AI in Piloting (AI-Augmented flight) and in Aircraft Inspection and Maintenance

@AIAeroscan see...

For now.

Realistically, the way things are going, I absolutely see single pilot airliners in the fairly near future not like 5 years from now, but soon, I hesitate to prognosticate too much, but yeah... sooner than most of us would like. It would simply be too lucrative to not do it if it ends up being possible. I'm personally not even entirely sure it will be possible, but if the robot was good enough I actually think it's "ok." The thing flightcrews have on their side is that airplanes are expensive, and designing new ones is really expensive, so maybe the Airbus A1050 or whatever will be single pilot, but it's going to be a little bit yet. Furthermore, I doubt the number will ever go below 1, if it even gets to 1. Someone has to be legally responsible for the vehicle; by law and by tradition the captain will always be responsible for the ship.

That said, with current automation? I absolutely agree with you, but things are going crazy right now and I doubt this will remain out of the realm of possibility. In particular, this paper about a simulated hospital is astoundingly impressive as is Google's weird AI that learns from videogames and simulations. If we can learn in the simulator why can't the robots.

I've been pretty keyed into this for the last 2 years as I've been doing grad school. My research was in computer vision but basically the entire curriculum and research was on AI or AI adjacent topics. Now that I'm done with grad school I'm kind of at a loss with what kinds of jobs to actually even look for, because the more I see the more I suspect that all jobs have a pretty significant risk of automation in the next decade or two and picking the winners is going to be hard.

If I could still fly, I would continue to do that and ride that in - happily even, flying is a great career and I miss it immensely. Real jobs suck and I cannot think of anything more satisfying, but yeah, it's going to be a lot different and I expect flying will be influenced a lot by AI. I see dispatch and ATC to be more likely to be disrupted in the relative near term, I guess with the staffing and workload that actually might be nice to the ATC people, but I don't really know enough about the ins-and-outs of ATC to be too doom-and-gloom. It's going to be a different world, and that's ok. I am 99.9% we'll always have pilots in the airplane, but I do suspect that for some transport category airplanes that will drop to only one. At the very minimum the rules around being an IRO will probably change and we might get a robot IRO or something similar. If I was going to try to automate away a pilot that is exactly where I would start.

The really cool applications I can think of are things like augmented reality for flying (which needs AI to identify things), ATC assistance tools, and a lot of other cool things. In a perfect world, we should use these tools to make flying safer, but that will take advocating for using the tools for good from the start.
 
Let's have AI replace low-skill workers like CEOs and VCs first, and we'll evaluate from there.
You may laugh, but I fully expect AI to take a huge chunk out of the mid-tier or higher managers very soon.

I spent several hours today writing code for this open source product my friend is working on to make an AI project manager... Marshall Brain's "Manna" short story is probably closer than we think.
 
You may laugh, but I fully expect AI to take a huge chunk out of the mid-tier or higher managers very soon.

I spent several hours today writing code for this open source product my friend is working on to make an AI project manager... Marshall Brain's "Manna" short story is probably closer than we think.

Really, I'm just patiently waiting for someone to explain capitalism's end goal to me.
 
capitalism has no intrinsic goals other than to seek equilibrium points in the market. the capitalists have goals, and all of their goals are to become monopolists.
Work hard, Make money. What is your solution? Do you like communism, because that's been tried repeatedly and always ends up in totalitarianism. What we are currently seeing is meritocracy being stripped away as a founding principle. Smart people that work really hard should be rewarded not canceled. But you've also said you think there should be open borders, how's that working out?
 
Last edited:
I fail to see where/ how AI will aid in maintenance, true hands on maintenance. Is ChatGPT going to crawl up into a center section with a torque wrench, insert the SD card for a 28 day database update, or leak check a manifold, pack wheel bearings in its palm.

Maybe upstream there are ways to utilize it with respect to writing manuals, or development of procedures, but to the technician that’s getting their hands dirty on the hangar floor or out on a handstand doing line maintenance, I just don’t see it.
 
I fail to see where/ how AI will aid in maintenance, true hands on maintenance. Is ChatGPT going to crawl up into a center section with a torque wrench, insert the SD card for a 28 day database update, or leak check a manifold, pack wheel bearings in its palm.

Maybe upstream there are ways to utilize it with respect to writing manuals, or development of procedures, but to the technician that’s getting their hands dirty on the hangar floor or out on a handstand doing line maintenance, I just don’t see it.
AI could be transformational in troubleshooting thought.
 
I fail to see where/ how AI will aid in maintenance, true hands on maintenance.

One use case is allowing pilots to send a picture of a bald spot on a tire or a dent on the airframe to the AI program and it will tell them if it's within tolerances.
 
One use case is allowing pilots to send a picture of a bald spot on a tire or a dent on the airframe to the AI program and it will tell them if it's within tolerances.

That I could see working, digital analysis of components to determine within limits. Although with how sensitive tire damage is given the Concorde and Nigeria 2120, I don’t know if we’re there yet.
 
I fail to see where/ how AI will aid in maintenance, true hands on maintenance. Is ChatGPT going to crawl up into a center section with a torque wrench, insert the SD card for a 28 day database update, or leak check a manifold, pack wheel bearings in its palm.

Maybe upstream there are ways to utilize it with respect to writing manuals, or development of procedures, but to the technician that’s getting their hands dirty on the hangar floor or out on a handstand doing line maintenance, I just don’t see it.
Troubleshooting is a big one.

