Friends have been asking me to comment on the latest news regarding the June 1, 2009 crash of Air France 447, e.g., this CNN article. I dug up my June 9, 2009 posting on the subject (pure speculation at the time) and, as the black box story unfolds, would appreciate readers grading my success.
28 thoughts on “Revisiting my conjecture about Air France 447”
Comments are closed.
Sadly, it’s so long that I flew in anything but my taxi, ….
And I’ve never flown anything more complicated than a 172. But your theory was more of overspeed, my understanding is this accident is seen more as stall and underspeed.
But I am not qualified to question you given how long ago it was since I flew at all. But let me ask you this question about your theory and this accident.
I understand they lost speed indications. But they didn’t lose gyros, attitude indicator, and artificial horizon. I find it interesting that having stalled, having fallen out of 35,000 feet, that they couldn’t recover based on the attitude indicator.
If I recall unusual attitude recovery under a hood, it was to trust the attitude indicator. I understand that attitude indicator and a bad airspeed indicator is still very dangerous at altitude and at a high airspeed. And what do I know anyway about it.
That said, I’ve also wondered for a few years why we have to rely on pressure driven airspeed indicators alone. I have wondered if a laser couldn’t be used to directly measure the speed of air over the wing by measuring the speed of the particles caught in a laser beam.
Or even worse comes to worse approximating air speed with GPS and accelerometer derived absolute speed.
I would think I could constantly measure pitot determined airspeed, compare it to GPS and accelerometer derived absolute speed, and use the trendline of that comparison to determine when pitot speed has failed.
This is part of my daydreaming of how an iPad or similar tablet might be used to replace the conventional systems in a next generation suite.
(But could be, probably is, just non-physical unbounded thought salad.)
Can airspeed be measured by measuring the doppler shift of a fixed frequency tone (and correcting with altitude, temperature, … information)?
Your earlier speculation seems basically accurate.
It’s worth going beyond the rather superficial CNN article and taking a look at the original report (English translation at http://www.avweb.com/other/447may27report.pdf ), which contains some chilling observations such as “The angle of attack, when it was valid, always remained above 35 degrees.” (Though this is apparently not displayed directly for crew consumption — and because the aircraft entered alternate law presumably due to the airspeed sensor discrepancies, normal stall-protection measures were disabled.)
Well, well, well, dr. Greenspun, looks like ur 2009 post on the subject was spot on given the preliminary revelation by the investigators so far. Good job. I guess u might not have been so eager for evaluation by us unwashed masses if u had been totally off, ;-).
Here’s a comment about the AP disengaging when the redundant avionic systems disagree: it seems like a cop out and a death warrant to me. When do such systems fail? When conditions are so bad that no human can expect to know what’s going on. If humans can have a visual reference in flight, relying on the senses to orient them for control input is guaranteed disaster! Recall JFK jr. There has to be a better response than for the avionics to throw in the towel when it can’t make sense of the data, don’t you think?
Jerry: They could have flown, obviously, purely based on the attitude indicator and a constant power setting (nose-up about 2 degrees and thrust levers in a normal place). But recovering a high performance wing from a deep stall is not necessarily possible and there is really no training for it at airlines, as far as I know. A single-engine plane has to have demonstrated stall recoverability, but a multi-engine plane need not be demonstrated as such to the FAA.
JT: I think your autopilot question is a good one. Even competent instrument pilots are often disoriented and in trouble if there is a sudden and unexpected encounter with instrument conditions (e.g., they were planning to fly somewhere visually and entered the clouds inadvertently). At the very least, when the autopilot trips off it could give a situation report to the pilot taking over, e.g., “before my sensors failed, I was flying happily at 3 degrees nose up and 90 percent N1 [power setting] and it seems as though you should refer to the backup attitude and airspeed indicators.”.
Looks like the general agreement was unusual attitude leading to an overspeed at the time. Now it seems even more elementary. The pilots stalled the aircraft in IMC after the autopilot disengaged, and never even identified the stall in order to correct it.
http://www.avweb.com/avwebflash/news/air_france_447_investigators_stall_crash_204730-1.html
Some quotes: “Data shows that the pilot flying held the sidestick at the full left and nose-up stops for the entirety of one 30-second span.”
I recall a similar finding during the Colgan Air crash in Buffalo. The aircraft stalled, and the captain pulled entirely back on the stick. At 2000 feet, the aircraft stalled and spun, and recovery at that altitude was impossible regardless of pilot inputs, it seemed.
However, here the stall occurred at 38000 feet. Even at the 11,000 ft/min descent rate that the aircraft experienced, there was still three minutes for the pilots to act.
