Hacker Newsnew | past | comments | ask | show | jobs | submitlogin
Qantas flight plunge blamed on computer (stuff.co.nz)
36 points by pwg on Dec 20, 2011 | hide | past | favorite | 22 comments


TL;DR: This incident was, as usual, the result of multiple failures happening together.

1. An airspeed sensor starts outputting intermittent data spikes. This is a software failure, but they still don't know what caused it. There is some theory about high energy particles hitting the CPU. It seems very strange to me that a critical system like this can output garbage data and nobody knows why.

2. The flight control computer usually detects these spikes and filters them out by using the last good data for the next 1.2 seconds. However, at the end of the 1.2s, it assumes that the input from the sensor is good. If another spike is occurring at the moment it switches back, it isn't detected. That seems like a rookie mistake to me. I've definitely run into the same sort of issue in other contexts -- handling a second event while reacting to the first one.


Happened in November 2010 in the timespan of 4 days. (I think there were even more but this is what I remembered I can quickly grab from http://goo.gl/FgcAv)

11/4: Quantas Airlines Grounds All Airbus A380s After Engine Fire Over Indonesia http://www.huffingtonpost.com/2010/11/04/qantas-airbus-probl...

11/5: Engine Problems Hit Second Quantas Aircraft -- This Time a Boeing 747-400 http://www.reuters.com/article/idUSTRE6A435P20101105

11/8: Quantas Uncovers More Engine Problems http://www.abc.net.au/news/stories/2010/11/08/3059568.htm


http://www.atsb.gov.au/media/3532398/ao2008070.pdf is the final report. Quite a scary read. 30 warnings on the displays with master caution and master warning going on and off. All of it because of a bug where the binary representation of the altitude was interpreted as angle of attack and a design problem in the flight computer which does filtering on the signals but couldn't cope with spikes exactly 1.2 seconds apart. Result - the vertical acceleration went to -1.5G in 3s and then to 1.7G in the next 2s. That can look something like this http://www.faa.gov/​other_visit/aviation_industry/​airline_o...

I don't understand why the flight computers don't perform more aggresive smoothing on the signals from the sensors - AoA just can't go from 4 deg to 50 deg in 1/25s.


Your link doesn't seem to work: here's a retry: http://www.faa.gov/other_visit/aviation_industry/airline_ope...


I think the biggest problem is simply that it was a software bug that occured infrequently enough. To change a working piece of software that has thousands if not millions of problem free flying hours behind it is not something you do "just because". It requires lots of analysis and testing to make sure that they don't introduce new problems.


> I don't understand why the flight computers don't perform more aggresive smoothing on the signals from the sensors

In this case, you can solve the hard AI problem, and you're just getting started. Witness Air France.



> The air-speed sensor malfunction was one of only three such malfunctions known worldwide in 128 million operating hours ... In a strange coincidence, all three sensor malfunction events occurred on Qantas flights off the coast of Western Australia.

This line in the article seems designed to elicit comments about "the Christmas Island triangle" and "Canberra conspiracies".


There's a very powerful transmitter from a US spy base in Exmouth that was suspected at one point: http://www.takesontravel.com/web/index.php?option=com_k2&...


I don't know anything about antennas, but that thing looks freakin' massive:

http://maps.google.ca/maps?q=exmouth+western+australia+harol...


And here I was just thinking Quantas had found the island from Lost...


Or somebody who didn't turn their cellphone off while in flight?


I've got great faith in planes as a whole, but these sudden vertical accelerations (wind pockets etc) is one area where they don't seem to have much of a plan aside from hoping you're wearing a seat-belt at that particular point in time.

Maybe they should add something crumple-zone like in the panel above the passengers heads. Or put something air-bag like there.


I don't think they can add much crumple zone up there - that area has to be strong enough to hold people's luggage.

(On another note, I'm a very nervous flyer, and I put my seatbelt on as soon as I sit down, and don't take it off unless I'm using the restroom, or until we're at the gate. Hurray, paranoia!)


Make sure to take it off if they are refueling though. You'd want to get out of there if things light up.


How often does that happen?


Effectively never. The fuel is kerosene which has a high flash point, and the plane is grounded to the refuelling apparatus to prevent any static sparks.


It's probably fairly comparable to a trifecta of software bugs causing a plane to suddenly plummet twice, causing injuries.


Does anyone know why an investigation like this takes three years? I would have guessed that since the plane landed safely it would be a pretty quick diagnosis.


Debugging a bug that has showed up 3 times ever doesn't seem exactly like it would be easy...


Airbus could easily be the most unsafe major airline in history. No surprise when I read this article that it said this happened on Airbus. Those things have a nasty habit of randomly crashing for no reason whatsoever. Unfortunately planes cannot be booted back up after a crash. I would not be caught dead on an Airbus because that is probably what I would end up being, dead.





Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: