Have you read the BEA report?

HazelNuts39

The final report Bea, I’ve printed 224 pages, with my printer, I did not deliberately read yet. But, I’ve read the other Af447 Bea report.

nikplane,

if you've read the 3rd interim report, you should know that the pitot-tubes did not act like an altimeter. The final report has a figure 64 that shows the vertical and horizontal movements of the air.

Switching off 2 FCCs came pretty late, they were trough 10000 feet already.
But "on comprend rien on a tout tenté" came much earlier, close to 2 minutes earlier, in fact it was part of the briefing that the PNF gave to the captain at his return.

There is some history on that especially at AF which was one very early operator and at great scale of the new protected FBW technology.
Came Habsheim (I don’t want to discuss it here) but with Habsheim came already suspicions of computer interference. Then came in France the air inter accident where again the investigation was lacking of transparency, FDR was labeled as destroyed … QAR data were not extensively published, and multiple time references were used indicating partial 'loss' of CVR data.

Also the FCOM has this interesting note :

http://i35.servimg.com/u/f35/11/75/17/84/med_tc10.png

Did it happen already at AF on the early days of the 340 … I’m not sure but I’ve heard something in that direction where the only way to get rid of unwanted protections was to force direct law by switching some FCC off.
Was it unofficially briefed at AF during sim training ... ?

Listing
 

Concerning the AF447 and BEA final report .. it's unfortunate (a shame ?) that the complete FDR LISTING is not public

CONF iture: In the said case (re: your #164), switching off one of the two consistent-but-wrong ADRs would be enough?

Turning one of those ADR off could be one way to deactivate the protections, but turning all FCC but one off would be a radical way to switch to direct law and deactivate the protections as the same time.

Such initiative to switch 2 FCC off without any ECAM command did not come out of the blue. I am questioning the BEA how they did not investigate this episode to find out what may be hiding behind …

This action also shows IMO there was not any panic on board as many like us to believe.

Pitching up to 15 degrees at FL350 indicates panic to me. No airliner can do that without stalling.

As I understand the system it doesn't work quite like this. The BEA report says:

From all the diagrams I have seen describing how C* works the pitch rate/g mix is a feedback term. S/S movement always commands a load factor as the steady state output just as the BEA report says. What the pitch rate/g term does is change the shape of the transient response between initial stick movement and achievement of the final desired 'g'. The origins lie way back in aircraft handling research when observers noticed that pilots tended to base their opinions on what constituted 'good' behaviour on how the pitch rate transient varied at low speeds and how the 'g' transient varied at high speed. Then somebody thought if that is what they rate as good why can't we give it to them?

So when in Alt 2B the system gains default to their 330 kt (Vmo) values the effect would be that the system would be trying to achieve the 'ideal' transient response (remembering that the Airbus intent was to make the pitch response the same at all speeds) using high speed 'g' base gains but at much lower speeds. This is what I think BEA are getting at when they say the aircraft has an unusual response dynamic in this situation.

Since the transient 'g' response to elevator movement would be much crisper at high speed I think this would lead the system to apply less elevator (to drive the pitch acceleration) than it would normally use at these low speeds, so the aircraft response would be more sluggish.

Anyway, that is how I see it working :8

@Owain Glyndwr

Thank you for this input, there was an argument about this some posts back and thankfully we have straightened that one out. Your explanation is sound and easy to understand ( sorry that i couldn´t explain it that way), and now the result of using those fixed high speed gains gets clearer. This sluggish behaviour in pitch would be a new expierience to PF, especiually when the roll channel behaved quite differently (more agile).

Interesting would be, at what point those default values come into action. Is it the point where the speeds are lost or is there some time-delay / internal checking before this changeover takes place? And is this changeover sudden or gradual? Could the initial NU input be under the false speeds ( agressive) and the later correcting attempts under the default gains (sluggish)? When would the default gains be replaced again by the actual speeds?

Sure, and I understand you would prefer that (IIRC you are +/- advocating the Alt laws are useless, if not dangerous).
My point was: it's not the only way.

