Abort options for Starship and Starship/SuperHeavy

[TheRadicalModerate]

...
What is the confidence level required for the pLOC parameter, and what is the acceptable lower bound on confidence interval, given that confidence level?
...

Unknown. If you wade through the CCP docs, the only published LOC, LOV, LOM, whatever numbers are those 1:270, 1:500, etc. Confidence level and interval are not public to my knowledge. The details are not public; all we have is (to paraphrase): to NASA's satisfaction; and that a PRA was required.

There appears to be undue concern over the confidence interval.  Unless they are weaseling it, pLOC is probability of loss of crew.  The same pLOC is the same risk to the crew, regardless of the confidence level.   The confidence interval is a measure of the amount of data, perhaps a stand in for the maturity of the program

The problem is that there's no such thing as a single probability.  There's merely a confidence interval at a specified confidence level.  The probability itself is illusory.

If you have a PRA Monte Carlo model, you can run it a zillion times and make the interval arbitrarily narrow.  But you can't do that if you're working empirically.

[chopsticks]

What do we think are the most dangerous phases of flight where an abort system would be the most needed? If we can define that, we can work from there and figure out what sort of system could be designed to fit these scenarios.

In my opinion, I see the liftoff and stage sep as the most dangerous parts of flight during ascent, maybe max-q. This being said, it seems to me that the whole EDL is quite a bit more dangerous than the ascent, so that portion of flight should probably get priority. If we assume this and also assume that after stage sep until orbit insertion no separate abort is possible (meaning some kind of separation event like ejection seats, a pod, etc.) this simplifies things a lot. You can basically get away with the same sort of systems used on aircraft. I was checking this video out:



The ejection pod mechanism is pretty interesting. I think it is conceivable to have 4 of these that each seat 5 people (20 crew members). The pods would be arranged in a grid.

A system like this would rely on already known systems and would be effective for the most dangerous parts of flight (the most dangerous being the flip and landing burn IMO).

It would also be nice to have some sort of abort capability during re-entry (with the abort pod having ablative TPS) in case the vehicle started breaking up, but I don't know how such a system could be designed.

There's actually a small Wikipedia article about escape pods as well: https://en.wikipedia.org/wiki/Escape_pod

[sevenperforce]

What do we think are the most dangerous phases of flight where an abort system would be the most needed? If we can define that, we can work from there and figure out what sort of system could be designed to fit these scenarios.

That's definitely the right starting point.

The ejection pod mechanism is pretty interesting. I think it is conceivable to have 4 of these that each seat 5 people (20 crew members). The pods would be arranged in a grid.

A system like this would rely on already known systems and would be effective for the most dangerous parts of flight (the most dangerous being the flip and landing burn IMO).

It would also be nice to have some sort of abort capability during re-entry (with the abort pod having ablative TPS) in case the vehicle started breaking up, but I don't know how such a system could be designed.

The biggest first-order problem for doing both things is going to be aerodynamic stability. You need something that will be aerodynamically stable during a sideways or partially-sideways ejection at Max-Q, but which will also be aerodynamically stable enough to survive re-entry and face heat-shield-down. That's a pretty inherently challenging problem, because a three-dimensional solid that is passively aerodynamically stable in one orientation is simply not going to be passively aerodynamically stable in any other orientation.

[Barley]

...
What is the confidence level required for the pLOC parameter, and what is the acceptable lower bound on confidence interval, given that confidence level?
...

Unknown. If you wade through the CCP docs, the only published LOC, LOV, LOM, whatever numbers are those 1:270, 1:500, etc. Confidence level and interval are not public to my knowledge. The details are not public; all we have is (to paraphrase): to NASA's satisfaction; and that a PRA was required.

There appears to be undue concern over the confidence interval.  Unless they are weaseling it, pLOC is probability of loss of crew.  The same pLOC is the same risk to the crew, regardless of the confidence level.   The confidence interval is a measure of the amount of data, perhaps a stand in for the maturity of the program

The problem is that there's no such thing as a single probability.  There's merely a confidence interval at a specified confidence level.  The probability itself is illusory.

