Why do pilots still die from not knowing which end is up?

Re:Why do pilots still die from not knowing which end is up? by apaprocki on Saturday October 22, 2011 @05:13PM (#3144345)

I think a more robust model of N completely different systems (both mechanical and electrical) to compute the same variable could be found so that the probability that it is computing a value outside of x% of the actual real value is lower than the probability that the human could make a mistake. Even if a speed indicator model in the event of pitot tube loss had a 25% margin of error, autopilot could have done the right thing or it could have possibly influenced their decision to pull back.

EDIT: You can down-vote, but this is already the direction that cars have gone. For example, there is no way to stop any Mercedes after 2009 from automatically applying full brakes in an impending collision.

http://www.daimler.com/dccom/0-5-1210220-1-1210348-1-0-0-121...

[ Parent | on HN ]

Re:Why do pilots still die from not knowing which end is up? by gnaffle on Saturday October 22, 2011 @06:53PM (#3144530)

I'm not downvoting you, but I think this is a kind of problem that seems solvable on the surface, but is very, very difficult to get right. You will always have situations where all speed indicators fail in a similar way leading the software to believe that the speed is correct. With all the erring on the side of caution that aircraft makes have used, we've had a couple of cases where the software "decided" something and caused incidents.

Studies in N-version programming (such as used in Airbus planes) show that software written by different teams often fail in the same corner cases (with differing outputs).

Eventually you might have software with "situational awareness" that can be trusted to make fewer mistakes than humans, but we're not there yet.

[ Parent | on HN ]

Re:Why do pilots still die from not knowing which end is up? by apaprocki on Saturday October 22, 2011 @07:14PM (#3144579)

I agree it will take a long time to develop automated systems that can be trusted. It took 26 years for tech to develop from the first ILS assisted passenger landing to the first fully automated ILS landing.

If you were to take this specific speed sensor issue, I'm more concerned there aren't different types of systems being used to try to cross-validate the numbers. The pitot-static system has its own backup on the plane, but if there is an inherent weakness of that particular system (e.g. ice), there is no other way to measure speed that is not susceptible to icing. Surely that particular problem is within our reach, no?

[ Parent | on HN ]

Re:Why do pilots still die from not knowing which end is up? by gnaffle on Saturday October 22, 2011 @07:38PM (#3144631)

To measure airspeed I guess something has to have air flowing through it or past it, and that something would be susceptible to icing. That said, current pitot have very good ice protection, and AF447 was an extreme example with a known problematic type that Air France were in the process of exchanging.

Perhaps GPS ground speed data with previous wind estimates could at least help with establishing that "something must be wrong with the airspeeds".

[ Parent | on HN ]

Re:Why do pilots still die from not knowing which end is up? by apaprocki on Saturday October 22, 2011 @07:43PM (#3144645)

That is what this captain suggests:

http://travel.usatoday.com/experts/cox/story/2011/05/Ask-the...

Something scary from a Boeing pilot in the comments:

  Loss of all airspeed indications is not something we train 
  for. It's always assumed in transport category aircraft 
  that redundancies built in will always give us at least 
  attitude, heading, and airspeed.

[ Parent | on HN ]

Re:Why do pilots still die from not knowing which end is up? by caf on Sunday October 23, 2011 @01:53AM (#3145329)

There are meterological radars that use doppler shift to measure wind speed, I wonder if you could use the same basic idea to construct a radar airspeed indicator?

[ Parent | on HN ]

Re:Why do pilots still die from not knowing which end is up? by tjoff on Sunday October 23, 2011 @05:58AM (#3145718)

"For example, there is no way to stop any Mercedes after 2009 from automatically applying full brakes in an impending collision."

Which is sad.

From your link: [quote]Brake Assist PLUS uses two radar systems. Both of these radar systems are very good at not only registering objects up ahead but also calculating the distance from these objects and their relative speed. If the distance between the vehicle and the registered object is so small that there is a real danger of collision, Brake Assist PLUS issues a visual warning on the instrument cluster display and sounds an alarm over the car speakers. While this is happening, the electronic system calculates the braking deceleration necessary to avert the collision. If the driver then hits the brakes, the braking pressure needed to bring the vehicle to a stop before impact is immediately provided automatically. Ensuring that an appropriate braking force is not exceeded gives those driving behind more time to react.[/quote]

A human would be quick to notice that the reason for why the car in front is braking is because of the impeding concrete wall. The computer however doesn't know that and would only brake in order to not hit the car in front and thus ensuring that both cars hit the concrete wall at equal speeds, great! (I truly hope that it respects my choice if I really wan't full brakes).

What if there are slippery conditions? The collision might easily be avoided by changing lanes but changing lanes in muddy snow while braking could be suicide. What if I just started changing lanes when the computer feels like it needs to brake?

What if the reason for why the air-speed indicator is malfunctioning (reporting wrong, a complete loss of velocity is too easy to check for) is because the plane just flew into a flock of birds, it might be obvious for a pilot why the stall-warning was issued and the pilot might opt to to pull back the stick - just to avoid an collision...

Unless the sensors truly get the whole picture and the complete state of the aircraft/car (and I doubt we can do that before being able to implement a true AI) the pilot/driver must always have the last word. Assistance can be great, especially in cases where a human wouldn't have the time to react but you should always be able to override it.

The assistance should be clear and specific enough so that the pilot/driver really thinks twice before overriding anything (this requires that the pilot/driver respects the system). And if the pilot/driver can't handle the vehicle despite that then he/she shouldn't be allowed to drive it.

[ Parent | on HN ]

Re:Why do pilots still die from not knowing which end is up? by vacri on Saturday October 22, 2011 @07:05PM (#3144560)

Aircraft are a lot more deadly than roads. An air collision is pretty much fatal, whereas automobiles hit things all the time, often with no real damage to the occupants.

If your software is making the decisions, then the pilot can't make an educated response to unusual decisions. Big commercial aircraft are already largely run by software, but the pilot is there for unusual situations because there's often a lot of conflicting issues.

[ Parent | on HN ]

Re:Why do pilots still die from not knowing which end is up? by CamperBob on Sunday October 23, 2011 @12:43AM (#3145229)

Keep in mind that unlike auto accidents, incidents like AF447 are incredibly rare. That particular problem could've been fixed with better pitot heaters, no need for HAL-like software gimmickery. By implementing large-scale changes in an attempt to eliminate corner cases like that, it is all too easy to introduce vulnerabilities that will lead to completely different accidents, no less unpredictable.

There's a lot I don't understand about AF447, like, why couldn't a simple GPS receiver have saved them? GPS does a relatively poor job at reporting altitude, and is obviously useless at reporting airspeed, but it's certainly capable of telling you that you've dropped 10,000 feet over the past minute or two. I can understand how pilots could become disoriented, such that they don't know "which end is up," but I can't fathom why they shouldn't even know whether they are climbing or descending, which apparently was part of the AF447 scenario.

[ Parent | on HN ]