SpaceX Starship/Super Heavy Engineering General Thread 4

[OTV Booster]

Perhaps this paper is relevant to this discussion.

It discusses the use of electric linear actuators in submarine control surfaces.

Also, regarding the force and torque requirements, in submarine control surfaces we use mechanical linkages that act as force amplifiers to allow these very large surfaces to move very quickly underwater. The submarines where I served used hydraulic linear actuators, but I have no reason to believe that the same could not be achieved with electric actuators these days, though I no longer have first-hand knowledge of what is current (plus I would probably not be able to discuss).

I am not an expert in aerodynamics or rockets, so please forgive my constant submarine analogies, it is one of the few things that I feel somewhat knowledgeable about and hopefully I am contributing to this discussion at least a bit.

A very interesting paper. And yes, ISTM that submarine experience is relevant, except you guys put the pressure on the wrong side of the hull.    Seriously, submarine and spacecraft face many parallel issues.

One detail difference is that as long as SS doesn’t come down through a storm there won’t be an analog of sea or wave slap, or at least nothing approaching that magnitude.

Phil

[Eka]

SpaceX will have the budget for the highest accuracy encoders and drives, but from a motor control standpoint, control is much, much easier if there is some gear reduction between the motor and the load.

On a car or a washing machine, you don't need stepper motor behavior, don't need to control wheel rotation accurately to degrees to drive the car. Direct drive can work fine.
But on a control surface, position must be accurate, and aero forces on the flaps pushing back against the motor can create a lot of encoder noise.  When encoder values are changing the opposite direction of what the drive is commanding, the drive either has to shut down or the drive has to be able to interpret and ride through the disturbance.  Even a 10:1 gearbox makes a big difference on reducing backlash to the encoder.

I would not be surprised if they end up with some gearing.

I can give you that control and I can reliably tell you the phase angle the motor is at WITHOUT a fine pitch encoder. I only need an encoder to tell which phase the motor is in. May not be precise enough for robotics, but it will work fine for a SS flap. I suggest reading up on sensorless control of switched reluctance motors. You can apply it to any motor where you drive each coil separately. Yes, it is possible to know if the motor has been overloaded and has slipped from one pole set to the next. That can be detected.

Gut response without numbers: the control will not need to be all that tight. Once the fins have aero loading all the backlash, if any, will be taken up. Direct drive has little or no backlash.

To hold the fins fixed against load will require power. Slippage should result in some detectable back EMF (probably the wrong term). If the slippage is too slow to detect this way there will eventually be an impact on attitude. The computer will not care why the attitude is changing. It will throw in a correction.

Encoders will be there for coarse positioning and maybe limit switches for ‘if all else fails’.

In small aircraft there are degree settings for flaps. For everything else there is airspeed, rate of descent and maybe pitch angle. The pilot doesn’t know or care what angle the control surfaces are at.

It would be interesting if anybody, active or lurking, has experience programming autopilots. The old analog autopilots had to work with what the bird was doing, not control surface angles.

Phil

I agree.

The important point is an AI is controlling the system. The control decisions the AI will produce are much like we'd produce. Those commands will then be handed to a hardware abstraction layer that will implement them. Nowhere in the system is high accuracy needed except for knowing the real trajectory and SS attitude achieved. Maybe the most important thing is the inconsistency of the atmosphere. It will make any high precision control of the flaps moot.

[capoman]

A lot of the above is dependent on how stable the flight configuration is, and how much movement and force is required to affect the required change, and how fast you need to react. This will determine if direct drive will work or what gearing is required.

This really doesn't require AI for control. It's PID control. It may look like AI, but PID control has been at the heart of control systems for decades.

Edit: corrected to as PID, not PDI control.

[JamesH65]

I would have thought the torque requirements if using direct drive would rip the motor apart, or rip it off its mountings. I cannot see how this could be  driven without gearing. Note that if you need to apply a huge torque to the flap, that same torque is also present in the opposite direction in the motor itself when directly driven.

And remember that electric motors are more power efficient at higher RPM.

And Tesla cars are single gear, not No gear  - they have a reduction gear on them of about 10:1 IIRC.

