Flight Management

[EconomyClass]

Well, getting back to what my original thought was, you manage elements with the idea of achieving an outcome. Or you can manage by operating controls, attempting to hit targets. I think management science calls it "management by objectives". Define objectives, then change what is controllable and measure. The flight management system designers probably have an architecture that emulates one or another management philosophy. I'm just wondering which. Or maybe they were forced to make some sort of mix. I think obviously the pilots manage by objectives. They have a bottom line which is a safe flight and landing. But they also have to have mileposts to achieve in that pursuit.

[T.O.G.A.]

I'm not sure what the difference is between the two forms of management.

There is a goal, and targets to reach the goal. Those are dialed in. The FMS, in conjuction with the autopilot, manage the variables encountered along the way with live information from the air data computers, inertial reference units, lateral and longitudinal accelerometers, feedback transducers on the control surfaces as well as the navigation equipment.

I would even go as far as calling it artifical intelligence.

For full autoland systems, there are three of everything and they cross talk. They have to agree that they are recieving th same information and if one system disagrees, the autoland system degrades or disengages.

Hope this helps. Tell me if the fills one form of management or is a hybrid of both.

[Gabriel]

Management?

It's Control Theory. The most difficult subject I had in the university.

In the second year we had Math Analysis IV. Gauss transformate, Laplace transformate, vectorial second-order differential equation with non-linear coefficient and non-zero independent term, and other spooky mathematical articrafts. And all the students wondering: why on Earth would we want to know all these surreal things?

In the third year we had Rational Mechanics. It was a strange subject. Basically, the same than Physics I: Mechanics of the particle, of the rigid body, and of oscillating systems. The difference was that the particles, bodies and systems were much more complex than the box down the slope that we used in Physics I. Everything was still F=m.a and M=I.alpha, except that we had things like bound trajectories, variable forces, variable masses, and the tensor of inertia instead of just the moment of inertia. We used things like Euler angles, Lagrange free-space physics, non-euclidean geometry... In the end, the problem was reduced to 1) Make a mathematical model of the system and what we got was: a vectorial second order differential equation with non-linear coefficients and non-zero independent term, and b) forget about Physics and solve the Math (which was the most complex part). Again, where on Earth would we use all that?

Then in the 5th year we had Automatic Control and Guiding Systems. And we said Ahhhh... So this is why we saw what we saw 2 and 3 years earlier. We even had to learn more mathematics specific for this subject, like the Z transformate for digital control.

Just to name some titles within the control philosophies you have:
Classic control
Open-loop control
Closed-loop control
Modern control
Vectorial control
Robust control
Adaptive control
Learning control
Fuzzy control
Detectability
Observability
Stability
Controllability

If you thought that some of my threads regarding aerodynamics, performance and Physics were long and complex, you never saw one of control theory. And you will not, not from me at least, because actually I didn't manage to get the level of understanding that I got with the other subject.

Just let me tell you that it is a VERY COMPLEX ISSUE.

But, EC, let me give you one example:

You are driving a car. You have to brake somehow strongly but no need to apply full brakes. For example, the traffic light turned red and you are close to the crossing and at a good speed.

What do you do?
a) Start to apply increasing amounts of brakes until you get the desired level of desceleration, and then make small corrections to keep that desceleration at target? Or
b) You know how to drive, so you know how much brake to apply, so you just apply that much amount of brake and close your eyes knowing that you'll stop exactly where you want?

I don't know you, but I apply a combination of both. I don't just start investigating the right amount of brake needed, and I don't apply what I know will do it and close my eyes. I apply what I know is needed, and then make adjustments as needed.

Now, this is a VERY simple system. Your target is an acceleration and said acceleration is simply proportional to the application of brakes. You are going at 50MPH with no brakes and the gear in neutral and the acceleration is nearly zero (there is some due to the drag). You apply one amount of brakes and you get that acceleration, and that acceleration remains until you change your brake setting. You release the brake, again zero acceleration and constant speed.

