SpaceX F9 : Starlink v1.0 L19 : CCSFS SLC-40 : 15/16 Feb 2021 (0359 UTC)

[LouScheffer]

Here's something that's been bugging me about the L19 failure:

I've always assumed the re-entry burn was a closed loop burn under guidance control, ending when the booster (assuming normal rest-of-landing trajectory) was on a path to meet the landing ship.  So if an engine fails (as it did here) the remaining two engines would burn for longer until the target trajectory is met, just like on ascent.  Of course this would not guarantee mission success, since the longer burn has more gravity losses and eats into the fuel reserves, but the opposite seems like sure failure (if your trajectory does not intersect the barge, you are doomed anyway).

So I would have expected that when one engine failed during re-entry, the remaining two would have fired longer until the same final velocity was attained.  But looking at the circled area in the telemetry below, this was not the case - the remaining engines may have fired a second or two longer (not completely clear since they started a second or so later), but the final velocity was higher than in a normal mission, so the entry burn did not fully compensate.

This seems very odd to me.  The other booster burns are clearly closed loop, both for engine-out on ascent, and landing.  So surely the booster could do the same for the entry burn, and I would expect it to.  However, the telemetry indicates that it did not do so.  Anyone got any suggestions as to why?

[edkyle99]

This seems very odd to me.  The other booster burns are clearly closed loop, both for engine-out on ascent, and landing.  So surely the booster could do the same for the entry burn, and I would expect it to.  However, the telemetry indicates that it did not do so.  Anyone got any suggestions as to why?

Propellant budget for the landing burn?

 - Ed Kyle

[cscott]

Here's something that's been bugging me about the L19 failure:

I've always assumed the re-entry burn was a closed loop burn under guidance control, ending when the booster (assuming normal rest-of-landing trajectory) was on a path to meet the landing ship.  So if an engine fails (as it did here) the remaining two engines would burn for longer until the target trajectory is met, just like on ascent.  Of course this would not guarantee mission success, since the longer burn has more gravity losses and eats into the fuel reserves, but the opposite seems like sure failure (if your trajectory does not intersect the barge, you are doomed anyway).

So I would have expected that when one engine failed during re-entry, the remaining two would have fired longer until the same final velocity was attained.  But looking at the circled area in the telemetry below, this was not the case - the remaining engines may have fired a second or two longer (not completely clear since they started a second or so later), but the final velocity was higher than in a normal mission, so the entry burn did not fully compensate.

This seems very odd to me.  The other booster burns are clearly closed loop, both for engine-out on ascent, and landing.  So surely the booster could do the same for the entry burn, and I would expect it to.  However, the telemetry indicates that it did not do so.  Anyone got any suggestions as to why?

The answer may lie just to the right of your circle: the velocities quickly converge.  The drag depends on velocity squared, so the faster rocket slows down more until they are both at terminal velocity.

The re-entry burn isn't to bleed off speed as much as it is to control heating. So the failed burn leaves the stage with more heating, but you don't have to burn longer to make it up, the heat damage has already been done. Just do the best you can, let the atmosphere take care of the rest, and cross your fingers that your structure has enough margin to take the extra heat.  And as Ed says, don't blow your propellant margin trying to close the barn door after the heat horse is already out.

I am not a rocket engineer, just my opinion, etc.

[LouScheffer]

This seems very odd to me.  The other booster burns are clearly closed loop, both for engine-out on ascent, and landing.  So surely the booster could do the same for the entry burn, and I would expect it to.  However, the telemetry indicates that it did not do so.  Anyone got any suggestions as to why?

Propellant budget for the landing burn?

 - Ed Kyle

I think they'll end up with extra landing fuel in this case.  By shutting one engine shut off early, SpaceX would now have more fuel than they need (or perhaps want) for the landing burn.  So if they are targeting the "right" amount of fuel remaining (so the have usual mass at the start of the landing sequence) then that's another reason to extend the burn of the remaining two engines.

[LouScheffer]

Here's something that's been bugging me about the L19 failure:

[...]

This seems very odd to me.  The other booster burns are clearly closed loop, both for engine-out on ascent, and landing.  So surely the booster could do the same for the entry burn, and I would expect it to.  However, the telemetry indicates that it did not do so.  Anyone got any suggestions as to why?

