wing levelers

In article .com,
Predictor wrote:
Ernest Christley wrote:
"Dan, did you ever get a chance to work with fuzzy logic?"

Pete Schaefer responded:
"I have. It's been a while. I'd never, ever use it on an airplane.
Maybe a washing machine controller or something like that. I'm not
saying that it can't be done, but just that it's a risky design
approach."


Why is fuzzy logic "risky"?

BECAUSE it _is_ fuzzy. grin

Seriously, the nature of fuzzy logic is that it it _not_ deterministic.
Given a specific set of inputs, you cannot predict exactly what the
fuzzy logic will do for every occurrence of those inputs.

Identifying and analyzing "boundary" conditions in fuzzy logic is
"difficult, at best". _at_ a boundary condition, there is no telling
how far back one must trace to find the 'bias' that changes the output,
when all 'intermediate' inputs are identical. Even worse, the decision
may be based on 'noise' in the system.

This like this can lead to "unexpected" behavior in "unusual" circumstances.



Pete Schaefer continues:
"There are other methodologies much better suited for aircraft."


Why are other methodologies "much better suited for aircraft"?


Because they're more "predictable". see above.

You *really* want to be able to predict what the control system will
do, under every possible combination of inputs.

While 'strange things' may happen, at least you can rely on the fact
that "given the same circumstances again", the *same* "strange thing"
will happen.

..

Robert Bonomi wrote:

In article .com,
Predictor wrote:

Ernest Christley wrote:
"Dan, did you ever get a chance to work with fuzzy logic?"

Pete Schaefer responded:
"I have. It's been a while. I'd never, ever use it on an airplane.
Maybe a washing machine controller or something like that. I'm not
saying that it can't be done, but just that it's a risky design
approach."


Why is fuzzy logic "risky"?


BECAUSE it _is_ fuzzy. grin

Seriously, the nature of fuzzy logic is that it it _not_ deterministic.
Given a specific set of inputs, you cannot predict exactly what the
fuzzy logic will do for every occurrence of those inputs.

Fuzzy logic is deterministic. The rules are well defined, there is no
random number generator in any fuzzy logic implementation that I've seen.


Identifying and analyzing "boundary" conditions in fuzzy logic is
"difficult, at best". _at_ a boundary condition, there is no telling
how far back one must trace to find the 'bias' that changes the output,
when all 'intermediate' inputs are identical. Even worse, the decision
may be based on 'noise' in the system.

Noise will make any system behave randomly, but that is because you are
providing random inputs. Fuzzy logic is no different than PID or any
other control algorithm in this regard. However, the identical inputs
will produce the same outputs if the software is designed correctly.


This like this can lead to "unexpected" behavior in "unusual" circumstances.

Baloney.

Matt

Ernest Christley wrote:
"Dan, did you ever get a chance to work with fuzzy logic?"


Pete Schaefer responded:
"I have. It's been a while. I'd never, ever use it on an airplane.
Maybe a washing machine controller or something like that. I'm not
saying that it can't be done, but just that it's a risky design
approach."


Predictor asked:
"Why is fuzzy logic "risky"?"


Robert Bonomi answered:
"BECAUSE it _is_ fuzzy. grin

Seriously, the nature of fuzzy logic is that it it _not_ deterministic.
Given a specific set of inputs, you cannot predict exactly what the
fuzzy logic will do for every occurrence of those inputs."


Fuzzy logic is completely deterministic. Given the same inputs, fuzzy
systems will produce exactly the same outputs.

Here are links to some reasonably good introductory material on fuzzy
logic:

http://www.austinlinks.com/Fuzzy/overview.html
http://www.ncst.ernet.in/education/a...nt/fuzzy/fuzzy...
http://www.fpk.tu-berlin.de/~anderl/...uzzyintro4.pdf
http://www.phys.ufl.edu/docs/matlab/...y/fuzzy_tb.pdf
http://www.faqs.org/faqs/fuzzy-logic/part1/
http://www.fuzzy-logic.com/ch3.htm


-Will Dwinnell
http://will.dwinnell.com

"Predictor" wrote in message
oups.com...
Ernest Christley wrote:
"Dan, did you ever get a chance to work with fuzzy logic?"


Why overcomplicate things. This is just too easy an application with a
BasicX board and an accerleometer.

http://www.basicx.com/
http://www.analog.com/en/prod/0%2C28...L203%2C00.html

Couple this with a cheap GPS, and big servos modelled after the ones used in
RC aircraft (the BasicX computer is designed to drive them directly), and a
complete autopilot is an easy design project.

I;ve been playing with this for a while, but I'll admit i've spent more time
thinking than soldering.

Anybody seriously interested in teaming up?



Ron Webb..

Ron Webb wrote:

"Predictor" wrote in message
oups.com...

Ernest Christley wrote:
"Dan, did you ever get a chance to work with fuzzy logic?"



Why overcomplicate things. This is just too easy an application with a
BasicX board and an accerleometer.

http://www.basicx.com/
http://www.analog.com/en/prod/0%2C28...L203%2C00.html

Couple this with a cheap GPS, and big servos modelled after the ones used in
RC aircraft (the BasicX computer is designed to drive them directly), and a
complete autopilot is an easy design project.

