SaturnV

The English version seems, on first reading, to be fairly good English.

Having learned my lesson of extrapolating from a David Learmont post from the Paris Air Show, I'll not revisit the cockpit door.

I will note however that the location of the captain's seat and the co-pilot's seat look to be about 30-35 meters apart on the ocean floor. The fourth seat was found about 10 meters from the captain's seat.

With respect to the belts, the English version says this:

A question is whether there is a missing phrase: i.e., "On the left side seat the lap belts were attached [to the occupant],.." or is the more accurate interpretation that the belts were no longer with the seat?

If the latter, that throws into question the assumption that the first two bodies recovered by the Ile de Sein were those of the PF and PNF in their seats.

Also,
The recording starts at 0009.
____________

Smilin_Ed

Of course I want the aircraft to respond to control inputs but I wouldn't want it to change the trim. That is a very basic thing when flying in turbulence. In this case, it followed what was clearly an erroneous input by the PNF and trimmed them up into a stall, a no-no taught to every student pilot.

gums

Thank you, Wozzo, very interesting press release. Seems that some BEA investigators are concerned about some of the same things that several here are, as well. And it's not all strictly pilot error or strictly aircraft system design.

Thank you A33Z for the link to the English version of the report.

- The lack of discussion on the CVR concerning the stall warning puzzles me.

- An AoA indicator is "nice" to have, but we must examine the flight conditions it is intended to support. 'bird, Retired, Smilin' and Gums can testify that the AoA "bracket" in the HUD or the "indexer" lights were extremely valuable for approaches ( especially on the weaving deck of a large boat, heh heh). We didn't need to calculate airspeed to a knot, and we routinely landed at various weights depending upon our external loads. 'nuff said about that.

For a commercial airliner, we don't have the same requirement. Nevertheless, the AoA is very important for stall warning and recovery. The plane produces lift according to AoA and dynamic pressure!!! You can stall at many knots faster than the manual numbers for one gee.

So placing a new indicator in the displays is not a biggie for this ol' dinosaur. seems the 'bus has enough confusing displays as it is.

Use of AoA other than a display is another matter.

- Ask the above pilots if the AoA sensors ( vanes or cones) worked below 60 knots. Even if the AF447 suckers were bouncing around a bit due to an extreme AoA, I'll bet they showed a high AoA, and were not flipping back and forth from plus 30 deg to minus 30 degrees.

For the FBW system to ignore AoA below 60 knots does not seem right. Most military planes use the weight-on-wheels switch to display or even use AoA.

Further, how was the stall warning being sounded if the system had disregarded both speed and AoA?

No doubt training and manufacturer claims will come to the fore here. After all, we're "protected", right.

"We're going down, sir"

"O.K., command the plane to go up"

" I am doing that, sir, but she keeps on descending"

mm43

I would suspect she knows nothing of Rugby Union supporters from Toulouse!

DozyWannabe

Hi Ed,

Is that because you can't see any instance when it would be useful or because it goes against the methods you were taught flying aircraft with more "conventional" controls?

IMO (for what that's worth) there's nothing wrong with the autotrim setup as long as the way it works is taught properly. If you look at the traces the trim moves minimally under automatic control. What caused it to deviate so strongly was a series of inputs that trended towards nose-up, the majority of which were around half the stick's rearward travel limit, that were sustained for the best part of 45 seconds. What caused it to continue the movement to the stops was a full nose-up deflection that lasted between a further 30-40 seconds - that's more than 1 minute and 20 seconds of nose-up input at a deflection that ranges from halfway to the stops - at cruise level!

The trim doesn't move in any noticeable way as long as the inputs are relevant to the flight regime. The PNF notices that the PF's lateral inputs seem to be extreme, and admonishes his counterpart. He then further upbraids the PF for commanding a climb when it is unnecessary. This is why the BEA recommend training for manual aircraft handling at altitude, because in this case the sidestick inputs are repeatedly way beyond what is reasonable at that altitude and airspeed.

