Not in my lifetime; probably not in yours
Consider how much effort it takes to get a machine working, with some sort of operator interface attached.
Now double that, and you'll have the effort needed to get some sort of AR functional (not necessarily good) for the same machine.
I did a project 20 years ago where there was a packaging machine with ~100 photoeyes and prox switches, any one of which could hold up the boxing sequence if it failed to read something that was
ALMOST in place, or if the sensor got knocked a smidge out of alignment.
I found a "artificial intelligence" program (actually an "expert system" that could ask Yes/No questions, and connect to other systems via DDE), and came up with a methodology to reverse-code the PLC ladder logic into the AI's structure, to figure out "Why has the machine stopped?". When it came up with the answer, it would show a page with the troublesome sensor identified, a layout of the machine where the sensor where it was located, allowed the user to add notes, or say, "No, that's not it, keep trying".
I didn't see the project to completion (layoffs), but it was hitting ~80% accuracy on the first hit, and not all the PLC program had yet been entered.
It was a fun, cool project. But the only reason that it was undertaken was that it required knowledge of the machine that wasn't apparent. If you were to try to troubleshoot it, you might think that the reason that the machine had stopped was that it was waiting for this prox switch to be made. But that prox switch (DI) requires that a solenoid (DO) fire, and it hadn't because it was waiting for something else, which was in turn waiting for something else. Without knowing the relationship between the inputs and outputs, an average sparky would have no chance.
So the ROI was there. Most companies / machines don't have that ROI, and management doesn't care if it takes them a little longer to use their brains to figure out something, rather than building an AR app for a specialized machine whose life-span is only 5-7 years anyway. The maintenance guy is getting paid the same with or without the app, so how do you justify building it?
Consider how slowly "situational awareness" graphics are coming into the factory floor. And that's for
normal operation to alert the operator to signs of trouble (i.e., $$$ loss), not for ABnormal operation, which is what the maintenance guy is dealing with.
The QR code thing that mk42 talked about is cheap and easy to implement: the app probably already exists in the Apple Store; the PDFs came from the vendor. AR, not so much. The way you are describing it, it needs to know where in the machine you are. Half the time, I can't even get phone service in the factory, let alone GPS. And having the app recognize which part is which in it?
https://xkcd.com/1425/