If you have a PRA Monte Carlo model, you can run it a zillion times and make the interval arbitrarily narrow.  But you can't do that if you're working empirically.

Assuming you believe in probability at all:
The probability of LoC is itself a distribution.  As a distribution on a finite interval [0,1] it has a mean.  That mean is the probability of LoC.  It's single valued.  It's the best estimate of the probability of loss of crew.  The confidence interval does not change that.  Given all available information there is a single probability.

The confidence interval tells you nothing about the fate of the next launch.  It does tell you something about the range of possibilities for say a hundred launches and perhaps the fate of the program, but that should not be either a comfort or distress to the next crew.

[chopsticks]

The biggest first-order problem for doing both things is going to be aerodynamic stability. You need something that will be aerodynamically stable during a sideways or partially-sideways ejection at Max-Q, but which will also be aerodynamically stable enough to survive re-entry and face heat-shield-down. That's a pretty inherently challenging problem, because a three-dimensional solid that is passively aerodynamically stable in one orientation is simply not going to be passively aerodynamically stable in any other orientation.

I actually don't think that an ejection in the horizontal orientation would be problem though (during the skydive) if the same sort of ejection is used as is on those bombers. Those eject forwards and up, so even with zero forward airspeed the concept should still work I think.

Or perhaps I misunderstood your statement..?

[sevenperforce]

The biggest first-order problem for doing both things is going to be aerodynamic stability. You need something that will be aerodynamically stable during a sideways or partially-sideways ejection at Max-Q, but which will also be aerodynamically stable enough to survive re-entry and face heat-shield-down. That's a pretty inherently challenging problem, because a three-dimensional solid that is passively aerodynamically stable in one orientation is simply not going to be passively aerodynamically stable in any other orientation.

I actually don't think that an ejection in the horizontal orientation would be problem though (during the skydive) if the same sort of ejection is used as is on those bombers. Those eject forwards and up, so even with zero forward airspeed the concept should still work I think.

Or perhaps I misunderstood your statement..?

I'm saying that there's in inherent challenge in trying to make a single ejection system work for all the high-risk phases of flight.

Assuming the same sort of design as is used on the F111, consider what has to happen at each point.

> On the launch pad, the ejection pod needs to eject laterally away from the vehicle in order to clear the vehicle and the pad, but then it needs to propel itself upward vertically so that it goes far enough away from the vehicle to be safe from any explosion. It needs to be passively aerodynamically stable in the vertical axis so that it doesn't start to tumble as it builds up airspeed. It then needs to self-stabilize in an orientation that is suitable for the deployment of drogues and mains.
> During boost, at Max-Q, the ejection pod still needs to eject laterally to clear the vehicle, but now it needs to do so while already experiencing massive aerodynamic forces from the vertical axis. It doesn't need to gain any more vertical speed, though. It has to maintain aerodynamic stability to avoid tumbling until drogues and mains can be deployed.
> If the ejection is to also be functional during an orbital breakup, then it is deploying into an extremely harsh thermal environment with a low-density, high-velocity airflow coming from a completely different direction. It has maintain its aerodynamic orientation perfectly the entire time so that its heat shield takes the brunt of heating.
> During descent and approach, it's facing a high-density, medium-velocity airflow coming from the same direction as re-entry, and an abort requires that it separate, then gain lateral clearance.
> During a failed flip, it's back to the same situation as a pad abort, except that it now needs MORE thrust to gain altitude since it's already dropping.

There's no single solid object which can have the requisite aerodynamic properties to be passively stable in all of those orientations.

[chopsticks]

The biggest first-order problem for doing both things is going to be aerodynamic stability. You need something that will be aerodynamically stable during a sideways or partially-sideways ejection at Max-Q, but which will also be aerodynamically stable enough to survive re-entry and face heat-shield-down. That's a pretty inherently challenging problem, because a three-dimensional solid that is passively aerodynamically stable in one orientation is simply not going to be passively aerodynamically stable in any other orientation.

I actually don't think that an ejection in the horizontal orientation would be problem though (during the skydive) if the same sort of ejection is used as is on those bombers. Those eject forwards and up, so even with zero forward airspeed the concept should still work I think.

