Abort options for Starship and Starship/SuperHeavy

[Robotbeat]

In 1919, the probability of dying while flying for the air mail service was one death per 115,000 miles flown.
https://www.atlasobscura.com/articles/the-pilots-who-risked-their-lives-to-deliver-the-mail

If you used a Falcon 9 to do long haul travel from one side of the world to the other, it’d be far safer per mile than that. About 10 times safer (not counting how particularly perilous long distance flight was at the time). Launch failure (full failure) probability of F9 is about 1 in 187 (and likely Dragon would’ve survived if it had had someone on board to command the parachute to deploy, even without an actual LAS… or even if crew would be able to bail out with parachutes, as was done in the 20s and very late 1910s, although they were primitive and unreliable early on).

In terms of per-flight, the 1919 airplanes had ranges of between like 120 to 500 miles. So if we say each air mail flight was around 115 miles, the per flight fatal failure rate was 1:1000.


So overall, I’d say this is a good comparison. Today’s rocket tech could probably be made at least as safe as airplanes in 1920, even without a full LAS. There haven’t been enough flights to do this, yet, tho, unless you count per mile.

(Wikipedia claims the fatalities per passenger mile as being 1:1million in 1924 for airlines, and F9 would be about a factor of 2 lower if used for transporting someone to the other side of the world. But the Wikipedia claim doesn’t have a good source, although is very plausible and fits with other statistics.)

[sevenperforce]

While I generally agree with your points in here, I will note that this isn't quite telling the whole story. Many crew capsules have historically used monopropellant only for the actual re-entry module, keeping the hypergolic bipropellants in a separate service module separated from the crew. I believe that the only crew vehicles to ever contain bipropellants inside the OML of the re-entry vehicle were Gemini, Apollo CM, Shuttle, and Crew Dragon. In contrast, Orion and Starliner and Mercury, plus all of the Soyuz and Soyuz-derived capsules, use or used only monoprop in the actual crew vehicle, keeping bipropellant RCS in the service module. I'm not sure what was/is planned for Dream Chaser and Orel.

Fair point, but it begs a question:  Is the big risk a hypergolic explosion, or a leak that poisons the crew?

ISTM the biggest deal is a poisonous leak, and monoprop doesn't help you very much with that.  A hypergolic explosion, even in a separate service module, probably isn't a survivable event.

Well, we have two instances of explosions happening inside a crew vehicle. One was Apollo 13; the other was Crew Dragon during that one ground test.

The explosion during Apollo 13 was obviously survivable. If the same explosion that wrecked Crew Dragon during that ground test had taken place in a separate module, separated by the heat shield, there would at least have been a chance for the crew to make it.

Being able to blow the entire pressurized fairing is...well, yikes. Definitely a huge engineering challenge. It might almost be better to have the crew Starship use a different OML where the LES module is attached so as to be entirely clear of the fairing.

Yeah, this is definitely the green weanie in the pack.  Just to clarify:  The internal fairing is probably neutral static pressure.  But you have to be able to blow it clear at max-q.

You can't have the LES module clear of the fairing because of the canards.  The canards have to be well forward, and there's no way you get a stable separation with those puppies in place.

It's all a hideous kludge.  And a fully agree with everybody that the better solution, if time and the design permit, is to increase the empirical reliability to the point where everybody's comfy with no escape system.  I just don't believe that time and the design will permit, at least not before when it becomes really awkward for SpaceX not to have a crewed Starship launch/EDL capability.

What I meant by the OML change was something like what is attached (which is of course cursed but hopefully it communicates the point). A completely separate OML for the capsule.

[InterestedEngineer]

Falcon 9 is less than 1% right now for full failure.

That's the pessimistic naive Bayasian model result.   Which means it's probably the real number (0.7% pLOC per Bayesian model posted above).

SLS was 2/135 so 2.2% pLOC.

We don't have any visibility on the updated PRA model, AFAICT.  Did NASA publish it and we missed it?  I've been surprised about what details one can get out of e.g. Starship from NASA publications.  I suspect it's far better with that model.

After all, if one came claim with a dozen flights that the pLOC on Dragon-2 is 1:1000, why can't one claim it for Falcon-9 with 150+ consecutive successful flights?