[RoboGoofers]

could it still be considered direct drive if the gearing was 1:1?
I'm thinking 4 motors at each hing end in an x pattern with the hinge pin in the middle. 8 motors per wing.

it'd be like a planetary gear but with the motors directly driving 4 planets and the wing hinge as the sun gear.

[Coastal Ron]

The important point is an AI is controlling the system. The control decisions the AI will produce are much like we'd produce. Those commands will then be handed to a hardware abstraction layer that will implement them. Nowhere in the system is high accuracy needed except for knowing the real trajectory and SS attitude achieved. Maybe the most important thing is the inconsistency of the atmosphere. It will make any high precision control of the flaps moot.

While I agree with your logic, especially in the early days of the Starship testing they will want as much data as possible in order to refine their control systems, and some of that data would be knowing exactly what position the flaps are at.

For instance, while an AI system could control the Starship by making adjustments and not caring about exact positioning, the engineers will want to understand how well the vehicle was performing. Was it balanced? Did both flaps provide equal responses? Did both systems require the same power input for the same responses? And so on.

[Draggendrop]

Just my opinion...nothing more and nothing less.
AI is nice and all...but we do not need AI for present landings or future ones. AI still requires modeling. What we need is Data..lots of data to improve mathematical modeling.

As many have said before here...engineering is done with numbers. Here it is mathematical contructs for particular devices being controlled for operation along a path within boundaries. Each mechanical device has a mathematical statement that takes into acount position, displacement, the rate of change required for the transit and the characteristics of the device...an example being the power required over full range of movement and rates of change. The level of accuracy is defined by the number and type of characteristics modeled into a devices "statement" and by the integrity of the data for these devices. All these are tied into a system of hardwired and processor control. They need to know as much as possible for every subsystem and the environment that it works in. This is why "flight time" is so valuable.
If you get a chance, take a look at Lars Blackmores papers...particularly the requirements they strive for with future Martian landings. What they do is extremely complicated and a bit convoluted at times to those not in the field. When in doubt...treat it as a 50 year old PI or PID for generalization as stated in a post above...No AI is required but realize that what SpaceX has done for landing control systems is a work of mathematical art and the best is yet to come.

// change rolls out of meter...

edit...after reading this, I sound like a grumpy old barn door. I think it may be that too much speculation can drown situations at times. SpaceX has already decided on starting equipment, has done the preliminary modeling but needs data for refinement and another flight for more data...fly/iterate/fly/iterate...we know the routine. The most difficult time is the initial data gathering.
Don't mind me...back to speculation.

[edzieba]

Remember that unless Starship is intended to start flying loop-de-loops, the force on a given body flap during re-entry is going to be in one direction and varying in magnitude around a desired average value (from 0 if Starship has rolled to take that flap completely out of the airflow, to a maximum where Starship has rolled 'onto' that flap). A spring to counter that average force connected to the flap means you have now halved the load the motor needs to deal with, and minimises the force for displacement around that desired average (i.e. fast response in the fine control region). Connecting the spring by a 'slow' jackscrew or similar so it can be engaged and disengaged avoids the motor needing to remain active to keep the fins stowed in the centre position.

[Blueshift]

Remember that unless Starship is intended to start flying loop-de-loops, the force on a given body flap during re-entry is going to be in one direction and varying in magnitude around a desired average value (from 0 if Starship has rolled to take that flap completely out of the airflow, to a maximum where Starship has rolled 'onto' that flap). A spring to counter that average force connected to the flap means you have now halved the load the motor needs to deal with, and minimises the force for displacement around that desired average (i.e. fast response in the fine control region). Connecting the spring by a 'slow' jackscrew or similar so it can be engaged and disengaged avoids the motor needing to remain active to keep the fins stowed in the centre position.

The more I think about that idea, the more I like it. It would be similar to the control of an aircraft horizontal stabilizer. A slow but powerful trim motor adjusts the angle of the whole stabilizer flap in the neutral position required for each flight phase (well, as long as MCAS doesn't think otherwise ). Smaller elevator surfaces perform high dynamic corrections.

For Starship, a slow but highly geared spring actuator could release the e-motors from static loads, which I guess will be considerable depending on the reentry profile.