Or perhaps I misunderstood your statement..?

I'm saying that there's in inherent challenge in trying to make a single ejection system work for all the high-risk phases of flight.

Assuming the same sort of design as is used on the F111, consider what has to happen at each point.

> On the launch pad, the ejection pod needs to eject laterally away from the vehicle in order to clear the vehicle and the pad, but then it needs to propel itself upward vertically so that it goes far enough away from the vehicle to be safe from any explosion. It needs to be passively aerodynamically stable in the vertical axis so that it doesn't start to tumble as it builds up airspeed. It then needs to self-stabilize in an orientation that is suitable for the deployment of drogues and mains.

I believe the F111 used a small drogue immediately after ejection to keep it aerodynamically stable in the forward direction. Presumably a small drogue or streamer could be used in the case of Starship. As long as it stable in the forward axis, it should be able to pop the main chutes from the top and land belly first as did the ejection capsule on the F111.

> During boost, at Max-Q, the ejection pod still needs to eject laterally to clear the vehicle, but now it needs to do so while already experiencing massive aerodynamic forces from the vertical axis. It doesn't need to gain any more vertical speed, though. It has to maintain aerodynamic stability to avoid tumbling until drogues and mains can be deployed.

I don't think that aerodynamic loads would really (need to) be much of a concern for a small capsule like this. Presumably the steamer or small drogue would provide the aerodynamic stability.

> If the ejection is to also be functional during an orbital breakup, then it is deploying into an extremely harsh thermal environment with a low-density, high-velocity airflow coming from a completely different direction. It has maintain its aerodynamic orientation perfectly the entire time so that its heat shield takes the brunt of heating.

Yeah, I doubt that an ejection capsule (bomber style anyway) would be an easy solution to surviving an orbital breakup.

> During descent and approach, it's facing a high-density, medium-velocity airflow coming from the same direction as re-entry, and an abort requires that it separate, then gain lateral clearance.
> During a failed flip, it's back to the same situation as a pad abort, except that it now needs MORE thrust to gain altitude since it's already dropping.

Don't really see this being an issue, it seems to me that this would be about the easiest regime for a capsule ejection to deal with - pretty similar scenarios to what the bombers might have dealt with (loss of control).

There's no single solid object which can have the requisite aerodynamic properties to be passively stable in all of those orientations.

A sphere?

[TheRadicalModerate]

Assuming you believe in probability at all:
The probability of LoC is itself a distribution.  As a distribution on a finite interval [0,1] it has a mean.  That mean is the probability of LoC.  It's single valued.  It's the best estimate of the probability of loss of crew.  The confidence interval does not change that.  Given all available information there is a single probability.

The confidence interval tells you nothing about the fate of the next launch.  It does tell you something about the range of possibilities for say a hundred launches and perhaps the fate of the program, but that should not be either a comfort or distress to the next crew.

First, I think you're right that placing the pLOx value in the center of the confidence interval, rather than at the lower bound as I was doing, is the correct strategy.  Now I'm going to nitpick with you, not to give you a hard time but because it'll help me get all these concepts straight in my head.

Definitions:

Bernoulli trial:  Any experiment with exactly two outcomes, "success" or "failure".  Each trial is independent of the others.

Bernoulli process:  A sequence of Bernoulli trials, of finite length.  The length is effectively equivalent to the sample size for finding non-boolean statistics.  A Bernoulli process where each outcome has a 50% chance of occurring per trial has a binomial distribution.

n:  The number of trials in a process.

p: The number of successful trials in a process.

q: The number of failed trials in a process.

Statistic:  A parameter to be estimated by means of a process (Bernoulli or otherwise).  In our case, pLOC, pLOM, and pLOV are all statistics.  In any given process, we'll derive each of them from q.  Note that we can derive three different processes from the same set of data, simply by assigning different criteria for "failure".

Confidence level:  (And here's where your definition goes somewhat awry.)  A fraction of the number of Bernoulli processes (not trials!) whose pLOx statistics are expected to fall in the confidence interval.

