Falcon first stage finished

Alan F wrote:

Vince Cate wrote:
So far so good!

Unless I missed it being discussed in this thread, did SpaceX recover
the first stage? Reusing the first stage - or perhaps parts from it -
was part of their original plan to cut costs. For this flight, it would
presumably be useful to inspect the engine and pumps for any unexpected
damage.

Reusing the first stage is a capability to be acquired at some
indefinite point in the future.

D.
--
Touch-twice life. Eat. Drink. Laugh.

-Resolved: To be more temperate in my postings.
Oct 5th, 2004 JDL

wrote:

On Mar 22, 7:29 am, Craig Fink wrote:

Sure looked like what one would expect from a resonate frequency between
the control system and slosh. [...]

I still think it's slosh. Either way, auto-gains in there flight control
would have fixed it.

It is not clear to me that a simple auto-gain would necessarily work
in this case. Some systems work below resonance, and rely on the
resonance averaging out (when you balance a meter stick on your
finger, you don't need to worry about bending modes). Some work above
resonance, and depend on control forces being able to compensate for
the resonance effects (Balancing the rocket at launch is in this
mode). But some systems, especially where the system changes during
operation, need to operate through resonances. The basic idea here is
to get through resonance quickly enough that the vibrations cannot
build to a destructive value before you reach the other side and
regain control. Drive shafts for big ships, and spin-polarized
synchrotrons, both use this idea.

It would not surprise me if a rocket needs to work in this regime. As
the rocket empties, the performance of the engine gimbals remains the
same, while the resonant frequencies (and the Qs) will vary
considerably as the tanks empty. Here a simple auto gain will not
work (as I understand it) - as the resonance approaches, it will just
keep lowering the gain to keep the system stable, possibly resulting
in in-adequate control. It would not realize that it can increase the
gain, allow the system to go temporarily unstable, then regain control
on the other side of the resonance.

A stability goodness test lowers the gain (stability), but there is another
goodness test on control that can raise the gain. It would jump over the
resonance frequency when the gain has been lowered to the point of not
enough control (really the test is before lose of control, just sloppy
control). The gain can go both ways depending on what is needed.

Something like;

If (test 1 fail) gain = .8 * gain
If (test 2 fail) gain = 1.5 * gain


Programmable gains vs time can
deal with things like this, but can screw up if your model is wrong.
(IIRC, this is what happened to Delta-III -there was an un-predicted
resonance in their control band, they ate up all their fluid trying to
control it, then lost all control when fluid ran out).

Given modern computer power, and distributed small accelerometers, you
could imagine a smart auto-gain that cross-correlated the engine
position and all the accelerations, measuring and modeling in real
time all the resonances and their Qs and their rates of change, and
then adopting an optimum control strategy. This could deal with
unanticipated effects and physical damage. But I suspect this would
be really hard to qualify, since you'd have lots of very complex
software in the control loop, worries about convergence of the
algorithms, prove the computed control strategy converges in time, and
is stable, and so on.

So my guess would be they used fixed gains vs time (or tank level, or
something else simple). There was something wrong with their model -
perhaps the engine was dinged, Qs are different in vacuum, the modeler
thought some effect was negligible but it is not, some effect was
stronger than anticipated (this got the Ariane 5 upgrade - their main
engine generated more torque around the long axis than anticipated),
and so on. Then the system became unstable and boom....

My guess as to what happened,

Have your balancing stick handy, helps with the words.

Yeah, a modeling problem. There are three modes to balancing a stick. Finger
moving in and out, pitch, using pitch attitude, rate, acceleration, jerk...
and left and right, yaw, again attitude, rate, acc... and a third mode roll
(but not the roll one normally thinks of, rolling the stick between your
fingers). The roll in terms of moving your finger in a big circle to
balance the stick. The stick isn't rolling but is going in a big circle,
finger moving in pitch and yaw.

This seems to be the one that failed, the engine was moving in circle that
spiraled out. Looking at the gimbal motion in Polar Coordinates, the circle
has a radius R and an angular rate w. The cause of this circular motion was
most likely slosh, a wave at the surface of the liquid in the tank moving
around the wall of the tank. Like moving a bucket of water in a circle, one
side of the bucket has the crest, the other side has the trough.