The answer may lie just to the right of your circle: the velocities quickly converge.  The drag depends on velocity squared, so the faster rocket slows down more until they are both at terminal velocity.

The re-entry burn isn't to bleed off speed as much as it is to control heating. So the failed burn leaves the stage with more heating, but you don't have to burn longer to make it up, the heat damage has already been done. Just do the best you can, let the atmosphere take care of the rest, and cross your fingers that your structure has enough margin to take the extra heat.  And as Ed says, don't blow your propellant margin trying to close the barn door after the heat horse is already out.

I am not a rocket engineer, just my opinion, etc.

If the rocket was coming straight down, your answer makes perfect sense.  Then the applied delta-V does not affect targeting, so just take off as much velocity as you can, and hope the heating is OK.

However, the rocket is coming in at something like a 45 degree angle.  So if you don't slow it enough, the higher velocity will take it further downrange until terminal velocity kicks in, and you'll overshoot the barge.

The re-entry burn has two purposes:  one you mentioned, is to get the velocity down to minimize heat damage.  The second is to leave the trajectory in such a state that the grid fins and landing burn combined can hit the target.  It's this second reason that I'd think SpaceX would want closed loop control, ending at the same speed.  Presumably a normal re-entry burn puts them smack in the middle of their box of correctable errors.  So anything different than the usual delta-V will bias them more towards the edge of the acceptable region, and could (if bad enough) even make it impossible to hit the target.   This targeting, rather than heat removal, is why I expected them to extend the re-entry burn to get the usual delta-V.

As far as Ed's point about the amount remaining fuel, in this case I'd expect a compromise.  If you just shut off the failed engine and do nothing else, you'll end up with extra fuel.  Conversely, if you extend the burn to the usual delta-V, you'll have less fuel remaining than usual because of increased gravity losses.   It seems to me that the best you could do would be to run the two remaining engines for a few seconds longer, getting as close as possible to the nominal trajectory while keeping the usual amount of landing fuel.  But there is no indication they compensated like this.

[Barley]

There are at least two different ways to close a loop.

One way is to have a nominal trajectory and close the loop to keep on the nominal trajectory.

Another method, which is usually more efficient in physical resources, but more computationally expensive, is to continual recalculate the best trajectory from where you are* to where you want to end up.  If you follow this rule perturbations may not be immediately corrected.  After a perturbation the actual trajectory may not converge to the nominal trajectory until right at the end.  It's possible for the the corrected trajectory to be radically different from the nominal one (e.g. Hiten)

* In phase space, not just position.

[LouScheffer]

There are at least two different ways to close a loop.

One way is to have a nominal trajectory and close the loop to keep on the nominal trajectory.

Another method, which is usually more efficient in physical resources, but more computationally expensive, is to continual recalculate the best trajectory from where you are* to where you want to end up.  If you follow this rule perturbations may not be immediately corrected.  After a perturbation the actual trajectory may not converge to the nominal trajectory until right at the end.  It's possible for the the corrected trajectory to be radically different from the nominal one (e.g. Hiten)

* In phase space, not just position.

The problem is that they did not follow either of these two sensible strategies.

They did not follow the nominal trajectory, since they ended up with a higher velocity, which they could have fixed by burning the two remaining engines longer.   And I can't believe they used the "optimize from here on out" strategy, since they did nothing different after one engine failed.  Whatever the optimum strategy is after losing 1/3 of your thrust, just doing the same thing minus one engine is not it.

My current thinking is that they simply don't have the software to account for an engine loss during re-entry burn.  Landing is single-string anyway, so perhaps the extra software work was deemed not worth the effort.  How often will this particular case happen?  Not often at all, I suspect, and even if it does it does not affect their primary mission.  So maybe they have some "simple" algorithm with (say) an orientation, a delta-V target, and a time limit for backup.  So they lose an engine during the burn, and then time out before reaching their delta-V target.

[Comga]

There are at least two different ways to close a loop.

One way is to have a nominal trajectory and close the loop to keep on the nominal trajectory.

Another method, which is usually more efficient in physical resources, but more computationally expensive, is to continual recalculate the best trajectory from where you are* to where you want to end up.  If you follow this rule perturbations may not be immediately corrected.  After a perturbation the actual trajectory may not converge to the nominal trajectory until right at the end.  It's possible for the the corrected trajectory to be radically different from the nominal one (e.g. Hiten)

* In phase space, not just position.