Perhaps the BEA didn't investigate enough. Or perhaps they hadn't found anything consistant to write about it.
In intermediate report 2, the BEA wasn't sure if the PRIM1 and SEC1 were commanded off (by the crew) or faulted off (by the "system"). Ref IR2 §1.16.2.4.1.
With the benefit of the CVR transcript, one can conclude that PRIM1 & SEC1 were commanded off:

2 h 13 min 28,2
PNF : essaye de trouver ce que tu peux faire avec tes commandes là-haut (my note: the PRIM/SEC commands are on the overhead panel)
2 h 13 min 30,4
PNF : les primaires et cetera (my note: "primaires etc." may refer to the PRIMary and other computers)
CPT : (* fera rien)
2 h 13 min 31,5
CPT : on (fera /verra) rien

And then:
2 h 13 min 45 F/CTL PRIM 1 FAULT
2 h 13 min 51 F/CTL SEC 1 FAULT


I disagree. The transcript show the PNF to want to "try" something with the "primary" (computers?) then the captain answering it will be no use, then the PRIM1 & SEC1 to be switched OFF.
This crew did enough "strange" actions for me to believe they were following a really logical path. It seems to be a case of "what about trying that?" (even with no clue/agreement) meaning: they didn't understand, they tried this or that, because they didn't realised the root cause of their incapability to control the aircraft (i.e. the stalled condition).

Owain: thank you, a light went on as you described the gains and input relationships, and responses.

Retired F4:
Some years ago, I experienced a somewhat disorienting moment in the cockpit when the SAS and trim in the lateral channel went awry while the pitch channel remained just fine. We disabled AFCS and flew AFCS and SAS off (which required a gentle touch ... but is also something we used to practice!) once we figured out what was wrong. It "felt" wrong to say the least at onset.

I suspect that PF in AF447 encountered a similar "wrong" feeling, if the graphs depicting his efforts in the roll control are any indication.

"What we have here is...(a) failure to communicate." :(

One would want to check the trace record of elevators position at Law change with PF's first inputs in Pitch. With very low indicated airspeed wouldn't the response of elevators via FCC be extreme? relative to .82Mach? The aircraft exhibited a brisk increase in VS just at handover. Also just prior to the First Stall Warn, the PNF says: "What was that?". Could that have been a sudden reaction of the a/c to inappropriate "response"? Here, as in other areas, the CVR would be helpful, but is not published. The explanation for the comment was offered that PNF was noticing the SW, certainly, but it is not proven.

"Unusual Response-sluggish" Or "Unusual Response-Sudden"

BEA could clear this up with a memo, rather than relying on someone's best guess?

One does not expect a Stall Warn from a pull on the stick that is modulated through a computer to correct for load? Sluggish? NOT

Can this be related to Airbus OEB not to reselect A/P they released after 447?

Were they fearful the ap would "Climb Uncommanded?" with low speed inputs in this control law? Alternate Law2B? why wouldn't the ap be locked out by the system when UAS obtains? Not to mention Flight Directors....

lyman #151

re: "HazelNuts39" and "sluggish"
"If the responses were sluggish, would that not possibly explain PF's dogged pursuit of some G? Might he have been seeking the twitchiness in Pitch that he found in Roll? If he thought the aircraft was unresponsive in Pitch, would he not command as much as he could get? If he sussed sluggish, and got no immediate "response" would he be tempted to trade rate for continuation of input?"

i consider this comment pertinent to Pilot's possible reactions derived from known or unpublished evidence. iow Human Factors...

"Pilot continued his NU commands, and the aircraft 'started to climb...' " from BEA, IR3....

HN39 & OG may be entirely correct here.

However, from my experience, the 'usual' response of aircraft at abnormally low speeds for the configuration IS sluggishness.

Unusual would be 'responsive' to a higher degree, much like 'stick force lightening' in some aircraft. :confused:

OK465

"OK465 "HN39 & OG may be entirely correct here."

"However, from my experience, the 'usual' response of aircraft at abnormally low speeds for the configuration IS sluggishness.

Unusual would be 'responsive' to a higher degree, much like 'stick force lightening' in some aircraft."



447 was not at abnormally low speed. She was moving right along. Her indicated airspeed was erroneously low.

So it would not be unusual for her to respond with a too-far excursion of elevator If at high speed but the computer sensed low? Something is behind Airbus' concern for uncommanded climb.

Again, this should have been addressed by BEA...

@Retired F4

I'm entirely with you regarding the relative changes in lateral and pitch control, although I believe the change from S/S demanding roll acceleration instead of roll rate to have been the important factor. But pilots who have tried it say that although the aircraft is more sensitive in roll it is by no means difficult - just like a non-FBW aircraft in fact, and the fact that the AF447 pilot got it sussed in 30 seconds or so seems to confirm that.