I see something like this being the case for difficult to troubleshoot weird intermittent problems: "Hi there Mechanic Bob, I've looked at the current statistics and 92.5% of similar squawks in my database were related to the primary buffer panel (see figure 7-8). In fact, one PIC even said, 'I can't believe that the primary buffer panel has fallen off of my Gorram ship!' shortly before similar electrical issues started to occur. Since all the problems recently have been seemingly unrelated, that is statistically the best choice; there is one wiring harness and cable (see figure 9-4) that goes through the primary buffer panel that connects to most of the systems that have been recently written up multiple times. Maybe you should check to see if it's worn out?"

Another thing I could see is walking you through a repair you haven't done in awhile. "Nice to see you again, ok, I know you haven't done this in a while, but do you need help remembering what to do first? I can give you a checklist whenever you like!"

Mostly, I think what you'll end up doing will be largely the same, but the access you have to information be increased amazingly.

One use case is allowing pilots to send a picture of a bald spot on a tire or a dent on the airframe to the AI program and it will tell them if it's within tolerances.
This is a fantastic use case, there's a lot of stuff like that that could be similar.
Can you design an AI based system to design the AI based system you are designing?
Some really smart people are working on this, and while I am not clever enough to do that, I routinely use AI to write boilerplate code and to help me refactor code I've used to generate other AI stuff.
That I could see working, digital analysis of components to determine within limits. Although with how sensitive tire damage is given the Concorde and Nigeria 2120, I don’t know if we’re there yet.
Most radical case would be where the AI has looked at data for the type of airplane you're operating and can give you better analytics that can help you reduce down time early:

"Hi Mr Director of Maintenance. The N12345 is not scheduled to have <a weird component> replaced for another 50 hours, however, when I consult my whole fleet records, 99.6% of them have failed within 200 hours or more before they were scheduled to be replaced. You should have a technician check it or replace it early on night shift soon, that's possible on the weekend of 5-18. That should save you AOG time as that aircraft is not scheduled to fly the following morning if you find bigger problems.

Looking at next weeks schedule and the typical scenarios when other airlines have had to squawk that component, that airplane is likely to be at an outstation where we have contract maintenance. Doing that repair at an outstation would likely be twice as expensive as doing it this weekend."

I know mechanics are always doing that sort of calculus in their mind, but imagine having a polite old-timer advice guy who'd seen everything 100 times and while he was too old to reliably turn wrenches, he knew practically everything about the airplane and would patiently walk a mechanic through the most complicated repairs and could also advise management on the best times to do maintenance, etc.
 
Last edited:
Troubleshooting is a big one.

I see something like this being the case for difficult to troubleshoot weird intermittent problems: "Hi there Mechanic Bob, I've looked at the current statistics and 92.5% of similar squawks in my database were related to the primary buffer panel (see figure 7-8). In fact, one PIC even said, 'I can't believe that the primary buffer panel has fallen off of my Gorram ship!' shortly before similar electrical issues started to occur. Since all the problems recently have been seemingly unrelated, that is statistically the best choice; there is one wiring harness and cable (see figure 9-4) that goes through the primary buffer panel that connects to most of the systems that have been recently written up multiple times. Maybe you should check to see if it's worn out?"

Another thing I could see is walking you through a repair you haven't done in awhile. "Nice to see you again, ok, I know you haven't done this in a while, but do you need help remembering what to do first? I can give you a checklist whenever you like!"

Mostly, I think what you'll end up doing will be largely the same, but the access you have to information be increased amazingly.


This is a fantastic use case, there's a lot of stuff like that that could be similar.

Some really smart people are working on this, and while I am not clever enough to do that, I routinely use AI to write boilerplate code and to help me refactor code I've used to generate other AI stuff.

Most radical case would be where the AI has looked at data for the type of airplane you're operating and can give you better analytics that can help you reduce down time early:

"Hi Mr Director of Maintenance. The N12345 is not scheduled to have <a weird component> replaced for another 50 hours, however, when I consult my whole fleet records, 99.6% of them have failed within 200 hours or more before they were scheduled to be replaced. You should have a technician check it or replace it early on night shift soon, that's possible on the weekend of 5-18. That should save you AOG time as that aircraft is not scheduled to fly the following morning if you find bigger problems.

Looking at next weeks schedule and the typical scenarios when other airlines have had to squawk that component, that airplane is likely to be at an outstation where we have contract maintenance. Doing that repair at an outstation would likely be twice as expensive as doing it this weekend."

I know mechanics are always doing that sort of calculus in their mind, but imagine having a polite old-timer advice guy who'd seen everything 100 times and while he was too old to reliably turn wrenches, he knew practically everything about the airplane and would patiently walk a mechanic through the most complicated repairs and could also advise management on the best times to do maintenance, etc.
Have you ever worked on or managed a fleet of airplanes? A lot of your great ideas already exist and are not only used, but depended on daily. We already have what you're talking about, including reminders about upcoming inspections and alerts about the critical dispatch component list (it's a list the manufacturer publishes that lists which components have caused canceled flights fleetwide and the pertinent details). Like @CFI A&P said there is no replacement for the people on the floor, I can get a lot more info from 10 minutes on the floor than 2 hours online or on the phone. I've told the story of pulling a G-IV PDB and finding a 1/4' drive 1/4" socket floating around inside it, how would AI anticipate that failure? If we had taken the time to tell AI what the issue was, it would've taken the same amount of time as it did to look at the WDM and come to the same conclusion, something's wonky in the PDB and we better pull it out and have a look. The other part is the human factor, if you start to depend on AI to do your critical thinking for you your mind will atrophy. I hate inspections and routine MX, I love troubleshooting.
 
Last edited:
Back
Top