Timeline for Why do most space probes survive for far longer than they were designed for?
Current License: CC BY-SA 4.0
20 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Sep 8, 2020 at 21:37 | comment | added | Russell Borogove | @MasonWheeler The "five year mission" was Kirk's initial assignment to the Enterprise, not the starship's design life. Enterprise was commissioned in 2245 (commanded by Robert April); Kirk took command in 2265. The ship was largely rebuilt in an 18-month refit in the 2270s, and was destroyed in 2285, after 40 years of service. | |
| Feb 19, 2019 at 12:06 | comment | added | Mason Wheeler | "So half the time, you expect this spacecraft designed for a 5-year mission to last 13 years." Hmm... anyone know how long the Enterprise canonically lasted before Kirk blew it up in The Search for Spock(? | |
| Feb 18, 2019 at 23:23 | history | edited | Tom Spilker | CC BY-SA 4.0 |
Fix typo
|
| Feb 18, 2019 at 6:28 | comment | added | Mooing Duck | @Barmar: Check again. Not all the parts did last that long. Many of the parts failed. | |
| Feb 17, 2019 at 2:13 | comment | added | Tom Spilker | @Barmar For the reasons given above, and by ShadoCat, the individual components' (parts') expected lifetimes must be much longer that the desired mission duration. | |
| Feb 16, 2019 at 20:25 | comment | added | Barmar | Don't failure statistics suggest that devices are less likely to last a long time? We're not talking about the distribution of failure among thousands of Mars rovers, because there aren't that many. But when a probe has thousands of parts, the chance that all of them will last many years beyond their rated times seems very unlikely. | |
| Feb 16, 2019 at 18:03 | comment | added | Mark Adler | Random electronics failures are considered to have a Poisson distribution. That needs just one rate parameter, for which there is a fair amount of published data, e.g. MIL-HDBK-217F. Sometimes they are argued to have a Weibull distribution with some memory, which requires two parameters, and so much more data to get a useful estimate. I rarely see those used. | |
| Feb 16, 2019 at 15:08 | comment | added | alephzero | @NuclearWang you are trying to oversimplify the situation down to a single number. In reality every different failure scenario is assessed against not only its likelihood, but also the consequences of it happening. For example if 5 experiments out of 10 on the lander "fail", that has less effect on the mission than if all 10 "work" but there is no communication link to get any of the results back to earth - the overall success rating would be "50%" in one case, but "0%" in the other. | |
| Feb 16, 2019 at 13:53 | vote | accept | Hrach | ||
| Feb 16, 2019 at 7:44 | comment | added | Arthur | Having "within two standard deviations of the mean" correspond to 95% probability is for a two-tailed deviation, though. More than two standard deviations (or really closer to 1.96) below the mean happens in only 2.5% of cases. | |
| Feb 15, 2019 at 19:50 | comment | added | Tom Spilker | @Mindwin In addition to Christoph's comment, the probability of success of an entire mission is the product of the probabilities of a large number of components, each of which has its own mean and standard deviation. That composite is not exactly a normal distribution. | |
| Feb 15, 2019 at 15:25 | comment | added | Dietrich Epp | @Christoph: Isn't a beta negative binomial distribution a discrete distribution? That doesn't make sense to me. I would probably use something like a Weibull for lifetime. But using a normal distribution as an approximation is appropriate when you're just doing some back-of-the-envelope math. | |
| Feb 15, 2019 at 13:49 | comment | added | Nuclear Hoagie | Does the operational lifespan of a probe take into account those "infant mortality" events? If there's a 20% chance the probe goes splat upon landing, but a near-certainty of surviving 10 years after that if it doesn't, is it expected to operate for 10 years, or 8? | |
| Feb 15, 2019 at 13:05 | comment | added | Christoph | @Mindwin normal distribution is defined from -infinity to +infinity. A negative life time doesn't really make sense so it can't be normally distributed. I think a better match is probably the Beta negative binominal distribution. The average to +infinity side of both looks pretty similar but the -infinite side to average is "compressed to" zero to average which makes a lot more sense. Actually we rarely have a full normal distribution (you can't be -1cm tall) but it closely matches many different observations. | |
| S Feb 15, 2019 at 12:27 | history | suggested | CommunityBot | CC BY-SA 4.0 |
Fixed typos
|
| Feb 15, 2019 at 12:08 | review | Suggested edits | |||
| S Feb 15, 2019 at 12:27 | |||||
| Feb 15, 2019 at 11:46 | comment | added | Michael | How do you calculate probability of success? If you knew all the risks, surely you could avoid/mitigate them? | |
| Feb 15, 2019 at 11:22 | comment | added | Mindwin Remember Monica | I'm curious, why are probability of random failures not normally distributed? | |
| Feb 15, 2019 at 2:12 | comment | added | Tom Spilker | After posting I see that @ShadoCat has posted a very useful graph of a normal distribution. | |
| Feb 15, 2019 at 2:08 | history | answered | Tom Spilker | CC BY-SA 4.0 |