kalvado wrote:

You don't have to feel superior, I am referring to the fact that pilot is supposed to take over in case things become nonstandard. Yes, older generation computers are not setup for anything else; they possibly could have more authority, though. Next generation could definitely be more capable. Would it?
There are many issues here. One is that human error is an accident, computer error is a liability. That alone is enough to accept a few crashes...
Process more information... In case of AF447 computer and pilot had same information from same instruments. And pilot could not digest that information.
I agree that human can think out if the box. AF447 situation could be handled within the box, though. Actually having checklist, memory items and SOPs means pilots are forced to stay inside the box most times...

I have the same feeling. It's actually a form of insurance for OEMs to give control back to humans once outside the very clearly defined norm. Maybe a computer could do better a lot of times, but maybe not always. If it is one of the cases where it cannot, then the OEM is on deep trouble. ("If only the machine had given control back to the humans!")

But there is one additional issue. Before we have AI, the computer in an airplane is basically a collection of what-ifs that other humans (with nerdy glasses sitting firmly on the ground) have defined. In an abnormal situation you basically trade off the on-the-spot decision making of pilots (with the help of standard procedures) vs. a pre-coded troubleshooting routine developed by a group of humans with less time pressure but also without the ability to "feel" the airplane and look out of the window.

And then with AI, who the hell knows...

Starlionblue wrote:

You can certainly add more and more logic to the autoflight system for edge cases, but that means more cost and more potential for misprogramming and other edge cases. Note that we're talking early 90s computer tech for the A330. A generation later, the A350 is more robust when it comes to autoflight edge cases. For example, if the aircraft slows to alpha prot on the A350, the autopilot will not disengage. It will go into "AP IN PROT" mode.

If I'm not mistaken, the A350 also tries automatic engine relights and such. Do you know more? Interesting to learn what A thinks they can handle by pre-defined code and what they shy away from.

Is B more conservative with respect to automation/AP scope/computer flying? The 787 was the latest clean sheet and I haven't heard those things. But also older than 350. What about 777x?

xl0hr wrote:

kalvado wrote:

You don't have to feel superior, I am referring to the fact that pilot is supposed to take over in case things become nonstandard. Yes, older generation computers are not setup for anything else; they possibly could have more authority, though. Next generation could definitely be more capable. Would it?
There are many issues here. One is that human error is an accident, computer error is a liability. That alone is enough to accept a few crashes...
Process more information... In case of AF447 computer and pilot had same information from same instruments. And pilot could not digest that information.
I agree that human can think out if the box. AF447 situation could be handled within the box, though. Actually having checklist, memory items and SOPs means pilots are forced to stay inside the box most times...

I have the same feeling. It's actually a form of insurance for OEMs to give control back to humans once outside the very clearly defined norm. Maybe a computer could do better a lot of times, but maybe not always. If it is one of the cases where it cannot, then the OEM is on deep trouble. ("If only the machine had given control back to the humans!")

But there is one additional issue. Before we have AI, the computer in an airplane is basically a collection of what-ifs that other humans (with nerdy glasses sitting firmly on the ground) have defined. In an abnormal situation you basically trade off the on-the-spot decision making of pilots (with the help of standard procedures) vs. a pre-coded troubleshooting routine developed by a group of humans with less time pressure but also without the ability to "feel" the airplane and look out of the window.

And then with AI, who the hell knows...

WHat I was getting at, is that in most cases (outside aviation at least) automation is not designed to drop dead in case of a problem. There is usually a range of fallback modes to keep things under control on one hand and a warning to allow human troubleshooting without instantly assuming total control on the other.
What are the possible fallback modes for an aircraft? Pitch and thrust is one getting constantly mentioned. Anything else?
Of course, P&T is not a panacea, but are there any cases where trying that on an autopilot while asking pilot to see whats going on would make things worse? Would trying and giving up (if things are really bad ) on P&T before full autopilot disconnect make things worse?

There are really people equivocating about the need for razor sharp, hand flying, stick and rudder skill?

GalaxyFlyer is right - in the end, that is what it comes to. Know the systems, know how to fly the airplane if the systems won't do it. Very simple.

N1120A wrote:

There are really people equivocating about the need for razor sharp, hand flying, stick and rudder skill?

GalaxyFlyer is right - in the end, that is what it comes to. Know the systems, know how to fly the airplane if the systems won't do it. Very simple.

