Starship On-orbit refueling - Options and Discussion

[Greg Hullender]

My mental image was of one of those "incoming prop" geysers interacting with the receiver tank prop in such a way as to induce a vortex in the receiver tank, potentially torquing the entire structure in chaotic ways each time.  BoE says 1400 ton of prop swirling at 1 rpm is roughly equivalent to a reaction wheel with 1,800,000 Nms of angular momentum, so if that kind of induced swirling is a real problem, then that's a heck of a design challenge for your pump connectors. Presumably there are simple ways to baffle/disrupt these effects, but it could mean pumping slower or increasing tank mass. Maybe you have to pump slower anyway, and not expending a prohibitive amount of prop on linear G acceleration would be the problem a spin-G prop transfer scheme would be solving. I think I'll noodle a bit more.

I just had an off-the-wall thought; what if you induced the swirling on purpose? If the propellant were swirling around the walls of the tank, at least you'd know where it was; instead of sloshing randomly, it'd be held to the walls by centrifugal force. Between that and the ullage acceleration, that ought to let the depot avoid sucking liquid out of the ullage pipe until the tank really was full.

To get the fluid to swirl, I'm visualizing replacing the nozzle at the bottom of the propellant tanks with something like a three-arm lawn sprinkler head, except that it wouldn't spin.

This would cause the connected ships to counterrotate, of course, until the pumping stopped and the liquids settled down. To reduce this effect, you could have the methane tank swirl in the opposite direction from the oxygen tank, but, since most of the mass is oxygen, this would only help a little. You could also have the source tanks swirl in the opposite direction from the depot tanks (e.g., via matching sprinkler heads at the bottoms). There would still be net torque, since the center of mass wouldn't usually be symmetrical (the depot usually will have a lot more mass, if only because it's usually partly full), but at least it would reduce the amount of it. As a possible plus, when fueling is almost complete, the joined vehicles will be rotating almost on the depot's main axis (until the propellants settle down), so the last dregs of prop in the tanker should settle to the side and bottom. If one of the nozzles ended there, it would make it easier to suck out the last few drops. (Assuming it were okay for the other two to suck ullage gas.)

Apologies if this is unclear. Extra apologies if someone else already thought of this and I just see it!

[TheRadicalModerate]

I just had an off-the-wall thought...

Well, it's really an on-the-wall thought...

As a meta-observation, note that what we're doing here is trading some combination of ullage propellant consumption and thruster complexity against some combination of transfer time (a proxy for flow rate and the momentum that comes with it) and complexity (in terms of funky inlet and outlet schemes and attitude stability). 

Seems to me that the KISS solution is to start with higher ullage accelerations and make the flow patterns be either axial and laminar or, alternatively, so turbid that the pool winds up with very few surface tensions breaks that could cause blobs on the receiving side and/or inlet uncovering on the sending side.

Things we haven't yet tried to compute and/or don't know:

1) What are the actual dynamics of the pool around the inlets and outlets as a function of flow rate? 
1a) Would knowing the maximum forward velocity that a blob could leave the pool without hitting the top of tank be a useful figure of merit?
1b) Does anybody know how to compute what's needed to break surface tension as a function of laminar flow speed and surface area of the flow?  (This doesn't help very much if the flow is highly turbulent, but might be interesting if it stays mostly laminar coming out of the outlet.)

2) Are there going to be combusting gas thrusters available?

3) A random one:  How much hydrostatic head does a pump need to function reliably?  I'm tempted to say that the answer to this is "zero", but I'm pump-illiterate.  Based on some calculations, zero is a pretty good approximation to the head at the bottom of both tanks, even at 5mm/s².

Given that we have some evidence that SpaceX is doing something new with combusting gas thrusters, and that combusting gas, if it's available in low enough thrusts or impulse bits, solves the problem, that seems the most likely solution to me.  But I'd guess that flow dynamics are a big unknown, and likely one that can only really be known via on-orbit experiments.

