Given the significant cost saving in sending up second hand rockets, and given SpaceX have demonstrated the capability, why are they still sending up so many new rockets?
That, plus the market is still somewhat skeptical of the idea of reusable boosters. Many customers still want brand new ones, and are willing to pay for them. (Rocket people tend not to jump on bandwagons.)
Give them time; there will be more used rockets launched.
According to that Wiki page, the June 23, 2017 BulgariaSat-1 mission was also on a reused booster.
I would guess that in the eight launches since the first reuse, the main reason you haven’t see more reused boosters is contracting. Whoever is paying for the flight can opt for a reused booster but for reliability and insurance reasons the customers may not be ready to do that yet. I am not sure that SpaceX has made public how often they are offering reused boosters yet. The refurb process is probably still somewhat slow.
I believe that Iridium has said they will begin using the “flight tested” boosters.
They are going to be engaged in the delicate process of working out how to get the time for the refurb process down. Which is tricky. Right now I would not be surprised if the refurb process was actually not economical, partly because not only do they need to do the refurb to the current spec, but then check that the refurb they have done is actually good enough for reflight, and then update their process, which adds even more time. Rinse and repeat. But that is how you make progress. That the launches on refurb’ed boosters have been good validates their process, but the goal of getting the refurb time down to 24 hours is still a long way off, and the desire to attain that will cost in both time and money. But costs that will be amortised. Once they get to a 24 hour turnaround we would reasonably expect that the entire game has changed.
There are always going to be launches where the booster is not recovered. Sometimes the margins are too slim to have any fuel left for the return. If the customer wants their payload in orbit, that is the cost.
It’s where hype meets reality. They probably at a minimum have to take all the engines apart and clean the components of combustion residue. (that coking you get from burning kerosene). They probably have to change all kinds of seals and bearings that actually can only handle a single flight. Longer life components are maybe possible - keep in mind that with rockets, any added weight reduces performance, and longer life parts tend to be heavier.
The previous experience with refurbishing has of course been dreadful. A Space Shuttle required insane amounts of work. A lot was due to compromises made in the design to save up-front money, and an element of paranoia about losing a craft. They were in a situation for years where they regularly cannibalised one orbiter for parts to enable another to fly. They had all sorts of unexpected and unpleasant issues. Some of which conceivably could turn up for Space-X. Things like cavitation damage to turbopump blades. Requires a total teardown to fix, and lots of care with borescopes to check for.
The first ever recovered Falcon-9 booster will never fly again, and was probably subject to a range of destructive tests. This would have guided later refurbs, and guided the checks make on those refurbs.
In principle all you should need to do is check for damage, replace the one shot components (explosives, burst diaphragms and the rest of the engine startup consumables, ablative protection, batteries etc) refill the fluids, and truck it out. The aviation industry has become very good at using trending analysis to look for problems rather than defined maintenance activities. Space-X may well be able to do much the same. But they need to obtain the experience to know what to look for and to work out what the margins actually are.
It is the unknown unknowns that are the problem. Finding a path where you economically close down those is the trick.
As everyone else has said, the economics of the concept makes it difficult. Stranger has weighed in on this several times. I note that he hasn’t said a lot lately on the subject-perhaps he has had contracts with Space-X that precludes him commenting? Some of the anticipated problems have turned out to be less severe? I have no idea. The reusable concept has been looked at since the sixties. Up until now, it has always been rejected an not worth it. Musk, as is his habit, has come to a different conclusion and perhaps will be proven right. Heck with the rest of the world-that is how progress is made.
The economics are tricky though. If you make lots of single-shot boosters the per unit cost goes down. So by reusing your own boosters, you are competing with yourself. The difference between the cost of a new booster vs the cost of a carefully inspected and refurbed booster, as everyone has said, may not be that great. Perhaps the customers are willing to pay new car prices to get the security, perhaps they will like the savings enough to buy used. Only time will tell.
As I understand it and as others have said in this thread, what Musk really wants/needs is the 24 hour turnaround time. Land the first stage, move it quickly to the pad, mount a new payload on top, refuel and launch. THEN the cost savings will become significant. But it is going to be a while getting there. But one can’t get there without taking the first step.
The short answer is that SpaceX has a huge backlog of orders, and the majority of them paid for a new rocket (because that’s all that was available at the time). It’s most likely difficult for many of these to transition to a CPF rocket (Certified Pre-Flown–thanks, LSLGuy), due to implications for insurance, the fact that the flight was already budgeted, etc. So we’ll see new cores dominating for a while yet.
