...but the forces they would put on their own bearings and other structures, which could cause problems or even failures.
Right. If you lay the typical turbo pump on its side and split it lengthwise (which happens in lots of museum exhibits), you can identify the bearing points typically as larger-diameter cylindrical regions of the rotor. There are typically two. Where they lie along the shaft is a matter of considerable engineering tradeoff. To reduce eccentric loads from gyroscopic behavior, the bearings want to be as close to the inertial center of the shaft as possible. But to meet flexure tolerances for the rest of the assembly, they want to be spaced farther apart.
At one end you have the turbine, which is just a pinwheel. At the other end you have the pump impeller(s), which is just a propeller whose primary job is moving fluid. There are parasitical thrust forces generated in the shaft. The pinwheel, in addition to spinning, also pulls on its end of the shaft. The impeller often wants to pull or push on its end of the shaft too. So you either have to balance those forces, or provide for them in your bearing design.
Every part of any articulated assembly behaves inertially. So even a "thing attached to another thing" has to accommodate inertial responses if the assembly is suddenly accelerated or rotated. But spinning masses amplify the reaction magnitude. If you try to alter the orientation of a spinning mass, the reaction -- applied at the bearing points, at right angles to the shaft -- is very strong. The spinning mass
really wants to stay in the same orientation.
In most bearing designs, you expect the distribution of mechanical load around the circumference of the bearing race to be consistent -- and ideally equal. Of course in something like an earth vehicle wheel bearing you know that there's also a gravity load involved. But the way the load across the communicating-surface boundary changes over time informs the bearing design. In some cases, shifting the load to one side of the bearing -- and thus reducing it on the other side -- can change the overall friction as measured in resistance to the rotation. For liquid-fueled engines, turbo pump speed has to be be rock steady, otherwise it induces combustion instability.
Roller bearings are like ball bearings. The communicating surfaces are separated by rollers so that there's no sliding involved, just rolling. This is the sledge-on-logs principle. The upside of this type of bearing is that the measured friction doesn't vary much with gyroscopic load. But the measured friction of this type of bearing is, on average, higher than with other kinds of bearings. Normally that would just mean sizing the pump load appropriately to the turbine shaft power. But roller bearings incur mechanical wear and so are not the best choice for rockets you hope to be reusable. They also require more attention to mechanical tolerances and precision to engineer, since they have more moving parts. Plus, you really need efficiency. The F-1 turbo pump had to deliver as much power as the entire engine on an F-16 fighter.
The more popular choice is fluid bearings. The broad theory here is that if you have two communicating surfaces operating in shear, and there is a film of fluid between them that possesses a certain viscosity, very magical things happen in that fluid depending on shear velocity and normal pressure, and induced fluid pressure. The result is that the fluid film is substantially resistant to normal force (a force trying to mash the two surfaces together) and thus prevents the surfaces from mechanically touching. And at the same time it offers very little resistance to the shear motion -- the plates sliding across each other.
Fluid bearings are divided into two types, depending on what's keeping the fluid between the surfaces. Your car engine probably uses hydrostatic bearings. A separate oil pump sucks oil from the oil pan, through the oil filter, and pumps it through passages in the engine block where it emerges from ports appropriately spaced in the the outer bearing surface. The pressure from this pump is what keeps the rotor centered. But the oil escapes freely from the bearing in these cases, to be replaced by new oil from the pump. It's captured and allowed to flow back into the pan to be recirculated.
If the pump fails, or if the moment of inertia in the rotor exceeds pump-supplied pressure, the rotor's bearing surface will grind against the stationary bearing surface. The result is a dramatic, sudden increase in rotational friction, damage to the bearing surfaces, and a sudden increase in differential torque along the bearing shaft. You now also likely have metal particles in the oil, which may block smaller lubricant passages. The additional problem for rocket engines is where to get the power to drive the external oil pump while the motor is starting up. Your car solves this problem by designing the bearing so that a little bit of residual oil remains in the bearing to lubricate all its shafts for a few seconds until the oil pump supplies suitable pressure. And since gravity isn't available to help capture the free oil, you need to plan for how to get the oil back to whatever reservoir your oil pump is drawing from. You have to borrow solutions from high-performance aircraft engines.
Hydrostatic bearings simply operate in the presence of an ambient fluid. The communicating faces are designed to create a standing fluid "wedge" as they shear across each other. You may have experienced this effect also in your car in the more terrifying form of hydroplaning. Your car tires interact with the road surface and the intervening fluid to create a wedge of water that is literally capable of supporting the weight of your car and preventing the rubber from meeting the road. A hydrostatic bearing is in a perpetual state of hydroplaning. But as you can imagine, the effect happens only at high shear velocities, so hydrostatic bearings have to carefully match the viscosity of the lubricant to the expected operating speed.
Maintaining the "wedge" requires a delicate balance of distance and force between the communicating surfaces, fluid viscosity, and shear velocity (which, in a throttlable engine, can vary greatly). Liquid lubricants change viscosity as they change temperature. The reactive strength of the "wedge" changes with velocity and normal force.