[OTV Booster]

I would have thought the torque requirements if using direct drive would rip the motor apart, or rip it off its mountings. I cannot see how this could be  driven without gearing. Note that if you need to apply a huge torque to the flap, that same torque is also present in the opposite direction in the motor itself when directly driven.

And remember that electric motors are more power efficient at higher RPM.

And Tesla cars are single gear, not No gear  - they have a reduction gear on them of about 10:1 IIRC.

If the motors are concentric to the shaft the loading through the mounting is simple and predictable unlike driving through a gear box (concentric planetary excepted) where there is slop and each gear both rotates and tries to climb the next gear. If the load is enough to shear or warp the mounting, or RUD the motor, it’s under built. The fix is straight forward. Beef up the mount and add motors.

ISTM that power efficiency, a true concern for a rocket, might have to take a back seat to raw torque here.  Unless I’m way off (somebody straighten me out if so) electric motors develop max torque at 0 rpm.

In the end it’s not really that simple. Weight of gearbox and whatever power it sucks up on one side of the equation and weight of extra motors and mounting on the other.

One data point that we, and maybe SX, don’t  have is how fast the fins realistically have to move. I’m not comfortable estimating this from a sim renders. I’m sure the gravity was properly characterized along with the general atmospherics but there’s no way to know the fidelity they used for atmospheric randomness.

Phil

[capoman]

I would have thought the torque requirements if using direct drive would rip the motor apart, or rip it off its mountings. I cannot see how this could be  driven without gearing. Note that if you need to apply a huge torque to the flap, that same torque is also present in the opposite direction in the motor itself when directly driven.

And remember that electric motors are more power efficient at higher RPM.

And Tesla cars are single gear, not No gear  - they have a reduction gear on them of about 10:1 IIRC.

If the motors are concentric to the shaft the loading through the mounting is simple and predictable unlike driving through a gear box (concentric planetary excepted) where there is slop and each gear both rotates and tries to climb the next gear. If the load is enough to shear or warp the mounting, or RUD the motor, it’s under built. The fix is straight forward. Beef up the mount and add motors.

ISTM that power efficiency, a true concern for a rocket, might have to take a back seat to raw torque here.  Unless I’m way off (somebody straighten me out if so) electric motors develop max torque at 0 rpm.

In the end it’s not really that simple. Weight of gearbox and whatever power it sucks up on one side of the equation and weight of extra motors and mounting on the other.

One data point that we, and maybe SX, don’t  have is how fast the fins realistically have to move. I’m not comfortable estimating this from a sim renders. I’m sure the gravity was properly characterized along with the general atmospherics but there’s no way to know the fidelity they used for atmospheric randomness.

Phil

The sheer mass and surface area of Starship likely means that large, but not fast movements will likely be required to control the ship. A similar comparison, would be comparing a small boat with a large ship in rough waves. The large ship rides through the turbulence and takes a big rudder to change its movement, with the associated response delay to control. In other words, Starship will need to likely set itself up far in advance and will likely not be able to make fast or large adjustments easily on it’s trajectory. It will need predictive control due to response delays, a perfect scenario for PID control.

[eriblo]

Regarding "Direct drive using several Tesla Plaid motors in parallel..." and "Rear flaps each need ~1.5 megawatts.":

The specs of the Plaid motors are not known so lets use the Model 3 permanent magnet motors as an example: maxing out at 450 Nm, 250 kW and weighting 46 kg according to some quick googling. Lets also assume 100% efficiency, spherical cows and all that. If we could just run the motors at max power that would be 6 motors at ~280 kg per rear flap.
The motors are however limited by max torque below ~5300 rpm so if we assume a maximum actuation speed (angular velocity) of 2 rad/s (~full range of flap in 1 s) we get a maximum power draw of 6*450 Nm*2 rad/s= 5.4 kW per flap. If we go by 1.5 MW that would require 1.5 MW/(450 Nm*2 rad/s)=1670 motors for a total of ~77 t per rear flap.

These numbers are just a result of the different rotational speeds in the two cases and I am personally ok with assuming that Elon meant "direct [mechanical linkage]" as in the somewhat poorly formulated "direct drive/electromechanical" option instead of a hydraulic system in the question he replied to...

