[SeanB]

Airbus does not allow any switch to manual operation unless the triply redundant control system cannot agree, and declares itself failed. This removes all of the augmentation and limit controls built into the system. It seems a failing was having all the air data feeds coming from a single probe, it would have been better to have had 2 ( even if one was not ideally placed, but was able to provide some meaningful data) or more input sensors, 3 better, provided the input is provided to each component of the flight computer and each can vote as to what it considers a valid input. Still not a good idea to rely on this instead of having a proper rated heater, and even better a 2 stage heater and thermostat, that is capable of detecting potential icing conditions and switch to a high power deice mode for a short while ( hey, I should patent this, but now it is Public Domain) to clear the ice.

As to training for stall on real aircraft, remember that certain aircraft ( MD80 AFAIK, but any T tail will do) is quite capable of stalling and staying stalled in the "right" wrong stall area, as the control surfaces are then in a turbulent airflow, and have no or little effect. Better to train in a simulator where you can practice recovery using engine power, as these stalls often are going to require more altitude than what they have to recover.

[graham.d]

Sean, I had a look at the training manual and the situation they were in seems quite well covered. It says that the airspeed indication will be unreliable in a stall (I think there are three sensors) and actually recommends using GPS data as I suggested (I had not read this before). It has got a de-ice facility and has a procedure for getting out of a stall situation. I feel these must have been practiced if they are in the training manual but, in any case, they do not seem unusual.

Some aircraft are difficult in a stall as you say, and powering out is usually recommended. I am not sure whether that is necessary in this aircraft but thrust is recommended in any case.

[Geezer (OP)]

I noticed on that website there's is a bit of a rethink going on regarding stalls. It says that, traditionally, too much emphasis was put on maintaining altitude. Perhaps that explains why the junior pilot on 447 initially tried to pull the nose up.

It also strikes me that there is a lot of circular thinking going on. If simulators could not simulate a stall, how were the crew expected to be properly trained without stall training in a real aircraft? Seems to suggest a "well, we know that's never going to happen" mindset.

[graham.d]

Where did you see that simulators can't emulate a stall? I find that hard to believe as it would not seem difficult (relatively speaking). When the aircraft stalled it was at considerable altitude so fairly safe from a ground impact. Well one would have thought so! I think actually that, for this aircraft, the recommendation is actually to increase thrust, level the wings and try to achieve a slight nose up attitude (this from memory of the lengthy training manual but have not checked). It looked like the manual did cover this situation but that the pilot did not follow procedure.

[rosy]

Seems to suggest a "well, we know that's never going to happen" mindset.

Well, maybe.

To do live training they'd presumably have to genuinely stall a genuine plane. Probably repeatedly. The risk of accidental stalling is clearly a real one, and a stall of a loaded passanger plane which leads to a crash will likely lead to the death of several hundred people... but that doesn't necessarily mean that live training is the right thing to do.

If stalls in modern, highly automated planes are very rare, if for example there is a reasonable expectation that most pilots wouldn't encounter a stall in their career otherwise (and I have no idea what the stats are, but it must be something for which stats exist), more live training would not necessarily be expected to save more lives in the long run.. because if stall training killed a significant number of crews that would add up! (Leaving aside the fact that if pilots dropped several planes out of the sky in training, it might obviate the need for training entirely.. because it would put the public right off those models, I'd think!)

[SeanB]

Most stalls occur in the region in which recovery is impossible - right near the ground, generally with some bad outcomes. Mostly too slow, to low and too little power. Most dangerous part of flying is take off and landing, and the part in between is generally pretty quiet, may be a little turbulent, but weather is worse nearer the ground and nearer the top of the flight envelope.

The AF plane was unfortunately operating at the edge, and the accident pushed it into a region that was supposed to never be entered as being a known danger. A combination of poor route selection and a known flaw in the plane design, which in the end resulted in the crash.

[Geezer (OP)]

Where did you see that simulators can't emulate a stall?

I thought it was on that site, but I can't find now it either! I'll try to retrace my steps and dig it up. I was quite surprised when I read it too.

I'm pretty sure it's easy enough to simulate the stall. I got the impression that what was lacking was the simulators' ability to handle the recovery accurately.

[Geezer (OP)]

Seems to suggest a "well, we know that's never going to happen" mindset.

Well, maybe.

