Was that a case of extremely engine-rich exhaust?
It’s certainly commonplace in the aerospace industry to speak of “losing an engine”. But that doesn’t usually mean misplacing the whole darn thing. I suppose they were going too fast to go back and look for it like you would for a lost hubcap.
Thank you for your indulgence. 
I would say so. Usually “engine-rich exhaust” means the insides vaporizing away. But flinging the entire nozzle overboard is a whole other thing!
The picture came from unknown sources with no details. So I am curious what exactly happened. Was the detachment of the nozzle the origin (or immediate consequence) of the failure? Or was there some other failure which caused the spinning, and centrifugal forces from that caused the detachment? Well, we’ll probably find out fairly soon.
One thing I notice is that even through it the camera appears to be in the same location between the two images above (though maybe physically different since they have a different exposure), the center engines appear to be in a different position. This says to me they’re gimballing like hell to maintain attitude. Wasn’t enough, though.
Indeed. Arguably, they still aren’t failing rapidly enough for optimal progress. One could model iterative development as an implementation of the Metropolis-Hastings algorithm for sampling a probability distribution. Basically, a random walk around a probability space where the samples are either accepted or rejected as they arrive. The optimal “acceptance” (success) rate for high-dimensionality distributions is actually only around 25%!
Whether it’s fair to model rocket development as a Markov chain is arguable, but there’s at least some theoretical underpinning for actually targeting a high failure rate.
Unfortunately if a rocket component fails too early you don’t get to test any of the other components.
That there’s ever a partial success argues for an even higher failure rate! The Metropolis-Hastings algorithm throws out samples completely. Where as a failed rocket still imparts some information. Maybe not a lot, but something.
25% of launches being a 100% success and 75% being a 99% is still too slow a learning rate. You should expect some failures that don’t tell you much other than “try something else next time”. Maybe there’s a way to model partial successes as a sequence of events rather than a grouped one.
Fires of hell:
What did they make the lens cover out of? General Products hull material?
Put it this way: Let’s say that you have 100 different components that can fail catastrophically, and each individual one has a low failure rate (say, 1%), but because there are so many, there’s still a high overall chance for one of them to fail or another. Well, you need to improve that. You need to improve each and every one of those 100 components, to get each of them to a failure rate of, maybe, 0.01% (or lower-- This depends on what you’re ultimately using the thing for).
On any flight where a failure happens, you figure out what caused that failure, and can use that information to improve that component. That’s good. But on any flight where a failure doesn’t happen, that just means that you got lucky, and still don’t know what causes any of the failures, and your next flight will have the same chance of success as this one, and might not get a good die roll.
Now, eventually, in this process, you get to the point where your flights are successful because you really have fixed all of the components. That, of course, is good. But you can never be sure you’ve gotten to that until you’ve had many, many successful flights.
Yeah, of course the thing about a rocket intended to be reusable is that unlike an expendable rocket (e.g. the Saturn V), lasts-just-barely-long-enough isn’t good enough. They’re going to be trouble-shooting bugs for as long as Starship is going to be around; the interim goal is to reduce mission kill below an acceptable standard.
Building off @Chronos’ points. …
There’s testing of components and testing of the entire system. You can do a lot of valid testing on components without needing to throw away a whole rocket ship each time a particular gizmo fails due to overtemp. If you can accurately model & reproduce the actual flight conditions.
As to sampling, etc., is the relevant sample space that of all the subsystem tests or only the all-up round tests?
In the case of Starship, the relevant space is that of all plausible design variations of a reusable super-heavy-lift rocket. The design isn’t frozen!
It’s still at the point where they are exploring variations on dimensions, structure, plumbing, heat shielding, control algorithms, and so on. Which is what you need for a very advanced system that no one has built before.
Of course, on top of that is the possibility of component failure. Which has a fuzzy boundary with design failure. Are the recent plumbing problems a fundamental design defect, or are just minor revisions required, or was it actually just a manufacturing defect? But then, manufacturing defects can be caused by a design that’s difficult or impossible to manufacture correctly. It’s all interrelated.
Holy shit that’s awesome! It’s hard to imagine engineering materials that can withstand these forces. Thinking about the surface area each Raptor is pushing against, the area it is making contact with not counting the area of fuel/ox flow, is so small it’s seems unreal that a structure could withstand it.
It’s got to be regeneratively cooled somehow. Maybe water flowing between two fused quartz plates? And then a periscope system to keep the camera farther away, and probably a filter/mirror to keep the infrared light out. Regardless, amazing that it could film the whole thing. Though you can see the incredible vibration levels from the part of the housing coming in on the left… they must have stabilized it.
The levels of power are really unimaginable. There’s more thermal power in a Starship launch than in every nuclear plant in the US combined.
Flight 9 coming up soon. Maybe May 27:
Also, they gave an update on the Flight 8 failure:
The most probable root cause for the loss of Starship was identified as a hardware failure in one of the upper stage’s center Raptor engines that resulted in inadvertent propellant mixing and ignition. Extensive ground testing has taken place since the flight test to better understand the failure, including more than 100 long-duration Raptor firings at SpaceX’s McGregor test facility.
So, not the same failure as Flight 7 (resonance in the fuel transfer tubes). Really just a generic Raptor failure that was a little too energetic to be survivable.
They’re still using Raptor 2s. Raptor 3 should be more robust. The Ship probably could have survived failure of a center engine had it not damaged surrounding engines in the process.
IMHO, Starship desperately needs an unqualified success. “Move fast and break things” used to be an admirable paradigm of being willing to take risks and learn from failure, and avoiding the “must get it perfect the first time, whatever the cost” paradigm that drove Apollo and the Shuttle. Now it’s mocked as an inability to ever get anything right and a failure to deliver on promises. People are now comparing Starship to Cybertruck, and denigrating the repeated failures as testaments to Musk’s delusions.
The first stage Super Heavy booster builds on SpaceX’s experience with Falcon 9; to an extent it’s just scaling up what they already know how to do, and it’s been the most successful part of Starship. By contrast the upper stage has been problematic all the way back to SN8 in Dec 2020. Even when the upper stages have made it to splash down, they’ve done so in damaged condition that could never have been refurbished.
I don’t think SpaceX gives the tiniest microshit about whether terminally online folk “mock” them. They can actually just do things independently of what anyone online thinks.
There’s zero evidence that any of their problems have been any kind of real-world obstacle. Starbase is now a city. The FAA continues giving their approval for launches. NASA still gives them money. The military is disappointed with everyone but SpaceX. They still get billions from private equity rounds when they need it, which isn’t often since Starlink and now Starshield are doing so well.
If some of this online ire actually fed back into real-world difficulties, that might be something else, but Redditors don’t actually have any political power. SpaceX can keep iterating in the way that makes the most sense to them.
You need to test all of the subsystems, and you need to test all of the relevant interactions between systems. But there are a combinatorically-large number of interactions between systems, and the only way to know which interactions are relevant is to test all of them. Which means tests of the entire craft all together.
Though, of course, you also still test individual components, and all of the combinations of components that you already know are relevant.