The problem is that they did not follow either of these two sensible strategies.

They did not follow the nominal trajectory, since they ended up with a higher velocity, which they could have fixed by burning the two remaining engines longer.   And I can't believe they used the "optimize from here on out" strategy, since they did nothing different after one engine failed.  Whatever the optimum strategy is after losing 1/3 of your thrust, just doing the same thing minus one engine is not it.

My current thinking is that they simply don't have the software to account for an engine loss during re-entry burn.  Landing is single-string anyway, so perhaps the extra software work was deemed not worth the effort.  How often will this particular case happen?  Not often at all, I suspect, and even if it does it does not affect their primary mission.  So maybe they have some "simple" algorithm with (say) an orientation, a delta-V target, and a time limit for backup.  So they lose an engine during the burn, and then time out before reaching their delta-V target.

But is that “the SpaceX way”?
After the first stage failure on CRS-7, (IIRC). Musk stated that the software would be modified to deploy the parachutes in any similar event in the future.
That’s a low gain remedy for a low probability event, but Musk said they would work on it.
Recovering a first stage after an engine failure at entry burn seems more value and higher probability.

[AndrewRG10]

Today's launch of L20 had a re entry burn cut off of about 5920km/h which is faster than L19 yet came back to earth beautifully. Kinda makes me think L19 had some damage to its heatshield as well and not just the boot having a bit of damage.

[cscott]

Technically the boot is part of the heat shield.

[AndrewRG10]

Technically the boot is part of the heat shield.

Oh you're right it just seems to me the damage might've been a bit more damaged than just an engine cover having a hole in it.

[Vettedrmr]

Technically the boot is part of the heat shield.

Oh you're right it just seems to me the damage might've been a bit more damaged than just an engine cover having a hole in it.

What they said was that a hole in the boot let hot gas into areas it shouldn't have been.  IMO that's where the actual landing-killing failure occurred.

[LouScheffer]

Hmm.  Up above, I was drawing conclusions about the re-entry burn aiming based on the higher re-entry velocity.  But as pointed out above, the entry speed was even higher on the recent L20, which was a successful entry.  So it looks like the L19 entry burn was not out of family, and could well have done what was intended in terms of delta-V.  This in turns means my speculation about control algorithms was likely based on within-family launch-to-launch differences, not a result of the engine failure, and is very likely wrong.

[Barley]

If the rocket was coming straight down, your answer makes perfect sense.  Then the applied delta-V does not affect targeting, so just take off as much velocity as you can, and hope the heating is OK.

However, the rocket is coming in at something like a 45 degree angle.  So if you don't slow it enough, the higher velocity will take it further downrange until terminal velocity kicks in, and you'll overshoot the barge.

Speed is a scalar.  The "acceleration" discussed here appears to be the derivative of speed so is also a scalar.

Do we actually know enough about the directions of these to make sensible statements?   Do we know that the deorbit burn accelerations are anti-parallel to the velocity?  There could be all sorts of things going on w.r.t. yaw and pitch that don't show up.

You might be able to do something using the altitude, rate of change of altitude and speed to place some limits on what the directions actually are.  We can certainly see that at the start of the reentry burn the angle of the velocity to the horizontal is far less than 45 degrees.  (The ratio of the minimum speed at the top of the arc to the speed at the start of the burn is less than sqrt(2) ).

[LouScheffer]

You might be able to do something using the altitude, rate of change of altitude and speed to place some limits on what the directions actually are.  We can certainly see that at the start of the reentry burn the angle of the velocity to the horizontal is far less than 45 degrees.  (The ratio of the minimum speed at the top of the arc to the speed at the start of the burn is less than sqrt(2) ).

Not just might - you can do it straightforwardly.  Assume the launch is aimed straight at the droneship (it's less efficient to do anything else).  Then motion is contained in one vertical plane.  What's plotted as acceleration on the graph is actually the derivative of the scalar velocity. and has a value of 4 m/s at the start of the re-entry burn.  What's the angle to the horizontal that gives this?