To do live training they'd presumably have to genuinely stall a genuine plane. Probably repeatedly. The risk of accidental stalling is clearly a real one, and a stall of a loaded passanger plane which leads to a crash will likely lead to the death of several hundred people... but that doesn't necessarily mean that live training is the right thing to do.

If stalls in modern, highly automated planes are very rare, if for example there is a reasonable expectation that most pilots wouldn't encounter a stall in their career otherwise (and I have no idea what the stats are, but it must be something for which stats exist), more live training would not necessarily be expected to save more lives in the long run.. because if stall training killed a significant number of crews that would add up! (Leaving aside the fact that if pilots dropped several planes out of the sky in training, it might obviate the need for training entirely.. because it would put the public right off those models, I'd think!)

There does not seem to be much real stall training at all. I see some people are suggesting that there ought to at least be live training in trainer aircraft. (I can understand why the airlines would not be too keen to run the risk of pranging one of their nice aircraft.)

Is it true that an airliner at cruising altitude has to operate within a narrow window of parameters to avoid stalling?

[Geezer (OP)]

Where did you see that simulators can't emulate a stall?

I thought it was on that site, but I can't find now it either! I'll try to retrace my steps and dig it up. I was quite surprised when I read it too.

I'm pretty sure it's easy enough to simulate the stall. I got the impression that what was lacking was the simulators' ability to handle the recovery accurately.

This is it, but it's not very clear how new the technology actually is.

"That’s because pilots are supposed to never allow a plane to reach a stall in the first place, and because earlier-generation simulators were not able to mimic a full stall.

But thanks to new simulator technology, “now you can put someone in a stall scenario and let them recover,” Babbitt said."

http://uprta.org/faa-to-seek-major-changes-in-pilot-training/

[Geezer (OP)]

Sorry if I'm becoming a bit obsessed with this one!

If I remember correctly, accurate airspeed data becomes more critical at greater altitudes (because of reduced air density). That being the case, there should be no circumstances under which the automatic systems, or the crew, ever get suspect airspeed data.

Why don't they calculate it based on what the engines are doing? Would it not be possible to estimate airspeed from air density, turbine speeds and fuel consumption, or maybe they already do that?

[graham.d]

The manual recommends (as I guessed at) using GPS data to get, at least, ground speed. The only problem with this is that at high altitude there can be a large difference between airspeed and ground speed because of the very high winds that can be encountered. In any case, if the stall warnings are going off then it is perfectly possible to increase the thrust and use the artificial horizon instruments to level the craft and to restore a safe flying attitude. It seems like the pilot did not believe the aircraft was in a stall but in a dive. It is a very fundamental error in any aircraft to believe your senses rather than the instruments. It is very basic training in light aircraft to NOT depend on your senses when recovering from a spin, for example, especially in low visibility situations.

Rosy, every pilot would have had some live training on a stall recovery in some aircraft though maybe not in every aircraft he is expected to fly. It should have been done in a simulator though but it looks like the early simulators did not cover this fully (from Geezer's post), which is very surprising, though the omission seems to have been corrected now. Every pilot should be aware of what to do in a stall though. It does look like a case of insufficient practice; the stall would have put the aircraft into an intermediate mode (alternate law, as opposed to normal law and direct law) between fully automatic and manual modes, as the feeds from the airspeed pitot tubes had become unreliable. I suspect he had too many things to think about - OK if you practice enough but not good if you have to look up what to do! We will probably never know exactly.

[Geezer (OP)]

It ain't just me  []

http://news.yahoo.com/ap-impact-automation-air-dulls-pilot-skill-070507795.html

[alancalverd]

Can't imagine why I didn't get stuck in to this one earlier! A couple of quick points:

Elementary stall training always takes the form of "nose down, power up",and every pilot will have done a fair bit of this in a trainer before progressing to first solo. The PA38 "Tomahawk" trainer was even designed to enter the deep stall and spin condition of a T-tail. Problem with the AF flight was failure to recognise the stall condition when flying on standard instruments, and possibly relying on autopilot altitude hold in turbulent conditions (this will stall most trainers). ALT HOLD is hardly modern automation, though.

The essence of all stalls is excessive angle of attack, hence AOA indicators on modern airliners - optimum lift/drag ratio (i.e. economic cruise) can occur quite close to the critical AOA.