The majority of landed boosters have been with the “Block 3” * iteration of the Falcon 9. Block 3 can be reused but there are some suboptimal aspects. The latest version is “Block 5”, which fixes many of these problems–among other things, the legs will have a different and more easily reusable design (they’re thrown away right now). These aren’t flying yet, so I suspect SpaceX is just slow-rolling Block 3 reusabililty, learning the lessons they can for a few flights but not pushing it very hard. Block 5 will (allegedly) have 24-hour turnaround (theoretical: they don’t plan on actually using this), which basically means that whatever the refurb/inspection requirements are, they’re either trivial or automated.
No one outside of SpaceX or their customers really knows how core revisions are tracked. The “Block X” terminology is something Musk brought up once but probably doesn’t directly translate to anything internal. Instead, it’s probably a bundle of semi-independent features that are tracked separately.
I understand that having a place to store a rocket for a week costs money,
but I think they can probably afford some land to let it cool down before reloading it. Because actually cost is reduced by setting up multiple rockets at one time.
So you have 14 rockets … and half the fleet is reset in one week and half in the second week. So why does it a take a week ? you get the expert to do job 1 on Monday, Job 2 on Tuesday, job 3 on Wednesday. These may be outside contractors, or using equipment that has to come and go , in order to make room …
Well anyway my point is that a pipeline strategy for rockets, eg 14, might be enough that that one rocket becomes ready to fly each day, on average.
I think Musk is smart enough to recognize that his buyer’s market will *grossly *overreact to any failure of a CPF booster. Even if the failure is later attributed to something unrelated to its earlier flight(s).
So Musk needs a long past manifest of successful new booster flights, and needs provably robust reusability *before *he puts very many eggs into a CPF basket. They can get a bunch of “provably robust” by doing intensive post-flight disassembly and testing. Something most conventional designs could never do since if anything was recovered at all, it was from the bottom of the sea.
As to the refurb process …
As the learned **Stranger **has said so often, the key is to all but eliminate refurb and reinspection, not just reduce it a couple percent from the thousands and thousands of man-hours it takes with conventional designs.
Which means designing from the git-go for less complexity, greater robustness, and therefore a bit more weight. And including ways to know, or reliably predict, health of all these parts without having to take them apart each time to look at them. With all the attendant risk of damage in disassembly or reassembly. And with the risk of making assembly errors that convert a component that worked perfectly on flight #1 and would have worked perfectly on flight #2 into a component that absolutely *will *fail on flight #2.
My interpretation is that extensive postflight work on early block 1s gave them the knowledge to improve to block 3. And extensive postflight work on early block 3s gave them the knowledge to improve to block 5.
Will 5 be the “production” model that gets heavy re-use? Or will that wait until 7? Only the future knows for sure.
It *is *a virtuous circle engineering-wise. Making that into a virtuous circle economics-wise is still an open question.
You know, what we actually need is much, much better robotics. With this new machine learning stuff, it now appears like it will eventually be feasible to have robotics software that is vastly more flexible and smart than anything seen in the past. That is, the robots wouldn’t use custom, ‘one-off’ software solutions for a given task in the factory. Nor would they have their motions programmed for a given task.
Instead, they’d be given the schematics of what they are trying to accomplish, and design data, and the robotic systems themselves would be able to figure out the motions needed. As failures and unusual situations happen, all robots would be networked, so in theory robots would be able to “call the cloud” and get help from a vastly larger pool of software tools that would analyze the problem the robot is having and use information shared by other robots.
I sort of imagine a robot cleaning the floors of a factory and encountering a Mr. Goodbar wrapper. It “calls the cloud” and a robot in a factory 3 states over had been told by a human operator that wrapper was trash, and the knowledge of “classification : trash” is shared by a machine learning network running on the cloud. So the robot, pretty confident it’s trash, picks it up and throws it away without asking a human.
Similarly, a robot in Mr. Musk’s Tesla plant is having a heck of a time with a tiny bolt and washer. It “calls the cloud” and a robot in a different plant has discovered by accident a way to grab that kind of bolt reliably. So it uses the same technique.
So these smarter robots could eventually put rockets together. And then the recovered rockets could get recycled by having the robots disassemble the entire rocket and send every mechanical part through laser scanners and X-ray machines.
Any part that doesn’t meet new part tolerances gets recycled, and so every rocket is “new”, having 100% automated inspection of every component.
Since the robotic systems don’t get bored, and develop their inspection checklists automatically, every rocket goes through a several million point inspection, done as it is assembled. This eliminates nearly all manufacturing failures.
For a one off custom design such deep learning might someday be of value. For anything that we are making more than one of, it doesn’t help much. You work out how to make it, and then program the dumb robots to just keep doing it. Elon is pretty good at that, just look at the Tesla factory.
The thing about mechanical contrivances is that bathtub curve. The time a mature booster design is most likely to fail is on its first launch. The trouble is that getting to a mature design is not something that comes quickly. The aviation industry spends a lot of time working through a new aircraft design before it is considered passenger ready. Even very old booster designs are still really just beta stage development items in comparison. Probably more like alpha. The nature of the industry has never really encouraged efforts to drive the technology to more like mass production.