Or for SLS or any other new rocket / capsule you care to name.

[Lee Jay]

In 1919, the probability of dying while flying for the air mail service was one death per 115,000 miles flown.
https://www.atlasobscura.com/articles/the-pilots-who-risked-their-lives-to-deliver-the-mail

If you used a Falcon 9 to do long haul travel from one side of the world to the other, it’d be far safer per mile than that. About 10 times safer (not counting how particularly perilous long distance flight was at the time). Launch failure (full failure) probability of F9 is about 1 in 187 (and likely Dragon would’ve survived if it had had someone on board to command the parachute to deploy, even without an actual LAS).

In terms of per-flight, the 1919 airplanes had ranges of between like 120 to 500 miles. So if we say each air mail flight was around 115 miles, the per flight fatal failure rate was 1:1000.


So overall, I’d say this is a good comparison.

No it's not.  That's paid professionals, which is equivalent to the NASA astronaut requirement (1:270?), not the general public.  That's nowhere near good enough for the general public, even then, which led to regulation.

"The safety benefits were obvious: the fatality rate for the Air Mail Service was one per 789,000 miles flown between 1922-1925, while the comparable figure for itinerant commercial fliers (for 1924 only) was one per 13,500 [3]."


Obviously, that wasn't good enough, so they regulated and fixed it.


https://www.tc.faa.gov/its/worldpac/techrpt/ar0839.pdf

[Robotbeat]

That process will happen for RLVs, too, unless we kill it by regulating too early. I’d say regulation proceeded at the right rate for airplanes.

Currently at nearly 1:200, could get to 1:1000, at which point cargo delivery becomes viable.

[InterestedEngineer]

In 1919, the probability of dying while flying for the air mail service was one death per 115,000 miles flown.
https://www.atlasobscura.com/articles/the-pilots-who-risked-their-lives-to-deliver-the-mail

If you used a Falcon 9 to do long haul travel from one side of the world to the other, it’d be far safer per mile than that. About 10 times safer (not counting how particularly perilous long distance flight was at the time). Launch failure (full failure) probability of F9 is about 1 in 187 (and likely Dragon would’ve survived if it had had someone on board to command the parachute to deploy, even without an actual LAS).

In terms of per-flight, the 1919 airplanes had ranges of between like 120 to 500 miles. So if we say each air mail flight was around 115 miles, the per flight fatal failure rate was 1:1000.


So overall, I’d say this is a good comparison.

No it's not.  That's paid professionals, which is equivalent to the NASA astronaut requirement (1:270?), not the general public.  That's nowhere near good enough for the general public, even then, which led to regulation.

"The safety benefits were obvious: the fatality rate for the Air Mail Service was one per 789,000 miles flown between 1922-1925, while the comparable figure for itinerant commercial fliers (for 1924 only) was one per 13,500 [3]."


Obviously, that wasn't good enough, so they regulated and fixed it.


https://www.tc.faa.gov/its/worldpac/techrpt/ar0839.pdf

Nice data.  Backs up the point we are at about 1920 level at least with F9.

can we be at 1940 level in 2042?    Possibly.   Complicated abort systems won't get you there.   They certainly didn't for airplanes.   What did they do?  Iterated, improved the technology.

I note the CAA wasn't formed until 1938, which is quite a ways down the improvement curve of your graph.  So "regulation" didn't help improve the first 20 years of commercial aviation.   Just plain old greed and common sense.

[InterestedEngineer]

That process will happen for RLVs, too, unless we kill it by regulating too early. I’d say regulation proceeded at the right rate for airplanes.

Civilian Aviation Authority was formed in 1938.   So the first order magnitude of safety happened in an unregulated environment, the second order of magnitude in a regulated environment.

Which fits most new tech curves quite nicely.   You start by moving fast and breaking things, and then get the last bit by good project management.   Logistics "S" curve.

Rocketry has stalled at the lowest part of the logistics S-curve for 50+ years.  It's time to get it passed the middle.  When we are passed the middle, regulation starts to make more sense.

https://en.wikipedia.org/wiki/Logistic_function

[Robotbeat]

The point at which the expected lifetime loss (assuming you’re good for at least another 80 years, which is on the high side) equals the time savings is about a fatality rate of 1:100,000 for rocket flights which save at least 7 hours of time vs other modes of transport. 1:25,000 if you assume 40 years of life left with 14 hours saved (ultra long haul aircraft flights). Of course, regulators/rulemakers don’t value lifetime wasted, otherwise we wouldn’t have the insanity of TSA, but I digress.