Take an airplane flying straight and level with just enough thrust to keep the speed constant. Apply one elevator setting (different from the one you have) and what you get is a dual-mode oscillation with oscillating changes in angle of attack, pitch, load factor, vertical speed, altitude and airspeed. Return the elevator to the original position and everything stabilizers, tight? Wrong. You get another oscillation similar to the previous one. So, you see, you don't have a direct relationship between elevator setting and any parameter that you might want to control with it (angle of attack, pitch, load factor, vertical speed, you name it) like the brake-acceleration relationship. The elavator has to be modulated along the maneuver.

All the autopilots know the dynamics of the airplane that they are flying. It's part of their program. So the autopilot (or the flight control computers for FBW planes) already know how much input is needed along the maneuver to achieve the objective, and of course it gets feedback from the sensors to make adjustments and not blindly follow the predicted input needed. The mass and CG is part of that dynamic. But how does it work if the mass and CG are not fixed? There are different approaches. The first one, which was and still is used with great success, is robust control. Robust control means that the system is still controllable if there are differences between the model and the reality. So the system is modeled with a "standard" mass and CG and the system is robust to differences between the modeled values and the real ones, but those differences must remain within a range. Now, modern jet can have a MTOW that more than doubles the empty weight, and fly at cruise at speeds that are 4 times the landing speed and Mach numbers that approach very much to the sound speed and make major changes in the aerodynamic behavior of the airplane. So here simple robust control is not enough and adaptive control comes into play. In adaptive control, parameters of the system that are not the ones that you want to control are part of the input, and the system model changes based on that input. Things like airspeed, density altitude and Mach number have been part of these inputs for a long time. Not so mass and CG, but then learning control comes int play. With the standard mass and CG programmed in the system model, the control system applies a given control input "knowing" what the result will be, except it's not so it reverse-engineer the calculation to deduct what the mass and CG must be to get the result it got, and then inputs that info in the system model (as seen in the adaptive system). Of course, if the automation already knew what the mass and CG really are, it wouldn't have to play "trial and error" and would achieve the expected result more efficiently. Is this being done? As I said in my second post here, I really don't know but I suspect that, for some things, it is.

But please understand that programming an automation saying "you have these controls, you have these variables, and you have these targets for them, now figure out how to make it work" doesn't work. It can be done, but the system may brake a few planes in the learning process.

[3WE]

While Gabriel goes complicated (because it is complicated), I'm thinking Economizer was thinking more big picture...is some computer looking at fuel load and doing crazy complicated calculations as to how the plane needs to be controlled.

I don't think that's the case.

Instead Otto is flying the plane...period (ok, fundamentally period).

Now, is something shuffling fuel and analyzing the effect on WB- yeah, sure, why not...But it's "goal" is not to "control the plane".

It's goal is to keep the fuel load balanced.

And....with the plane being balanced, then Otto deals with wind gusts and the really fat passenger who goes from his bulkhead seat back to the Lav versus the hostie and the drink cart which redistributes soda weight from the front galley throughout the cabin.

As to whether computers should manage fuel- the one key difference is that fuel transfer is nowhere near the split second deal of controlling the plane...so heck yeah- have some "gauges" and some off switches for the pilots to "monitor" and over ride if Kerosene HAL gets his own ideas, but so what if HAL goes crazy for a few minutes there.

Now as to what HAL does during a low pass and if he gives off an whole different vibe as you TRY to power up while airspeed is decaying and the plane is settling into trees....

[Gabriel]

While Gabriel goes complicated (because it is complicated), I'm thinking Economizer was thinking more big picture...is some computer looking at fuel load and doing crazy complicated calculations as to how the plane needs to be controlled.

I don't think that's the case.

EC said:

I wasn't specifically referring to autopilot. If the plane was on autopilot, it would have to be running some sort of routine to keep course and altitude. But if the software stands between the pilot and the controls, then it would have to interpret whatever the pilot does to achieve the same result as autopilot. That is, not just literally do whatever the pilot tells it to do or why have software at all?).

We are running in circles here.
I've already said that I expect the modern flight control computers that control the FBW systems to take into account the actual known weight and CG of the plane to compute an initial target control input, and then monitors the result to "fine tune" said control input. If the information is there, why not use it? I don't even mean "depend on it".