[Eka]

The motors are however limited by max torque below ~5300 rpm

Wouldn't this low speed torque limit be due to wheel slip on the car. When both the rotor and stator are fixed to flap and body, the motor can output full power at 0 RPM. It just needs sturdy mounts and a good cooling system to get rid of waste heat. Tesla motors have excellent cooling. So, where will they be dumping that heat during reentry?

[envy887]

The motors are however limited by max torque below ~5300 rpm

Wouldn't this low speed torque limit be due to wheel slip on the car. When both the rotor and stator are fixed to flap and body, the motor can output full power at 0 RPM. It just needs sturdy mounts and a good cooling system to get rid of waste heat. Tesla motors have excellent cooling. So, where will they be dumping that heat during reentry?

That's... uh... not how shaft power works. Power is the product of torque and rotation rate. At low rotation rate you need proportionally more torque to transmit the same power. At zero rotation rate you get zero power.

[Barley]

The motors are however limited by max torque below ~5300 rpm

Wouldn't this low speed torque limit be due to wheel slip on the car.

A torque limit is a property of the motor.  With proper control you have constant torque between zero and the base speed, in this motor apparently 5300rpm.

The torque limit is because of a current limit in the motor (or the motor controller -- ideally they are the same, but they never quite match).  The current limits are thermal, but better cooling does not change things, if you have better cooling at 0 rpm you also have better cooling at 5300rpm and you still have constant torque over that range.

[OTV Booster]

The motors are however limited by max torque below ~5300 rpm

Wouldn't this low speed torque limit be due to wheel slip on the car. When both the rotor and stator are fixed to flap and body, the motor can output full power at 0 RPM. It just needs sturdy mounts and a good cooling system to get rid of waste heat. Tesla motors have excellent cooling. So, where will they be dumping that heat during reentry?

That's... uh... not how shaft power works. Power is the product of torque and rotation rate. At low rotation rate you need proportionally more torque to transmit the same power. At zero rotation rate you get zero power.

Ah, but if the motor is loading at 0 rpm it is consuming power. Every motor start begins at 0 rpm and power draw will be ~6-10 x steady state. Maybe 1.5mW is peak.

[Eka]

The motors are however limited by max torque below ~5300 rpm

Wouldn't this low speed torque limit be due to wheel slip on the car. When both the rotor and stator are fixed to flap and body, the motor can output full power at 0 RPM. It just needs sturdy mounts and a good cooling system to get rid of waste heat. Tesla motors have excellent cooling. So, where will they be dumping that heat during reentry?

That's... uh... not how shaft power works. Power is the product of torque and rotation rate. At low rotation rate you need proportionally more torque to transmit the same power. At zero rotation rate you get zero power.

Ah, but if the motor is loading at 0 rpm it is consuming power. Every motor start begins at 0 rpm and power draw will be ~6-10 x steady state. Maybe 1.5mW is peak.

Yep. Place your car on a 45 degree slope, and use the motor to hold it in place. That motor is outputting power despite not moving the car. It is resisting gravity trying to roll the car down the slope.

And SS will never use the motor at 5300 RPM, even if geared down. The RPM will be low due to the mass of the flap. Let's assume one moves the flap 90 degrees in 0.1 second. That's 150 RPM if direct drive, and 1500 if 10 to 1 geared. I doubt they will move those flaps anywhere near that fast. What matters for SS is torque near and at stall. Why at stall, the motor will be resisting the air pressure on the flap. The current going into the motor coils is lowered to allow the air flow to lift the flap up, and increased to push the flap down against the air flow. Yeah, switching to the next set of coils is needed for larger motions, but the principal is the same. As for cooling, it is critical for both the motor and the drive circuits.

[envy887]

The motors are however limited by max torque below ~5300 rpm

Wouldn't this low speed torque limit be due to wheel slip on the car. When both the rotor and stator are fixed to flap and body, the motor can output full power at 0 RPM. It just needs sturdy mounts and a good cooling system to get rid of waste heat. Tesla motors have excellent cooling. So, where will they be dumping that heat during reentry?

That's... uh... not how shaft power works. Power is the product of torque and rotation rate. At low rotation rate you need proportionally more torque to transmit the same power. At zero rotation rate you get zero power.