This. So very much this.

As mentioned, also enshrined at the very start of the FCTM.

xl0hr wrote:

Starlionblue wrote:

You can certainly add more and more logic to the autoflight system for edge cases, but that means more cost and more potential for misprogramming and other edge cases. Note that we're talking early 90s computer tech for the A330. A generation later, the A350 is more robust when it comes to autoflight edge cases. For example, if the aircraft slows to alpha prot on the A350, the autopilot will not disengage. It will go into "AP IN PROT" mode.

If I'm not mistaken, the A350 also tries automatic engine relights and such. Do you know more? Interesting to learn what A thinks they can handle by pre-defined code and what they shy away from.

Is B more conservative with respect to automation/AP scope/computer flying? The 787 was the latest clean sheet and I haven't heard those things. But also older than 350. What about 777x?

Automatic engine relights aren't really an A350 thing. The A330 Trent engines have the same feature. I'm assuming it's more of an engine manufacturer feature than an airframe feature.

I don't really think B is more conservative, to be honest. Just different.

kalvado wrote:

xl0hr wrote:

kalvado wrote:

You don't have to feel superior, I am referring to the fact that pilot is supposed to take over in case things become nonstandard. Yes, older generation computers are not setup for anything else; they possibly could have more authority, though. Next generation could definitely be more capable. Would it?
There are many issues here. One is that human error is an accident, computer error is a liability. That alone is enough to accept a few crashes...
Process more information... In case of AF447 computer and pilot had same information from same instruments. And pilot could not digest that information.
I agree that human can think out if the box. AF447 situation could be handled within the box, though. Actually having checklist, memory items and SOPs means pilots are forced to stay inside the box most times...

I have the same feeling. It's actually a form of insurance for OEMs to give control back to humans once outside the very clearly defined norm. Maybe a computer could do better a lot of times, but maybe not always. If it is one of the cases where it cannot, then the OEM is on deep trouble. ("If only the machine had given control back to the humans!")

But there is one additional issue. Before we have AI, the computer in an airplane is basically a collection of what-ifs that other humans (with nerdy glasses sitting firmly on the ground) have defined. In an abnormal situation you basically trade off the on-the-spot decision making of pilots (with the help of standard procedures) vs. a pre-coded troubleshooting routine developed by a group of humans with less time pressure but also without the ability to "feel" the airplane and look out of the window.

And then with AI, who the hell knows...

WHat I was getting at, is that in most cases (outside aviation at least) automation is not designed to drop dead in case of a problem. There is usually a range of fallback modes to keep things under control on one hand and a warning to allow human troubleshooting without instantly assuming total control on the other.
What are the possible fallback modes for an aircraft? Pitch and thrust is one getting constantly mentioned. Anything else?
Of course, P&T is not a panacea, but are there any cases where trying that on an autopilot while asking pilot to see whats going on would make things worse? Would trying and giving up (if things are really bad ) on P&T before full autopilot disconnect make things worse?

The automation on an Airbus doesn't "drop dead" if it doesn't work. There is a wide range of degradations depending on the specific failure. For exanple, even if the AP disconnects, you're most likely still in Normal Law with all protections.

xl0hr wrote:

If I'm not mistaken, the A350 also tries automatic engine relights and such. Do you know more? Interesting to learn what A thinks they can handle by pre-defined code and what they shy away from.

Is B more conservative with respect to automation/AP scope/computer flying? The 787 was the latest clean sheet and I haven't heard those things. But also older than 350. What about 777x?

Auto relight is a FADEC function, it the computer that is bolted on the side of the engine. It is a feature the engine manufacturer brings to the party, you find auto relight on gulfstreams, Boeings, Airbus.

Airbus has three flight control law, normal, alternate, and direct. The 777 has similar “modes”, normal, secondary, and direct. What each law/mode does is different, however the overall concept of different levels of redundancy is similar.

Starlionblue wrote:

The automation on an Airbus doesn't "drop dead" if it doesn't work. There is a wide range of degradations depending on the specific failure. For exanple, even if the AP disconnects, you're most likely still in Normal Law with all protections.

In that sense, FBW automation has no right to power off completely, so it should never drop truly dead (i know, manual trim wheel as last-ditch)
It is more about pilot has to assume control at the first step of degrading situation.
One may argue it's a must for any moving vehicle, but my impression it's not the best way.