[Greg Hullender]

Well, it's really an on-the-wall thought...

It's so off-the-wall it's actually on-the-wall! :-)

Given that we have some evidence that SpaceX is doing something new with combusting gas thrusters, and that combusting gas, if it's available in low enough thrusts or impulse bits, solves the problem, that seems the most likely solution to me.  But I'd guess that flow dynamics are a big unknown, and likely one that can only really be known via on-orbit experiments.

That leads me to wonder why SpaceX isn't already trying to do experiments to settle these questions. They put something in orbit every week. You'd think they could find room for a few payloads that could answer these questions. It seems to make a lot more sense than learning them using entire Starship upper stages.

[Paul451]

It seems to make a lot more sense than learning them using entire Starship upper stages.

They need to launch those Starships anyway in order to test everything else.

[Twark_Main]

The tolerances would have to be pretty tight for something like this to work, but it is completely different than standard pumping in gravity or spinning ships around

Surely this could be simplified from two moveable baffles down to one moveable baffle and one fixed baffle.

[Thrustpuzzle]

I just had an off-the-wall thought; what if you induced the swirling on purpose? If the propellant were swirling around the walls of the tank, at least you'd know where it was; instead of sloshing randomly, it'd be held to the walls by centrifugal force. Between that and the ullage acceleration, that ought to let the depot avoid sucking liquid out of the ullage pipe until the tank really was full.

Centrifugal seperation can work in zero G, but has losses due to wall friction so it's really useful if the walls are smooth to reduce friction and turbulence.  There's a nifty benefit in that the pumps (and power generation) can all be on the depot and not in tankers, so you don't have to transport that machinery mass every launch, just once for the depot.

My feeling is that this could work but might have high energy expense needed to accelerate and maintain that momentum. Of course maybe that's more acceptable compared that to tons of ullage gas used to give microthrust setting. It would be fun to determine the mechanical stability of two coupled spacecraft with huge liquid gyroscope flywheels inside each.

[skyflyer81]

The tolerances would have to be pretty tight for something like this to work, but it is completely different than standard pumping in gravity or spinning ships around

Surely this could be simplified from two moveable baffles down to one moveable baffle and one fixed baffle.

lol, so obvious. definitely would simplify it a ton.

[Greg Hullender]

Centrifugal seperation can work in zero G, but has losses due to wall friction so it's really useful if the walls are smooth to reduce friction and turbulence.  There's a nifty benefit in that the pumps (and power generation) can all be on the depot and not in tankers, so you don't have to transport that machinery mass every launch, just once for the depot.

My feeling is that this could work but might have high energy expense needed to accelerate and maintain that momentum. Of course maybe that's more acceptable compared that to tons of ullage gas used to give microthrust setting. It would be fun to determine the mechanical stability of two coupled spacecraft with huge liquid gyroscope flywheels inside each.

I couldn't find the actual paper(s) from that link, but it gave me search terms that turned up this publicly available paper: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5884133/

I've only skimmed it so far, but I think it contains the equations we'd want to actually calculate whether this will work or not. I did notice that the paper is only concerned with very small systems, but it does work out a general formula that includes both swirling and microgravity.

[frith01]

The tolerances would have to be pretty tight for something like this to work, but it is completely different than standard pumping in gravity or spinning ships around

The image looks cool, but I can't quite make sense of it. Can you elaborate a little? :-)

Just thinking that some baffles that start on one side of the interior of a tank could rotate around to the other side and "squeeze" out propellant

Interesting idea for a zero-G door-hinge type fluid pusher, but this would only work if the tanks were symmetrical, and didn't have anti-slosh baffles, pipes, and other stuff routing through the tanks.   Cryogenic temperatures would prevent the use of any flexible surface from being used on the outer edges.

