other hand, robust designs don't usually fail in test unless you try really, really hard, and sometimes not even then, because you're relying on safety factors built in by other people, and you don't know how much margin you really have, or how it will change when a component vendor changes their process in some unanticipated way, which provides yet another argument for not getting too creative.
Mike Halloran
Pembroke Pines, FL, USA