Do we need to require “simple” recurrent training for all professional pilots? That is, every 6 months or so, every 747 captain out there needs to strap into a Cessna 172 and do some stalls, spins, unusual attitudes, etc.? I specifically mean outside the simulator, which is a sterile, comfortable environment. Get them in a small airplane with a new instructor and be outside their normal, professional comfort zone.
I am not convinced of the value of unusual attitude recovery training for airline pilots. Doing it during day VFR in a 172 has almost nothing in common with doing it in a A330 at night in a thunderstorm.
In a documentary on the AF447 crash about a year ago, they put a test pilot into fixed-base sim and made it behave aerodynamically like a large airliner. He was barely able to do it and that was in day VFR. The test pilot in questions flies jet fighters to over the edge of their envelope and then works out how to recover them in has day job…
This also isn’t the first time that a flight crew decided to ride a stalled jet all the way to the ground with the controls all the way back. Sometimes with and sometimes without pitot/static in tact. Sometimes a FBW Airbus, sometimes a Boeing jet or an Dash-8 turboprop. Blaming it on the computer error message overloading the pilots is irrelevant, it happens in all aircraft from dumbest to smartest.
I vote for *more* automation in these cases. I reckon a smart computer programmer is a far better person to trust the controls of a plane to when flying without pitot/static system or compromised aerodynamics.
Phil: You surmised “the pilots failed to meet the challenge [of unusual attitude recovery] and their control inputs were not helpful in stabilizing the airplane.” Whether ultimately it was an overspeed or a stall that led to loss of the aircraft is a mere detail. They were hopelessly confused and were unable to recover the aircraft given the conditions they were in, despite having a significant amount of time within which to do so.
I think you did pretty well.
Why is it not possible to recover an airline jet when it has stalled, or what are the conditions when this is not possible? If a stall occurs at say 30’000 feet, why is it not possible to save the airline when it has descended to 10’000 feet?
Michael: It is a design trade-off. The easiest planes to recover from a stall or spin are the slowest and have the most inefficient wings, e.g., a Cessna 172 can almost always be recovered merely by the pilot taking his or her hands off the controls and letting the airplane fly itself.
Due to the fact that, even if both wings are stalled, one will tend to be generating more lift than the other, a prolonged stall can turn into a spin and recovery from a spin is not always possible (see http://en.wikipedia.org/wiki/Spin_(flight) for a reasonably good explanation).
From reading the English translation of the interim report, it is unclear whether the airplane entered a spin or not. The fact that the airspeeds were very low towards the end (when presumably the pitot tube would no longer have been iced over) implies that the plane was spinning. Also the very high descent rate of 11,000 feet per minute.
From the report it is hard to understand why the airplane was so difficult to recover. They were flying along at 2.5 degrees nose up and everything was good. Then the airspeed indicators went down to a crazy low number (60 knots indicated) and the pilots responded to that by yanking back on the stick (if they truly believed that they were at 60 knots, the correct input would be stick forward to push the nose down).
One significant bit of data from the report “The angle of attack is the angle between the airflow and longitudinal axis of the airplane. This information is not presented to pilots.” Angle of attack (between the wing and the relative wind) is the best information about whether the airplane is still flying. Military fighters have angle of attack indicators as do some kit planes and the new Icon A5. The AOA information was available to the autopilot, so it could have kept flying the airplane by holding its old attitude and maintaining a reasonable AOA. But instead it tripped off suddenly and surprised the pilots.
Bas,
That’s a fair point about unusual attitudes not being the same, but I think it may still have some value in instilling “basic pilot stuff” instincts. This is the second ATP in a few years to respond to a stall situation by yanking back on the stick, as I pointed out. It may be that just strapping into the Cessna will get some of those basic instincts, like “nose down during a stall” back into the short-term memory.
Joshua: As my friend Dirk, who has a couple of jet type ratings says, “pilots are notoriously stupid”.
All I know about airliners is that these are always too small to offer reasonable space for my not-so-tiny body.
So, sorry if this sound stupid, but why is a plane’s hight, location, speed, orientation not also computed from (several) GPS devices data?
These values would be more reliable than physical sensors, I guess. Am I wrong?
Reinhard: GPS data is useful, but not the most critical information for aerodynamics. For example, suppose that you’re flying a slow airplane at 80 knots of airspeed. You’re unfortunate enough to have encountered an 80-knot headwind. The GPS will accurately report a ground speed of 0 knots. A 0-knot airspeed would be cause for serious concern due to the impossibility of generating any lift at that speed. But the wing doesn’t know that it isn’t moving over the ground, so the aerodynamics continue to function normally.