Could be used in cargo area for fuel Blivet  type usage though.

https://www.atlinc.com/images/atl_dropdrum-group.png

[Twark_Main]

The tolerances would have to be pretty tight for something like this to work, but it is completely different than standard pumping in gravity or spinning ships around

Surely this could be simplified from two moveable baffles down to one moveable baffle and one fixed baffle.

lol, so obvious. definitely would simplify it a ton.

Yeah that's basically all I'm good for. Call me Captain Apparent! 

[edzieba]

Even ignoring all the cryogenic sliding seal issues, plunger mass, plunger stiffness and jamming, etc, issues: radial plungers would mean that halfway through a burn, your vehicle would have 85% of its mass one one side and 5% on the other.

[OTV Booster]

Conclusion: bottom fill works best and whatever provides the transfer pressure needs to be variable.

This was why we stopped thinking very much about dorsal-to-dorsal, nose-to-tail configurations--or inline tail-to-tail configurations for that matter.

I still think that equalizing ullage pressures (by physically connecting the two ullage spaces) and using a very low power pump solves all the transfer pressure management problems.

One other problem, which is sort of related to your "don't splash" problem:  Prop will "geyser" into the receiving tank, even if it's pumped from the bottom.  At microgee accelerations, that geysering effect is going to cause all kinds of sloshing.  That may be OK, because uncovering the pump outlet on the receiving tank shouldn't be a problem.  But if you get blobs of prop bouncing off the receiving tank wallsı, they may mess up microgee ullage accelerations enough to cause sloshing on the sending side, which could uncover the inlet.

If you're using low-power pumps with equalized ullage pressures, the pump can probably be made robust enough to work with the sending side's inlet uncovered temporarily.  The pump will have to be able to re-prime itself with tiny head pressures, though.

Note that this is yet another reason not to use ullage pressure differences to transfer prop:  If you ever uncover the sending side inlet, the higher pressure ullage gas will instantly blow through the line, equalizing pressures, and then you have yourself a problem.  You can recover from this, but it'll require some combination of venting the receiving side and heating the sending side to restore the pressure differential.  It's wasteful, slow, and could potentially happen so often that the system wouldn't work at all.

_____________
ıYet another problem related to geysering:  If you have unsettled blobs on the receiving side, they'll occasionally get sucked into the ullage pressure equalization line.  That could be made to be OK, but you may wind up pumping some of the prop through the system multiple times.  Presumably, as the tank gets full it will geyser less.  That's important, because otherwise you could have a case where the recirculated prop problem gets worse just before you're completely full.

Yup. Geysering (good word) is why variable pump pressure seem a good idea. Pumping into a dry tank would face minimum head with geysering at max risk. As the head increases the pump pressure would need to be higher, but not so high as to geyser. Any hiccup that allows the sending outlet to ingest gas and send it on to the receiver just kicks off a positive feedback loop.


You've got me three quarters convinced that equalizing ullage pressure is the way to go. There is something about not equalizing ullage and venting from the receiver that's attractive. ISTM that vented ullage thrust would take too long for initial settling but might work well to maintain a settled state. But then, the lower the g (minimizing consumption) the easier things splash.


Counter to this is that during later transfers, as the sending ship is going empty and the receiving ship is filling up, the head pressure is high and pressure fed transfer would slow and maybe even stall out. Down the road the time it takes for transfer will become important. That probably means active pumping and if ya gotta pump, embrace KISS and don't screw around with two ways to power the transfer.


Once their testing advances to transfer between two ships, pressure fed would be the standard SpaceX 'just good enough' method but maybe they'll have a pump in parallel and move propellants back and forth and try it every which way.

[Greg Hullender]

Pumping into a dry tank would face minimum head with geysering at max risk. As the head increases the pump pressure would need to be higher, but not so high as to geyser.

Is the head pressure going to increase? If the ullage acceleration is just a few tens of micro-g, it doesn't seem like there'll be very much holding the propellants in place. That's why I was thinking about deliberately making them swirl. If you know they're going to slow around, at least be sure they're sloshing the way you expect them to.