The average C.G. is still along the centerline of the vehicle, but the
instantaneous C.G. is moving in a circle with some radius R and the same
angular rate w as the engine gimbal's angular rate. If the angular position
of the gimbal is behind the angular position of the C.G., the wave is
flowing down hill and getting larger. The instantaneous C.G. shift R will
grow. If the angular position of the gimbal is ahead of the angular
position of the C.G. then the wave is trying to flow up hill, energy is
taken away and the instantaneous C.G. shift R will be reduced.

The normal pitch and yaw should only be used with respect to the average
C.G. and not used to chase transients around circlecaused by the slosh
instantaneous C.G. moving in a circle.

It wouldn't be too hard to add the a slosh roll control model to the
pitch/yaw control of the engine. Or, they could just add some more slosh
baffles.

Just a guess.

--
Craig Fink
Courtesy E-Mail Welcome @

Like moving a bucket of water in a circle, one
side of the bucket has the crest, the other side has the trough.

The average C.G. is still along the centerline of the vehicle, but the
instantaneous C.G. is moving in a circle with some radius R and the same
angular rate w as the engine gimbal's angular rate. If the angular
position of the gimbal is behind the angular position of the C.G., the
wave is flowing down hill and getting larger. The instantaneous C.G. shift
R will grow. If the angular position of the gimbal is ahead of the angular
position of the C.G. then the wave is trying to flow up hill, energy is
taken away and the instantaneous C.G. shift R will be reduced.


And then again, when moving a bucket of water around in a circle, a vortex
also form, making the instantaneous C.G. shift even worst. So moving the
gimbals in a circle in the opposite direction of the rotating C.G. is
probably much better. To get ride of the forming vortex. Chasing the
instantaneous C.G. around reinforces the vortex.

If the vortex reaches the inlet to the engine, whoosshhhh, out the Helium
goes. no pressure in the tank.

Or, add more baffles.

Just a guess.

--
Craig Fink
Courtesy E-Mail Welcome @

kT wrote:

I suppose as a US citizen it never occurred to you to write your own
software. Oh, the horror. Better to let India write software for you.


Or, I could just use yours. Were can I get a copy?

kT wrote:

India

Yeah, they probably have a copy too, by now.

Humm, reinvent the wheel, gees doesn't sound like too much fun to me.

My impression was that it was too early for the helium to run out.
There was no apparent reason to run out of helium, there weren't any
excessive control actions.

The pitch and yaw excursions were noticable but did not seem fatal.
The roll must have been caused by a stuck valve or burnt out FET or
IGBT. From a control strategy point of view it might be better to
intentionally roll the vehicle and control the roll rate or just not
worry about it too much once it is rolling.

The instability can be a result of too low servo rates, or time delays
in the control loop. The use of high level languages and real time
operating systems can easily eat up the CPU time and result in
unstable control loops. The time delay in the pneumatic system may
also be the culprit for the oscillations.

I hope there is enough data to determine why the roll took off.

Craig Fink wrote:

Like moving a bucket of water in a circle, one
side of the bucket has the crest, the other side has the trough.

The average C.G. is still along the centerline of the vehicle, but the
instantaneous C.G. is moving in a circle with some radius R and the same
angular rate w as the engine gimbal's angular rate. If the angular
position of the gimbal is behind the angular position of the C.G., the
wave is flowing down hill and getting larger. The instantaneous C.G.
shift R will grow. If the angular position of the gimbal is ahead of the
angular position of the C.G. then the wave is trying to flow up hill,
energy is taken away and the instantaneous C.G. shift R will be reduced.


And then again, when moving a bucket of water around in a circle, a vortex
also form, making the instantaneous C.G. shift even worst. So moving the
gimbals in a circle in the opposite direction of the rotating C.G. is
probably much better. To get ride of the forming vortex. Chasing the
instantaneous C.G. around reinforces the vortex.

If the vortex reaches the inlet to the engine, whoosshhhh, out the Helium
goes. no pressure in the tank.

The Vortex is like a big Gyro of rotating fluid, Pitch becomes Yaw and Yaw
becomes Pitch. Helium loss through the engine, Helium pressurization valve
opens wide open, Helium is exhausted and Vehicle begins to Roll. Pressure
drops the Tank and thrust terminates.

