In Atmosphere Computer Navigation

[razark]

Let's pretend Flight 447 wasn't the only incident in the entire history of air flight, though, and imagine another situation.

Let's go with the previously mentioned, completely hypothetical incident and consider how a computer reacts in such a case.

[magnemoe]

17 hours ago, YNM said:

That's the crew problem. If they were to be scrapped altogether the system would continue running. The crash were purely because of the crew's mistakes.

More examples :

https://en.m.wikipedia.org/wiki/TAM_Airlines_Flight_3054 - crash due to incorrect setting of the autopilot.

https://en.m.wikipedia.org/wiki/Philippine_Airlines_Flight_137 - another one.

There are a lot more of crashes caused by improper usage or communication. The amount that's caused by the automatic system itself without any intervention is much lower, even closer to non-existant.

On the other hand UAV has very high loss rates, not talking about the small tactical ones but the plane sized raven and predator drones have high losses.
Much because the operator has no feeling for the plane and limited situation awareness. Data link fails are another major issue. 
Yes the UAV are used hard in an combat environment there you are used to equipment getting destroyed on the other hand have fun trying to get permission to operate one over an large city even if you are an governmental organization, landing one on an busy airport would be a bit more difficult to get. 

[kerbiloid]

On 24.08.2017 at 8:20 PM, Snark said:

The humans had to make a judgment call:  land the plane on water, or try to make it to the airport?

"My Bonnie flies over the ocean.
My Bonnie flies over the sea."

And there is a hundred of bonnies over the water at once.
An autopilot must have emergency water landing mode, otherwise it just must not be certified to fly over the sea.
And this is a human fault, not autopilot's.

Radiocontact doesn't matter here, as any navigation computer has/must have a map of a place where it currently flies.
Even without a radio it should know is it a water or what, and nobody expects flying between the bridge supports from a human pilot, too.

On 24.08.2017 at 8:20 PM, Snark said:

The humans understood what happened, i.e. that the engines failed because of a bird strike, not something internal to the plane that might affect how to respond.

"Engine is hit by a bird" is so common situation, that it's definitely not a question of intuition.
Also there can be other situations when an engine is damaged by external/unknown reasons (sucked a stone into the intake while taxiing, or the same with a technician (real situation)).
So, this is just an "unexpected engine failure" situation, which doesn't require any intuitive efforts. The only things autopilot has to know: does it work? does its temperature raise? does it have leaks?

On 24.08.2017 at 8:20 PM, Snark said:

"Something smells like a burning wire"

On 24.08.2017 at 8:20 PM, Snark said:

"Hey, smoke appears to be coming out of the wing"

Fire sensors do this for decades. Any smell means a fluid concentration. Any fire means carbon oxides (even without a fire)..
Also if the crew in a pressurized cabin can smell a burning wire inside a wing, it looks too late to smell. A soulless fire sensor inside a wing would do this earlier.

On 24.08.2017 at 8:20 PM, Snark said:

"What's that funny sound?"

On 24.08.2017 at 8:20 PM, Snark said:

"Passengers are reporting a loud bang"

Any sound means a mechanical movement. A vibration, a hit, a metal tearing apart.
Mechanical movements better feel with mechanical sensors, rather than ears. Also, stresses can be easily measured by over-cheap tension sensors which you can just stick on the metal.
If a tension drops to zero, something is teared off.

On 24.08.2017 at 8:20 PM, Snark said:

various such imminently-need-to-be-dealt-with problems that might not have an automated sensor that catches it.

So, they should have sensors. And there can be no unexpected situations after a century of flights, there can be only not implemented.
Also (as aviation guides teach us) any aircraft fatal incident is always caused by at least two problems, not one.

On 24.08.2017 at 11:03 PM, Snark said:

When you ride on a plane, you're trusting your life to a machine with zillions of moving parts

While the only moving part in a computer is a cooler. And it's not really necessary.

 

On 24.08.2017 at 10:26 PM, wumpus said:

terrorist attacks/hacking are a better justification

Is a ground security problem, not a flight computer's one.
Also if an autopilot doesn't react on terrorists' claims, there is no reason to claim something in flight. So, again the situation moves from sky to ground.

On 24.08.2017 at 11:03 PM, Snark said:

A pilot is trusting his own life, along with the passengers'.

And dies with them several times per year, as an importance mostly means a stress, rather than an accurateness.
Computer is never frightened, tired or stressed.