Should be doable. That sort of utilitarian safety argument might be okay for private chartered flights, etc. Got to expect an order of magnitude higher for regularly scheduled passenger flights…. Which could be doable if the safety already was proven on chartered flights and cargo flights.

There’s no way to make it safe if the flight rate isn’t hundreds to thousands to tens of thousands per year. To first order, you’re not going to be able to prove safety beyond one fatal failure per ten years. So your per-flight safety will be terrible if the cost is so high you can only fly a few times per year.

[InterestedEngineer]

I believe wholeheartedly in iteration, but it's not a panacea.  You have to have your architecture sorta-kinda right to begin with, or all iteration does is pile kludge on top of kludge.

Yep. I've seen iterative projects fail from bad architecture.

I've also seen well architected projects fail from bad requirements.

I've also seen iterative development fail due to too much project management.

Architecture and iteration are each necessary, but not sufficient.

In the end, do I trust Elon Musk as a system architect?

Yes, because I was a system architect for two successful product lines ($1B+/year turnover total), and I recognize a good one when I see one.   Starship seems to have the right mix of good architects, rejecting bad requirements, project managers relegated to information support role, and iterative development.

So I don't see Starship as a bad reusable second stage architecture.   Does it meet 1930s-era civilian air flight pLOC rates?  Absolutely not, until at least 2035 or so.   But there's a decent probability it can before SpaceX runs out of money.

And a decent probability is all we can ask for.

[Robotbeat]

It’s important that SpaceX has a huge potential revenue stream and launch demand source in the form of Starlink, which gives them a hope of being able to bootstrap to the high launch rates needed for full reuse and high intrinsic reliability/safety that comes with full, rapid reuse. This is true even if Elon wastes the rest of his life trying to make Twitter profitable or whatever. Starship, if it can get even a halfway decent launch rate (comparable to Falcon 9), will have pushed launch capability up past a significant energy barrier where it can settle to find a new optimum over the next few decades. It doesn’t necessarily need new, aggressive systems engineering leadership like Elon any more, just execution.

First Starlink and trips to deep space using lots of refueling enabling ultimately 99.9% reliability, then point to point cargo enabling 99.99%, then point to point chartered flights and mass trips to LEO and deep space for thousands of people enabling 99.999%, then hopefully point to point flights for millions of people, and millions of people  and millions of tons to orbit and deep space enabling 99.9999+%.

[Lee Jay]

First Starlink and trips to deep space using lots of refueling enabling ultimately 99.9% reliability, then point to point cargo enabling 99.99%, then point to point chartered flights and mass trips to LEO and deep space for thousands of people enabling 99.999%, then hopefully point to point flights for millions of people, and millions of people  and millions of tons to orbit and deep space enabling 99.9999+%.

It's just an opinion, but I think those numbers are completely delusional.  Never, ever going to get there with this architecture.

[chopsticks]

First Starlink and trips to deep space using lots of refueling enabling ultimately 99.9% reliability, then point to point cargo enabling 99.99%, then point to point chartered flights and mass trips to LEO and deep space for thousands of people enabling 99.999%, then hopefully point to point flights for millions of people, and millions of people  and millions of tons to orbit and deep space enabling 99.9999+%.

It's just an opinion, but I think those numbers are completely delusional.  Never, ever going to get there with this architecture.

I don't think you're going to get those numbers no matter the architecture. Well maybe a space elevator might work.

[TheRadicalModerate]

Does it meet 1930s-era civilian air flight pLOC rates?  Absolutely not, until at least 2035 or so.

But this is my whole point:  2035 is too late.

It's too late to put private individuals into cislunar, which requires being able to defray the costs across 12+ people.  Without the private cislunar traffic, NASA won't suffer the embarrassment they need to, and will stand pat on SLS/Orion, continuing to use LSS only as a nuclear weapon to swat a fly.  If NASA stands pat on SLS/Orion, then Artemis will fail and be cancelled.  If Artemis is cancelled, then SpaceX has no platform to subsidize its testing to get ready for Mars--or any other crewed BEO destination, for that matter.