Well, the press release appears to be in response to the articles that appeared to have sources with in AF who are clearly not happy about the stall warning situation. Whether that relates to a well-researched hypothesis, or whether it is an attempt to muddy the waters in anticipation of the division of responsibility in the coming criminal and civil litigation it is impossible to say.

And it is indeed in the recommendations that have been agreed thus far.

Confusing how? I'd say they're pretty well laid out for mid-80s technology!

Well here's your traces (corrected values and raw) :



Looks pretty spiky to me after 02:11:45...

With all due respect, they never said anything even close to that...

Machinbird

Dozy, The spiky traces are merely an indication of how confused the computers were.
I can promise you that those AOA vanes were essentially pegged to the high AOA position after the stall, moving down slightly perhaps only when the crew tried some nose down stick.
If the traces show the vanes moving that much and that regularly, it has to be bogus.
Probably a good argument for routing the signal to an AOA indicator first, and then to the computers/ADRs as a derived signal. The computers/ADRs, it appears, cannot be trusted with the data in extreme conditions.

jcjeant

Hi,

A nice exercice is to experiment some reverse engineering ...
Just forget all you know about the FDR and the plane datas as released by BEA (this is the most difficult part of the execise)
Just study the CVR and from there .. try to imagine what make the plane ......

gums

You are correct, Dozy, the crew never said that. I was being sarcastic about the system design that "protects" hapless pilots from getting into trouble. Do we have "pinball wizards" or real pilots flying these things?

An "aggressive" design would have used whatever the hell was causing the stall warning to sound to move the nose of the jet down!!! It would have some limits regarding mach, but the main operational function at the time would be to break an impending stall. "We can worry about mach later", says HAL.

The AoA traces show a fairly smooth increase, then we have what looks like a lotta noise. Could be electrical noise, could be effects from being on the bottom of the ocean for two years, but I cannot fathom the vanes/cones moving that much due to basic engineering practices of mechanical dampening and the inertia of the probes themselves. In any case, the "system" can use many techniques to "smooth" the data that is provided the main FCS confuser. Talk with me about data reduction on a test system that had crappy electrical wiring, and took us weeks to figure out the problem.

I am still not convinced that the crew had clear warnings and indications of the situation. Given that the line pilot has not been presented the combination of events and such, I can understand a bit of confusion. But in the end, I see the claims about enhanced safety and "protections" and then I see a situation where all the engineering/laws/protections didn't help at all.

Shadoko

Somebody has suggested the curbs were Excel made. We know from BEA that "void" AOA values are not "transmitted". So, in the BB memories, they are absent or replaced by a non numeric value. With Excel graphes, if you don't tick the option "ignore absent values", they are considered as zero. Perhaps it is simply that?

takata

Hi gums,
It is due to the effect of projecting this graph at this resolution.
What one would see, at a better resolution, is a pike [max-zero] each time the value of one probe is invalid (NCD: no computed data). It looks so "noisy" because the value of each probe is not sampled at the same time.

e.g. if the sampling rate of the recorder is 1 per second:
0.33 s - alpha 1
0.66 s - alpha 2
0.99 s - alpha 3
1.33 s - alpha 1
1.66 s - alpha 2
1.99 s - alpha 3
etc.

With a lot of NCDs, it will become unreadable pretty fast.

gums

Thank you TK and Shad. Makes my day.

We must be careful looking at the recorded data unless we have put the test vehicle and recorders thru similar environments and exerted a lotta effort to ensure we have "good data" at the end. This incident did not lend itself to a disciplined engineering testing approach. It may well have been a "one of a kind" test point.

From my experience as a data reduction engineer, after I hung up my gee suit, the most noisy data from the sensors was pneumatic pressures. So I can see some spikes in the air data and such. Really great sensors and some clever filtering by the data reduction gnomes can help. But we still have to look at the raw data and make a value judgement.

