Obviously AA587's tail fell off after it was sorely abused. There's but one candidate for said abuse (IMHO). It's just the process that they now need to hunt down.
This 13 Nov 01 incident below may be relevant (happened quite remarkably the day after the AA587 crash).
FACTUAL INFORMATION
A Boeing 747-SP38 aircraft was maintaining Flight Level (FL) 430 with autopilot `A' engaged, when the aircraft yawed abruptly to the right and rolled to a bank angle of approximately 20 degrees. The autopilot was disengaged and the aircraft stabilised in a straight and level attitude. The uncommanded yaw occurred again. The flight crew broadcast a PAN (radio code indicating uncertainty or alert, not yet the level of a Mayday) and received a descent authorisation to FL380.
The upper rudder position indicator showed a rudder displacement of 5-degrees right and the lower rudder indicator showed zero degrees deflection. The flight crew began activating and de-activating the upper and lower yaw damper switches attempting to isolate the problem. During those actions, the aircraft commenced to `Dutch roll' (lateral oscillations with both rolling and yawing components). The crew then successfully isolated the problem to the upper damper and turned the upper damper switch off. With the aircraft at FL380, normal operations ensued. Autopilot `B' was then engaged and the flight proceeded without further incident.
Investigation by company maintenance personnel
confirmed an anomaly of the upper yaw damper computer. The unit was replaced and the system tested. Normal operations ensued.
Analysis of Flight Data Recorder information revealed that during the event the upper rudder displaced 4.7 degrees. The data also indicated that the maximum roll encountered was 13 degrees to the right.
System redundancy had operated as required to limit the effect of the upper yaw damper anomaly.
http://www.atsb.gov.au/aviation/occu...ail.cfm?ID=381
I believe the A300-600 has two yaw damper actuators and two yaw damper computers. My gut feeling is that if there was to be a software "bug" latently in the software of one system it would be very likely repeated in the other. For there then to be a conflict between the two, a significant trigger threshold (such as the wake turbulence upset) may be required to trigger any vehement algorithmic disagreement. The interesting comparison is that in the 747SP case above, the computer anomaly was able to act upon its own (upper) rudder (only) - whereas in the AA587 accident it may have been the case that a resonance (or electronic echo) was set up between the two computers (each acting through separate actuators - but upon the same rudder). In that scenario, both may have been accepting feedback from the other's reaction to the yaw induced by the wake turbulence encounter. That sort of thing in acoustics (and digital and analogue electronics) soon sets up a very annoying superheterodyne squeal. In a computer-controlled flight-control system the equivalent outcome may be capable of chaotically driving the rudder to its limits. It might explain A300 tail-wagging in toto.
We know that the A300-600's FCS is capable of very rapid sampling and reaction times (such that the cockpit display info is necessarily filtered, as are the DFDR data-feeds also -unfortunately). One notable aspect of a yaw damper system is that it is by design a very reactive system i.e. it pauses (before acting) for a finite period to sense what is happening (from yaw-rate gyros) so as to calculate its appropriate response. It may even reset and resample if it doesn't believe its first answer. Because Dutch Roll is normally an escalating phenomenon, the yaw damper software may be "trained" to reject and resample any sharp disturbance of a large initial amplitude and rate (such as wake). So let us suppose, in a very dynamic scenario, that this "sensing and resetting" of each computer is able to become ever so slightly out of synch - perhaps sufficiently that the two systems are actually reacting to each other's inputs. Therein would lie the seeds of mayhem and destruction. Computers in this type of arrangement are normally "tied" (and constrained to agree), but perhaps not to the same extent as autopilots in an autoland configuration.
It would also be interesting to learn whether each yaw damper computer gets individual feeds (if any) from the CADC. My theory all along has been that, once things get dynamic, you cannot trust pressure sensors within a highly responsive and reactive system.
Once again it's just a theory. But you will recall that a yaw-damper failed its BITE test on start-up for Flt AA587 and had to be reset. Whilst not uncommon, it might at least be indicative. The 747SP anomaly at least proves that two identical systems, even though tied together and fed the same info, can still slug it out like a couple of boxers (once provoked sufficiently).