Slashdot Mirror
Elon Musk Explains Why SpaceX Prefers Clusters of Small Engines (arstechnica.com)
An anonymous reader quotes a report from Ars Technica: The company's development of the Falcon 9 rocket, with nine engines, had given Musk confidence that SpaceX could scale up to 27 engines in flight, and he believed this was a better overall solution for the thrust needed to escape Earth's gravity. To explain why, the former computer scientist used a computer metaphor. "It's sort of like the way modern computer systems are set up," Musk said. "With Google or Amazon they have large numbers of small computers, such that if one of the computers goes down it doesn't really affect your use of Google or Amazon. That's different from the old model of the mainframe approach, when you have one big mainframe and if it goes down, the whole system goes down."
For computers, Musk said, using large numbers of small computers ends up being a more efficient, smarter, and faster approach than using a few larger, more powerful computers. So it was with rocket engines. "It's better to use a large number of small engines," Musk said. With the Falcon Heavy rocket, he added, up to half a dozen engines could fail and the rocket would still make it to orbit. The flight of the Falcon Heavy likely bodes well for SpaceX's next rocket, the much larger Big Falcon Rocket (or BFR), now being designed at the company's Hawthorne, California-based headquarters. This booster will use 31 engines, four more than the Falcon Heavy. But it will also use larger, more powerful engines. The proposed Raptor engine has 380,000 pounds of thrust at sea level, compared to 190,000 pounds of thrust for the Merlin 1-D engine.
For computers, Musk said, using large numbers of small computers ends up being a more efficient, smarter, and faster approach than using a few larger, more powerful computers. So it was with rocket engines. "It's better to use a large number of small engines," Musk said. With the Falcon Heavy rocket, he added, up to half a dozen engines could fail and the rocket would still make it to orbit. The flight of the Falcon Heavy likely bodes well for SpaceX's next rocket, the much larger Big Falcon Rocket (or BFR), now being designed at the company's Hawthorne, California-based headquarters. This booster will use 31 engines, four more than the Falcon Heavy. But it will also use larger, more powerful engines. The proposed Raptor engine has 380,000 pounds of thrust at sea level, compared to 190,000 pounds of thrust for the Merlin 1-D engine.
Seems like a good idea to me but I'm no rocket scientist.
Of course, it has to be actual redundancy. The Soviet N1 moon rocket had a problem that when its engines failed, they tended to take out adjacent engines. You have to be absolutely sure that failures aren't going to spread (pieces of shattering turbopumps, fires, backpressure, etc), or you're actually making the problem worse.
Of course, everyone working on rockets today knows the lessons of the N1 and it'd be incompetence not to exhaustively test for resilience against cascading failures.
Beyond redundancy, one neat thing about engine clusters is that you can create a virtual aerospike effect to some degree.
It's time for Operation Crazy Plan.
I thought the F in BFR stood for something else than Falcon...?
Yes but Falcon doesn't get censored in interviews
Pain is merely failure leaving the body
Yes. Redundancy is always good.
Obviously SpaceX has calculated this but Id like to see a graph of the probability of flight failure of a rocket with 5 big engines and a rocket of 31 small engines. The more engines the higher the chance one will not work but also the higher the redundancy. The fewer engines the less chance one will not work but also the greater the chance one going out dooms the flight.
I came to the datacenter drunk with a fake ID, don't you want to be just like me?
Yes. Redundancy is always good.
Yes. Redundancy is always good.
Yes. Redundancy is always good.
Yes. Redundancy is always good.
Yes. Redundancy is always good.
CLI paste? paste.pr0.tips!
Yes, but at the same time it's more complex. I think today we have the tech for such a thing, but if we look back at the Soviet who tried (first? don't quote me) this approach with the N1 moon LV - well it failed miserably. There are a lot more risks and much more complexity, which is to the credit of SpaceX!
So saying "Duh, it's obvious" is a bit shortsighted. Redundancy and scaling is hard, especially when you're talking about a rocket. Pumps, fuel, precooling, spin up, and all that are non-trivial. Even if you take like 9 engines in a square, if one fails, yeah you have 8 remaining, but the thrust is reduced by 1/9, the LV is now unbalanced especially if more than 1 goes down on the same side (e.g. engines themselves must compensate with gimbal or vernier needs to do this, I'd like to see the software to control this, surely a beast!!), to achieve the same final delta-v you need to burn the engines longer, so they need to be rated for much longer burn time if you still want to make it to orbit, which means more chances of failures, engines overheating, and then if an engine explodes it could take out others with shrapnel, and I go on...
