Naked Science Forum
Non Life Sciences => Technology => Topic started by: syhprum on 24/01/2015 19:51:55
-
Analysis of radar sightings and black box recordings seem to indicate that the air Asia plane QZ8501 got into a stall condition that the pilots were not able to control after attempting a too steep climb to avoid turbulent weather.
Could a computer have done better ? could there be a switch for the pilots to throw saying "we can't fix it you have a go"!
The pilots could well be reluctant to throw such a switch could there be some mechanism to take the decision out of their hands?
-
A computer could probably already do a better job of recovering from that situation, but it could probably also ensure that the plane doesn't get into that situation in the first place. It may have been too late by the time they started climbing though, so ultimately the problem that most needs fixing is with air traffic control which refused to let them ascend when they first saw the need to do so. We need to get away from centralised air traffic control to a system where individual planes do most of the work instead, keeping track of all the planes around them and ensuring that they don't get too close together even if one suddenly needs to make emergency maneouvres through the space near to them.
-
Hmmm, interesting question.
Stalling an Airbus A320 in the middle of a storm isn't something that a lot of pilots do frequently. Perhaps it is like explaining to someone from central Africa how to handle a car on ice. So, with that in mind, perhaps a good autopilot would be better than a pilot. Obviously preventing the stall would have been better than going into a stall.
One of the issues, of course, was what happened to Air France 447 (http://en.wikipedia.org/wiki/Air_France_Flight_447#Final_report) in which the plane apparently had inconsistent sensor readings, so the pilots reacted improperly.
I suppose if a computer could determine that some sensors had malfunctioned, and could identify the faulty sensors, then it may be able to compensate, perhaps better than a pilot that is doing instrument-only flight. And one could presumably teach a computer not to stall an airplane.
One of the problems, of course, with replacing the pilots with computers is that there are times when one just needs a good pilot. Without practice, the pilots may not be prepared to react in an emergency.
Read the list of the Top 10 Emergency Landings (http://www.historynet.com/the-10-greatest-emergency-landings.htm), plus undoubtedly many others that didn't make the list, and one has to wonder if a computer would do better than a seasoned pilot. For example, would the computer have ejected when it lost the wing off of the F-15? Or, perhaps the acrobat pilot that flew (and nearly landed) his airplane upside-down to compensate for a broken wing spar.
Perhaps the difference between a pilot that can safely land an F-15 with one wing, and a pilot that stalls and crashes a presumably intact A320 is pilot training, and perhaps an innate sense of reasoning in an emergency.
-
Perhaps the lost Malaysian Airlines flight, and the recent Air Asia Crash are signs that our air traffic control system is now stressed to the max, and is ready for a major revamp. Of course, if one computerizes the traffic controllers, that might allow more planes in the air with closer tolerances... which is good until something goes wrong.
-
Ever since the F16, computers have increasingly controlled the flight of an aircraft, via Fly-by-Wire (http://en.wikipedia.org/wiki/General_Dynamics_F-16_Fighting_Falcon#Negative_stability_and_fly-by-wire) control systems.
With computer control, it is possible to modify the design of the plane away from one which is easy for a human to control, to one which is quieter, more maneuverable, faster or more fuel efficient (not all at the same time, unfortunately).
This can have good or bad effects:
- At an airshow, an early computerised commercial airliner was giving the crowd a spectacular view of the new airplane, with a long, slow pass over the runway. The computer gave several warnings about the plane's wheels being up, despite the low altitude; the fact that the plane was approaching stall speed, etc; the test pilots disabled the warnings. As I recall, the plane ploughed into the trees at the end of the runway, which did not leave a good impression of the new aircraft.
- The F16 has a high-power engine, and the computer software is designed so that the pilot can't tear the airframe apart in mid-air. Part of flight testing is to put the plane into various kinds of spins and stalls, and check that the plane recovers. One anecdote from the early F16 flight tests is that when the test pilot put the plane into a particular spin, the computer blocked him from getting it out of the spin on the basis that it would stress the airframe too much. The plane crashed. After that, they put a reset button on the computer during flight tests.
- An incident where an onboard collision avoidance system, communicating with nearby planes, gave better advice than a ground controller.
The good news is that once they have been flight tested, and used within the design specifications, computerised planes are safer and more efficient than a purely human-controlled plane. In theory, a computerised plane could take off from one airport, and land at another, without the pilot touching the controls - but I don't think that commercial airlines are quite ready to do that routinely. So the pilot is there in case something goes wrong with the computer(s), and the co-pilot is there in case something goes wrong with the pilot.
There have been some talk about further reductions in the flight crew to a single pilot, a computer, and a cross-trained cabin crew member (as a last resort).
-
The problem may well have been caused by a computer.