Ah, but if the motor is loading at 0 rpm it is consuming power. Every motor start begins at 0 rpm and power draw will be ~6-10 x steady state. Maybe 1.5mW is peak.

Yep. Place your car on a 45 degree slope, and use the motor to hold it in place. That motor is outputting power despite not moving the car. It is resisting gravity trying to roll the car down the slope.

In that scenario the motor is consuming electrical power but outputting only heat and no shaft power

And SS will never use the motor at 5300 RPM, even if geared down. The RPM will be low due to the mass of the flap. Let's assume one moves the flap 90 degrees in 0.1 second. That's 150 RPM if direct drive, and 1500 if 10 to 1 geared. I doubt they will move those flaps anywhere near that fast. What matters for SS is torque near and at stall. Why at stall, the motor will be resisting the air pressure on the flap. The current going into the motor coils is lowered to allow the air flow to lift the flap up, and increased to push the flap down against the air flow. Yeah, switching to the next set of coils is needed for larger motions, but the principal is the same. As for cooling, it is critical for both the motor and the drive circuits.

That's rather inefficient. If the flap is driving the motor, then the motor should be in regen mode and sending that power to the opposite corners. If it's not moving it should be mechanically locked.

[niwax]

https://twitter.com/SpaceXFleet/status/1213528017248481280

Presenting the Class of 2020. #SpaceXFleet

This list of ships is a great illustration of all the fixed costs that a system like Starship saves. This alone lowers the floor for reusable operation massively.

[Eka]

The motors are however limited by max torque below ~5300 rpm

Wouldn't this low speed torque limit be due to wheel slip on the car. When both the rotor and stator are fixed to flap and body, the motor can output full power at 0 RPM. It just needs sturdy mounts and a good cooling system to get rid of waste heat. Tesla motors have excellent cooling. So, where will they be dumping that heat during reentry?

That's... uh... not how shaft power works. Power is the product of torque and rotation rate. At low rotation rate you need proportionally more torque to transmit the same power. At zero rotation rate you get zero power.

Ah, but if the motor is loading at 0 rpm it is consuming power. Every motor start begins at 0 rpm and power draw will be ~6-10 x steady state. Maybe 1.5mW is peak.

Yep. Place your car on a 45 degree slope, and use the motor to hold it in place. That motor is outputting power despite not moving the car. It is resisting gravity trying to roll the car down the slope.

In that scenario the motor is consuming electrical power but outputting only heat and no shaft power

How isn't it transmitting power? Please explain the physics.

And SS will never use the motor at 5300 RPM, even if geared down. The RPM will be low due to the mass of the flap. Let's assume one moves the flap 90 degrees in 0.1 second. That's 150 RPM if direct drive, and 1500 if 10 to 1 geared. I doubt they will move those flaps anywhere near that fast. What matters for SS is torque near and at stall. Why at stall, the motor will be resisting the air pressure on the flap. The current going into the motor coils is lowered to allow the air flow to lift the flap up, and increased to push the flap down against the air flow. Yeah, switching to the next set of coils is needed for larger motions, but the principal is the same. As for cooling, it is critical for both the motor and the drive circuits.

That's rather inefficient. If the flap is driving the motor, then the motor should be in regen mode and sending that power to the opposite corners. If it's not moving it should be mechanically locked.

Now you just added very fast acting brakes and regen circuits to the system. More weight and complexity. Regen??? Because of the air pressure on the windward side of the flap, to move it a controlled manner to either direction, the motor will need to be providing force in the direction of turning it to windward. It is much like controlling the main sail on a larger ail boat. You never let go of the rope. It is always wound around the winch, even when the sail is being let out(equivalent to the flap moving to windward). That is because if you let it go, it wildly swings out of control due to all the force on the sail. To let the flap freely swing, or even use regen, it will swing way to fast. It needs the force from the motor to keep its motion under control. Regen can't be applied everywhere. Anyways consider how long the system will be operating. Reentry doesn't drag on for hours. It would not be prohibitive to have a car's worth of batteries per motor, though I suspect half that is way more than enough. I suspect the battery pack will have to be sized for maximum current draw rate, and capacity will be overkill.