And that does not take into account the R&D necessary to relight the engines twice AFTER the initial burn for the re-entry and landing! So yeah, quite an R&D achievement for SpaceX to have such reliability!
Mad props.
Us Gen Xers just say it: big fucking rocket...
Wow Grandpa, that's so badass. You're using that .. word .. like it's just a word. I'm terribly impressed.
when kids are in ear shot.
Friendly reminder, millenials aren't kids anymore.
CLI paste? paste.pr0.tips!
To a certain extent, SpaceX's architectural approach of many engines has arguably reduced costs. By making more copies of single engine design, the cost per engine has dropped significantly.. The manufacturing reliability is better, too. (What would the failure rate of a Model T have been if Ford was only building one per week? Building lots of something continuously brings you up the learning curve faster, reduces mistakes, and forces you to invest in tooling and fixturing that ensures each step is successful and repeatable.)
In this case, I think it is likely that the cost/kg and the reliability of a 9-engine rocket is better than a rocket that had a single engine of comparable power.
As you say - there are limits to this approach. (I'd call it modularity, rather than redundancy.) The efficiency of rocket engines doesn't scale down well and, as you point out, requires a build up of all the ancillary equipment.
If he meant fail catastrophically, he would have said fail catastrophically, you fucking troll.
Pot Kettle Black.
If one fails catastrophically you basically have a very large bomb on your hands, of course you've had it at that point. If you're going to chime in here, at least try to advance the discussion with something that's worthwhile.
Let me explain it in simple steps.
The point is that it's worth considering that the chances of one of 31 smaller engines failing could be larger than the chances of one larger engine failing catastrophically. Note I am not saying that it is, the engineering considerations could make larger ones more prone to failure, I'm just saying that it's worth considering. There have been a number of catastrophic rocket engine failures in space history, so it is certainly a possibility. Also, this is different from the data-centre analogy as it is very unlikely that a server failing will destroy the whole data centre.
Friendly reminder, millenials aren't kids anymore.
Yes they are. Get off my lawn!
The rocket is built to contain engine explosions. We don't know if that'll be effective for all engine failures, but they've already had at least one engine failure on a F9 flight without consequences for the mission.
Do you have to shut down computers opposite to the one that fails to maintain "computing balance"?
The Soviets did that on the N-1 because it allowed them to install the engines without gimbaling hardware, simplifying the design. The F9 does have gimbals, so it doesn't need to shut down the opposing engine.
Does a failing computer fundamentally alter your mission profile to the point that you have to change the computations for ALL other computers?
So what? That's what computers are really good at.
The N1 failed through a combination of lack of money, lack of political will and losing the space race.
The plan was to skip building a test stand for the first stage (which would be large and expensive, to cope with 5000 tons of thrust, and take lots of time to buil). Instead, they'd do test flights, fully expecting a number of initial failures. 14 test flights were planned.
After the Apollo 11 landing, the urgency was lost and funding slowed. The last straw was appointing Valentin Glushko as head of the Soviet space program. He was a known opponent of the N-1, favoring his own design.
A lot of the Soviet plans were based around the expectation of failures. All of their (numerous) Venus missions, for example, were launched in pairs. The idea was that the incremental cost was low but the initial costs high, so you might as well send two. And if both work, you collect two separate datasets, from different locations. Usually when one failed they pretended it was an experimental or military launch - for example, Venera 4's twin was Kosmos-167, while Venera 7's was Kosmos-359.
It's hard to call one approach the right approach and one the wrong approach. The Soviet approach certainly paid dividends on Venus, but their Mars programme was a miserable failure compared to the US.
It's time for Operation Crazy Plan.
They said that failure was because of lack of fuel, so more engines wouldn't help that.
IIRC it was a chemical that starts up the engines that ran out TEA-TEB (Triethylaluminum-Triethylborane) So they could not fire the others
Actually, they had an engine control computer called KORD. While simple, it was an electronic computer. It was fed 4 types of measurements from each of the engines, and based on a simple algorithm would decide if they were out of acceptable operating parameters, and if so issue a command to shut down the offending engine (in theory, before a catastrophic failure) and its opposing counterpart. It would then ramp up the good engines to compensate for the loss of the dead engines (they defaulted to operating at 75% throttle to allow for this, as well as to reduce stress on the poorly tested engines).