[OTV Booster]

If you intend to fill a depot for a specific mission, then number of tankers is really the right metric, and overall prop efficiency isn't very important.  But if you're continuously filling and partially drawing down a depot, as you would in very high cadence ops, then the efficiency becomes more important.  Note also that increasing the power of the transfer pumps, and thereby shortening the transfer time, can also change things.  As usual, it's a pretty rich trade space.

So, to answer your question:  In high cadence, there's probably a number where rotational settling makes more sense than ullage thrust.  But it doesn't make any sense early on, when low cadence will almost certainly boil the depot dry between missions.  And even at high cadence, you'll have to show your work on why the extra complexity is worth it.

I didn't respond to this earlier post, but seeing this, as well as your comment about 5mm/s² being conservative and finding the Kutter research paper had me all but give up on the spin-G prop transfer concept I was noodling with, because I think it probably only comes out on top of linear acceleration in scenarios where you actually need considerably more than 5mm/s² (like say 1% of G = 100mm/s²).

One other problem, which is sort of related to your "don't splash" problem:  Prop will "geyser" into the receiving tank, even if it's pumped from the bottom.  At microgee accelerations, that geysering effect is going to cause all kinds of sloshing.  That may be OK, because uncovering the pump outlet on the receiving tank shouldn't be a problem.  But if you get blobs of prop bouncing off the receiving tank wallsı, they may mess up microgee ullage accelerations enough to cause sloshing on the sending side, which could uncover the inlet.

If this is the only use for the fill pipe, they can put diverter such as a mushroom cap over the outlet.

At the kind of accelerations and head pressures we're talking about, even turbulence after the diversion is likely to cause some slosh.  But it might be enough to prevent blobs of prop from slamming into stuff with enough force to cause slosh in the sending tank.  And a mushroom cap would probably help for the "nearly full" case, too.

...But then I see these posts, and I wonder if 5mm/s² is really that conservative all things considered? The environment inside those prop tanks sounds like it will be quite dynamic if you want to pump propellant at any significant rate, and fluids are gonna fluid. In your earlier post you were assuming 150 ton of prop transferred in 2000s (75kg/s) which seems more than enough to disrupt surface tension if those 75kg only weigh ~40g, but still have the inertia of 75kg. You wouldn't want to get up there and find you can only pump at 15kg/s if you want to avoid chaotic effects. Playing with your spreadsheet to increase transfer time to 10,000s really does blow out the number of tanker trips.

My mental image was of one of those "incoming prop" geysers interacting with the receiver tank prop in such a way as to induce a vortex in the receiver tank, potentially torquing the entire structure in chaotic ways each time.  BoE says 1400 ton of prop swirling at 1 rpm is roughly equivalent to a reaction wheel with 1,800,000 Nms of angular momentum, so if that kind of induced swirling is a real problem, then that's a heck of a design challenge for your pump connectors. Presumably there are simple ways to baffle/disrupt these effects, but it could mean pumping slower or increasing tank mass. Maybe you have to pump slower anyway, and not expending a prohibitive amount of prop on linear G acceleration would be the problem a spin-G prop transfer scheme would be solving. I think I'll noodle a bit more.

I'm assuming that an 'ideal' mature transfer scenario allows 6 hours (21,600 s) for transfer. The depots orbit can be matched up every 12 hours, assuming a single launch site. Figure three hours for approach and docking and the same to undock and clear.


I'm also assuming that only the depot will have active cooling and that PV and maybe radiators will have to tuck away during transfer ops. These assumptions point to wanting a loading campaign done either as fast as possible (a tanker every 12 hours) or very relaxed with a chance to literally chill between loads. There is no aspect of Elon that is relaxed so a fast cadence looks like a good bet. Fast also means less wear and tear on the panel deployment mechanism.


There's a lot of assumptions here but you might think about tucking some of them into your noodling.