I think if SpaceX could talk NASA into D2-assisted LSS, then they probably could wait until 2035 for a crewed launch/EDL capability.  But I don't think they can do that without embarrassing NASA and their congressional patrons, and that requires private flights, starting somewhere in the 2028 timeframe.

As I said, way up-thread now, an alternative strategy is to subsidize a couple of D2-LSS cislunar flights.  If that gets NASA to move, it can be SpaceX's anchor customer for all kinds of R&D for Mars.  But the much better solution is to get bigger crews launched, sooner.  (This also keeps Maezawa from being really angry and/or litigious.)  I think it's so much better a solution that it's worth doing the escape engineering.

[Robotbeat]

First Starlink and trips to deep space using lots of refueling enabling ultimately 99.9% reliability, then point to point cargo enabling 99.99%, then point to point chartered flights and mass trips to LEO and deep space for thousands of people enabling 99.999%, then hopefully point to point flights for millions of people, and millions of people  and millions of tons to orbit and deep space enabling 99.9999+%.

It's just an opinion, but I think those numbers are completely delusional.  Never, ever going to get there with this architecture.

Right, your arguments have been opinion or begging the question for the vast majority of this thread, you don’t need to keep reminding us of that.

I’m old enough to remember when people claimed droneship landing was delusional. Or, actually, the actual word a space industry professional I know used was “impossible”, and they were better qualified than either of us.

Droneship landing has been FAR more reliable in operational use than just about anyone expected, even people fully on board with VTVL RLVs.

Sometimes the laws of physics don’t care about the status quo.

[Robotbeat]

First Starlink and trips to deep space using lots of refueling enabling ultimately 99.9% reliability, then point to point cargo enabling 99.99%, then point to point chartered flights and mass trips to LEO and deep space for thousands of people enabling 99.999%, then hopefully point to point flights for millions of people, and millions of people  and millions of tons to orbit and deep space enabling 99.9999+%.

It's just an opinion, but I think those numbers are completely delusional.  Never, ever going to get there with this architecture.

I don't think you're going to get those numbers no matter the architecture. Well maybe a space elevator might work.

Space elevator is worse. Space elevator is just what people say when they want to wave their hands a lot.

[Robotbeat]

Does it meet 1930s-era civilian air flight pLOC rates?  Absolutely not, until at least 2035 or so.

But this is my whole point:  2035 is too late.

It's too late to put private individuals into cislunar, which requires being able to defray the costs across 12+ people.  Without the private cislunar traffic, NASA won't suffer the embarrassment they need to, and will stand pat on SLS/Orion, continuing to use LSS only as a nuclear weapon to swat a fly.  If NASA stands pat on SLS/Orion, then Artemis will fail and be cancelled.  If Artemis is cancelled, then SpaceX has no platform to subsidize its testing to get ready for Mars--or any other crewed BEO destination, for that matter.

I think if SpaceX could talk NASA into D2-assisted LSS, then they probably could wait until 2035 for a crewed launch/EDL capability.  But I don't think they can do that without embarrassing NASA and their congressional patrons, and that requires private flights, starting somewhere in the 2028 timeframe.

As I said, way up-thread now, an alternative strategy is to subsidize a couple of D2-LSS cislunar flights.  If that gets NASA to move, it can be SpaceX's anchor customer for all kinds of R&D for Mars.  But the much better solution is to get bigger crews launched, sooner.  (This also keeps Maezawa from being really angry and/or litigious.)  I think it's so much better a solution that it's worth doing the escape engineering.

the bar isn’t to get to 1920s pLOC figures before sending anyone to space in Starship. Just need to beat pLOC of Artemis 2, probably around 5% (in *reality*).

40 consecutive launches, re-entries, and landings of starship (and then adding extra margin and quality assurance for crewed flights) would absolutely crush the pLOC levels that NASA will ever get with Orion and SLS. That’s not at all unreasonable to expect by Artemis 3, say.