'Sluggishness' in pitch is a relative term and it it always difficult to find words that give the desired impression. Based on experience with other large aircraft I was thinking of numbers like these for cruise conditions (the times are approximate indications only)

For a demand of Xg
Normal: after 2 seconds Xg; 4 seconds 1.2Xg; 6 seconds and onwards Xg
Sluggish: after 2 seconds 0.7~0.8Xg; 4 seconds 0.9Xg; 6 seconds and onwards Xg

In other words the initial response would be less rapid but achieving the final desired 'g' not appreciably different.

I don't know of course, but I would think the default gains kicked in at once when the internal checking logic said go to Alt2B. That would place the whole sequence with the default gains. I can't see any change to these gains unless and until Alt2B got replaced by some other law - which didn't happen.

Where did I mean 'dangerous' – Any quote ?
Alternate law on the pitch but direct on the roll brings confusion.
Yes, direct law all the way, that crew would have been better served for all the reasons mentioned earlier.

The BEA did not positively report as such.

Asking questions and trying something is not my definition of panic.

Thank you, OG. A great explanation of some fine points of the control laws.

As a matter of credibility, I was one of the pilots 33 years ago that talked with the engineers and flew the simulators that had different control laws and such. In our primitive system, we had a very clear "warning" that we were in "standby gains". If we were at 400 or 500 knots CAS, then the thing was sensitive! Once below 350 knots or so, the gains felt good.

Most of us would have liked a blend of AoA and gee command, with pitch rates to prevent overshoots. But the AoA sensors were less reliable if our speed got too slow. Nevertheless, above 120 knots CAS or so there was no problem.

In normal gains, we only saw a significant dampening of pitch rates when AoA was approaching the max limit. At low AoA we could get unbelieveable rates until we got to 20 degrees or so of AoA, regardless of the gee command. The so-called "bat turn" at 9 gees. Remember that we could maintain 25 degrees per second of sustained turn rate. The "corner" was at about 360 knots CAS, with 9 gees available and AoA of about 15 degrees.

Again, than you OG for a great explanation of the gains.

Hang on, I never said it was a strict quote (hence my use of the "+/-"), but my understanding of your PoV (which I understand & respect). Something that bring confusion is dangerous in a cockpit.
I may differ of your PoV in a broader picture:
- I'm not sure Alt laws bring that much confusion
- I'm not sure Direct law is always the best fail mode (in AF447, it may have been, as you very well demonstrated; but this is one accident, what about other events/situations?)

I don't know if you read french? I do, and that's simply my reading of the CVR transcript around 2 h 13 min 30.

You said "IMO there was not any panic on board".
I agree that no mention of "shouts" or breakdowns are on the CVR. I agree that the crew tried until the last second to do its job.
The fact that they "tried everything", without much logic, without agreement between crew members, with the serious concern of altitude passing FL100, with the PF & PNF "fighting" for the controls, with rambling, chopped talks... is IMO symptomatic of at last some kind of panicked state. Or should I say deeply concerned?

The laws and protections on the Airbus are very much dependant on the reliability of the numerous probes and sensors. Airbus has opted to keep as much as automation as possible depending of the circumstances. My proposition would be different : As soon there is a doubt on any data, first advise, second degrade to the basic.
That way, any Airbus simulator training would invariably lead to direct law with manual trim in case of manual flight and the crews would be used to that known and unambiguous state.

In Perpignan, the only reliable AoA data was silently discarded by the system, the crew proceeded with the test at an unreasonable low altitude, but knowing what we know now on AF447, nothing positively indicates that the test would have ended differently if the crew had tested at the minimum recommended altitude …

The simple and logic thing would have been to merely advise the crew of the loss of certainty regarding the data : AOA DISCREPANCY
My bet is that the crew would not have proceeded with the test at all as the test was specifically based on the accuracy of the AoA readings.
And why not reverse to direct law with still the capability for the AP ?
Many airliners fly around that way, is it a problem ?

Lets keep the magic of the Airbus when everything is known at 100% not lower.

The BEA transcript is not :
Ça ne servira à rien

The BEA transcript is :
(* fera rien)
on (fera /verra) rien

with the BEA notes :
( ) Words or group of words whose meaning has not been identified with certainty. The “ / “ symbol gives various proposals.
(*) Words or groups of words not understood

Conf, I am not sure I agree with you on the all or nothing point you are making.

Let me compare what Airbus does with the cascading control laws when various inputs go wrong to what Sikorsky does with the levels of flight control automation for a Seahawk helicopter. (Note: Not FBW in a general sense).

