4 ms·
> Except that this is ALSO a failure mode Yes, it is. But in the Air France incident, there was faulty data from the airspeed indicators; the computers were in
by pdonis 10y ago
> Except that this is ALSO a failure mode
Yes, it is. But in the Air France incident, there was faulty data from the airspeed indicators; the computers were incapable of flying the aircraft under those conditions, so there was no choice but to let the humans try. They failed, but that's not an argument for not letting them try if the computers would have also failed.
In the Quantas incident, there was no faulty instrument data as far as I can tell; it was just a computer crash. So the human would have done fine if only given the chance.
> The programming needs to start assuming that it gets no help but needs to recover anyway.
How can that happen if the computer crashes? Or if the instruments are giving faulty data? Unless and until we have artificial intelligence equal to that of humans, I don't see how you can not have a way for the human to have some last ditch way of overriding the computer, since the human at least has some chance of handling conditions that the computer simply can't.
- bsder 10y ago> so there was no choice but to let the humans try. Keeping status quo without an explicit human override while notifying humans would have been FAR more preferable. Once common thing in all these failures is that the derivative of these readings/settings has a spike. These are passenger jets. They have envelopes; they behave according to physics. They don't do anything suddenly. Sudden anything should get flagged as a mistake and ignored until corroborated. > How can that happen if the computer crashes? Or if the instruments are giving faulty data? You have voting and redundancy for a reason. 3 votes is not enough. The space shuttle has 5 computers of which 4 are active and voting. Embedded computers can and do reboot within milliseconds if that is part of the spec. The real problem, I suspect, is lack of testing. Especially of a fully assembled system whose parts have come from disparate suppliers. Once you take the attitude "The human cannot help you out" your testing regime needs to get a LOT more stringent.
- rlpb 10y ago> the computers were incapable of flying the aircraft under those conditions Perhaps so, but a safe method of flying the aircraft under those conditions were known, and could have been programmed. "When airspeed indication fails, fly at this power setting". The computers handed over to humans, knowing exactly what indicators were faulty, as programmed with the intention being that they're better at dealing with this kind of situation; it turns out that they are not.
- pdonis 10y ago> a safe method of flying the aircraft under those conditions were known, and could have been programmed. "When airspeed indication fails, fly at this power setting". That might be safe as an immediate response, but is it safe for long term control of the plane under changing conditions with no airspeed indication?
- rlpb 10y agoWhen flying straight and level, with the airframe set up for cruise (eg. no flaps, gear up, etc): yes. Weather conditions don't matter; the plane's speed through the air will remain the same, and it is the airspeed that matters when it comes to keeping the plane flying. Windshear can temporarily change the airspeed due to inertia, but keep the same power setting and the plane will catch up by itself. When doing anything else, such as climbing, descending, etc: the required power settings will be different, but can be predetermined. Turning will reduce airspeed for a given power setting, but if the fault is known, the turn can be done with a small enough angle of bank that it won't matter. This can be done by the pilot, or for the sake of this line of argument, a computer could easily be programmed to limit the bank angle when an airspeed indication is known to be unavailable.
- cmurf 10y agoIn the 447 incident, the pilots were often giving contravening inputs, and nothing was communicating either the contradiction or the computer's resolution to the contradiction. Bad design. And the user interface for the computer indicating its own confusion, was a bogus overspeed warning when in fact the plane was stalling. There was no sensor failure causing that miscommunication, it's how the system was designed, but it only behaved that way in a particular software abstraction mode (one of the alternate laws), and there's no unambiguous UI to know what law applied. The cockpit was also dark, and there was severe turbulence. Anyway, there is no such thing as an override when it's fly by wire. An "override" is just a different level of software abstraction, it's yet another mode, with yet another set of rules. And that's what I think is bullshit, are the layered rules. It's such bullshit it makes me angry. This is how the myth started that Airbus's can't be stalled, the computer won't let you do that. True in normal mode, and complete bull fucking shit in at least one of its alternate modes.
- colde 10y agoDuring Alternate Law, the ECAM displays "ALTN LAW: PROT LOST" I think that is fairly unambiguous.
- cmurf 10y agoFrom the Air France 447 findings: "Between the autopilot disconnection and the triggering of the STALL 2 warning, numerous messages were displayed on the ECAM. None of these messages helped the crew to identify the problem associated with the anomalous airspeed." Knowing merely that you're in alternate law is not sufficient to know the entire consequences of the mode you're in.
- zkms 10y ago> so there was no choice but to let the humans try. They failed, but that's not an argument for not letting them try if the computers would have also failed. The major issue in "letting the humans try" in the AF447 incident is that the pilots did not have the angle-of-attack measurement shown to them. AOA is the most critical flight parameter (it determines whether the wing is lifting or stalled) and while the flight computers have access to it, AOA is not displayed in the flight displays. This is a huge flaw; forcing the pilots to guesstimate AOA from airspeed readings is a horrid idea, especially if the airspeed sensors might be malfunctioning. There are already dedicated AOA sensors on the airframe and only a minor software change is needed to display AOA on the primary flight displays. This design decision of hiding critical information from the humans (what AOA is, what the other pilot is doing with their stick) is what scares me most about fly-by-wire. I get that displaying all the information that's on all the avionics busses would overwhelm the crew, but displaying AOA is not pointless verbosity, it's the most crucial air data measurement on a fixed-wing aircraft. Sullenberger agrees, "We have to infer angle of attack indirectly by referencing speed. That makes stall recognition and recovery that much more difficult. For more than half a century, we've had the capability to display Angle of Attack in the cockpits of most jet transports, one of the most critical parameters, yet we choose not to do it." So does the AF447 report, "It is essential in order to ensure flight safety to reduce the angle of attack when a stall is imminent. Only a direct readout of the angle of attack could enable crews to rapidly identify the aerodynamic situation of the aeroplane and take the actions that may be required."
- pdonis 10y ago> The major issue in "letting the humans try" in the AF447 incident is that the pilots did not have the angle-of-attack measurement shown to them. Yes, I agree, but that's not a reason not to give the humans an override option. It's a reason to provide AOA data directly to the pilots. I am actually flabbergasted that that is not done, since, as you say, AOA is the most critical flight parameter. > This design decision of hiding critical information from the humans (what AOA is, what the other pilot is doing with their stick) is what scares me most about fly-by-wire. I agree 100%.
- 10y ago