[chopsticks]

First Starlink and trips to deep space using lots of refueling enabling ultimately 99.9% reliability, then point to point cargo enabling 99.99%, then point to point chartered flights and mass trips to LEO and deep space for thousands of people enabling 99.999%, then hopefully point to point flights for millions of people, and millions of people  and millions of tons to orbit and deep space enabling 99.9999+%.

It's just an opinion, but I think those numbers are completely delusional.  Never, ever going to get there with this architecture.

I don't think you're going to get those numbers no matter the architecture. Well maybe a space elevator might work.

Space elevator is worse. Space elevator is just what people say when they want to wave their hands a lot.

Don't overthink it, I just said space elevator because it was the first thing that came to my mind in terms of what would theoretically be the safest way to get to space. Maybe there are other far out options there that I haven't heard of that don't rely on chemical rocket engines.

Don't know what you mean about waving hands a lot.

[Lee Jay]

Right, your arguments have been opinion or begging the question for the vast majority of this thread, you don’t need to keep reminding us of that.

1) I don't think you know what begging the question means.
2) Everything you've posted is an opinion too.  We don't have any data because, if you recall, this vehicle has never made orbit.

[TheRadicalModerate]

Does it meet 1930s-era civilian air flight pLOC rates?  Absolutely not, until at least 2035 or so.

But this is my whole point:  2035 is too late.

It's too late to put private individuals into cislunar, which requires being able to defray the costs across 12+ people.  Without the private cislunar traffic, NASA won't suffer the embarrassment they need to, and will stand pat on SLS/Orion, continuing to use LSS only as a nuclear weapon to swat a fly.  If NASA stands pat on SLS/Orion, then Artemis will fail and be cancelled.  If Artemis is cancelled, then SpaceX has no platform to subsidize its testing to get ready for Mars--or any other crewed BEO destination, for that matter.

I think if SpaceX could talk NASA into D2-assisted LSS, then they probably could wait until 2035 for a crewed launch/EDL capability.  But I don't think they can do that without embarrassing NASA and their congressional patrons, and that requires private flights, starting somewhere in the 2028 timeframe.

As I said, way up-thread now, an alternative strategy is to subsidize a couple of D2-LSS cislunar flights.  If that gets NASA to move, it can be SpaceX's anchor customer for all kinds of R&D for Mars.  But the much better solution is to get bigger crews launched, sooner.  (This also keeps Maezawa from being really angry and/or litigious.)  I think it's so much better a solution that it's worth doing the escape engineering.

the bar isn’t to get to 1920s pLOC figures before sending anyone to space in Starship. Just need to beat pLOC of Artemis 2, probably around 5% (in *reality*).

40 consecutive launches, re-entries, and landings of starship (and then adding extra margin and quality assurance for crewed flights) would absolutely crush the pLOC levels that NASA will ever get with Orion and SLS. That’s not at all unreasonable to expect by Artemis 3, say.

I'm too lazy and stupid to wade through all the statistical methods, so I just cribbed the Lewis points and the adjusted-Wald score from Ed Kyle's old reliability metrics at the late lamented spacelaunchreport.  (It may be gone, but its spirit carries on in the Wayback Machine...)

40 launches, no failures:
Lewis score: 97.6%
Adjusted-Wald 95% confidence interval = [89.6%, 100%] with a mean of 95.6%

You can get the Adjusted-Wald to a mean of 99% with 189 consecutive successful missions, 95% CI=[97.6%, 100%].  Lewis score would be 99.5%.

If you want the lower bound of the 95% CI to be at least 99%, you need 460 consecutive successes.

Just for grins and giggles, 2 failures in 500 missions would give you a 95% CI=[98.5%, 100.0%], mean 99.2%.

And, since we're also talking about successful landings, from the time the F9 first did a successful landing, its record is (if I tallied it right), 155/164.  95% CI=[89.8%, 97.2%] with a mean of 93.5%.  Lewis point: 94.4%.

[Robotbeat]

I also would assume at least a factor of 3 (half an order of magnitude) improvement in reliability/safety from launches occurring with crew due to higher standard of quality assurance and margin. Not counting high thrust abort systems.

Soyuz launches with crew have a higher reliability than an R7/Soyuz family launch without crew. (This is something we can check on, and I’ll do so later tonight I think.)