P.S.
Afaik, almost all known big crashes of last years were caused by human pilots struggling against an autopilot who was trying to prevent their suicide.
Crashing in a mountain because "we are not in this area, computer lies" (Russia/Indonesia).
Intentionally crashing down because pilot was feeling blue (Spain/Germany).
Several overweighted flights (a good old classic).
Struggling against an autopilot during the landing when a plane has a tricky manual sequence when you have to move a handle in a counter-untuitive way.
Manual landing attempt to land on a runway in clouds (when an autopilot would just move to another runway) - for political reasons.
Unexpected manual braking on take off.
Other subjective and intuitive human reasons.
And none of these crashes was caused by an autopilot itself.

Also, an autopilot doesn't give a piece of waste:

• whether boss orders to the pilot something unwise
• whether the company tries to overweight tthe plane for economy,
• whether the compant tries to limit the fuel to save money
• political necessity
• other human so important games.

If an on-wheel mechanical sensor feels that the plane is overweghted, the plane will just stay on land, whether angry and greedy humans cry on it, or not. An autopilot doesn't care, it would just lock the ignition.

Edited August 26, 2017 by kerbiloid

[Green Baron]

I don't know the movie Scully, but it seems to me as if several situations were showily exaggerated (American style :-)). Yes, bird strikes aren't uncommon and a pilot has a water landing in mind, it is not ingenious to come up with the idea, it is (was ?) even discussed in PPL courses in the 90s.

But it would probably be difficult to tell an autopilot "You are in the initial climb phase, stall speed is so close, the power plants have both gone on strike, you must maintain control over the flight path no matter what, the airfield is out of reach or the margin is too small if only the slightest incident happens. So look for next free area including water bodies and golf ranges and touchdown the plane there with minimum speed. Leave the rest to physics."

If a plane has the necessary sensors (thinking of ground radar or so to judge the last 20 meters) that might work but i can imagine that a well trained pilot has the better repertoire of courses of action to deal with unforeseen tiny things.

OTOH the situation with the smoke in the cockpit is different, it is much easier: simply land the plane on the next airport that allows an automatic landing. There have been accidents where very experienced pilots flew the plane until nothing could be seen from the cockpit due to smoke. The was a Swiss air flight that crashed near Newfoundland/Greenland because of this.

There an autopilot would have been the better choice.

 

[YNM]

7 hours ago, magnemoe said:

On the other hand UAV has very high loss rates, not talking about the small tactical ones but the plane sized raven and predator drones have high losses.
Much because the operator has no feeling for the plane and limited situation awareness. Data link fails are another major issue.

You're describing remote-controlled UAVs. What about the fully autonomous ones ?

Also, because the systems and computers made for these flying contraptions is also made by the same apes who's used to fly them, it's limited by what the apes think necessary for them to have. A perfect computer would be able to completely do the job.

So, yes, it's not easy to have a perfect one, but it's because the people who made them doesn't even know to all detail what goes into flying the contraptions in the first place.

[wumpus]

5 hours ago, Green Baron said:

I don't know the movie Scully, but it seems to me as if several situations were showily exaggerated (American style :-)). Yes, bird strikes aren't uncommon and a pilot has a water landing in mind, it is not ingenious to come up with the idea, it is (was ?) even discussed in PPL courses in the 90s.

At the time (i.e. well before the movie.  I know I've heard more of it directly from real (small plane) pilots than people who saw the movie), my memory is that the air traffic controllers were surprised that Scully picked the Hudson as his place to land (they assumed an airport even if their screens didn't indicate sufficient velocity or altitude).  Obviously you should never look to movies (especially the Hollywood variety) for facts.

3 hours ago, YNM said:

You're describing remote-controlled UAVs. What about the fully autonomous ones ?

How many fully autonomous UAVs are made with the budget and requirements of an Airbus autopilot?  If we get sufficiently large UAVs flying over populated areas (the UAV company I worked for moved from Washington DC to Texas for a reason.  Much easier to pay a rancher for a cow collision than deal with landing on a school) we may see such software.

8 hours ago, kerbiloid said:

Also, an autopilot doesn't give a piece of waste:

Ask the good folks at the University of WV and others who discovered less than ethical software in VW diesel controls.  An autopilot does what it is programmed to do, and certainly doesn't value its life if the plane manufacturers (or whoever sources the autopilot) want it to look like it is saving money.

