FAILURE: SpaceX Falcon I Launch III - August 2

[Antares]

It seemed like the roll control just ignored it until it hit a threshold, then overcompensated slightly. Hopefully just a minor tuning of the controller?

Yeah, basically.  Once the roll angle error hit a limit, the FCS rolled it back to zero.  That repeated several times.

My guess is they will just set the null on the roll control nozzle to a slight angle rather than straight down.

[antonioe]

My guess is they will just set the null on the roll control nozzle to a slight angle rather than straight down.

Yes, but what slight angle?  Will it be the same next flight?  If I knew that was the source of the limit cycle, I would simply make the control loop a Type I rather than a Type zero and be done with it.  That said, in general, limit cycles are not bad unless they cause excessive use of a consummable.

However, I'm not fully convinced that is what is happening in this case.  I'd love to see the telemetry, I've fixed many a GN&C/aero/flight dynamics problem in my life from telemetry.  I've never had a case where the video helped (Joe, how about you?)

[Lee Jay]

I had a situation once that I just couldn't figure out from data or analysis of the parts (and I had great data).  I figured it out from the video.  (This was not rocketry).

I agree that switching from a P controller to a PI controller might be a good solution if that is, in fact, the problem.

[antonioe]

I had a situation once that I just couldn't figure out from data or analysis of the parts (and I had great data).  I figured it out from the video.  (This was not rocketry).

Well, I must admit that video would be nice in a partial deployment (antenna or solar panel) situation; fortunately we've never had a situation quite like that on one of our own spacecraft (knock on wood - statements like these are only good until the next flight...)

[just-nick]

My guess is they will just set the null on the roll control nozzle to a slight angle rather than straight down.

Yes, but what slight angle?  Will it be the same next flight?  If I knew that was the source of the limit cycle, I would simply make the control loop a Type I rather than a Type zero and be done with it.  That said, in general, limit cycles are not bad unless they cause excessive use of a consummable.

Wow.  Would you care to elucidate your verbiage?

I've worked with a few PID controllers, but never really understood the theory behind them (attach wires, play with settings until desired outcome achieved).  Seems like a fantastic chance to learn...

  --Nick

[Lee Jay]

I had a situation once that I just couldn't figure out from data or analysis of the parts (and I had great data).  I figured it out from the video.  (This was not rocketry).

Well, I must admit that video would be nice in a partial deployment (antenna or solar panel) situation; fortunately we've never had a situation quite like that on one of our own spacecraft (knock on wood - statements like these are only good until the next flight...)

Mine was not too dissimilar from the current cargo bay door hose issue on the orbiters.  I had a flexible hose flexing in an unexpected way, but that was very difficult to determine from pressures, temperatures and pump speeds.  When the video showed a "kink" forming in the hose under operating conditions, my low-pressure fault got really easy to understand.

[Lee Jay]

My guess is they will just set the null on the roll control nozzle to a slight angle rather than straight down.

Yes, but what slight angle?  Will it be the same next flight?  If I knew that was the source of the limit cycle, I would simply make the control loop a Type I rather than a Type zero and be done with it.  That said, in general, limit cycles are not bad unless they cause excessive use of a consummable.

Wow.  Would you care to elucidate your verbiage?

I've worked with a few PID controllers, but never really understood the theory behind them (attach wires, play with settings until desired outcome achieved).  Seems like a fantastic chance to learn...

  --Nick

A "P" controller won't "push" against the system unless there's an error.  So, if there's a steady "push" from the system (i.e. roll torque on this rocket), there will be a steady error with the P system before it can compensate for the steady "push" (assuming the system is stable, of course).

The "I" in a "PI" controller will "Integrate" that error, effectively "pushing" harder and harder until it nulls out the error, and then it will hold that setting.  Thus, a system with a steady "push" (roll torque) can still have zero mean error if controlled by a stable PI controller.

[antonioe]

I've worked with a few PID controllers, but never really understood the theory behind them (attach wires, play with settings until desired outcome achieved).  Seems like a fantastic chance to learn...