It's pretty normal on any moderately sophisticated plane to set the required rate of climb and target altitude into the autopilot, which can indeed fly the aircraft very accurately and deliver you to your assigned flight level. This works fine in moderate conditions but in extreme turbulence and icing you can get into a condition where the plane just won't climb at the set rate, even at full power, due to the ice load and the downdraught. The autopilot tries to increase the angle of attack to maintain climb rate, but it doesn't know that the stall speed has been increased by the weight of ice and the altered aerofoil shape, and the air you are flying through is actually descending faster than your best rate of climb. The result can be anything from a sharp lesson in aerodynamics to a complete disaster.
Flying a small plane on autopilot in turbulent visual conditions can be pretty spectacular if you just set "altitude hold", never mind maximum climb - you are pointing at the sky one moment and the ground the next, as the servos try to fight the vertical gusts. The rule is never to attempt an autopilot climb in a thunderstorm for all the reasons above - at some point it will stall and quite likely spin. You have to choose a power setting high enough to stop ice forming on the propellor or the air intake, and low enough not to stress the wings as you hit the gusts, then hand-fly to the assigned altitude, accepting that from time to time you may actually be going downwards, whilst feeling for the effect of ice on the aerofoils and knowing all the time that you are on the edge of a serious problem.
Without prejudice to the unfortunate Air Asia crew, I can well imagine that, at night, sitting in a quiet, airconditioned office with a zillion horsepower and a panel full of weather radar and other gadgets, it's very tempting to punch in the numbers and wait for George to take you a less turbulent place. Knowing that there's other traffic around tends to concentrate the mind on maintaining track and level, to the exclusion of actually flying the plane, and by the time the stall warning has penetrated the brain, you are in seriously bad shape, possibly with limited control surface moment, snow in the fans, and wings like ice lollies. The transition from marginal flight to pandemonium can then take less than a second.
The conclusion from a number of recent disasters is that modern airliners, and indeed modern training schedules, are so sophisticated that pilots lose the habit of hand flying, and need to be encouraged to chuck an Aerobat or Tiger Moth around from time to time, and indeed hand fly an airway sector occasionally, "just in case".
Birds don't fly in icing conditions. Despite billions of years of evolution, they know it's dangerous.
-
One of the big problems for computers (and humans) is "sensor fusion (http://en.wikipedia.org/wiki/Sensor_fusion)".
You can make flying safer by providing multiple modules on the plane (computers, pilots, engines, tires, methods of sensing fuel, speed, altitude and location), so if one module fails, you can continue using the others (at least, in theory) .
Modern planes have so many sensors that nobody could possibly look at all of them, so you need a computer to scan the sensors for out-of-tolerance values and draw attention to them. But humans can (and have) focused on minor problems (like a malfunctioning indicator light) while missing major problems (like the approaching ground). In other cases, pilots were flooded by alarms indicating that so many sensors were out of tolerance that the pilots could not determine the root cause. So you have to reliably prioritize them, too.
And what happens if two sensors disagree? How do you identify that something is malfunctioning, and identify the most reliable source to draw accurate conclusions?
Humans have problems with this - sometimes pilots get disoriented, and think they are flying straight and level, while their instruments tell them that they are in a spin. Statistically, the instruments are more reliable than the human sense of balance (especially at night or in fog/cloud or turbulence).
There was a case where the pilots were transferring fuel from one tank to another, and the indicators showed that the fuel was flowing properly. Unknown to the pilots and computer, there was a hole in the transfer pipe, and much of the fuel was streaming into the air. The plane had sensors to measure the rate that the fuel was being pumped out of the source tank, and sensors to indicate how much fuel was present in the destination tank, but the information was presented on different screens, and the computer was not programmed to know that the fuel coming out of one tank should equal the fuel going into the other tank (these checks have subsequently been added to that aircraft).
Unless you have a specific sensor for ice on the wing, you won't know that the shape of the wing has changed (but in this case, perhaps ground control knew that high altitudes were a risk in the current weather conditions).
So properly controlling an aircraft requires many sensors, and a complete picture of the health of the aircraft requires the pilot and/or computer looking for instances where the behavior of the plane diverges from your expectations, and reliably taking remedial action.
-
Thinking about computer control in aircraft, the new V22-Osprey (http://en.wikipedia.org/wiki/Bell_Boeing_V-22_Osprey#Avionics) is supposed to be quite an engineering feat. The helicopter apparently has some significant stability issues that are compensated by the computer (as well as pilot and maintenance training). No doubt the computer system will be stressed sometime in the future with damage and inconsistent inputs, but since the production model rolled out, it has done remarkably well.
