the avionics are usually real, being fed dummy data from a software model with regards to the plane.

Here's a PDF about it.[0]

Training simulators are typically convertible, meaning that the flight characteristics are entirely fluid and made by the simulator using physics models and data provided by the airplane manufacturer. This makes it possible to train multiple platforms on a single simulator.

I don't know how corporate simulators work. In automotive fields they utilize all OE automotive hardware and the simulator is only in charge of feeding data to the automotive systems. I would hope that it's similar for corporate plane simulators -- a real plane ECM/brain and accompanying systems being fed dummy data.

I doubt the corporate simulators are at all convertible -- they're likely brain-in-a-jar simulators; planes without engines or hydraulics, being fed dummy data.

[1]: https://www.faa.gov/about/initiatives/nsp/ac/media/AC_120-40...

If you're looking at the highest fidelity level D simulators, the instruments and controls in the cockpit are either the same parts as the aircraft, or functionally identical (but cheaper).

> If you simulated a broken AoA sensor, would the simulated plane behave similar to the real plane? Would the MCAS system have the same bugs in the simulator as in the one in the real aircraft?

One of the big costs in building a simulator is buying the data package from the aircraft manufacturer, with the aero model and details of system internals, things like electrical and hydraulic schematics. Sim makers build a software model of these internals at a pretty low level. For the most part, if you introduce a fault in some part of the system it will behave the right way as an emerging property, not because you're forcing the system to have the right outputs.

Some software components from the aircraft get installed on the simulator with the same hardware platform from the aircraft, others get run as executables on the simulator's computers, and others get re-implemented from scratch (lots of FORTRAN and C).

That kind of detail comes into play when the instructors introduce multiple failures at the same time - pilots have to take corrective actions to make the faults go away or manage them - if you don't model the systems at a pretty low level you'll never high fidelity.

> Can you try new scenarios in a simulator, or can you just try scenarios that the simulator was designed to run?

There is a list of malfunctions available to the instructor, who runs the session from the back of the "cockpit" on touch panels. For the most part, these malfunctions cover failures that are anticipated by the aircraft manufacturers, and the corrective actions / system behavior are well understood. Each fault is tested to make sure it works properly. You don't go and fail some random component in the system.

When an important failure happens in the real aircraft, it might get added as a training scenario to simulators already in operation.

I always wondered if they did that, something akin to fuzzing tests in SW. Wouldn't it be useful to detect unexpected situations that'd be catastrophic? Or the benefits from it wouldn't outweigh the cost/time loss?

Also, imagine you're an airline with thousands of pilots and dozens of instructors: you're running an airline and a school at the same time. You need to build a curriculum of training and testing that will standardize your pilots. There's room for thinking outside the box but not too much.

Well Boeing certainly doesn't fuzz their software as evidenced by the major bugs in the 737 NG and 747-400's flight displays. Both had bugs that would black out all instruments under specific conditions. That got fixed fairly quickly on the 747, but apparently Boeing didn't learn their lesson with the NG.

The bug blacked out all six display units. What other instruments are you thinking of?

On the other hand, if you're simply looking for some training hours on some specific basic scenario and/or aircraft, the simulators can be a lot rougher and still be certified. I "flew" on an airliner manufacturer designed simulator, and everything from the instrument panel to the hydraulics simulating the small impact when rolling between plates on a concrete runway felt pretty damn real.

For more basic stuff, even X-Plane exists in an FAA-approved version.

http://www.flyipilot.com/

”iPILOT is currently available in London, Munich, Berlin, Dresden, Hamburg, Düsseldorf, Prague, Doha, Dubai, Basel, Zurich”

I have a little background in this myself and the requirements to certify as Level C are tough l, but Level D requirements are quite stringent and rather hard to replicate. ( I did a detailed proposal a decade ago on using cutting edge VR and Haptics to try and cut the operational cost of maintaining multiple simulators for a flight training school. Level D was quite pedantic with respect to the “realism” requirements, we kept running into the requirement to not just look real, sound real, heck even feel real (switch actuation forces), but be real for that exact type of plane. Our argument was that we could reach sufficient fidelity with the haptics and graphics. This was all crazy expensive stuff, but ours would be cheaper crazy expensive stuff ;-)

To put it into perspective, in a couple of cases we found it was cheaper to build a Level D simulator by cutting off the entire front of a real second hand plane (finding one with an airframe issue that made it no longer airworthy was the dream) of the desired type, and then wiring up all the switches and displays and electrical stuff to the simulator driving hardware and mounting the entire thing on a huge platform. The cost to buy the plane, cut the front off, tap and splice existing cable harnesses stiffen the cockpit section and overbuild the motion platform was cheaper than reverse engineering the layout with sufficient documentation and wiring a “fake” cockpit with sufficient accuracy including sourcing all the correct parts with the associated paperwork to prove they were all correct.

It never went anywhere but it was very educational and has served as a useful perspective as I have observed the rise of the modern VR ecosystem with respect to input and haptics. :-)

https://www.youtube.com/watch?v=aoNOVlxJmow