The heading reflects the combined result of a yawing tendency AND the rudder surface deflection together.
If you step back from the imminent main upset, and look at the early part of the takeoff roll instead, the pilot already had to input stronger left rudder force than he expected with the wind data. So there must have been an additional right yawing tendency during the early part of the takeoff roll, that he - successfully - compensated with stronger push on the left pedal.

(If there had been no such tendency, the aircraft would have ran off the left runway edge before becoming airborne. But it didn't. Something was causing a greater tendency to yaw to the right than expected, which disappeared around the moment when the aircraft rotated. The continued pressure of the left pedal remained, led to an even greater deflection of the rudder pedals and surface, and the over-compensation of the disappearing-reducing right yawing tendency resulted in a susprise turn to the left and yaw to the left instead, with practically no input force change from the pilot flying.)

The last appearence of that "hidden" right yawing tendency may be between 20:40:26 and 20:40:28. You can see a slight heading deviation to the right, followed by a significant 10 degree left rudder deflection, and a subsequent return to the normal heading. Would be good to understand, what "conditioned" the pilot to input larger-than-usual left rudder forces, because that pre-conditioning may be partially the answer why the deflection increased into the wrong direction at rotation.

I see what you mean. Yes, that 10 deg rudder seem to fix a right deviation back to centre-line. It's almost 2 seconds and markedly larger than normally expected.

The plot is missing LOC deviation to see the lateral offset. Following your line of thinking, another would be the brake pressure for wheels 3&4 and their temperature.

To my experience, without a small aileron input the plane has a tendency to go downwind. And theirs (during the roll) didn't despite the left rudder input.

Allright, here Special FDR's from start of T/O roll (FDR tabular data starts at 20:40:00 anyway)
(Edit: replaced Plots with better ones)

Overview:
https://cimg9.ibsrv.net/gimg/pprune....50e76e3937.jpg


Following Plot with Heading in higher range, Rwy Hdg marked with yellow line, Rudder Zero marked with blue line
I reversed HDG here so +y is left turn to match rudder orientation:

https://cimg5.ibsrv.net/gimg/pprune....ca32cc5442.jpg

Especially looking at those new plots of mine,

• from 20:40:10 the left rudder peak values show a decreasing line over speed, as we expect.
• This is a trend up to 20:40:30, with exception reacting for 2 heading bumps to the right.
• Only at and after 20:40:30 this known "violent left push" leading to the mess is outstanding.

Now if a zig-zagging between -5 and -13° rudder in initial roll, then moving btwn -3 and -10° in the later roll is typical on that xwind values, I have no idea.
How much inch of travel is 10° rudder pedals?

engines seem stable
speed has unsteadiness at 4-5 points, looks like going along with those heading variations - gust or wake turbs ... so it was not totally steady air, but also not drastic.
no brake line values, no spoiler readouts to compare a right assymetry

Don't think we find other hard facts from it. But how do I know...

Capt. Occam suggests the mysterious right tendency might just be the cross-wind acting on the empennage after all.

In the interviews, they mention a computer simulation playback of the incident, so I think there will be some more information on the actual deviation from the runway centerline.
And yes, perhaps it was nothing but higher than expected crosswind gusts Captain Occam is often right. It's just good to have all the failure modes on the table, so that we can discard 99% of them with confidence :)

Thank you!
So there was indeed at the beginning on average a relevant left deflection of the rudder while heading was more or less stable and straight. The average Rudder deflection almost linearly decreases from slightly below 10° to around 5° at takeoff speed before suddenly and continuously being deflected to 20°.
Reducing of deflection would match the correction of a cross wind (or the nose wheel not pointing exactly aligned with the rudder, but this would have surely been identified since).

I have measured the deflection in a D-level FSTD. Well, it's not 10 inches. :rolleyes: More like 11 cm.

Thus the 1/2 deflection, oscillating and reducing to 1/3 (5 to 2 cm travel) during the stable roll matches personal experience (post #153)

The National Transportation Safety Board determines the probable cause(s) of this accident to be:

The captain’s excessive left rudder pedal input during the takeoff ground roll, which caused a large heading deviation and a left roll upon rotation that resulted in the left wingtip striking the ground.

Also from the NTSB final report:

Further examination of the airplane determined that the left wing had a permanent upward deflection starting about midspan of the left wing, resulting in the left wingtip being about 6 inches higher than the right wingtip.

Thank you BFSgrad. I assume the permanent upward warp was the result of the accident, not it's cause.

Summary on AVHERALD tells it was no aircraft related issue and I can't see any other influence than the PF excessive left rudder.

• It was a 14-17KT crosswind, stronger than wind check from tower suggested, but about half of companys' 35KT xwind limit
• comparison with 12 months data of other flights showed less rudder input, even on takeoffs with stronger crosswind
• simulation showed that the aircraft reaction was consistent with the input given. Expained the yaw to the left runway edge and the roll during liftoff. So no anomaly, and I guess no other contributing factors
• The only anomaly was the aileron deflection during wing scrape (as we found already in the FDR plot), again as a consequence, not cause.

Bad day, and yes, Human factor applying control force without being aware of it.

Answer to other questions:

After the accident, American Airlines examined runway 31L and documented a left main landing gear tire mark that was 564 ft long, a right main landing gear tire mark that was 216 ft long, and a scrape mark from the left wingtip that was 323 ft long. The wingtip scrape marks began to the left of the runway edge marking and extended to the edge of the paved surface. The airplane’s left wing impacted a distance remaining marker on the left edge of the paved runway surface

Means the scraping lasted for around 1.25 seconds.