I’ve had that thought about GPS as a backup source of primary flight information in an aircraft that had a truly integrated FMS, like the G1000. You could use GPS to get a ground speed and an altitude, plus XM or some other data link for weather to get wind speed. Using ground speed plus wind speed aloft, you could calculate an estimate of true airspeed, which, combined with altitude, could give a rough estimate of indicated airspeed, which is the aerodynamically useful figure.
I guess such an estimate would be both too slow to update and too inaccurate to really use to control the aircraft though.
Joshua: What are you going to use for “wind speed aloft” in the middle of the Atlantic ocean? On the G1000, the little wind vector exists as a consequence of a functioning and accurate airspeed sensor (plus the GPS ground speed).
GPS ground speed is most useful to a pilot when conducting an approach on a windy/gusty day with a lot of wind shear. If the pilot takes the normal approach airspeed and flies that fast over the ground, by the time the airplane arrives at the runway it will almost surely still have enough airspeed (since wind tends to diminish in intensity closer towards the ground). For example, if Vref is 100 knots and at 1000′ there is a 30-knot headwind, the pilot simply flies 130 knots indicated airspeed, which will be 100 knots over the ground. If the wind sudden drops to 0 at the runway threshold, the airplane’s inertia will ensure an airspeed of roughly 100 knots, which would be the desired number in calm wind conditions.
Phil,
Fair point. I’m not sure what data is used to compute winds aloft forecasts over terrain, and the accuracy of those is probably not going to be good enough anyhow.
Maybe, in the future, if there are enough ADS-B aircraft flying with functioning pitot tubes and GPS/IRU FMSs, approximate wind vectors can be calculated sent to nearby aircraft should one of them detect a failing pitot tube.
Given the pace of software development in aviation, perhaps a feature like that could be implemented by 2150 or so.
Besides your conjecture about AF447, this is also a good time to review your article about foreign airline safety (blog post too). The root cause in this fatality is not the failed instruments but rather the poor training of the pilots. It would be interesting to see the pilot history of the pilot actually at the controls (PF) who held the aircraft in an unusual attitude of 35 degrees nose up exacerbating the situation. Did he follow a simulator based professional crew syllabus with little to no realistic unusual attitude recoveries? As far back as 2002, the FAA recommended this: Selected Event Training is voluntary flight training in hazardous in-flight situations that are not specifically identified in FAA regulations or directives. Examples include: false stall warning (stick shaker) at rotation, full stalls, excessive roll attitudes (in excess of 90 degrees), high pitch attitudes (in excess of 35 degrees)
I haven’t found the EASA or Air France training standards on unusual attitudes, but I bet it doesn’t cover high pitch attitudes. I hope to proven wrong.
Josh, I understand your point, but the flight dynamics of a huge jet at altitude require different recovery techniques than a C172. The basic point is valid – there was insufficient training in this circumstance. Hopefully the flight standards will evolve to provide this to pilots worldwide, in a rigorous way.
You were pretty close, but I think your initial article understates the effect of weather. AF447 plowed into a pretty serious line of thunderstorms at a time when other flights were making significant deviations to avoid this line. The ride was probably not just ‘bumpy’, but downright violent.
Most significant is that the Airbus went into ‘Alternate Law’ mode ( http://www.airbusdriver.net/airbus_fltlaws.htm ), which basically makes it a regular hand flown airplane.
In the ‘normal law’ mode the action of pulling and holding the sidestick back and to the left would have put the plane into a nose high left turn. The flight computers would have prevented a stall.
These protections disappear in alternate law. The pilot can override the computers and stall the airplane. Which seems to be what happened.
The Airbus must have pretty stable stall characteristics since the airplane hit the water intact.
(Unrelated note: I got a flight in a new air conditioned R44 this weekend. The ASOS was reporting 41C. The very well designed cabin cooled very quickly, the visibility was amazing, and the aircraft was great fun to fly. My bottom line: ‘I gots to git me one of deese!’)
how come the newspapers quote the airlines as saying the pilots acted professionally and that they did their job well? and how come the pilot didnt jump right in from his resting spot and take charge? Cd a really good pilot like that Cpt. Solly have saved the plane? It seems like the two pilots who were in charge got thoroughly confused. Is there a good way to train for this situation???