Confidence interval:  The upper and lower bounds over which the same fraction of statistics (aka the confidence level) from each process (not trials) will fall.

A notation thing, just to beat the dead horse and make it even deader:  "95% CI = [x, y]" means, "The confidence interval, computed at a confidence level of 95%, has a lower bound of x and an upper bound of y."

For our purposes, there's exactly one process for each statistic.  That's a very common case, and the confidence interval is still valid.  As you said, the pLOx is indeed a distribution.  But the distribution comes from a set of processes, not individual trials.  This is the nit that I'm picking.

Note that q/n is the sample mean for the number of failures in our single process.

From here, the Central Limit Theorem applies, which says the mean values of set of samples (Bernoulli processes in this case) will form a normal distribution around the actual population mean.  Note that the lack of multiple processes now makes our confidence interval a bit suspect for the following reasons:

1) We don't really know if the population of failures is normally distributed.  Given that we'll have a fairly small sample size (i.e., a process with a limited number of trials), conventional wisdom says we should use a t-distribution, which has fatter tails than a normal distribution.

2) We're dealing with failure rates that are close to 0.  I can't figure out whether that's a reason only to use the lower bound of the confidence interval or not--opinions seem to vary.

This is why there are heuristics like Lewis points and the adjusted Wald scores instead of straight confidence intervals, which don't work well for extreme values and small sample sizes.  Unfortunately, these are all calibrated to 95% confidence levels, which, while they're probably good for pLOC, are probably two restrictive for pLOV or pLOM in non-crew cases.

FWIW, pLOC of 1:500 for ascent or descent, using adjusted Wald with the 95% CI, requires 957 consecutive successful flights.  For pLOC of 1:1000 (I've seen both as criteria in various documents), requires 1917 consecutive successes.

Moral of the story:  You're not gonna get to what NASA would consider acceptable criteria through purely empirical means.


My argument through all of this is that SpaceX has a clock running for when they have to have Starship crew-rated.  If there were no clock, it might be perfectly reasonable to wait until they have 1000 launches under their belts, build a PRA with everything they've learned, and prove to NASA that they had the requisite reliability.  Then it might easily be that they could engineer in enough reliability that no escape systems were necessary.

The best motivation for the clock is twofold:

1) They've got customers who've put down deposits for flights, and they're not going to be happy if they have to wait a decade or so.

2) Even though they're being good NASA contractors and keeping their mouths shut, they'd dearly love to put SLS out of business.  Not only would this be directly good for SpaceX, because a crew-rated Starship is really the only logical alternative to SLS/Orion, but it would be good for NASA, which would then have more money for lunar missions--which would be indirectly good for SpaceX, because more missions means more payloads, and Starship will wind up carrying a lot of those.

I suspect that Maezawa's, Isaacman's, and Tito's patience will start to wear thin by 2028 or 2029.  And the opportunity to strangle SLS in its crib runs out when EPOC phase 2 gets finalized.  Since EPOC 1 hasn't been finalized yet, there's some time yet, but my guess is that also starts to get sketchy by 2028 or 2029.

So:  They've got roughly 5-6 years to crew-certify.  I remain extremely skeptical that the range capacity, to say nothing of the Starship launch capacity, exists to do enough launches in 5-6 years to be able to crew-certify empirically.  I'm also skeptical that they'll have enough confidence in the validity of a PRA model by then.

The alternative is to engineer-in launch escape.  IMO, this is the only viable pathway, given the time restrictions.

[jimvela]

Is crew certification required if early crewed flights operate as development flights, and all crew and passengers sign proper waivers?

[InterestedEngineer]

The alternative is to engineer-in launch escape.  IMO, this is the only viable pathway, given the time restrictions.

You almost had me till this.

Tell me which launch escape system or capsule has been trialed 1000-ish times.   None?  Oh dear.

The answer is a combination of whatever they used to certify Dragon which has only a dozen flights or so under its belt (and just took civilians up).