  --Nick

Nick:

If there is ONE area in our profession where one microgram of math is work a megaton of experimentation, control systems is IT.  "Play with settings until desired outcome is achieved", while instructional (and I commend you for having done that!) has to immediately give way to differential equations (at least linear... it's OK, they don't bite...), block diagrams (very intuitive) and Lagrange transforms (a shorthand for the linear diff eqs.)  Formally, a "Type alpha" system is one that has "alpha more zeros than poles in its closed-loop transfer function".

OK, OK, so that's Greek to you...  it means that, mathematically, the error signal is being integrated (integral with respect to time) and added to the raw error signal in the feed back loop to zero the error signal...

If you are interested in control systems and you are just starting, I strongly recommend THIS book:

http://www.amazon.com/Schaums-Outline-Feedback-Control-Systems/dp/0070170525

for $19.89 and free shipping, you can't beat it!!!

By the way: learn ANALOG control first, because then DIGITAL control theory will be a cakewalk.  If you don't, you just get your head stuck in the math and never get the feel for the PHYSICS of the problem.

Oh, I forgot!  The NEXT best way to crash a rocket - after the "trial and error" method - is to apply pure math unadulterated by any feel for the physics of the problem whatsoever...

(By the way: great explanation, Lee Jay!  have you considered teaching, or are you already doing that?)

[Lee Jay]

(By the way: great explanation, Lee Jay!  have you considered teaching, or are you already doing that?)

Thanks!  I'm not (other than the occasional guest lecture) but I get that a lot.

[just-nick]

Never expected the free control systems course.  This is great.  Might have stuck with engineering with teachers like this...

And I think I'm starting to get it.  With a "P" system, if I set a certain degree of control response to error (e.g. 1 degree of nozzle deflection per degree of roll error) the system will null the roll rate against a steady-state roll force, but at a certain degree of error from the desired angle.

The "PI" will generate progressively more response until it not only nulls the roll rate but will work the vehicle back to the desired roll position.

I assume the same ideas could be used (perhaps in parallel) to achieve a desired rate about an axis, in addition to position as I've been thinking about it?  I assume the derivative part of a PID basically computes some "lead" on the rate of change and damps out the response in order to avoid overshoots?

I might have to pick up that book now.  My math isn't that bad -- though I never made it past 3rd quarter calculus (which was right around the time I was playing with PID's) -- I'm teach a pretty I pretty mean statistics class.

[Oberon_Command]

While we're on the subject of control systems, a question to those knowledgeable: is there any way a fuzzy controller would ever be feasible or advantageous for launch guidance?

[braddock]

[...]  I'd love to see the telemetry, I've fixed many a GN&C/aero/flight dynamics problem in my life from telemetry. [...]

I'd LOVE to see the look on Musk's face if you offered your services. 

[antonioe]

[...]  I'd love to see the telemetry, I've fixed many a GN&C/aero/flight dynamics problem in my life from telemetry. [...]

I'd LOVE to see the look on Musk's face if you offered your services. 

Only matched by the look DWT's face if I did that!...    Hmmm.. that might be worth the try...

[JonSBerndt]

A "P" controller won't "push" against the system unless there's an error.  So, if there's a steady "push" from the system (i.e. roll torque on this rocket), there will be a steady error with the P system before it can compensate for the steady "push" (assuming the system is stable, of course).

The "I" in a "PI" controller will "Integrate" that error, effectively "pushing" harder and harder until it nulls out the error, and then it will hold that setting.  Thus, a system with a steady "push" (roll torque) can still have zero mean error if controlled by a stable PI controller.

Flight control of rockets is interesting stuff. I'm certainly not an expert in this field (although "I did stay at a Holiday Inn Express once" ;-). I get to occasionally work with implementing some launch vehicle control schemes and tweaking them in the course of my work, and also at home using the open source simulation software (JSBSim) I've been developing with a small team for the past ten years (see www.jsbsim.org). I've been working on some hopefully instructive examples of creating autopilot functionality using PID controllers in the documentation for JSBSim. It might interest you. Visit the web site and click on the "Documentation" link at top. See the Case Study near the end, "Piston Aircraft with Autopilot and Scripting". I've also implemented a simple rocket simulation with first stage open loop control, but that example isn't complete, yet.