On 3/22/07 7:29 AM, in article
. net, "Craig Fink"
wrote:
Round and round the fluid went,
larger and larger circle. Stop moving the engine. Maybe, it did get bent,
asymmetrical thrust changing control frequencies and cross-coupling. But,
it sure recovered fast (very tight) from the separation attitude transient,
one quick movement of the nozzle and back on attitude.

I agree, it was clearly stable right off the booster, and it went unstable
later. The mostly likely controls effect of a bent nozzle is a thrust vector
misalignment, and that would be trimmed out quickly by the autopilot, or it
would diverge in very short order.

If you just "stop moving the engine" it goes ass-over-teakettle almost
immediately, as it will diverge rapidly if the thrust vector doesn't end up
going exactly through the CG.

To see how bad the later instability is, look how much the gimbals moved
to control the separation transient (just a little bit), and compare that to
even the early stages of the divergence -much more.


I still think it's slosh. Either way, auto-gains in there flight control
would have fixed it.

Maybe, but maybe not. If it went unstable from gain changes due to
changes in mass properties, a *programmed* (open-loop) gain change would
certainly fix it. Truly adaptive control (actively determining the system
poles from observer performance and altering the system to maintain them,
closed loop) is a dirty word and has had a very poor history.

But if it went unstable due to varying slosh characteristics, or more
likely not understood slosh characteristics, then there's no reason to think
that you would know how to correctly change the gains/filters to overcome
it. Bear in mind that even if everything else stays the same except the
propellant running out, the slosh frequencies change DRASTICALLY during the
burn. If for no other reason, the acceleration is going up, and that
directly affects the slosh frequency.

Of course it could be a plain old structural bending thing, too. The less
fuel the less rigidity in the tank and the more the weight gets to the ends
of the spring, er, tank.

BTW, it occurred to one of my buddies earlier that there is a good
chance the roll RCS fuel was depleted trying to remove the oscillation
coupled from the other axes going unstable.

Brett

On Mar 21, 8:52 pm, Brett Buck wrote:
On 3/21/07 8:15 AM, in article ,

I would have to guess that it's a fuel swirl/slosh issue.

It was clearly stable at separation and not later. This is not all that
unexpected. The stability margins typically go down as the fuel
runs out, because the inertia goes down, and the gimbal
torque/angle ratio goes up (since the CG moves forward as the
stage gets lighter). Both of these raise the effective system gain,
and anything like this will certainly go unstable if the gain goes
up sufficiently.

Here are a few interesting numbers. From SpaceX, the second stage
starts at 0.65 Gs and builds to 4.5 Gs, at starts at t=165 sec.
Ignoring the fairing, and assuming uniform burn, we get the
accelerations below. From timing 5 oscillations at 4:20, 4:40, 4:50,
and 5:00, I get the T5 numbers. This gives the oscillation period in
seconds, and in Hz.

Time sec M Acc(g) T5(sec) osc(sec) osc(Hz)
2:45 165 1 0.65
4:20 260 0.804 0.808 6.6 1.32 0.757
4:40 280 0.762 0.852 5.64 1.128 0.886
4:50 290 0.742 0.875 5.5 1.1 0.909
5:00 300 0.721 0.900 5.4 1.08 0.925

One interesting thing is that the oscillation frequency is going up
faster, or at least as fast, than the G force. Offhand I would think
slosh would only go up as sqrt(acc), but that's assuming it's only at
the surface of the fluid and not throughout the volume (which is
decreasing, clearly). It also does not fit a bending model where the
stiffness is constant and the mass is decreasing as the acceleration -
in this case you would also expect a sqrt(acc) as well.

Anyone have any ideas for a model that might fit this data??

Lou Scheffer

In message
(Derek Lyons) wrote:

Alan F wrote:

Unless I missed it being discussed in this thread, did SpaceX recover
the first stage? Reusing the first stage - or perhaps parts from it -
was part of their original plan to cut costs. For this flight, it would
presumably be useful to inspect the engine and pumps for any unexpected
damage.

Reusing the first stage is a capability to be acquired at some
indefinite point in the future.

Which avoids the question of has this first stage been recovered yet.
There was a recovery ship waiting in the expected landing zone for the
first stage on this launch. It being on station was stated in all three
countdowns.

Anthony