Originally Posted by
Engines
Where I would, very respectfully, part with him would be an assertion that an ILS/LSA analysis is 'stochastic'. In my experience, it's normally probability based, but if the newer analyses have found a way to include randomness then, hey, fill yer boots.
Probability-based events ARE "random"; that's essentially what a probability density function is telling you. Reliability analysis divides an items life into three phases:
1. The "infanty mortality" phase - early failures are presumed to be mostly due to manufacturing defects, and the probability of a manufacturing defect remaining present in-service decreases exponentially with time/usage/cycles etc. That's actually the technical purpose of burn-in testing (aka "Production Reliability Acceptance Testing"*)- to get the item through the infant mortality phase so that the probability of manufacturing defects remaining was acceptably low.
2. The "Wear-out" phase - the back end of the item's life when stuff is starting to exhibit wear/degradation. The probability of this happening is exponential with respect to time/cycles/usage etc. There are types of equipment that don't have a wear-out characteristic, but they are much rarer than most people think.
3. The bit between the two which is misleadingly called the "constant failure rate" phase. In this phase you expect to see a failure rate which looks fairly similar when measured over a long enough period - failures per year, failures per 100,000 cycles or whatever, so that when plotted it looks like a flat(ish) line. In this period we say the failures are "random" when what we actually mean is "the time between each failure is random". The total time covering any 100 failures may be a nearly constant figure, but those 100 failures will be randomly clustered into clumps rather than being evenly-spaced through the period.
All three of these phases model well using the exponential probability distribution. So we have an initial exponentially decreasing phase, then a flat-line phase, then an exponentially increasing phase. The probability of a failure at any time is given by the sum of the three phases, and when you sum the three probability densities you get a plot that looks like a bathtub (the infamous "reliability bathtub curve" - that's where it comes from).
HTH,
PDR
* yes, we do know what that looks like as an acronym. Blame the Americans - they put it in the mil-spec