Which I think is bottom up analysis of risks.  One could call this "organized wishful thinking", wishful being what bites you when you do it wrong (Challenger)

Which converges very quickly when there's a large sample size of say 100.  For example after 100 flights you know all the interactions and you very precisely know the true std deviation and even more the distribution shape of all your constituent components, instead of assuming a normal¹ distribution.

So the combination of empirical plus bottoms-up (please give me a good name or take a drink), which will converge to a truly safer system than Dragon in < 100 flights.

---
¹ As Taleb is always pointing out, the normal distribution is the most optimistic distribution, and that's why it's the wrong one a lot of the time.

[TheRadicalModerate]

The alternative is to engineer-in launch escape.  IMO, this is the only viable pathway, given the time restrictions.

You almost had me till this.

Tell me which launch escape system or capsule has been trialed 1000-ish times.   None?  Oh dear.

The answer is a combination of whatever they used to certify Dragon which has only a dozen flights or so under its belt (and just took civilians up).

You're making my point for me.  Pure empirical is way, way too expensive.  The alternative is PRA (which is the name you were looking for in your "bottoms-up" question).

But there are two potential problems with PRA:

1) First, PRA relies on extensive experience.  A lot of the PRA that applies to F9/D2 has its roots back in Mercury and Gemini, when we first learned to characterize staged, multi-engine launchers and the capsules that flew on them.  PRA for systems with the same architecture (and that includes F9/D2) get the benefit of all the weird stuff and odd corner cases we've discovered over that period.

Starship gets some of that benefit, but not all of it.  Even on ascent, Starship is a large, unitary vehicle, instead of a second stage and a capsule.  And on EDL, it shares a little bit of heritage with the Shuttle, but almost none with any other capsule-based EDL system.

So the PRA model starts out considerably more immature than the PRA of F9/D2, or SLS/Orion, or even A5/Starliner (although PRA doesn't really seem to be the problem with that last one).

2) Sometimes you put all the best-practice engineering into your system, account for all of it in your PRA model, verify all the component and subsystem reliabilities via tests--and you still can't reach the required reliability.

And that's why capsules have escape systems for ascent:  without them, they're simply not reliable enough to support what we now consider an acceptable pLOC.  It's also why parachute systems for capsules are so friggin' hard:  despite all of our best efforts, they're a single point of failure, and if that failure rate is greater than 1:500, so is the pLOC.  They therefore require all kinds of redundant engineering, and it's also why capsules simply can't be bigger than a certain mass.

So, to return to D2, it benefits from decades of PRA on similar systems--but it still needs an ascent escape system.¹

It's certainly possible that clever engineering and engine technology that improves reliability will eventually come up with an answer that doesn't need an escape system.  But it'll require more empirical data than usual (because the PRA model is immature), and it's also possible that, just like with all the ascent systems before it (except for the one that killed people), the math just won't add up (or multiply through, in this case).

And of course Starship EDL looks like nothing before it.  It can crib parts of its hypersonic PRA model from the Shuttle and other non-blunt-body vehicles, but its control surfaces don't work the same.  Bellyflop looks fairly benign, but its failure tree is still unknown.  And of course the flip-and-land maneuver is rife with complexity and imponderables, which will have to be learned the hard way.

Finally, note that an escape system is a kludge.  It's always a kludge.  But it's a very useful kludge, because it sits there, as the very last twig on the end of an iffy failure branch, saying, "Ah, screw it, if we get here there's no recovery, but maybe we can let the crew survive by blowing them clear of whatever mess we couldn't get them out of."  The escape system doesn't need terrific reliability, because it's a last resort.  But if PRA tells you that the pLOC along a branch is 1:50, and your escape system all on its own has a pLOC of 1:54, then plunking the escape onto the end gives the whole branch the required pLOC of 1:500.²

Sometimes kludges are better than nothing.

____________
¹I still wonder if, despite NASA telling SpaceX not to, there isn't a hunk of SuperDraco landing software lurking in the system, accessible only through some cheat code that the crew gets whispered to them during training.  If your parachute fails and you're hurtling at the ground at terminal speed, why wouldn't you try a Hail Mary using the SD's?