Elsa: Could Captain Sullenberger and First Officer Skiles (the US Air crew) have “saved” the plane? If the initial accounts are to be believed, a passenger with no flight training would have done better than the Air France crew. The passenger probably would have been reluctant to yank on the flight controls and would have let the airplane continue to fly at its trimmed pitch attitude. The passenger also would have left the thrust levers where they were, rather than pulling them back to idle. Airplanes aren’t that stable in roll (turning), but beginners are pretty good at recognizing when the plane isn’t level and rolling it back. Doing nothing would have been far better than taking the actions that the Air France crew took. So far, from the report, it doesn’t appear that the airplane needing “saving” until AFTER the Air France pilots pulled it up into a stall.
There are definitely ways to train for this, but airlines and the regulatory authorities are obsessed with training pilots to handle engines that fail right as the plane is rolling down the runway about to lift off. This almost never happens, especially with modern jet engines, but a tremendous amount of training is devoted to this one scenario (admittedly challenging).
All of the flight and simulator training that I’ve had has been very unrealistic regarding unusual attitudes. The instructor or examiner has you close your eyes. They put the airplane into a steep climb and bank and then say “open your eyes”. So you are fully prepared to concentrate your attention on the attitude indicator and then put in appropriate control inputs. In the real world you’re concentrating elsewhere, e.g., reading charts or leaning back to deal with a passenger, thinking that the autopilot is managing everything properly. When an alert comes, perhaps from the stall warning, you have no idea what led up to it (consider the Colgan crash in Buffalo a couple of years ago). That is much more difficult than closing your eyes and knowing that someone else is intentionally disturbing the airplane.
Hi Phil,
After watching the “Air Crash Investigation” series on the Discovery Channel, I noticed that there were many cases whereby pilots made incorrect decisions based on faulty instrument readings because they didn’t know or didn’t have time to find out what happened to the plane’s external structures, e.g. tail is missing, engine fire… etc. (Well, this might apply to the space shuttle Columbia as well)
Why can’t the airplane manufacturers install camera around a plane so that the pilot can SEE what’s happening to the airplane body?
I’m interested to know if there are any reasons why this is not useful?
Thanks.
Philip: Most recent accidents have involved pilots who were presented with too much data, rather than too little. In most cases, only one instrument needs to be looked at, i.e., the attitude indicator. If the plane is more or less level and at a reasonable power setting, it will fly safely (if not always in the desired direction or at the most efficient airspeed).
As far as engine fires are concerned, there are redundant fire sensing systems inside each turbine engine on an airliner and there is a big red fire light and bell that sounds in the cockpit.
Philip, even with all the instrumentation to tell them otherwise, pilots can still be stupid/untrained/ignorant enough to shut down the one engine that’s functioning normally:
http://en.wikipedia.org/wiki/Kegworth_air_disaster
And of course the passengers and cabin crew that knew the pilots shut down the wrong engine were too sheepish to to go bang on the cockpit door and tell them what idiots they were…
Philip, I wonder if you’re understating just how dire the situation was in the cockpit.
http://www.mye28.com/viewtopic.php?t=64381&postdays=0&postorder=asc&start=25
Check the post by TSMacNeil, a Boeing and Airbus pilot, on 5/24 at 11:43. What do you make of that? To me, this looks like a plane problem that could have been averted with a handheld GPS. They didn’t have usable redundant instruments.
David (above): “They didn’t have usable redundant instruments”? They didn’t need redundant instruments since their most important primary instruments were working perfectly. They had at least three working attitude indicators in the airplane. All that they needed to do was keep the airplane at about 2 degrees nose up and with whatever thrust setting is normal for that altitude. Were the airspeed indicators showing nonsensically slow data? Yes. But the pilots’ reaction to that information was opposite to the correct reaction, assuming that they believed it.
When in doubt, return the airplane to its “happy place” of slightly nose up with normal thrust. That’s all that they needed to know and do, though of course this is much easier to remember in a sim when you know that something bizarre is going to happen.
MacNeil suggests they didn’t have attitude data. I find it difficult to believe that three trained pilots would miss so simple a solution, were it available.
http://www.bea.aero/fr/enquetes/vol.af.447/point.enquete.af447.27mai2011.en.pdf is the latest from the French investigators. It doesn’t say anything about attitude indicator failures, nor do any earlier investigations.
http://www.aerospace-technology.com/projects/a330/a3301.html shows an Airbus A330 with a mechanical attitude indicator screwed into the dashboard (just to the right of the captain’s multi-function display). This is typical for a jet and there isn’t connection between this spinning gyro and the fancy computers (though as noted above, the attitude information from the fancy computers was correct at all times; it doesn’t depend on airspeed sensors).
http://www.techreview.com/blog/editors/26821/?p1=Blogs and http://www.theatlantic.com/technology/archive/2011/05/air-france-447-what-the-black-box-tells-us/239598/ have some additional explanation.