12
$\begingroup$

There are thousands of satellites orbitting our planet and everyone of them got there from a rocket launch. I know that when a launch fails it is practically always a catastrophic failure, otherwise its a minor or major setback and the launch is rescheduled.

What is the failure rate of space bound rocket launches, in general? I mean for all the nations that have tried.

$\endgroup$
  • 3
    $\begingroup$ en.wikipedia.org/wiki/Comparison_of_orbital_launchers_families $\endgroup$ – Deer Hunter Mar 27 '15 at 8:43
  • 3
    $\begingroup$ From that wikipedia link, 176 failures from 3024 launches = 5.8% failure. Various assumptions made as to what consititutes failure, but that's ballpark there $\endgroup$ – Rory Alsop Mar 27 '15 at 9:49
  • 1
    $\begingroup$ 94% success is over a long history of rocket development. More recently, launching agencies have refined their designs and processes to achieve really high reliabilities. Atlas II through Atlas V have had only one partial failure in 120 launches since 1991, for example. $\endgroup$ – Russell Borogove Mar 27 '15 at 20:42
  • 5
    $\begingroup$ > "I know that when a launch fails it is practically always a catastrophic failure" - not entirely true, there are a significant minority of launches that end up with noncatastrophic failures such as underperformance that leaves the payload in an undesirable orbit. Sometimes the spacecraft can boost itself to the correct orbit, usually at the expense of mission life due to the used propellants. $\endgroup$ – pericynthion Mar 27 '15 at 21:40
  • $\begingroup$ @Rory, where did you get your numbers? I assume you added up all the numbers in the 'total launches' column. But how did you measure 176 failures? $\endgroup$ – Octopus Mar 30 '15 at 22:50
13
$\begingroup$

To further add to the other answers, I found some raw data about this. Here's a visualization of it, broken down by successes/failures and manned/unmanned.

Orbital Launches Per Year Graph

Some descriptive statistics:

  • % Manned Failures in Manned, Entire Data Set = 1.64%
  • % Manned Failures in Manned, Last 20 Years = 0.79%
  • % Unmanned Failures in Unmanned, Entire Data Set = 8.08%
  • % Unmanned Failures in Unmanned, Last 20 Years = 6.68%

I'd go.

Edit: Updated the chart colors to better reflect unmanned failed launches.

$\endgroup$
  • 1
    $\begingroup$ Is there any explanation of what L(F) really means in the raw data? It might not accurately represent what was asked in the OP. I'm giving it a plus 1 anyway because the initial question didn't clearly state what was meant by a "failed launch" in the first place. $\endgroup$ – Octopus Oct 5 '15 at 21:20
  • $\begingroup$ There's this: "Launch vehicle failure counted when launch vehicle fails to inject payload into (or near) intended orbit." from a different page on the same site: spacelaunchreport.com/logdec.html $\endgroup$ – Radu Weiss Oct 5 '15 at 21:25
2
$\begingroup$

To qualify further Russell Borogove's comment:

"94% success is over a long history of rocket development. More recently, launching agencies have refined their designs and processes to achieve really high reliabilities. Atlas II through Atlas V have had only one partial failure in 120 launches since 1991, for example."

Let's take the one partial failure in 120 data and check whether this is statistically significantly lower than the $94\%$ success rate long term. One could apply the right same principles to the different launch vehicle categories on the Comparison of Orbital Launchers Families Wiki page

Assuming the true probability of partial failure were $p=6\%$, as in the figure quoted by Russell's comment, the probability of observing one partial failure or fewer in 120 launches is:

$$\binom{120}{1} p^1\,(1-p)^{119} + \binom{120}{0}\,p^0\,(1-p)^{120}\approx 0.0051$$

i.e. highly statistically significant. So the Atlas data betoken statistically significantly better performance than the $94\%$ long term data.

Now let's estimate a lower bound on the true reliability data. Suppose we reject the null hypothesis at 99% confidence level, then the highest failure rate in keeping with the data at this confidence level is the solution to:

$$\binom{120}{1} p^1\,(1-p)^{119} + \binom{120}{0}\,p^0\,(1-p)^{120}\approx 0.01$$

which comes out to about $p=5.4\%$. So this simple calculation shows that the true long failure rate of the Atlas family since 1991 is at most $5.4\%$ at 99 percent confidence.

$\endgroup$
  • 1
    $\begingroup$ I'm not sure what this is telling us? $\endgroup$ – Octopus Mar 30 '15 at 22:50
  • $\begingroup$ @Octopus I'm simply saying this is a way to relate Russell Borogove's data (sourced from Wiki) that the Atlas family have one partial failure in 120 launches to the "overall average" that one gets - 94% success rate in 3024 launches. Russell correctly pointed out that one might expect more recent rockets to be more reliable. So this statistical test says yes, for Atlas, there probably is a small improvement in reliability relative to the 94% success rate overall. But it is not large: a true 5.4% overall partial failure rate could in seldom instances (one in 100 sample lots of 120 rockets .... $\endgroup$ – WetSavannaAnimal aka Rod Vance Mar 31 '15 at 2:02
  • $\begingroup$ @Octopus ... each) produce the Atlas data. The above is a way to test this assertion by analysing subsets of the Comparison of Orbital Launchers Families Wiki page. $\endgroup$ – WetSavannaAnimal aka Rod Vance Mar 31 '15 at 2:02
  • 1
    $\begingroup$ The null hypothesis has been disproved already! $\endgroup$ – andy256 Apr 28 '15 at 7:42
  • $\begingroup$ What's with all this maths, so confusing!? $\endgroup$ – Zyrogen Aug 11 '16 at 9:43

protected by ForgeMonkey Aug 11 '16 at 12:11

Thank you for your interest in this question. Because it has attracted low-quality or spam answers that had to be removed, posting an answer now requires 10 reputation on this site (the association bonus does not count).

Would you like to answer one of these unanswered questions instead?

Not the answer you're looking for? Browse other questions tagged or ask your own question.