[TheRadicalModerate]

Pumping into a dry tank would face minimum head with geysering at max risk. As the head increases the pump pressure would need to be higher, but not so high as to geyser.

Is the head pressure going to increase? If the ullage acceleration is just a few tens of micro-g, it doesn't seem like there'll be very much holding the propellants in place. That's why I was thinking about deliberately making them swirl. If you know they're going to slow around, at least be sure they're sloshing the way you expect them to.

Head pressure doesn't increase very much, but viscosity should become your friend as things get deeper.  The deeper the pool of prop, the more any laminar jets coming from the outlet will degrade into turbulent flow and gradually damp out.

This was in the back of my mind when I asked the question about whether there's a formula for looking at what's required to break surface tension, as a function of a flow rate (more-or-less a flow velocity) and the cross section over which that rate was effective.  That's not perfect, because even if flow can't break surface tension, it's going to cause waves, which can then build up to break surface tension in other ways.  But at least it should give you a back-of-napkin estimate of the maximum tolerable rate.

[TheRadicalModerate]

I'm assuming that an 'ideal' mature transfer scenario allows 6 hours (21,600 s) for transfer. The depots orbit can be matched up every 12 hours, assuming a single launch site. Figure three hours for approach and docking and the same to undock and clear.

Unless you have an extremely well insulated transfer path, you'll have increased boil-off inefficiency during the transfer if you drop to very low rates, because there will be more time for the flow to warm up going from one tank to another.  This is probably solvable with more insulation, but that may increase the coupling complexity.

The other problem, of course, is that longer transfers take longer ullage "burns".  So unless the acceleration requirements decrease sub-linearly with lower flow rates (and I can't see a reason offhand why they would), then longer transfer times don't help.

[DanClemmensen]

I'm assuming that an 'ideal' mature transfer scenario allows 6 hours (21,600 s) for transfer. The depots orbit can be matched up every 12 hours, assuming a single launch site. Figure three hours for approach and docking and the same to undock and clear.

Unless you have an extremely well insulated transfer path, you'll have increased boil-off inefficiency during the transfer if you drop to very low rates, because there will be more time for the flow to warm up going from one tank to another.  This is probably solvable with more insulation, but that may increase the coupling complexity.

The other problem, of course, is that longer transfers take longer ullage "burns".  So unless the acceleration requirements decrease sub-linearly with lower flow rates (and I can't see a reason offhand why they would), then longer transfer times don't help.

Surely you mean sun shielding, not insulation, since the transfer is occurring in a vacuum?

[OTV Booster]

I just had an off-the-wall thought...

Well, it's really an on-the-wall thought...

As a meta-observation, note that what we're doing here is trading some combination of ullage propellant consumption and thruster complexity against some combination of transfer time (a proxy for flow rate and the momentum that comes with it) and complexity (in terms of funky inlet and outlet schemes and attitude stability). 

Seems to me that the KISS solution is to start with higher ullage accelerations and make the flow patterns be either axial and laminar or, alternatively, so turbid that the pool winds up with very few surface tensions breaks that could cause blobs on the receiving side and/or inlet uncovering on the sending side.

Things we haven't yet tried to compute and/or don't know:

1) What are the actual dynamics of the pool around the inlets and outlets as a function of flow rate? 
1a) Would knowing the maximum forward velocity that a blob could leave the pool without hitting the top of tank be a useful figure of merit?
1b) Does anybody know how to compute what's needed to break surface tension as a function of laminar flow speed and surface area of the flow?  (This doesn't help very much if the flow is highly turbulent, but might be interesting if it stays mostly laminar coming out of the outlet.)

2) Are there going to be combusting gas thrusters available?

3) A random one:  How much hydrostatic head does a pump need to function reliably?  I'm tempted to say that the answer to this is "zero", but I'm pump-illiterate.  Based on some calculations, zero is a pretty good approximation to the head at the bottom of both tanks, even at 5mm/s².