[YNM]

14 minutes ago, wumpus said:

How many fully autonomous UAVs are made with the budget and requirements of an Airbus autopilot?  If we get sufficiently large UAVs flying over populated areas (the UAV company I worked for moved from Washington DC to Texas for a reason.  Much easier to pay a rancher for a cow collision than deal with landing on a school) we may see such software.

We've seen endurance flights being flown autonomously nearly all the time. I could only imagine that for the time being such tech might still be under a fog-of-tech.

Edited August 26, 2017 by YNM

[kerbiloid]

1 hour ago, wumpus said:

Ask the good folks at the University of WV and others who discovered less than ethical software in VW diesel controls. 

Humans did it. Not autopilot itself.

[wumpus]

2 hours ago, kerbiloid said:

Humans did it. Not autopilot itself.

So does the autopilot writes its own software?  Or do humans?

[kerbiloid]

6 minutes ago, wumpus said:

So does the autopilot writes its own software?  Or do humans?

Humans certify what humans develop.

The VW software was proprietary and unlikely was studied by a government or international commission.
But software is a part of the aircraft, just invisible.
Can somebody build a plane on backyard and begin selling it to airlines without gathering numerous permissions?
The same with autopilot.

Edited August 26, 2017 by kerbiloid

[Exploro]

21 hours ago, razark said:

Let's pretend Flight 447 wasn't the only incident in the entire history of air flight, though, and imagine another situation.

Let's go with the previously mentioned, completely hypothetical incident and consider how a computer reacts in such a case.

Would a sufficently sophisticated automated flight control computer make the specific choice to ditch into the Hudson River? I'll concede it would probably would not. However would it have had to make such that choice?

From what I read about post-accident simulations, roughly one in two simulated runs ended with a successful landing at either LaGuardia or Teterboro. This at least demonstrates 1549 had the energy and altitude necessary to make a successful glide to landing possible. What matter seemed to be time to recognize the situation fast enough to make such a choice possible. While no specific reason was given for each failed run; I suspect that the other half that result in failure like result from much greater lag in response and applying action. In fact the one test had a more realistic delay between when the engines failed and when action was taken; 35 seconds. That test resulted in failure. Airmanship might also played factor in the failed test runs as well. But suffice it to say a computer would have been faster to determine the engines were inoperative. And a properly written control law might skip attempting a restart at such low altitude (The restart procedure itself assumes sufficient altitude to make the attempt anyway), saving time. Thus all that is required is to know position, altitude, energy state, the weather conditions (specifically wind) and what; if any, suitable landing field are available. Given that a computer could control an aircraft more consistently than a person and can troubleshoot and analyze the situation much faster than a person all the while being immune to mental fog that can befuddle even the sharpest of human minds in a crisis, the chance of making a successful glide to landing at a safe field would likely be much better than chance.

Edited August 27, 2017 by Exploro

[razark]

You're right.  Computer is always better than human.  Never a reason to have a human involved.

 

Incidentally, driving back from the eclipse, I used roads that did not appear on either the roadmap or the GPS system, yet I made it home and also avoided the large amount of traffic that the GPS plotted us into.

Curse my human frailty for leading me into such devastation.  If only I had followed the computer that always knows better.

[DerekL1963]

2 hours ago, Exploro said:

From what I read about post-accident simulations, roughly one in two simulated runs ended with a successful landing at either LaGuardia or Teterboro.

Even with an instantaneous decision to divert, roughly one in two ended in failure.  (And one-in-two isn't actually quite correct - IIRC it was one-in-two for LaGuardia and zero for Teterboro.)  However, that instantaneous decision requires knowledge that neither the pilots nor a computer had or would have - that the engines could not be restarted.  No sensible control process is going to skip a simple test (trying to restart the engines) in favor of a complex and risky maneuver (immediate divert).

Another factor is that IRL Sullenberger skipped ahead in the procedure and started the APU, ensuring the continued availability of hydraulic power.

[kerbiloid]

4 hours ago, Exploro said:

Would a sufficently sophisticated automated flight control computer make the specific choice to ditch into the Hudson River?

The flight control computer should have an emergency mode: "land on the nearest flat surface".
Flat surfaces can be solid (preferably) or liquid.

All flat surfaces, including runways, highway strips, water bodies etc,, are definitely designated on maps.
So, the plane brain just must select the nearest ones, sort them in preference order and try to land on the best of them.
If the best available flat surface is luquid, they make a new action movie.