First write V as V = sqrt(Vh^2 + Vv^2) where Vh = horizontal velocity (unchanged during coast) and Vv = vertical velocity, changing during coast.  Then dV/dt = dH/dVv x dVv/dt, where dVv/dt is known to be 10 m/s^2.  Expand the derivative (using the chain rule) to get dH/dVv = 0.5 * 1/sqrt(Vh^2 + Vv^2) * 2*Vv = Vv/sqrt(Vh^2 + Vv^2).   We need this value to be 0.4 since we are seeing about 0.4 of the real known 10 m/s^2.   We can't solve for either Vv or Vh directly, but we don't need to since we only want the angle.

Therefore express Vv as a multiple alpha of Vh.  This gives 0.4 = (alpha*Vh)/sqrt(Vh^2+(alpha*Vh)^2).  Now Vh cancels and we can solve the resulting quadratic for alpha, which is 0.5.  Therefore the vertical speed is 1/2 the horizontal, and the angle to the horizontal is then atan(0.5) or 27 degrees, all assuming I made no mistake in my algebra.

[LouScheffer]

My current thinking is that they simply don't have the software to account for an engine loss during re-entry burn.  Landing is single-string anyway, so perhaps the extra software work was deemed not worth the effort.  How often will this particular case happen?  Not often at all, I suspect, and even if it does it does not affect their primary mission.  So maybe they have some "simple" algorithm with (say) an orientation, a delta-V target, and a time limit for backup.  So they lose an engine during the burn, and then time out before reaching their delta-V target.

But is that “the SpaceX way”?
After the first stage failure on CRS-7, (IIRC). Musk stated that the software would be modified to deploy the parachutes in any similar event in the future.
That’s a low gain remedy for a low probability event, but Musk said they would work on it.
Recovering a first stage after an engine failure at entry burn seems more value and higher probability.

Yes, apparently it is the SpaceX way:   From SpaceX lost a rocket in the ocean last month. Here's why. :

The company is also working on future upgrades to the control systems on Falcon 9 that will help the vehicle land even in the event of an engine shutdown during flight like the one that occurred Feb. 15, Reed said.

[kdhilliard]

From SpaceX lost a rocket in the ocean last month. Here's why.:

The company is also working on future upgrades to the control systems on Falcon 9 that will help the vehicle land even in the event of an engine shutdown during flight like the one that occurred Feb. 15, Reed said.

I thought we had something new there, but that article appears to be based entirely on statements made by Benji Reed during the Crew-2 Mission Overview briefing from 1 March, and that excerpt above is Amy Thompson's interpretation of what Reed said about lessons learned from their high flight rate, from inspection of high flight count boosters, and from the investigation of this anomaly, specifically:

In terms of going from there.  We apply, well, here's where we need to do more inspections, here's where [garbled] kind of components we need to replace, and here's where we can actually upgrade the systems, the algorithms on the vehicle to even further detect and control what the vehicle needs to do.  And that's what we're doing as we move forward.

That may be a reasonable interpretation, but it should be noted that Reed did not explicitly say that such upgrades were targeted at landing a booster under those circumstances, and the tone of the article does not indicate that Thompson was working from any source other than the briefing.

Link to briefing at 52:15, where Reed discuses the anomaly.
The part I quoted above is from 54:30.

[Comga]

My current thinking is that they simply don't have the software to account for an engine loss during re-entry burn.  Landing is single-string anyway, so perhaps the extra software work was deemed not worth the effort.  How often will this particular case happen?  Not often at all, I suspect, and even if it does it does not affect their primary mission.  So maybe they have some "simple" algorithm with (say) an orientation, a delta-V target, and a time limit for backup.  So they lose an engine during the burn, and then time out before reaching their delta-V target.

But is that “the SpaceX way”?
After the first stage failure on CRS-7, (IIRC). Musk stated that the software would be modified to deploy the parachutes in any similar event in the future.
That’s a low gain remedy for a low probability event, but Musk said they would work on it.
Recovering a first stage after an engine failure at entry burn seems more value and higher probability.

Yes, apparently it is the SpaceX way:   From SpaceX lost a rocket in the ocean last month. Here's why. :

The company is also working on future upgrades to the control systems on Falcon 9 that will help the vehicle land even in the event of an engine shutdown during flight like the one that occurred Feb. 15, Reed said.

🙄
Yes =No
We are in violent agreement
It IS the SpaceX way to include software mods for unlikely events when the consequences of the software “patch” failing is no worse that not trying.
It IS NOT the SpaceX way to stop software updates because a new vehicle is being developed like Lou said.