When all of the buttons are green, you have AFCS which includes the trim, SAS and boost features. You can disable AFCS (which allows for a variety of "hold" autopilot features, like altitude hold, airspeed hold, heading hold, etcetera) and keep the SAS, boost, and trim. You can also slowly but surely disable or lose those servos and their features as well, and end up with a "boost off" flying mode. This is a challenging but flyable mode. You are your own SAS system, and your touch on the stick (cyclic) has to be very light, even though your feet (rudders) and left hand are working very hard. (Aside: "boost off" is a slight misnomer in the Seahawk, since there is hydraulic power still driving the gross inputs into the rotor system). The inputs into the flight controls that feed the rotor system controls have a boost piston that can be diabled if there are problems.

Given a choice, I would prefer not to go from "all systems on" to "boost off" in instrument condiditions where a chance of keeping a few features, depending upon failure mode, could be worked in.

Granted, not the same kind of aircraft.

I totally agree with the first item. :ok: This one is lacking (or at last was on june 1st, 2009). :=
For the second one, I still am not sure (I insist: I'm not saying "alt law is good", nor "alt law should be discarted"). There are pro and cons (e.g.: Lonewolf_50's answer). There are different situations/failures/combinaisons of those.
While I understand your reserves (in AF447's scenario), Airbus people chose to make an Alt law, they are better informed than I'm. Why is interesting. But I, as an outsider, prefer to restrain from an all-or-nothing position.
In french: Il ne faut pas jeter le bébé avec l'eau du bain.


Yes, I'm aware of that. Once again, that's simply my reading (my understanding/guess) of the meaning of the captain's words.
Do you have another (better?) explanation of the meaning of those words? As for myself, I see no other possible meaning of these words. It's not (and cannot be) a proven fact, as he's no more here to explain us what he meant.

Agreed.

This doesn't sound right (bolding), but I may not understand what you're getting at.

Amount of elevator deflection for a given SS longitudinal input is essentially the same for 330K in Normal Law, 330K in ALT2(B), and, for example, 200K in ALT2(B), an abnormally low speed for the clean configuration.

At 200K in Normal Law, elevator deflection for the above same SS input is about twice that of the ALT2 deflection.


As a result, I would think you would not ultimately achieve the same peak G value in ALT2 as in Normal, ALT2 value would continuously be lower over the period of the input and never catch up. Holding a less deflected elevator longer isn't going to eventually increase G....and the aircraft response difference would simply be one of the further reduced pitch rate for that speed. Longer time for a given FPA change, which is what a pilot would sense even though the SS 'spring' feel might be associated with different expectations.

My point is couldn't that paragraph in the report just refer to this directly? In the sense that the aircraft is even 'more sluggish' than 'normal sluggishness' at these speeds, I guess truly that is an 'unusual' response dynamic. But given that one is not normally at these speeds clean anyway, I would imagine a pilot assessment of 'unusual' behavior would be a bit superfluous. :}

Good point Okie.

The thing about a smaller and less rapid rate of elevator when on the backup gains is that HAL will keep trying to achieve the commanded gee. So after a minute of back stick, the THS starts trying to help, and as the smash runs out both elevator and THS keep moving to get to the gee command.

We dinosaurs didn't quite understand this aspect of our FBW control laws until we had line pilots get into the "deep stall" ( not what AF447 had). Same scenario...... relatively steep pitch and losing energy real fast. Elevators are trying to make the jet get to one gee ( corrected for attitude). If the time constants and such are not aggressive, you wind up with a high AoA that the control surfaces cannot handle.

So with the "standby gains" active, and below the default value with everything smoothly running, you are further screwed. The control surfaces don't move as fast or as far to keep you away from the stall AoA values and such.

To all here: I fully realize that the AF447 scenario will likely NEVER BE REPEATED. However, it should serve as a very serious lesson to all the "professional pilots" here that fly planes with FBW systems.

@OK465

I'm not sure if I can explain it easily, but essentially what I am saying is that the time to achieve the desired level of steady state 'g' will not be radically different whether the C* gains are high speed or low speed values. What will be different is the time history of how you get to that state.

With C* operative you cannot simply relate SS movement and elevator deflection in any dynamic sense. What is true is that for any given set of weight/CG/speed conditions the SS movement, elevator deflection and steady state load factor are uniquely related and independent of whether it is normal or alternate law in operation. In other words the steady state elevator angle/g required is independent of C*.