One of the most common applications of a hydrodynamic bearing is the thrust bearing on a ship. The propeller turns and the propulsive force occurs axially along the shaft. But where is it applied to the ship structure? Without the thrust bearing it would apply somewhere inside the engine, which is no good. To transmit the force explicitly to the ship structure, you attach a robust disc to the shaft, which turns along with it. The disc, while rotating, presses against a plate with a hole in it to allow the shaft to pass through untouched. The plate is attached to the ship frame. The area between the thrust disc and the thrust plate is lubricated with a hydrodynamic bearing, which bears the tremendous load of propulsion entirely by itself.
But to accommodate the ways in which the dynamics of the fluid film might change, the plate is actually replaced with a pie-slice assembly in which each of the pie-slice sectors is able to be angled and positioned to optimize the fluid behavior. This way the ship can operate at a wide variety of shaft speeds, a wide range of applied loads (e.g., what if the propeller comes half out of the water?), and a wide variety of fluid temperatures and pressures.
Obviously putting that sort of sophisticated assembly into a rocket engine turbo pump is problematic. Also, there is a big difference between a ship's propellor rotating at 180 rpm and a pump shaft rotating at tens (or even more than one hundred) thousand rpm. If you decide to go with hydrostatic bearings, you have to ensure that engine operation stays within the narrow range of "hydroplaning" afforded by your built-in parameters. This could be part of what Musk says when he wants to operate the Raptors slower: the pumps may literally not be able to build up the fluid "wedge" at such a slow speed.
Temperature is not trivial. The pinwheel end of the rotor will quickly reach temperatures of several hundred degrees Celsius. It's sitting in the flow of the preburners, which is as hot as a blowtorch. While there are many ways to cool the turbine end, borrowed from jet engine technology, it is inevitable that heat will travel down the bearing shaft and raise the temperature of the hot-end bearing substantially. Conversely, the pump impellers are sitting (at least in the LOX case) in cryogenic propellant. As the propellant passes through the pump, the cold end of the shaft drops to well below zero Celsius. This makes the mechanical engineering of the pump rotor... interesting. Remember what I said about differential torque in a bearing strike? Imagine that one end of the shaft is very hot (and therefore not as mechanically strong as it would be at room temperature). Imagine the other end is very cold, and likely brittle. Sometimes the shaft literally cannot absorb the sudden torque increase, and the pump wrings its own neck.
But it also complicates the choice of lubricants. What liquid lubricants stay viscous enough at blowtorch temperatures to act as a proper fluid film? What lubricants stay fluid enough at arctic temperatures to avoid turning into a useless paste.
In many cases the answer has been the propellants themselves, simply because no other fluid can coexist in the environment as a separate lubricant.. A surprising number of rocket engine pumps are designed to use their respective propellant fluids as the hydrostatic lubricant. But that means all the design parameters shift to the mechanical and operational parameters. The bearing faces have to be designed "just so," and the engine has to be operated "just so" to keep the bearing characteristics within the narrow envelope of the fluid behavior of LOX or RP-1 or liquid methane. So when you add violent engine gimbaling to the equation, it becomes a very hairy problem.
Pump bearing failure is what killed the Antares rocket that used the NK-33 engines [Jay spits on the floor]. This is an incredibly important part of engine design. And part of my early work as a graduate student, learning from the Rocketdyne H-1 and helping develop manufacturing methods for the rotor of the RS-25.
But the Raptor is a "full flow" engine. This is actually a blessing in disguise. Early engines were open-cycle. What powered the pumps just got dumped overboard non-propulsively (the early Atlas) or co-opted as a secondary heat barrier (the F-1). Closed cycle engines (the RS-25 and the NK-33 -- spit on floor again) burn
some of one half of the propellant mix and
all of the other to generate mechanical power, then directs the turbine exhaust to the injector where the rest of the other side of the propellant equation is added to complete the reaction propulsively in the thrust chamber. You don't lose any thermal energy this way, making the engine much more efficient.
A full-flow engine actually has two independent pump assemblies. Many prior designs pumped the fuel and oxidizer with impellers positioned on the same shaft. The Raptor (and other engines) have two instances of staged combustion -- one for methane and another for liquid oxygen. The problem with an oxygen-rich first combustion stage is that the hot oxygen attacks the metal of the engine itself. If you use a fuel-rich first combustion stage, you get a fuel-rich exhaust, which isn't a chemical danger to the plumbing. The Raptor does both, each in a separate preburner. The oxygen-rich exhaust doesn't have far to travel before it enters the thrust chamber, so you limit its exposure to engine materials.
But I'm sure you've guessed by now that if you have two pump assemblies, you have two shafts. And if you're smart, you arrange for those shafts to spin in opposite directions, so that the gyroscopic effects -- at least in terms of vehicle guidance -- cancel out. You still have to be very smart about how you design the pump bearings. But here the guidance problem I mentioned yesterday probably doesn't exist on Starship.