Given that we have some evidence that SpaceX is doing something new with combusting gas thrusters, and that combusting gas, if it's available in low enough thrusts or impulse bits, solves the problem, that seems the most likely solution to me.  But I'd guess that flow dynamics are a big unknown, and likely one that can only really be known via on-orbit experiments.

A couple conclusions and a bit of hands on lore and observations.

A rule of thumb for locating a flow meter is 10x the pipe ID from the last change of direction or obstruction and a minimum of 3x the ID downstream to a change of direction, obstruction or an open termination. The first is to allow the flow to become laminar, the last to avoid back impedance. This is probably all flow rate dependent but is the standard I've adhered to for instrumentation - none of which faced any radical flow rates. If transfer is direct connection between QD plates chances are high that the inlet to the O2 tank will be turbulent but I haven't thought this through for implications.

At the receiving inlet when empty, the head is zero and the only forces keeping the inrushing (in seeping?) propellant from going ballistic is g force and surface tension. Once there is any depth of fluid there will be impedance from viscosity. The impedance will be fluid velocity dependent, and sensitive to the fluid depth above the inlet. The outer annulus of fluid will be impeded the most and the fluid at the centerline of the outlet impeded the least. As fluid velocity increases the differential between the annulus and centerline will increase.

Not being capable of running the numbers I'll go out on a limb with a qualitative assessment. Higher thrust and lower transfer rate at the beginning of the transfer will keep everything playing nice. Lower thrust and higher transfer rate will work fine as the propellant level goes up. The key is to watch that centerline velocity and keep it low enough to keep the plume (right word?) from breaching the surface. The visuals would be a roiling in the propellant above the inlet with no spatter.

When filling the depot, each tanker will face different ullage thrust and transfer rates. This could well shoot down the steady 12 hour cadence I suggested earlier.

BTW, I tried looking up the viscosity of liquid methane and oxygen, and water for comparison. My talents lie elsewhere.


Edit to add: if centrifugal pumps have problems with a low head pressure a positive displacement piston pump might fill the bill.


Edit 2: Rad Mod got to viscosity before I did.

[DanClemmensen]

I'm assuming that an 'ideal' mature transfer scenario allows 6 hours (21,600 s) for transfer. The depots orbit can be matched up every 12 hours, assuming a single launch site. Figure three hours for approach and docking and the same to undock and clear.

I think the single site has two launch opportunities a day, but they are not 12 hours apart.   There are also two landing opportunities a day.  This assumes the Depot orbit crosses the latitude of the launch site. Am I missing something?  If those two launch opportunities are far enough apart, then you still get an average of 12 hours for the full transfer operation. Each tanker will need some time (maybe 24 hours?) to sync with Depot's location in its orbit, and will need maybe the same amount of time after completion of fueling to move to the correct spot along the orbit to start its deorbit.  With all of this, I guess each tanker mission is about three days, so you need six tankers to maintain a 12-hour cadence.  I'm pretty sure you also need separate launch and catch towers.

[OTV Booster]

I'm assuming that an 'ideal' mature transfer scenario allows 6 hours (21,600 s) for transfer. The depots orbit can be matched up every 12 hours, assuming a single launch site. Figure three hours for approach and docking and the same to undock and clear.

Unless you have an extremely well insulated transfer path, you'll have increased boil-off inefficiency during the transfer if you drop to very low rates, because there will be more time for the flow to warm up going from one tank to another.  This is probably solvable with more insulation, but that may increase the coupling complexity.

The other problem, of course, is that longer transfers take longer ullage "burns".  So unless the acceleration requirements decrease sub-linearly with lower flow rates (and I can't see a reason offhand why they would), then longer transfer times don't help.

To make things worse, the very act of pumping will raise the propellant temp. How does this scale with pumping rate?


Sooo many things to trade. It almost like going into a big box hardware store to pick out a new drill.