[EconomyClass]

Haha. Sorry if the subject is very, very deep. But going back to Gabriel's braking analogy. I know my brain is processing what my body feels and what my eyes see. Amazing programming is running. And since I've been driving since 1961, I've adjusted it to cars varying in weight and braking by at least 50 percent. In the end, the process becomes simply. Foot goes on brake. Eyes observe distance from me to the intersection or the car ahead of me. AUTOMATICALLY, my foot responds to brain signals to push harder or softer. This is management by objective. In any given car, any given weather situation. I need to brake once or twice, and then from then on, the braking management system does a near perfect job. So what I'm saying is that the artificial intelligence of the flight management system has to digest all the inputs at a given moment and then begin gradual approximation of the perfect set of control actions to achieve the goals it has, whether it is staying on course, maintaining speed and altitude. And when the pilot knows it is time to CHANGE those goals (the control tower says there is a delay to clearing for landing?), pilot feeds the goal change in and then the management system begins resetting control for the new goals. Anyone who has ever worked in a corporation has LIVED the experience of this management. No corporation ever pursues exactly the same goals from year to year. So for decades, I had to change how I worked at these businesses as the pilots in senior management adjusted company goals.

[Gabriel]

Haha. Sorry if the subject is very, very deep. But going back to Gabriel's braking analogy. I know my brain is processing what my body feels and what my eyes see. Amazing programming is running. And since I've been driving since 1961, I've adjusted it to cars varying in weight and braking by at least 50 percent. In the end, the process becomes simply. Foot goes on brake. Eyes observe distance from me to the intersection or the car ahead of me. AUTOMATICALLY, my foot responds to brain signals to push harder or softer. This is management by objective. In any given car, any given weather situation. I need to brake once or twice, and then from then on, the braking management system does a near perfect job. So what I'm saying is that the artificial intelligence of the flight management system has to digest all the inputs at a given moment and then begin gradual approximation of the perfect set of control actions to achieve the goals it has.

Again: Braking? 1-dimension problem (just a liner motion). One input (brake pedal input). One output (desceleration). Magnitude of the output directly proportional to the magnitude of the input.

Airplane? 6 degrees of freedom (3 linear movements and 3 rotation axis), 4 control, inputs (elevator, aileron, rudder and throttles), a myriad of outputs (bank, pitch, yaw, AoA, sideslip, load factor, airspeed, vertical speed, trajectory) and all the inputs and outputs are closely coupled (think applying an elevator input while in a bank and you'll change ALL of the above outputs).

Unlike you and the cars, the automation in an airplane NEEDS TO KNOW BEFOREHAND the dynamic model of the plane (including, among others, mass, moments of inertia, CG, control authority) or have an approximated model and be robust to deviations from that model withing a range.

As I said:

But please understand that programming an automation saying "you have these controls, you have these variables, and you have these targets for them, now figure out how to make it work" doesn't work. It can be done, but the system may break a few planes in the learning process.

A human can do much better than an autopilot or flight control computer to learn how to fly when conditions departed from normal, and that's why in those cases the AP disengage itself and the control laws revert to lower-level modes like direct law where it doesn't "interpret" the input from the pilot anymore, it just blindly applies it. Actually, some times the human pilot cannot do much better and, when the AP quits, they crash. (Colgan, AF, the turboprop with icing that rolled inverted...).

[EconomyClass]

Assuming since pilots still land and take off, the belief is that the airline would rather trust the human. As for driving a car, braking might be the simplest problem. There are many many other complex problems, and, so, cars do not yet drive themselves. One really complex problem is being in the same plane with an incalculable quantity of idiots. Airplanes are given a box in which to fly. No driver ever gets one. We all try to make space in front, but behind? No such luck. On either side? No such luck. We get fractions of a second to avoid crashes. Lucky a sober brain can actually take in all the variables and make a suitable maneuver. Mostly.

[brianw999]

Actually....cars do drive themselves, at least for the parking part.... http://m.youtube.com/watch?

And the totally self driving car is not far away..... http://en.m.wikipedia.org/wiki/Google_driverless_car

Bit scary really ?