The problem with a total tear down and reassembly is that more than a few times what kills a device is a fault in assembly. Once a part has survived one flight, it is more likely to survive another flight than a new part is. Also worth remembering, boosters are not flown as virgin builds. Minimally they have been subject to tanking, and an all up engine start, run up to full power, and shutdown. Many are run through their entire flight profile, just bolted to the ground. Then they get to do it again, but for real. The only difference then is that the vehicle is subject to the stresses of flight, but all the other stresses have been delivered more than once on that item. They don’t do anything to the booster after these tests. They fly as is, and for very good reason. It worked that time. Don’t mess with it.
This is part of why refurbishment isn’t so wildly insane as some worry. Rockets are not designed so close to their operational margins as to be wrecked after one flight. Rather they are designed with enough margin that they should survive a reasonable set of anomalies, and if flown within these margins there is no reason to think any part has been stressed past reuse. There is enough telemetry on the craft to have a very good idea if there has been an excursion outside these bounds.
So, like a jet airliner. Once you have a simple enough design that you have confidence in, and you have confidence it has not been pushed past its design limits (like a hard landing, or an engine has been running outside of nominal limits) there is no reason to think it can’t just go again. Airlines learnt long ago that the only reason to perform maintenance on much of an aircraft is when you have some indication it needs it. Eventually you reach things like structure lifetime limits - which are determined by things like crack propagation - and are possible to reason about in a sensible manner and schedule times for inspections. But they don’t need to do this after each flight, rather after years of service.
There are some evil limits on space systems that are designed in. Things like thermal limits that will be exceeded if a system runs longer than the flight profile expected. There is no reason to set up cooling for a subsystem that is able to keep it cool forever when it will only operate for 30 minutes. I was amused to note that the Apollo Lunar Module’s landing radar was only rated for 45 minutes use. After which it overheated. Not a problem. No doubt that radar system could have been used many times over if it were recoverable. Same with a Falcon-9 first stage.
Wait, what? You do realize that the only reason we can’t just automate all factories right now is that the process of making most complex goods is not quite smooth enough that it’s the same task every time. And, uh, it defeats the purpose to use deep learning for a single task. What you want is a vast network of reliable, well trained deep learning systems that are used as libraries.
To name something we humans do all the time - for the classic “separate cats and dogs in these images” challenge, part of the reason we can do it well, still better than machines, is we can use our past knowledge of what common environments look like to subtract the background, leaving our consideration just “cat or dog”. In addition, we can use general knowledge of what an animal looks like on the specific task of determining which kind of animal this is in front of us.
Robots dealing with machine parts in a factory can benefit from the same tricks.
Also, I name in my post another useful technique - automated disassembly. This is basically impossible today because disassembly is a far harder task that manufacturing because the object to be disassembled has worn parts, stripped screws, dirt…
It’s impossible to do without something like machine learning.
The biggest problem with automation is that robots are seriously expensive, and humans are vastly cheaper. Sometimes much cheaper. Humans have lots of degrees of freedom and proprioceptive abilities robots don’t. They can do messy things like thread wires and push circuit boards into small cases around lumps and bumps.
Something as trivial is working out how to better hold a bolt is something that needs programming once into a robot and that is it. It doesn’t need deep learning. A library of moves is fine, but why it needs massive AI is not clear.
AI still remains massively over-hyped. We seem to be having yet another round of hype about the ability of AI systems, and as usual the popular press is talking things up way beyond any reasonable limit. Once you get under the surface of a lot of the modern AI you will quickly become disappointed.
There is a big difference between a science fiction view of what the future may hold and what can be reasonably expected in the next decade.
SamuelA, ISTM you’re blithely assuming the robots can learn flawlessly how to create their own checklist for assembly and disassembly. How exactly do they go through the learning curve? How do we, or they, measure when they’re far enough along the curve that we, or they, can let them loose on flight hardware instead of test hardware?
Who writes the flawless specs for the goals for them to flawlessly learn to?
I agree that once the process has been utterly and completely routinized, then robots have a potential increment in accuracy & reliability and a potential decrement in cost vice using meat-based general purpose intelligences.
But that’s pretty large miracle that occurs there in your step #2. It’s the same miracle that we can’t quite achieve using our meat-based GPIs either.
How about a third option, Musk is a snake oil salesman, who might one day be in handcuffs being-escorted by FBI agents, and the reuse while technically impressive has nowhere the performance and savings that SpaceX PR people are claiming?
Yeah, you can drop the ‘snake oil’ bit. Musk is 10,000 times more intelligent and ambitious than you are and he has a pretty good track record. Even if the re-use thing never works out, Space X is producing rockets and putting things into space cheaper than their old-school nasa teat sucking rivals.