proof testing
If you cannot find the flaw that would kill a part, you can instead make the weak parts break on purpose, safely, before you ship them. Proof testing does exactly that: every single part is deliberately loaded once to a proof stress higher than it will ever meet in service. Any part hiding a flaw too big to survive that stress breaks there and then in the test rig and is thrown away, while every part that survives is certified fit. It is the same logic as pressure-testing every scuba tank or lifting-chain before it is trusted with a life.
The guarantee it gives is precise. Applying a proof stress sigma_p means that any part containing a flaw larger than the critical size for that stress, c_proof = (1/pi) times (K_IC / (Y times sigma_p))^2, will fail during the test. So every survivor is guaranteed to contain only flaws smaller than c_proof, which is the same as guaranteeing a minimum strength greater than sigma_p. In the language of statistics, proof testing truncates the weak lower tail of the Weibull strength distribution, chopping off the unreliable parts and leaving a population with a firm floor. To protect against delayed failure as well, the proof stress is set well above the service stress and the load is applied and removed quickly, so that subcritical crack growth during the test itself is limited; combined with a slow-crack-growth model this can even guarantee a minimum time to failure in service.
Proof testing matters because it is the only way to certify individual brittle parts when a single hidden flaw is fatal and non-destructive inspection cannot reliably find flaws only tens of microns across. It is used for the most safety-critical ceramics: rocket-nozzle components, optical fibres, and artificial heart valves. But it carries real costs and dangers, and honesty demands naming them. The proof test itself stresses every part hard and can grow subcritical cracks, which is why the unloading must be fast; too severe a proof stress can damage otherwise-good parts; and the guarantee holds only for flaws present at the instant of testing, so a part can still be degraded later by handling, corrosion, or slow crack growth. Proof testing is a powerful safety net, but it must be paired with a lifetime model, not trusted alone.
Every spun length of optical fibre is pulled over a set of wheels that strains it to a proof level, snapping any segment weak enough to fail there. The reel that emerges is guaranteed flaw-free above the proof strength, so a telecoms operator can trust a cable to survive decades of tension without a single break.
Loading every part past service stress weeds out the weak ones, guaranteeing a minimum strength for the survivors.
Proof testing has hidden costs: the test itself can grow subcritical cracks (so unloading must be fast), and too high a proof stress damages good parts. It guarantees a strength floor only against flaws present at the instant of testing, so it must be paired with a slow-crack-growth lifetime model, not trusted on its own.