What does change with C* law changes is how the elevators are moved between initial and final states. Suppose one has an aft CG condition where the steady state elevator angle/g is only say 2 degrees/g and you want an increment of 0.25g (these are just for instance numbers OK?). If you just apply 0.5 degree elevator you will eventually arrive at the new trimmed state, but 0.5 degree elevator isn't going to set the world alight in terms of pitch acceleration, so it will take a long time to get there.

To avoid this C* does what I think pilots would do instinctively in these circumstances - overdrive the elevators to get the aircraft moving and then back off to avoid any excessive overswing [correct me if I am wrong] and let the aircraft come gently to the final trimmed state.

In normal law, or alternate law with standard gains the aircraft response would be a fairly rapid g response, followed by a modest overswing and a damped recovery to the desired steady state g. I suggested that this whole process would take about 6 seconds, but that was a notional value and could be anything form 4 to 7 secs depending on aircraft.

In Alt2B with the default gains the elevator overdrive would be less, the initial pitch acceleration would follow that and there would be no overswing, just a gradual build up of 'g' to the final value. From reports I have read where C* gains have been varied (on large aircraft) this final state is arrived at in about the same time as with the 'normal' laws. The pilot's perception of the dynamic response however would be very different.

In this respect I agree that the dynamic elevator deflection in normal law could be twice what it is in Alt2B.

For the reasons given above I can't agree that the 'g' in Alt2B will never catch up or that holding a less deflected elevator longer won't eventually increase 'g'. There would be a slightly longer time for a given FPA change, but (question) does the pilot "see" FPA change first as a 'g'?

Agreed

@ gums

Gums. maybe it is just me, but when you write this I keep on thinking that you are inferring that the aircraft is not actually flown to 1g.

If you consider the whole range of genuine 1g level flight conditions the aircraft attitude could be anything from (say) 2 degrees to 6 degrees. An accelerometer fixed to the airframe could read between 0.9994 and 0.9945. If you are going to use that accelerometer to maintain 1g level flight then you must adjust the accelerometer output to correct for those attitudes.

Is this what you meant?, because if so the reference to attitude correction is superfluous as the aircraft is actually flown to 1g period.

The one gee command
 

Spot on, OG, that is exactly what I am saying based upon the documents provided by several here.

So if I am at 30 degrees of pitch, the actual Airbus command would be 0.87 gee. This is different than the Viper control law. Ours commanded "absolute" gee according to how we had the jet trimmed ( - one plus a bit or so, and + 3.5 gees). We had no bias for pitch attitude, so at an extreme pitch attitude, the jet would try to pull up to maintain the trimmed gee, which was mostly 1 gee.

The result of this control law was if you released the side stick the system would keep cranking in elevator to reach the one gee "command" that most of us used. The Airbus doesn't allow a gee command to be manually trimmed, and it looks to me that it always tries to achieve one gee corrected for pitch attitude. This also plays on the apparent speed stability, as the system has zero stick pressure changes to compensate for AoA changes due to speed changes - the basic "feeling" most of us had with the conventional systems and the original systems in the Chipmunk, Cub, Luscombe, et al.

Just some thots from an old FBW veteran.

Thank you OG & gums.

I wish you guys had taught my initial 'flight control' ground school, instead of CBT. :)

:ok:

(BTW: those elevator deflections were compared at the same weight/CG)

Lonewolf50

Thanks for the analogy you bring forward.
If I get you right, what you describe on the Seahawk helicopter is related to hydraulic issues and other servos malfunctions. An Airbus or any other more conventional aircraft is as much vulnerable to hydraulic issues. A dual HYD failure would trigger alternate law with fewer usable flight control surfaces and to switch straight to direct law for such malfunction is not the purpose of my proposition.

What I am after is everything related to probes and sensors, on which the Airbus is so dependable for its normal operation – This includes the law in force, the THS operation, the protection availability …
It is for that type of malfunctions known as unreliable data, that I would suggest a healthy degradation.

The guys are talking about ALT2, C*, one G command and so on, it is all very much interesting to try to understand what’s happening behind these words, and we all take such opportunity to educate ourselves, but in my book, direct law is the most understandable thing for a pilot who has to act now. For a given displacement of the stick there is a proportional displacement of the elevators. If not gentle enough the load factor will remind the pilot how fast his aircraft still is. If a large stick displacement don’t produce much, his aircraft is probably a lot slow.

Hi gums,

That is just my point! The system tries to achieve one gee. The pitch attitude input to the system is only to account for the fact that in 1g level flight the accelerometer doesn't read 1.0 and that the difference depends on attitude.