I find it remarkable that they decided to scrap the airframe 🤔

FlightGlobal reports that the aircraft was left with a permanent deformation midway along the port wing, such that one wingtip was 6 inches higher than the other.

Attempts to buff it out were unsuccessful. :O

Would love to say that there was a technical issue with this aircraft, can't find any justification for that however. The aircraft at all times responded correctly and promptly to the flight crew input.


Countering the stuck spoiler case, the aircraft yaws on the ground to follow the rudder,

"and the rudder is responding correctly to the rudder pedal"
Not so. Sensing rate for the data stream from the 429 bus is 8Hz. The timestamp is common to both data streams for the rudder and pedal.

The data shows that the rudder position appears to be feeding back to the pedals, in what should be an irreversible control system, yet, there are a number of data points that show the opposite, that the rudder has moved before there was a control input recorded. There is the potential for an artifact from the frequency of the sampling, however, there are a number of data points where it is not possible that the control input could have been made in the time to the record of the position that would result in an artifact, where the control is leading the pedal input.

I am surprised to see that in the data. I doubt that a normal person can make a control input that is so fast that it is recorded out of sequence with the following data.

The blue + is the rudder position, the orange X is the Rudder pedal position recorded in the DFDR. This should not be QAR data, so it should have no sampling errors from 429 to 717 or similar, this should be valid data and valid sampling times. In the expanded view below, there is a rational control position that could have a smoothed line on it, per the DFDR analysis done by the NTSB, but then the position of the control surface is preceding the control input, which is not something that should occur. This occurs in the mid acceleration on the runway, and again appears again in the middle of the upset. Is it an artifact? Probably, but it needs to be discounted before the firing squad lines up.

https://cimg6.ibsrv.net/gimg/pprune....8d845a53ef.png




Can a hydraulic control system feedback to the control input? Normally it does not, however, there are possible internal leak cases that could do that.

Does it look like that happened at the rotate? No, not immediately, however, if and it is a big if, the rudder system was compromised and feeding into the pedals, then how that would show would be dependent on how it has failed. The rudder pedal is fed back the position of the rudder as well, it also does so on both pedals.

It is possible that a rudder issue did actually occur. I doubt it, but the position of the rudder in the middle of the takeoff acceleration leading the recorded pedal position is concerning, and the same happening in the middle of the lateral excursion is also concerning.

Sensor saturation of the pedal, or rudder or both may be in play, but the data suggests a further look could be warranted. Equally, an error in the followup of the hydraulic actuator to the pedal input could be occurring.

All in all, it is an odd bit of data.



https://cimg3.ibsrv.net/gimg/pprune....42c3e5a38c.png

Figure 1: initial take-off acceleration

https://cimg6.ibsrv.net/gimg/pprune....f41f054905.png

Figure 2: Nod's wild ride

Super quick intervention by the RHS occupant, as in, wow!

https://cimg0.ibsrv.net/gimg/pprune....47b180a011.png

Figure 3: the last revenue landing of this A321

RHS landing. Interesting decision.


https://cimg6.ibsrv.net/gimg/pprune....5d32f07ed6.png

Figure 4. LH SSC gets a work out to 20:51

And then the FO does the landing with a damaged "plain". Interesting call.

I'll need to look it up, but there is a mechanical linkage of the pedals to the rudder as per FCOM. Should work both ways, but perhaps introducing some freeplay?

Would Yaw Damper function or just kinematic / aero lag (WTH, on a powered rudder, why?) explain a bit of what you are seeing? I can't really read it at first glance. I don't think there is any FBW filtering / gearing for the rudder at all, save for that Yaw Damer function (which is not fed back to the pedals - so much for a mechanical linkage).

Once I believed the link actually only goes as far as a hydraulic clutch, into the tail assembly, where it interfaces with a closed loop internal servo-booster of the rudder itself.

Otherwise the YD on the bus is underexplained in the FCOM and that is an understatement. But hey what, they need to say steady climb is a constant 1 g path for their audience.

I noticed this as well 1 month ago:
But now with the final report stating that no aircraft system was a cause, this would rather be a FDR related pecularity. Whatever, no undetected anomality influencing the events.

But keep digging, it's a funny detail.

I won't even try to dispute the technical part of your analysis. Having said that:
- The captain was (most probably) the only pilot with feet on the pedals.
- Those pedals ended up deflected to the left.
- The airplane turned to the left, and after rotation banked to the left.
- The captain commented the aircraft banked left, but not did comment the rudder pedals deflected uncommanded.
And while I cannot understand how a 20K hour pilot can make such a basic mistake, I feel it is more likely than a failure leading to an uncommanded input that wasn't recognized by the pilot.

Is not the aileron a give-away? When the heading starts to diverge, there is almost full counteracting stick. And just as you would in a cross-controlled slip landing to keep aligned, even more downwind rudder follows.

40:31 until 40:32,5 on FIGURE 2 from fdr. With the aircraft still on the ground, there is no corrective effect coming from the aileron deflection and the rudder makes the deviation worse. Once pitched up the effective span difference of the swept wing rolls the plane

The stick is back at neutral for rotation while the rudder input remains until after airborne.

Interesting that this is not the first AA Airbus departing JFK damaged by the pilot’s excessive rudder pedal input.

Waistline?