As far as the A320, the autopilot should be able to calculate things like lift, velocity, and wing turbulence as part of optimum climb calculations in all situations by evaluating real-time data. The autopilot can shut itself off when it fails, but as Evan mentioned, perhaps a well designed autopilot would also be best equipped to deal with tons of rapid-fire data. But, there may also be things like safety margins that an experienced pilot might be able to push to the limit. Although, a computer may still be able to best determine the true limits of plane's capabilities in a given set of circumstances.
Still, in an emergency, the trained pilot may come up with solutions that had not been put into the computer program (such as flying a plane upside-down with a broken wing, or hitting the afterburners when the F-15 lost its wing (see link above (http://www.historynet.com/the-10-greatest-emergency-landings.htm)).
How long were the pilots tracking the storm that they were headed into? A 2-minute window for a critical maneuver sounds quite short for a storm that should have been visible from the radar before the plane even took off.
-
Humans have problems with this - sometimes pilots get disoriented, and think they are flying straight and level, while their instruments tell them that they are in a spin. Statistically, the instruments are more reliable than the human sense of balance (especially at night or in fog/cloud or turbulence).
A classic problem, classic indications, and by all accounts the residual classic indicators in the Air France case were showing all the classic signs. But the likelihood is that the pilots had never actually flown a deep stall or spin recovery by hand and eye in an Airbus. Indeed since recovery from a fully developed spin has now been deleted from the EU Private Pilot syllabus (I think it's still compulsory in the USA, and as I recall, can lead to days of constipation as the buttocks slowly unclench) it's quite possible they may never have spun a plane in their entire careers. I suspect the same applied to Air Asia.
Interesting to note that the only person to have successfully ditched an airliner (Sullenberg) was a gliding champion, well versed in flying by hand and eye with no power. When the ice/bird/whatever hits the fan, basic aerodynamics overrides multicrew cooperation, compressibility corrections, aviation law, and all the other clever stuff in the airliner syllabus.
-
It ought to be possible to use GPS to determine how the plane is moving instead of relying on sensors which can be iced up, although there will still be difficulties when there's a high wind speed (which may be a powerful downdraught in a storm) or if you're flying through the jet stream. If you're climbing at three times the safe rate for avoiding a stall, that should be detected by a computer looking at the GPS data. Maybe it was though - from what I've heard, the black box data shows that there were alarms going off all over the place and the pilots might have had a hard job knowing which to act on first. Ultimately this kind of problem will be fixed by better computer systems, but they're only as good as the program code they're running, so it'll take time to get it right.
-
As far as the A320, the autopilot should be able to calculate things like lift, velocity, and wing turbulence as part of optimum climb calculations in all situations by evaluating real-time data.
That's the easy bit, and quite standard. But if your maximum climb in still air is, say, 3000 fpm, and the air is actually descending at 3500 fpm (no big deal fior a thundercloud), what does the computer do?
How long were the pilots tracking the storm that they were headed into? A 2-minute window for a critical maneuver sounds quite short for a storm that should have been visible from the radar before the plane even took off.
10 minutes can change a thundercloud from an adventure to a disaster. There's as much energy in a small thunderstorm as in the Nagasaki bomb.
I now learn that the Air Asia pitot and/or static heads were iced, so there was no airspeed indication available to either the pilot or the computers. Very nasty in the dark, but potentially hand-flyable or you can reduce the scope of the autopilot to "track" and concentrate on keeping just above the stall speed by power and attitude, whilst wondering why the pitot heaters aren't switched on.
-
It ought to be possible to use GPS to determine how the plane is moving instead of relying on sensors which can be iced up, although there will still be difficulties when there's a high wind speed (which may be a powerful downdraught in a storm) or if you're flying through the jet stream.
Indeed. In a full turning climb into a 200 kt jetstream, your groundspeed (what the GPS tells you) may well be below your stalling airspeed. Not a useful input.
-
How many near accidents do we never hear about.
For example China Airlines 006 (http://en.wikipedia.org/wiki/China_Airlines_Flight_006) dropped 30,000 feet in an uncontrolled fall due to an engine stall and pilot error. But it safely recovered.
-
Depends on your definition of a near-accident. My experience of commercial flights suggests about 1% involve something interesting (i.e requiring unplanned pilot intervention) happening, and pretty much the same ratio for private flying. Sounds a lot, but I doubt that you could spend 99% of your time driving a car or truck by just watching the countryside go past. As retiring a BA captain explained to me "it was 35 years of pleasant boredom punctuated by about 20 minutes of blind terror".
-
"Sensor function" is the problem I had with my semi computer controlled car after three months of misbehaviour and changing clutch plates etc BMW by remotely running diagnostic software suggested that two speed sensors be replaced which touch wood has cured the problem.