Computer controls based on sensors was a new thing for the team, and the difficulties in filtering out bad data came back to haunt them in the first flight; it misinterpreted pyro noise as a turbopump spinning out of control and shut down a pair of engines, then interpreted pogo and a different engine failure as all of the engines going bad - and shut down the engines for all of the stages, so they couldn't even test the upper stages.
KORD also turned out to have too long of a response time to prevent catastrophic failures in engines, which was one of the things the design team was counting on to overcome the known poor reliability of the engines. KORD's rushed schedule also left it with poor debugging and too few safety checks. On flight two, in addition to still being too overaggressive on engine shutdowns in general (in response to very real engine failures), it caused its widespread shutdowns while the rocket was still over the pad, rather than trying to keep it going long enough to clear the pad. The pad explosion was one of the largest manmade non-nuclear explosions in history and set the N1 project back a year and a half.
It's time for Operation Crazy Plan.
Damn straight! And while we're at it we should get back to single piston car and truck engines. Those Europeans are screwed with their finicky 12 cylinder sports cars or even (gasp!) 16 cylinders!!! It's madness I say!
Simplify it all to a more efficient single cylinder engine. And don't even get me started on all those crappy WWII airplane engine designs...
Has my point been made? No? Sometimes the cost and/or efficiency of the engine is not the biggest consideration in a project.
No! There are two extreme cases. One, where the probability of total system failure is the sum of the probabilities of failure of components (bad), the other where it is the product (good) 1% chance of failure + 1% chance of failure = 2% chance of total failure
You can't sum chance of failure like that. That's not how the math of it works. (think about it - if you take that to it's logical conclusion with 200 failure modes each at 1% chance of failure you can end up with a >100% chance of failure which isn't possible) First you have to determine whether the failure modes are genuinely independent or not. But even if you have two completely independent failure modes with a 1% chance of Failure A and a 1% chance of Failure B, the total chance of Failure is NOT F(A)+F(B) = 2% because there is a probability of both failure occurring simultaneously so the real probability will be less than 2%.
1% chance of failure * 1% chance of failure = %0.01 chance of total failure
It doesn't work like that either unless those failure modes are such that both have to occur for a failure to occur.
I don't think this article is particularly newsworthy for anyone who's familiar with the subject or even stopped for a few seconds to think about it.
But here are a few points why multiple smaller engines are better in this case.
* Mass production makes things cheaper and sometimes better.
If you look at the cost of an engine, the raw materials are a pretty small percentage of the total cost, It's more about the manpower and tolerances required.
If you have a guy performing a certain task maybe once every two months, he'll be slower and less proficient at doing it than say every few days.
And overall, economies of scale make more engines cheaper to manufacture.
* Redundancy, as mentioned already in the thread.
SpaceX has already lost one of the engines on one of the earlier flights and continued to complete the mission.
They have walls between them that prevent the explosion of one from damaging the next.
And it's only going to improve for Falcon Heavy and BFR, they'd be able to lost multiple engines and compensate for the imbalance.
* Telemetry collection.
You get to build up a history of past performance a lot faster with ten engines than you do with one or two.
After 54 flights, SpaceX has gathered operational telemetry on 486 first stage engine firings and 54 second stage ones (Not including all the test firings).
* Throttling, maneuverability, unique thrust characteristics.
In the early stages of landing R&D SpaceX had a rocket called the Grasshopper, which was modified Falcon 9, that was able to fly up, hover, and then land.
Most larger engines would not be able to throttle low enough to maintain a hover.
This one is just a guess, but I imagine it's much easier and faster to gimbal a smaller engine, and you don't have to put the smaller, cheaper gimbaling hardware on all the engines.
It seems that the exhaust from the multiple engines behaves somewhat similarly to an Aerospike engine, which gives the configuration some extra efficiency.
That's all I could think of, but I'm sure there are more reasons.
1) more engines is more parts, which is generally a recipe for more failures
Only if you hold everything else in the system constant which is clearly not the case in most real world systems. To use a car example, modern cars have a LOT more parts in them than cars from 40 years ago but they also are demonstrably more reliable. Same with jet engines. Modern ones are more complex and with (usually) more parts but they also are more reliable. The relationship between number of parts and reliability is not a simple linear one. Many of those added parts actually contribute to the reliability of the overall system.
The most extreme example of this sort of thing ever attempted was OTRAG.
John Carmack had some interesting things to say about that at his now-defunct Armadillo Aerospace website, some of which have been preserved at Wikipedia here.
Ceci n'est pas une signature.
The irony of the second flight was that KORD shut down every engine EXCEPT the one that was reporting a problem