Or maybe its me again :8

Which Nz accelerometer
 

The Nz command I am talking about would be that from a "strapdown" style sensor referenced to the aircraft, not a space-stabilized inertial sensor such as used for navigation and weapon delivery.

The difference between my FBW system of old and that in the 'bus is we did not correct for attitude WRT the Earth coordinate system. Ours was and is still body-oriented for the Nz command. Because the 'bus corrects for attitude, it appears to command an attitude. At small climb or descent angles, it's a small correction. Ours seemed the same when close to cruise attitudes, but we flew at extreme pitch attitudes, so our "one gee" command could help us get into the "deep stall" if we relaxed pressure on the stick when zooming up at 70 or 80 degrees. I don't see this as a factor with AF447.

The point I keep trying to make is that the 'bus appears to command an attitude, but it's a gee/rate command with bias for pitch attitude.

My second point is "hands off" and even at small pitch attitudes like 10 or 15 degrees, that trying to maintain a constant Nz will cause the elevator to trim nose up, and then the THS. Because the 'bus has more drag than our Viper, speed/energy decays quickly, and it is possible to reach a stall AoA even with "hands off" ( as BEA report asserts). With no AoA protections, you're on your own. Holding the stick back for minutes doesn't help at all.

Lastly, the "feel" of non-FBW planes with positive static stability is helpful, as trying to fly slower than the trimmed speed/AoA requires more back stick. Trying to fly faster requires forward stick pressure/movement. OTOH, most FBW systems provide "neutral speed stability" regardless of the AoA, cee gee and/or static stability margins. So we don't get that feedback like in the old days, even with fully hydraulic control valves and no artificial "feel" such as the B777 has.

Pretty good discussion on this aspect of the control laws, way I see it.

OK, A330 has a strapdown sensor so we are talking same language

That is an important difference, but I don't see how that implies that the 'bus commands an attitude. I see the pitch correction as a technical requirement brought about by the laws of physics, not a deliberate attempt to manipulate flight path. [Which is why I think that the phrase "tries to achieve one gee corrected for pitch attitude" might give the wrong impression].

This is the point I don't get. The AI system, as I understand it, seeks to maintain a commanded gee in earth axes not body axes as in your Viper. With that assumption the pitch attitude term is a necessary correction feature but not a command. The basic system is a simple gee demand with gee/pitch rate feedback arranged to optimise transient response throughout the envelope. [For completeness, the THS becomes the integral part of a standard P/I control system]

Agreed, except to remark that in A330 terms 15 degrees is not a "small" attitude. One needs to add a proviso though - the speed divergence only kicks in when the (thrust - drag) vs speed curve slope goes negative and even then the initial rate of divergence would be very low, although it gets more exciting as one nears stall conditions. The argument that the A330 system could take the aircraft into stall even "hands off" is a valid one, but really only becomes significant (in AF447 terms) because the aircraft was put in the vicinity of stall in the first place.

Agreed

Yup!

You mean in level flight?

@HN39

Yes of course - any gee puts the CL and Cd up and makes deceleration more rapid.

CONFiture, #182
I tend to agree with that, though perhaps not to that extreme. However, it would have the desirable effect of forcing airlines to put more emphasis on manual flying skills, which seem to have become neglected, fwics.

Simplify, simplify, where is my razor ?...

Owain Glydwr, #195
That sounds like a serious fault in the design, in that under such circumstances, the effect is a positive feedback loop that might be unexpected, not trained for and thus very difficult to recover from.

No info as to where it would get it's acceleration / g info from, but would assume the INS, which has a triad of accelerometers. It wouldn't be enough to just take the G (vertical) value alone, as the value would affected by both pitch and roll. The assumption would be that they are using the processed outputs, via a bit of trig etc, to resolve earth axis vertical from 2 of the 3 accelerometers.

Just edited to add some content, rather than the usual hobby horse criticism about fragile at the edges, which i'm still uncomfortable with :-)...

Keep it up - some of the best techlog/af447 stuff i've seen for months...

Yes, excellent discussions!

Owain Glyndwr, twenty years ago when checking out on the A320, one of the instructors observed that because the system maintains 1g, that it would gradually increase elevator deflection in the climb to cruise altitudes (where gravity is very slightly less, was the claim...), to maintain 1g and that a tiny ND input was required during the climb. I did a lot of thinking about the implications but never found a clarification. It seemed logical enough but as a pilot I nevertheless doubted that the statement was true. Is there any clarification that might put this notion to rest? Thanks...