²Robotbeat will now be preparing to tell me that the insertion of a system with a 1:54 probability of failure throughout the entire PRA model makes overall reliability worse.  But when you (properly) insert that component into all the necessary nodes in the tree, you have to couple it with the probability of being accidentally triggered.  I agree that system needs to be really, really reliable.  But it's not exactly an unsolved problem--nor is the probability of some kind of hypergolic accident.

Nonetheless, you have to do the work to characterize the reliability, and then crank the numbers through.  Unlike the good ol' days, you have configuration management software and massive amounts of compute power to run sims.  If your escape system winds up causing more problems than it solves, you'll know it.  But the real advantage  to escape systems is that they can solve problems that simply can't be solved any other way.  Getting them not to fire unless needed is usually a much easier problem than re-engineering the entire spacecraft system to get the necessary reliability--especially when your PRA models are immature.

[Robotbeat]

Pure empirical is actually cheaper than developing a LAS. I think Orion’s LAS has got to have cost at least $5 billion at this point. Starship is even bigger, the LAS would be even more expensive. If we assume a marginal cost of $10/kg, or about $1.5 million per flight, you can literally do 1000 flights with Starship for $1.5B.

…which is about how many they need to do anyway for the full Starlink constellation plus replacing Falcon for regular satellite launch. Considering how long it took to develop Orion’s LAS and to bring Dragon Crew to operations, and the fact that Starship should be capable of flying even more than Falcon when all is said and done, this is actually the cheaper and faster (and ultimately safer) option.

Also, they probably wouldn’t need to get a full 1000 flights. 100 consecutive successful flights gives you a ton of empirical data from close calls, etc, to make a proper judgement on reliability and safety.

…not to mention the development of alternative abort modes like using 9 engine upper stage for pad abort, maybe even ejection seats on early flights, etc. And if you can Engineer away the problems of an hypergolic abort system… maybe you can do the same thing for the upper stage propulsion system.

[TheRadicalModerate]

Pure empirical is actually cheaper than developing a LAS. I think Orion’s LAS has got to have cost at least $5 billion at this point. Starship is even bigger, the LAS would be even more expensive. If we assume a marginal cost of $10/kg, or about $1.5 million per flight, you can literally do 1000 flights with Starship for $1.5B.

Pure empirical implies that Starship's mission record is almost perfect after a while.  What if it's not?

I also think you're being wildly optimistic to assume that Starship is going to cost $1.5M/flight, and that there's capacity for 1000 flights in this decade. 

FWIW, my prediction on launches:

2023:  BC 2.  KSC 0.
2024:  BC 5.  KSC 0.
2025:  BC 5.  KSC 5.
2026:  BC 10.  KSC 20.
2027:  BC 20.  KSC 40.
2028:  BC 20.  KSC 80.
2029:  BC 20.  KSC 100.

Total this decade:  335.  Note that will be more than double F9/FH's cadence.

I'll also predict that SpaceX's cost/launch will start at $25M and ramp down to $10M by the end of the decade.  Average of maybe $12M over that period.  That would be $4B.

…which is about how many they need to do anyway for the full Starlink constellation plus replacing Falcon for regular satellite launch. Considering how long it took to develop Orion’s LAS and to bring Dragon Crew to operations, and the fact that Starship should be capable of flying even more than Falcon when all is said and done, this is actually the cheaper and faster (and ultimately safer) option.

As of right now, Starlink v2 is 7500 birds.  At 60 birds/Starship, that's 125 launches.  Even if they get 30,000, that's 500 launches.

And I don't know why you're going on about Orion's LAS.  Orion is a piece of pork; of course everything was insanely expensive.  That's a feature, not a bug.

You're also assuming that launch escape was a major chunk of the delays associated with D2.  I've seen no evidence that's true.  Even the gap between the SD explosion and the in-flight abort test was only 8 months, hardly a massive slowdown in test tempo.

…not to mention the development of alternative abort modes like using 9 engine upper stage for pad abort, maybe even ejection seats on early flights, etc. And if you can Engineer away the problems of an hypergolic abort system… maybe you can do the same thing for the upper stage propulsion system.

Raptors. Don't. Start. Quickly.

