[Info-vax] OT: Aircraft pitot tubes and clustering.

[JF Mezei]

Last June, an Air France A330 fell out of the sky between Rio and Paris.

Today, they released a second interinm report on the progress of their
investigation.

http://www.bea.aero/docspa/2009/f-cp090601e2.en/pdf/f-cp090601e2.en.pdf
for those who are interested.

One area of focus is the air data probes that are used to measure
altitide, airspeed, angle of attack.

The A330 is a highly computerized aircraft. The report explains many of
the logic used between the multiple computers and sensors. This has
interesting associations with  clustering issues.

For instance, they have 3 computers that calculate air parameters from
the air probbes. If the parameters of one are too different, then those
values are kicked out and unused. The average of the 2 others is used.

If, of the 2 remaining probes, the differences are too great, then the
flight computers declare both to be invalid (which one would be correct ?)



BUT, they found cases where 2 probes failed at the same time and by the
same magnitude. This caused the remaining one to be kicked out despite
being correct, and the 2 errorneous ones to be used because the
differences between them are within bounds. As a result, the aircraft
used very wrong values for the air sensors.


Sort of interesrting because I guess when they designed the aircraft,
they had not considered cases where 2 failures would happen at same time
and with the same amount of error, causing the one remaining valid
sensor to be kicked off and the erroneous values to be used since their
have "quorum" betwen themselves.