An Analysis of SpaceX’s Falcon 9 Crash Landing | Hacker News Reader

An Analysis of SpaceX’s Falcon 9 Crash Landing

wired.com

99 pointsjonah71 comments

A key piece of information we don't have, but Space-X does, is how much fuel was left at landing. Landing has to be done with minimum fuel, since it comes out of the fuel budget for putting payload into orbit. They have thrust vectoring on the main engines, and have attitude jets near the nose. They have to land vertical, at near zero velocity, with near zero angular rate, with an underactuated system. Underactuated means there are more degrees of freedom to control than there are controllable parameters. They have 5 controllable degrees of freedom - 2 axes of engine vectoring, main engine throttle, and 2 axes of relatively weak nose thrusters. They have to hit a point with three translation constraints, three linear velocity constraints, two angular constraints, and two angular rate constraints. So they have 5 DOF in and 10 DOF out. That, mathematically, is what "rocket science" is.

This under-actuated maneuver has to be solved as a two-point boundary problem, or, in practice, re-solved continuously to cope with problems such as wind gusts. You can see what happened; they had too much angular rate at landing, requiring more torque than the nose thrusters could impart to keep the rocket upright once the main engines could no longer help. It looks like they made a big correction shortly before landing to hit the target, and lost angular stability doing so.

If they'd had a bigger landing area, so they didn't have to apply big corrections to hit the tiny target, this would have worked.

You can't count position and velocity (i.e. d(position)/dt) as separate degrees of freedom. If you insisted on doing so, then you should also count rocketthrust and rocketthrust changes through time.

IMHO, the constraints here (as an armchair engineer) are more that the 'big rocket' isn't very responsive to requested changes in thrust, whereas the 'little nose rockets' may be responsive, but very weak compared to the mass of the thing they're trying to control.

"Degrees of freedom" is an oversimplified way of putting it. What the parent seems to be getting at is that the rocket, when landing, has 10 state variables while only having 5 control inputs, two of which have a very limited effect (the nose thrusters).

In this model, you don't count the rate of thrust changes as a separate input because your control system's job is to determine the required thrust at each instant. You also don't count thrust if you have already counted the main engine throttle (as has Animats). So the primary point of the parent comment, that the system is severely underactuated and thus difficult to solve, is correct.

>> You can't count position and velocity (i.e. d(position)/dt) as separate degrees of freedom.

Sure you can, and must. velocity is a controllable and it's also very important that it be controlled as close to zero as possible when Y=0. Parent is correct, technically they care about 11 degrees of freedom, but rotation rate about the length is either a non issue or well regulated somehow.

I'd really like to see the controller they're using. Wonder if it's fancy, or fairly simple with some large matrices.

I ignored rotation about the rocket's axis, since the rotational position at landing doesn't matter.

The controller for this almost has to be a predictive controller. You can't do this on feedback alone. The control goal is not stability; it's end state - stopped, vertical, and on target, all achieved at the same moment.

With enough fuel, you could do this step by step. Get into stable hover while vertical, then move slowly sideways over target, then descend slowly. NASA's Morpheus rocket works that way. Space-X is using much more aggressive control strategies; in the last seconds it's clear that all the goals are being sought simultaneously.

Landing should work much better on land, with a bigger, stationary target in a less windy place.

> I ignored rotation about the rocket's axis, since the rotational position at landing doesn't matter.

I agree the position at landing doesn't matter, but surely the rotation about the rocket's axis needs to be controlled so that it isn't spinning with any great speed - if it were, the RCS thrusters would have a hell of a time doing their job, as would the gimble of the main engine

>> I ignored rotation about the rocket's axis, since the rotational position at landing doesn't matter.

Right, which is why the care about 11 degrees of freedom and not 12.

No, you absolutely need velocity and momenta to uniquely specify a point in phase space (in addition to the other degrees of freedom)

Elon - "Looks like the issue was stiction in the biprop throttle valve, resulting in control system phase lag. Should be easy to fix."

AKA lost constrol of the engines. It could of landed in a field and it would of still crashed. Man there are so many armchair rocket scientists.

Aviation Leak says that message was "withdrawn."[1]

[1] http://aviationweek.com/space/spacex-checks-throttle-valve-a...

[2] ITAR.

as KSP player and armchair engineer:

when the rocket touched on the barge, thrust vectoring couldn't rotate the rocket anymore, since legs become a fixed pivot. and rcs didn't had enough trust to upright the rocket.

now I don't think the issue is as easy as 'rcs didn't had enough thrust'

what looks like is that the control system minimized vertical/horizontal velocity but allowed the rocket to touch down with too much rotational momentum.

whether this was caused by software or hardware has to be seen. if they are storing the landing fuel in the same storage as the ascent fuel, sloshing may be a big factor.

but I honestly have no idea, just like the speculation.

musk says throttle got stuck, in KSP terms the ctrl key didn't work as the computer expected it to.

Ugggh. StickyKeys activating and crashing my spaceship AGAIN.

Their wireless keyboard ran out of batteries.

Honestly, I've had several missions to Jove which have ended in disaster because my keyboard died during orbital insertion.

Jool, you mean? I think you are confusing Jupiter's adjective with the the KSP counterpart's name.

And that would be so annoying. It's no trivial task to get all the way out there and bleed off all that velocity.

Yeah, Jool. Has been a few months since I played - after pulling that stunt twice in a week. Both occasions missed critical burns to get out of aerobraking orbit, crashed - as best one can crash on Jool.

yeah I was so happy when they added X

I would lay claim to the idea its far easier to land in a field than on a floating barge. For one thing the field doesn't move, second its an awfully easier target to hit with greater room for error

The point was that this particular problem (oscillation due to control lag) would still have caused a crash even if it had been landing in a field.

That is a very interesting way of thinking about the landing. If you are going to use DOF, you should also mention the grid fins (4 DOF (1)), and how control over parameters can be "time-shifted". Early in the descent the rotation of the craft doesn't matter so much as its trajectory, and so the grid fins can control pitch to aim for the landing site (2 DOF). Once the trajectory for landing is mostly established, the grid fins can be used to eliminate the vehicle rotation (3 DOF). At the time of landing, the grid fins are no longer useful, but they have set one degree of freedom that none of the other controls will change (rotation about the central axis of the rocket), so essentially they have "6 DOF in" at the time of landing.

(1) In addition to control over rotation, the fins can be turned so that they cancel each other out overall, and thus slow down the vehicle without changing its rotation. I don't know whether the fins are actually used this way given the stresses involved.

They were able to do just fine with the F9R. I don't think this is a problem.

Guys please.

The sideways oscillation of the rocket is an obvious control loop problem. When it's working correctly, the booster will remain rock solid vertical and land gracefully. Elon's (now deleted?) tweet said as much.

Let the real rocket scientists fix their rocket. They know how. It's really unbelievably close to working. Amazing really.

Precisely. Looks like a classic control oscillation, possibly due to lag in the system. There are only a zillion ways to handle this sort of thing as it's an enormously common case. In fact it's within the realm of possibility that the "fix" for the next landing is just a tweak of some PID coefficients.

i'd guess the lag was accounted for, but due to something bad happening with the valve, it was different than expected. i don't do control theory but i'd assume there's a way of detecting this situation. maybe the valve needs an extra sensor or something?

> Let the real rocket scientists fix their rocket

By speculating and proposing questions and answers, we're all learning and increasing our understanding of the domain.

We're not causing any harm, so I think it's quite healthy for us to speculate and question what's going on and how it might be fixed. In fact, we even enjoy spending our time on it.

The fact that he deleted it implies they decided that wasn't the case after all, or at least wasn't the whole story.

They didn't handle media very well this launch, what with the fumbled video release and then a retracted tweet.

> The fact that he deleted it implies they decided that wasn't the case after all, or at least wasn't the whole story.

It doesn't. That's just your impression you are placing upon the tweet. Deleting it says nothing when you are the head of multi billion dollar organisations. Perhaps he didn't want to give their competitors information that could help them model F9.