And when they do start, each one generates almost 700kg/s of mass flow.  That's 6.2t/s, into an enclosed space that sits atop a massive methane tank.

But I'm glad to see you now entertaining abort options.  That's a step in the right direction.

[Robotbeat]

...

…not to mention the development of alternative abort modes like using 9 engine upper stage for pad abort, maybe even ejection seats on early flights, etc. And if you can Engineer away the problems of an hypergolic abort system… maybe you can do the same thing for the upper stage propulsion system.

Raptors. Don't. Start. Quickly.

If you say it dramatically, does that make it a law of physics? They'd have to make changes to the engine to make it start quickly. But none that violate any physical law. You're speaking of "Raptors" as if they're some static, unchangeable design when in fact the design is changing on nearly every vehicle they develop.

And when they do start, each one generates almost 700kg/s of mass flow.  That's 6.2t/s, into an enclosed space that sits atop a massive methane tank.

We've recently seen suggestions they could add blow-out panels or lattice there, actually.
https://twitter.com/TDSN19/status/1664305744676519938?s=20

But I'm glad to see you now entertaining abort options.  That's a step in the right direction.

I always have been entertaining these kind of abort options, because that's what SpaceX has been saying they're going to do. Keep up!

[Robotbeat]

Pure empirical is actually cheaper than developing a LAS. I think Orion’s LAS has got to have cost at least $5 billion at this point. Starship is even bigger, the LAS would be even more expensive. If we assume a marginal cost of $10/kg, or about $1.5 million per flight, you can literally do 1000 flights with Starship for $1.5B.

Pure empirical implies that Starship's mission record is almost perfect after a while.  What if it's not?

Falcon 9 is only partially reusable, but it has already DOUBLED the previous record for consecutive successful launches, achieving around 200. Falcon 9 has also successfully landed over 100 times in a row, which is even more impressive as that implies no engines out even after being used over 1000 times in aggregate. So, given Starship will be *fully* reusable, that is reason to believe it could eventually achieve that reliability level (after a few more failures in the next 100 flights).

I also think you're being wildly optimistic to assume that Starship is going to cost $1.5M/flight,

Well certainly not if they use your hyper-expensive abort system! Here you keep doing the "I assume they will not do what they claim they're attempting, therefore that counts as evidence they cannot do it" thing.

and that there's capacity for 1000 flights in this decade.  ....

"capacity" is not a fixed number, and they'll need most of those flights this decade JUST to deploy full Starlink 2. You can't just *decide* there's "not enough capacity" and use that as evidence.

[chopsticks]

I also think you're being wildly optimistic to assume that Starship is going to cost $1.5M/flight,

Well certainly not if they use your hyper-expensive abort system!

This double standard is what I find really annoying about these discussions. You said earlier that Orion's abort system cost billions of dollars and implied that it would be the same for Starship, meanwhile glowing about how dirt cheap everything else is/will be for the rest of the program. Really? That's the argument?

Why do you anti-abort people believe that somehow SpaceX is incapable of making an abort system for Starship that would be cheaper than systems that have come before, as they have been doing with the rest of the system? You have no problem having faith in SpaceX that they will shatter all sorts of records with Starship, but nope, an abort system is too hard and too expensive.

[spacenut]

An escape pod like the Aardvark aircraft could be made to eject on ascent at a 45 degree angle to the rocket on vertical.  Also, the rocket could turn with Starship TPS side facing down if an ejection is needed after turning in the upward arc. 

When descending, the same could be done even if the flip is not lining up vertical.  Ejection at a 45 degree angle would keep the pods at the minimum horizontal.  Parachutes could deploy immediately after the solid rockets complete their burns. 

I don't know how much this would add to the mass or take away from cargo or other crew needs, but this idea might work on early crew flights.  Pods could be placed on the opposite side of the TPS.  Access could be internal after entry into the crew Starship. 

Once a few hundred successful flights are under their belt, they might be done away with. 

Then again, even if NASA never wants to fly on a Starship for any reason, crews can be brought up with Dragons or other approved vehicles. 