But the problem with the 737MAX   was indeed excessive automation, coupling the faulty AOA probe directly to the stabiliser trim and not telling the pilot how to disengage it.

Success of the Hudson landing was nothing to do with automation but a lot to do with engaging the auxiliary power unit, thus maintaining electric and hydraulic services, and the fact that the pilot was a national gliding champion, with hundreds of "power off" landings to his name.

Not sure how far back other folks' memories go, but I have stalled some vintage simulators including an original US Navy Link  trainer and the last ever session in a Trident III. Very realistic, even into an incipient spin.

[paul cotter]

Quick question, alancalverd. How can you handle a deep stall if the elevators are effectively non functioning?

[evan_au]

One of the big problems with automation is called "sensor fusion"
- All critical aircraft systems are (at least) duplicated, so there are multiple possible sensor inputs
- Each input has its own random and systematic errors
- Knowing which inputs are valid, and which is invalid is not so obvious
- Some aircraft control systems solved this by just taking inputs from one of the redundant inputs (eg just one pitot tube or one angle-of-attack indicator). But if that single input fails, the control system won't be able to detect the fault, and will produce wrong instructions to control systems.

I heard of an air force fighter pilot who retired to become a commercial airline pilot. He described his new role as "hours of pure boredom, interspersed with seconds of sheer terror...".

[Bored chemist]

Looking at this  now, (I can't recall if I saw it earlier" I see the report into the incident said ". Both of these circumstances resulted from Alaska Airlines' attempts to cut costs".
(according to this)
https://en.wikipedia.org/wiki/Alaska_Airlines_Flight_261
If you have a big plane, with big control surfaces and big forces, then it's impossible for muscle power to move them (usefully).
Some level of automation becomes inevitable.
The automation itself wasn't the problem.
Cost-cutting was.

[alancalverd]

Quick question, alancalverd. How can you handle a deep stall if the elevators are effectively non functioning?

RTFM! A production aircraft will have been tested in all conceivable flight modes, load distribution limits will have been set, and recovery from "unusual attitudes" will be in the book (pace Boeing, perhaps).

If you haven't read the book, change something and see what happens. Full power may restore enough airflow to give elevator authority, or cut the power to idle and wait: if the center of gravity is within design limits the aircraft should eventually nose-down and start to fly again. In extremis, kick on full rudder and blip the throttle, which may initiate a spin and thus get one wing flying, then recover or abandon ship as per the flight manual.

Either way the condition is best avoided: all aircraft can bite if you mistreat them.

[alancalverd]

f you have a big plane, with big control surfaces and big forces, then it's impossible for muscle power to move them (usefully).
Some level of automation becomes inevitable.

There's a difference between automation (plane makes decisions or moves controls in an attempt to follow preset instructions - conventional autopilot or even autoland system) and servo assistance (power steering in one or more axes). But in either case most civil  aircraft can be "hand flown" with enough muscle power if there is no airframe damage. An independent trimmer circuit may allow slow changes if the primary hydraulics fail.

The difficult area is in "fly by wire" where the pilot's control input is not directly connected to the engines or control surfaces but interpreted and executed as efficiently as the airframe will allow, or an inherently unstable aircraft is maintained in a pilot-mandated condition by rapid servo action so it "feels" stable.  Problem with such systems is that the aircraft is not humanly flyable if you switch off the electronics. 

The weasel word is "airframe damage". The Alaska Airlines disaster seems to have resulted from a primary mechanical failure rather than an automation or servo fault.

[alancalverd]

I heard of an air force fighter pilot who retired to become a commercial airline pilot. He described his new role as "hours of pure boredom, interspersed with seconds of sheer terror...".

Much the same from a just-retired airline captain of my acquaintance, who described his career as 40 years of boredom and 20 minutes of panic, for which he was well paid. 

[evan_au]

How can you handle a deep stall?

The crash of a stalled commercial airline in Russia occurred because:
- The pilots knew that autopilot was on
- The pilots were not aware that autopilot is automatically disconnected if the human pilot makes a sudden change to the controls
- The stall would have been automatically recovered if the human pilot had just released the controls, allowing autopilot to resume
- Not helped by the fact that the pilot had one of his children at the controls...
30 minutes podcast:  https://timharford.com/2022/05/cautionary-tales-when-the-autopilot-switched-off/