[EconomyClass]

Actually....cars do drive themselves, at least for the parking part.... http://m.youtube.com/watch?

And the totally self driving car is not far away..... http://en.m.wikipedia.org/wiki/Google_driverless_car

Bit scary really ?

Getting electronics to guide a car is by far the easy part. Handling the unpredictable and totally unskilled people in other cars has too many unknowns to factor in any sort of software. I think all a program could do is FORCE the vehicle to obey fundamental rules of safety. Beyond that, forget it. Personal rapid transit will be the only personal transport that will ever work. Computerized cars on a paved motorway is a nonstarter.

[brianw999]

Getting electronics to guide a car is by far the easy part. Handling the unpredictable and totally unskilled people in other cars has too many unknowns to factor in any sort of software. I think all a program could do is FORCE the vehicle to obey fundamental rules of safety. Beyond that, forget it. Personal rapid transit will be the only personal transport that will ever work. Computerized cars on a paved motorway is a nonstarter.

.....and it was once said in the first days of motoring that the human body could not withstand more than 20mph. How fast does a Bugatti Veyron go ? I truly foresee the day when the automobile, in whatever form it develops into becomes nothing more than an automatically speed regulated means of transport programmed to a specific route.

[Gabriel]

Getting electronics to guide a car is by far the easy part. Handling the unpredictable and totally unskilled people in other cars has too many unknowns to factor in any sort of software. I think all a program could do is FORCE the vehicle to obey fundamental rules of safety. Beyond that, forget it.

And yet, they are doing quite well with the sudden situations too.



Of course, this needs tons of electronics, sensors, programming and, I bet, a fairly good model of the car dynamics rather than a "let's play trial an error" control inputs seeking for a goal.

[brianw999]

From the Wikipedia report on the Google cars. Two accidents in 700,000 test miles on public roads, both of which were the fault of humans...

In August 2012, the team announced that they have completed over 300,000 autonomous-driving miles (500,000 km) accident-free, typically have about a dozen cars on the road at any given time, and are starting to test them with single drivers instead of in pairs.[21] Four U.S. states have passed laws permitting autonomous cars as of December 2013: Nevada, Florida, California, and Michigan.[22] A law proposed in Texas would establish criteria for allowing "autonomous motor vehicles".[23][24]

In April 2014, the team announced that their vehicles have now logged nearly 700,000 autonomous miles (1.1 million km).[25]

In late May 2014, Google revealed a new prototype of its driverless car, which had no steering wheel, gas pedal, or brakes, being 100% autonomous. [26]

Incidents[edit]
In August 2011, a human-controlled Google driverless car was involved in a crash near Google headquarters in Mountain View, CA. Google has stated that the car was being driven manually at the time of the accident.[27] A previous incident involved a Google driverless car being rear-ended while stopped at a traffic light.[28] Google says that neither of these incidents were the fault of Google's car but the fault of other humans operating the cars.

[EconomyClass]

There's no way to judge the applicability of these "tests" to the real world. For example, what is the overall accident rate where the cars drive? I think the Catch-22 here is that if ALL cars are auto-driven, then you have the safest situation. But getting from the mode where one vehicle is machine-controlled to one where all are is the messy part. I'm quite sure the cybernetic car is by far the more intelligent and predictable vehicle. But it'll probably be like a Tesla. A toy for celebrities. Meanwhile the drinking, texting, weaving majority will continue and threaten anything cutting edge that ventures onto the roads. I'd expect the planes to be first to switch. Not that I'd be all that thrilled to BOARD an airliner totally run by computer. You know the saying: To err is human. To totally screw things up requires an advanced computer (we Americans just went through that with our healthcare enrollment system)

[Gabriel]

EC, you are right:

- Flight control computers have no clue of what a plane is or what control input will be needed to achieve a target, and instead play with the controls by successive approximations of trial and error until the find the combination that makes the variables meet the target.

- Self-driven cars are not tested in real-life traffic (which was not advised that a self-driven car was among them) and have not logged 700,000 miles with only one accident where they were rear-ended while stopped in a traffic light by a human-driven car.

My only question is...

Why do you come here and ask?