Starship will however have to land at least on Mars with it's atmosphere.  Will it come in belly first and flip to land like it is supposed to on earth?  Or will it be necessary on Mars. 

[zubenelgenubi]

Moderator:
Play the ball, not the man. 🏀 ⛹️‍♂️

[TheRadicalModerate]

Raptors. Don't. Start. Quickly.

If you say it dramatically, does that make it a law of physics? They'd have to make changes to the engine to make it start quickly. But none that violate any physical law. You're speaking of "Raptors" as if they're some static, unchangeable design when in fact the design is changing on nearly every vehicle they develop.

It violates a lot of physical laws.

What sorts of changes are you imagining?  Will these new Raptors have turbopumps?  Will they be FFSC engines?  There's a bootstrapping process involved in starting an FFSC turbopump that has a lot of physical limitations on it.  You have to:

1) Spin-start the turbopumps using pressurized gas.
2) Feed LCH4 into the regen cooling system and wait for it to flash to vapor and expand until it reaches a pressure great enough to inhibit backflow through the preburner injectors.
3) Cross-feed LOX into the methane preburner and GCH4 into the LOX preburner.
4) Ignite the preburners, but only when the pressure across the preburners is high enough to force the exhaust gases through the turbines, rather than back out the injectors.
5) Fiddle with the mixture so you don't produce a hard start in the chamber.
6) Damp out all of the various back-pressure oscillations for a stable start.
7) Ignite the main chamber.
8 ) Wait for things to stabilize again.
9) Throttle up.

Waving your arms and saying, "They'll just start them faster," doesn't really work.

Now, if you're going to have a 400bar pressure-fed engine with about 680kg/s of mass flow, that'll start almost immediately.  But it'll also only run for a few seconds before the pressure source that's feeding it is depleted.  Otherwise, the pesky laws of physics get in the way.

BTW:  How long do you estimate SpaceX can afford to wait to build this magic new engine?  5 years?  10?  15?  We're dealing with Starship abort options here, not some follow-on design.

And when they do start, each one generates almost 700kg/s of mass flow.  That's 6.2t/s, into an enclosed space that sits atop a massive methane tank.

We've recently seen suggestions they could add blow-out panels or lattice there, actually.

Those are very interesting.  But if they're actually blow-out relief for hot-starting 9 Raptors while the SS is still mated to the SH, we'd have to compute:

a) The area of the openings.  There are six of them around a 28.3m circumference.  If we model each one as a half-circle, each on has an area of 4.4m² (I'm assuming that the gridwork consumes half the area), so total exit area is 26.2m².
 
b) The mass flow rate (roughly 680kg/s * 9 = 6120kg/s--I was off by a bit the first time I did this).  This then needs to be converted to a volumetric flow rate.

c) The flow factor K_v.  This is usually something that's a property of a particular valve, and I have no clue how to derive it.  It's dependent on the opening area (26.2m²), though.

d) The specific gravity (ρ_exhaust/ρ_water of the exhaust.  We're dealing with highly compressible flow here, so this will depend on the pressure and temperature in the SS skirt and SH interstage.  (This seems like a pretty arbitrary metric.  I suspect that it's normalized this way because that's how the flow factor is written.)

e)  If we have all that, then a very rough estimate for the pressure differential in the skirt will be pInterstage - 1bar = flowRate²*specificGravity/flowFactor².

Show me that pInterstage is less than about 16bar and I'll believe that this is actually a blow-out system.  Not a sarcastic or even rhetorical statement:  I don't know how to do the computation of the flow factor.  But it's pretty clear that the interstage pressure will be the essential figure of merit for whether this will work.

FWIW:  My guess is that this gridwork is venting to avoid the buildup of gases during pump chill-down.

[Robotbeat]

Raptors. Don't. Start. Quickly.

If you say it dramatically, does that make it a law of physics? They'd have to make changes to the engine to make it start quickly. But none that violate any physical law. You're speaking of "Raptors" as if they're some static, unchangeable design when in fact the design is changing on nearly every vehicle they develop.

It violates a lot of physical laws....

You wrote absolutely no physical laws, there, just "it's hard." As if we didn't know.