> They didn't handle media very well this launch, what with the fumbled video release and then a retracted tweet.

This is just ridiculous nitpicking. SpaceX usually gives us camera views from inside the goddamn LOX tanks on the stages. 'Didn't handle media very well' my arse.

Didn't handle the media well? Some dickhead multi-millionaire who puts all his money into hedge funds and is pretty much just a negative drain on humanity can worry about their media image but how about we let the people putting their money into building space rockets do the space rocket thing.

He also deleted a tweet to John Carmack telling him to work for SpaceX. Don't read too much into a deletion.

My read on the deletion is that they didn't want bad press going into the next attempted landing, which is going to be on land.

Is it? Do you know which flight?

This is all I know: https://news.ycombinator.com/item?id=9387929

It's interesting to think about a case where Just Read The Instructions (the barge) might cooperate in the landing. Perhaps it could have an XY stage that could help compensate for residual X,Y or X,Y roll rates at touchdown.

Looking at the video, if JRTI could have slewed a platform quickly to the left after contact, it might have been able to return the tilting booster's center of mass to a stable position.

That's probably not much of an incremental engineering effort over what SpaceX has done already in this work. I'm completely awed by what they've accomplished so far.

(edited because I can't do word->initial strings late at night...)

I keep hearing lots of great ideas about how something other than the rocket could help the rocket land (catch it, move the platform, use a net, etc. etc.)

I honestly think SpaceX have no intention of learning how to do any of those (even though some might work great) because they want to learn how to land it in a place with no "smart" landing pad already setup.

i.e. Mars

Yes. That is also the reason they will not try to move the barge in unison with the rocket even if that was possible. Barge is trying to representing land.

The maneuver made by the rocket to land at an angle is intentional. The rocket wasn't trying to correct some unknown balance issue with it. The reason for too fast sideways speed came from the problem with the rockets response time, signal->engine.

The big sideways maneuver is made so the rocket will not hit the barge at high speeds in case of any problems.

Wellll, it's going to be a lot sooner than Mars. They inked a contract recently with the USAF to use a pad for landing. I don't know when that's supposed to happen, but the barge was never more than a stopgap until they could secure a spot on land and work out all the regulatory concerns.

> Wellll, it's going to be a lot sooner than Mars

For sure, but even then they could use some kind of tower or arm on their earth-based landing pad to "catch" it if they wanted. In theory they could even have the pad move some amount.....

If they wanted to of course, which they don't.

I would be shocked if any of that was actually easier than getting the landing right. Vertical landing without special arms or pads has been done a dozen times by now - DC-X, Armadillo, JAXA's RTV, Project Morpheus, and especially SpaceX's F9R... that's just off the top of my head.

The part that hadn't been done is getting from a fast, high altitude trajectory to the landing spot. Which they've managed to do.

Can a Falcon Heavy get the full center core of the rocket, including the first stage into orbit? Maybe with no payload? If so, this would mean transporting and landing a stage on Mars, for use as a return launcher, would be possible. Is this a possible long term plan?

I expect the more difficult part is getting the necessary fuel to Mars to escape. Obviously you could do it if you took enough trips, but that would likely be prohibitively expensive.

the plan that makes most sense is to produce fuel on site. allegedly there's water below Mars surface, so LOX/LH should be doable. maybe there's more stuff to synthesize methane, RP-1 or other fuels.

One plan I've seen for ISRU on the moon is to use a Al-Lox hybrid engine, since Al2O3 is all over the moon's surface. I wonder if there's enough on Mars to do the same thing, or if you could do something with iron oxide.

Definitely not. A fully fueled F9 first stage is about 500 tonnes, and can lift 13 tonnes to LEO; a Falcon Heavy is three first stages duct-taped together, and all three working together can only lift 53 tonnes.

> a Falcon Heavy is three first stages duct-taped together

I bet they didn't buy that cheap duct tape that stops sticking after 5 minutes :)

Odd that you say that --- that's about when the two booster stages are supposed to be dropped...