Machinbird

Takata.
If I understand you correctly, and read the trace correctly, when the system considers the value of the AOA to be invalid, it applies a zero value to the signal which is momentarily overridden by the periodic read of the probe, thus creating the spikes.
When the data is considered valid, the signal remains at its last value until the next read, thus only small steps, and no spikes.

takata

Hi Machinbird,
There is three curves (one per probe) represented at the same time, each with a small time offset. Hence, if three values are NCD, it will paint most of the background due to the three spikes. When there is only a single value, in other graphs, it doesn't produce the same effect as the spikes will be spaced.

airtren

Do you mean the case when the bogus mechanism delayed the recovery, to the point of the crash? As a frequent passenger, I would like to see the problem fixed asap, wouldn't you?

takata

@Machinbird
- first is mach
- second is speed
- third is attitude
- fourth and fifth are alpha
NCD values are either 0 or Max. The offset is clearly visible in speed graph, and the alpha graph is unreadable when NCD.

Machinbird

Takata,
Between the spikey areas of the AOA trace, there are brief interludes where there is a single smooth trace (apparently by all 3 probes). No spikes, and the AOA appears to briefly decrease slightly.

I'm interested in the usability of the AOA data in the system, and with weight off wheels it should be available full time. The 'system' appears to be applying a biasing signal to the data, bringing the signal to zero when the data is considered invalid, but leaving it at last value read when apparently valid. This has potential implications for future (and present) AOA installations on Airbus aircraft.

Essentially airspeed has been given authority to outvote AOA (which we already knew from the stall warning fiasco) but this is not a good situation for present and future AOA installations in Airbus aircraft. This I consider to be a fundamental engineering error, but one that can fixed, probably by better software.
I use the word error, because AOA and airspeed are both fundamental aircraft performance data, but independent of each other and derived independently. To then mix them together and prioritize them is simply bad logic.

takata

I don't understand what you meant by leaving it at last value read?
System doesn't use those "values" as depicted here, this is only a recorded sample of several sources used. Sensors are always sensing something, at any time, when they are not physically failed. Their own output rate is many times higher than that.
If, at one point, an output needs to be considered invalid below a certain level, there is obviously a potential logic about not using it. Now, there is certainly also some room to improve this logic as far as SW is concerned.

takata

Of course not. All air data sensors are influenced by a combination of airspeed and angle-of-attack, in addition to environemental factors which are not exactly the same at sea level or 41,000 ft.
During this stall sequence, what in fact outvoted the alpha probe was the very high alpha achieved ! Alphas outvoted themselves by making other probes displaying wrong values.
Certainly that this was never considered remotedly possible. But keep in mind that limits still apply for everything flying.
Say, maybe this aircraft is able to fly at Mach 1.2 without falling appart. Would you ask for its probes to be certified for not stopping an overspeed warning at such a speed, in the very remote case that someone would ever try it?

BOAC

Well, a quick look at the English version and it is really awful to read.

A couple of queries - the BEA constantly refer to 'co-pilot's stick' without saying which one. Why is it not 'Captain's Stick' etc? Without trawling through the traces it is difficult to decipher.

According to the CVR extract the spoilers were extended at one point and not 'noted' as being retracted. It is difficult to decipher the FDR trace on this. They appear to be extended for well over a minute which adds to the 'overspeed' mental scenario. Can anyone elaborate?

To anyone puzzled by the nose-up 'pull' at 4000', I ask what exactly would you suggest they did instead at 4000' with around 10,000 fpm down? I think we are down to pretty basic human instinct here.

takata

Hi BOAC,
It still make me sick while reading it today.
I did not noticed that. If it's part of the CVR/FDR table, "co-pilot" would apply for RHS (PF); PNF imputs are correctly labelled "Captain". If part of the narrative, where? (usually they add the seat?)
It makes
0212:04 - 12:07 - PF said "I have the impression that we have some crazy speed, no? what do you think?", and he deploys spoilers. PNF correct him "No! above all don't deploy (them)"; they were retracted immediately (~5 seconds from graph). Captain added nothing.