Another question here involved the use of fuzzy logic for rocket control. There's an interesting paper you might want to take a look at, called "Robust, Non-linear Control Using Neuroevolution". You can see that here: http://www.cs.utexas.edu/ftp/pub/neural-nets/papers/gomez.phdtr03.pdf.

Jon

[jimvela]

Only matched by the look DWT's face if I did that!...    Hmmm.. that might be worth the try...

Sounds too much like an invitation for a visit by a red-hot poker to me...

Back on topic, it seems to my naive point of view that you'd always want to keep your vehicle in a known state to prevent unexpected/unknown unknowns (e.g. slosh, spinning up fluids in tanks, unexpected loads on mechanical systems, etc.).

In some cases you might even want to spin up the LV/payload, but even then you'd want everything held to a steady state.

Is that not what is done with mature launch vehicles?

[jabe]

What was a bit of a surprise to me regarding the failure is that one reason that the Falcons' are a 2 stage rocket is that the second most common failure for rocket failure is Separation issues as quoted in here.

After propulsion, the second most common cause of launch failures was separation events, which were responsible for 28 percent of all failures.  Separation failures included staging, payload separation, or fairing separation.

With the expertise available I bet they are "kicking themselves" for missing this issue considering they know staging is a main problem for failure.  From the reading here it seems to be a "rookie" mistake to miss it or am I over simplifying the issue?cheers
jb

Sidenote, even though Falcon 1 doesn't need it for ullage why didn't they include a "ullage" type motor to aid in separation?  Or does that add in extra reliabilty issues if one of the ullage motors fails to fire... (I'm assuming that they are a "simple motor")
Does the Falcon 9 need a ullage motor?

[daver]

At stage separation would there be any atmospheric drag?  Could the 1st stage of slowed down as well as the 2nd stage holding speed or accelerating?   Hope this isn't a stupid question from watching to much NASCAR.

[antonioe]

Sidenote, even though Falcon 1 doesn't need it for ullage why didn't they include a "ullage" type motor to aid in separation?  Or does that add in extra reliabilty issues if one of the ullage motors fails to fire... (I'm assuming that they are a "simple motor")
Does the Falcon 9 need a ullage motor?

Strictly speaking, an ullage motor is a device to produce a small amount of acceleration for a relatively long period of time, long enough to allow gas bubbles in the upper stage's tanks or feed system to migrate away from lines and feed point(s).

Separation is traditionally accomplished either via springs (the favorite method from a reliability and cost standpoint) or retrorockets; neither of these methods provide ullage (spring accelerations usually do not last enough for practical gas separation).

One observation on the SpaceX video showing separation; it appears to me that the camera had a very wide angle lens.  This tends to exaggerate distances and therefore speeds away or towards the camera.  What appears to be a rather rapid separation and recontact is probably a lot slower (and the separation distance achieved before recontact much smaller) than it seems from watching the video.

As the Capitol Steps used to sing, "Objects in Mir are closer than they appear" 

[Comga]

On the video the first stage pushed the second up to 6 seconds after MECO.

Perhaps, but did you notice how well aligned the two stages remained? Unlike F1-002 with the yaw (or pitch) rate at staging where the second stage bell smacked the interstage, they seemed to stay remarkably well aligned. 

It appears that they did solve at least one of the problems of flight 2.  Perhaps if they had had an even larger rate, though, ....

PS. I like the three digits so they won't have to renumber after the first one hundred flights of the Falcon 1.  :-)

[jabe]

PS. I like the three digits so they won't have to renumber after the first one hundred flights of the Falcon 1.  :-)

I like that idea

but the $10 million question (or how however much you view the cost of the next launch and probaly including the reputation for Spacex) is the solution as simple as few seconds delay before separating?  Lets hope so..be great to see  F1!-100 some day soon
jb