I hadn't realised that the barge was called "Just Read The Instructions" and it's just brightened up my day to see life imitating art like this. The names of the spaceships in Iain M. Banks' Culture series is one of my favourite little details in those books.

A nice tribute to Bank's work. I just finished re-reading The Player Of Games a few weeks ago, so I also recognised the reference.

For those who are curious, there is a comprehensive list of spacecraft names at http://en.wikipedia.org/wiki/List_of_spacecraft_in_the_Cultu...

SpaceX's other barge (which I don't think has been used yet) is the Of Course I Still Love You.

They really should've gone with "You'll Clean That Up Before You Leave" for the Atlantic barge.

This was named after a GCU vessel. I wonder what the designation of this barge should be. I think it might be SDS where SDS means Spaceport Drone Ship.

The title is misleading, should say "analysis of video of F9 crash". I know the current tittle is the title wired put in it, but the crash is not actually analysed at all, just the video of it, and that analysis doesn't provide any information of importance.

I would not call such an article "an analysis".

Oh come on. Generating graphs of the rocket's vertical position and angular deflection from vertical over time, using a video analysis and physics modeling package, is not analysis? Sure, having the actual telemetry data from the rocket would be better, but suppose you were a competitor to SpaceX and wanted to understand what their capabilities were and whether this approach is feasible, this is exactly the sort of investigation you'd want to do.

Overall, I'm very impressed by SpaceX's efforts. However, I'm wondering why they haven't tested the Grasshopper out on a water barge? It may be no different than on land, but that seems like a large assumption to make and a crucial variable to ignore.

Would that save them much development effort? I would think they would have to do a lot of refit each time they dumped it in the ocean.

Very good point. It may not in terms of pure time or money. But in terms of eliminating "unknowns", it will certainly save them many headaches. Generally the unknowns are what cost you in development efforts. In this case, was it the variability of a water platform (however minute) that caused the over correction, or was it a fault of a piece (i.e. valve or something similar) in the rocket?

When doing something new it's much easier if you eliminate all unknown variables except for one. So first, develop a controlled rocket that can hover and land on land (grasshopper). Next develop one that can hover and land on water. Now all major unknowns are eliminated except for the big one, landing a rocket from space, on either a water platform or a land platform. However, you will now be able to discern whether the issue is related to the water platform or just the general difficulties of landing a rocket hurtling down from space.

Wouldn't saving either time or money (or both) be the entire point of using a specialized test vehicle? They are anyway working their way through the unknowns by trying to recover the boosters.

I think they are also pricing the launches so that at least all the direct costs are covered, so these tests are 'free'.

Or put a barge in a wave pool or lake to test the grasshopper. That would save on transport costs.

I didn't see a mention of Musk's tweet, which labeled the YouTube video as "slow motion". That will make a difference (though likely just a scalar?) in the calculations.

I'm interested in why they haven't built another grasshopper to primarily use for the pad abort and inflight abort to get some bonus testing with.

Does SpaceX have a location they could fly such a test rocket? They can't go over a km in their texas location.

The video shows a lot of white caps in the ocean so the wind was blowing pretty hard. If the wind was gusting on top of that it's not hard to see why it would be hard to keep the rocket from being blown over even if it landed properly.

The first stage will be empty of fuel and is basically a big empty metal balloon with a huge lump of metal at the bottom. The centre of gravity is very low, so while it looks really unstable it's much less bad than it looks --- like a double decker bus. OTOH, as you say, there'll also be a huge amount of windage from that big empty metal balloon...

SpaceX seem to be competent, so I assume they'd factor this in. I'd be interested to see an analysis, though.

There's some white caps but not a whole lot. Wind speed at sea level might have been around 8 to 11m/s.

http://en.wikipedia.org/wiki/Beaufort_scale

They were so close!

Previous landings where more elegant, this one "looks anxious" (like done in a hurry). If they leave more fuel towards the end for error margin maneuvering then it can aproach landing with less momentum

Landing attempt video: https://youtu.be/rAzwuEmZcmE

This doesn't seem to take into account that the video was in slow motion?

Reply on news.ycombinator.com