3 ms·
> It would be one thing if SpaceX tested the material coupon and found it within specs, but by sheer probability, the strut had a portion of its material struct
by hn_throwaway_99 5y ago
> It would be one thing if SpaceX tested the material coupon and found it within specs, but by sheer probability, the strut had a portion of its material structure that was anomalous. That's what you're talking about.
Actually, that's not what I'm talking about. What I'm talking about is lessons from QA processes and the chance that any particular bug escapes from one QA process to the next.
The idea being, suppose you have:
* one million different "components" in a system
* each of those components have, say, 10 possible failure modes
* There are 100 of these new complex products a year.
Obviously the numbers above are all made up, but the idea would be in that example you'd have a billion (1,000,000 X 10 X 100) possible failure modes. Any QA filter, might be expected to catch, say, 95% of remaining bugs.
So you run as many filters as you can, but at some point you're still left with the small probability for new bugs, and they only way to remediate those bugs is to figure out what went wrong after the fact.
- bumby 5y agoNot every bug creates a failure, so there's not necessarily a need to be completely bug-free. The point is to focus efforts on the bugs that can cause failure and design mitigations for them. If your system has a billion potential software failure modes, my hope is that a project manager will say the software is too complex and needs to be redesigned. I think this is where some in the software crowd get it wrong in their understanding of how software reliability in safety-critical applications is performed. There are a lot of project managers who deliberately avoid using software as a mitigation because of the complexity issue (see the 737Max use of software as a mitigation for a good example of where this practice can go wrong). So take the software example from above on CST-100. The valve mapping error should have been a known failure mode ("We command valve A and it doesn't respond." or "We command valve X and valve A responds"). Those failures are very straightforward to test for with simulations on the ground, yet somehow those errors made it into a flight configuration. Arguably, the software timer issue on that flight falls into a similar category, but the mitigations there are a bit more complex.