I;ve been playing with this for a while, but I'll admit i've spent more time
thinking than soldering.

Anybody seriously interested in teaming up?



Ron Webb..


Before anyone rushes into anything take if from someone who has worked
AFCS, stab aug and the like. I have been in rotor and fixed wing
aircraft when the systems have gone nuts during inflight ops checks.
Fortunately the systems could easily be overpowered manually and
disconnected either by a button on the stick grip.

In any event all the systems had a ton of engineering and flight testing
before being accepted for service.

On the other hand some of the nutso systems can be quite amusing in
hindsight. I had an H-3s aug system problem I couldn't duplicate on the
ground. I requested a flight. They sent me a pilot who had a habit of
placing his coffee cup on the floorboard near his foot. We hovered, he
engaged system which promptly went crazy. I'm standing in the back
looking over the FE's shoulder, holding on for dear life and collecting
dings and dents in my helmet. The pilot calmly reached down, picked up
his cup of coffee, took a swig, put it down and then disengaged. After
we disembarked I asked him why he had done that. He said something about
enjoying the ride. That was about 20 years ago. I still think it was a
bit funny

Dan, U.S. Air Force, retired

"Dan, U.S. Air Force, retired" wrote in message
news:wtJWd.24485

In any event all the systems had a ton of engineering and flight testing
before being accepted for service.

Dan

I agree 100% that any amateur designed auto pilot would be a risky
proposition. That's one reason I'm still in the parts accumulation stage 2
years after I started thinking about it. The part I'm having the most
trouble finding is a couple of big brass BALLS ;^}

Control systems engineering is perhaps THE most feared course in any
undergrad EE program, and for good reason. It's not an easy subject.

Having said that, I think it can be done. My own precautions will include:


1) The servos are coupled to the controls with springs - easily overpowered
by the manual controls which are left in place. And of course a big red
"DISENGAGE" button on the stick.

2) Thorough testing using X-Plane simulator. I'll use a custom designed
interface between the autopilot and the simulator.. That's one reason I'd
want more than one person on the project - the simulator test "pilot" should
be independant of the designer.

3) All control equasions will be radically over-damped. No "Auto-Pilot
Induced Oscillations".

4) The pitch control will only be controlling the trim tab.

On Sun, 6 Mar 2005, Ron Webb wrote:


Why overcomplicate things. This is just too easy an application with a
BasicX board and an accerleometer.
snipped
Anybody seriously interested in teaming up?


I started one, but then found a Navaid too cheap to pass up.
I bought the BX24 development kit, and had a great time programming
it to read the autopilot signal from my Terra, and then the NEMA code
from my GPS, but I got the navaid before buying a gyro.

Many pilots are upgrading the Navaid to a Trio solid state unit.

I can tell you that an acceleation chip reads nothing in the air.

There is some freeware for an altitude hold.

George Graham
RX-7 Powered Graham-EZ, N4449E
Homepage http://bfn.org/~ca266

Ron Webb wrote:
"Predictor" wrote in message
oups.com...

Ernest Christley wrote:
"Dan, did you ever get a chance to work with fuzzy logic?"



Why overcomplicate things. This is just too easy an application with a
BasicX board and an accerleometer.

Actually, its not that simple. Trust me, I worked on a similar problem
for 2 years. How can you tell whether you are straight and level? If
you're in a balanced turn, your accelerometer (which actually measures
net force) will believe you are straight and level.

And a GPS won't help much at all, mostly because its response rate is
too slow.

Frank

How can you tell whether you are straight and level? If
you're in a balanced turn, your accelerometer (which actually measures net
force) will believe you are straight and level.

And a GPS won't help much at all, mostly because its response rate is too
slow.

Frank

As I see it, accelerometers and GPS together are all that's needed. If the
vectors from all 3 accelerometers are in the right direction, and the
average GPS heading is not moving much- you're straight and level.

My GPS updates about once per second. That's plenty fast enough.Once every
10 seconds would work.When flying, do you react to every bounce, or just
ride with the flow and provide general guidance? I don't know about you, but
I try to stay relaxed.

The GPS does too make the differance between a really sticky problem and a
slam dunk.
I'm betting your practical experience was before $50 GPS and $3
accelerometers? (say, 5 years ago). Am I right?




Ron Webb

I'll take a few of those $3 accelerometers! Where may one find them,
please?

Thanks.


Ron Webb wrote:
How can you tell whether you are straight and level? If
you're in a balanced turn, your accelerometer (which actually measures net
force) will believe you are straight and level.

And a GPS won't help much at all, mostly because its response rate is too
slow.

Frank


As I see it, accelerometers and GPS together are all that's needed. If the
vectors from all 3 accelerometers are in the right direction, and the
average GPS heading is not moving much- you're straight and level.

My GPS updates about once per second. That's plenty fast enough.Once every
10 seconds would work.When flying, do you react to every bounce, or just
ride with the flow and provide general guidance? I don't know about you, but
I try to stay relaxed.

The GPS does too make the differance between a really sticky problem and a
slam dunk.
I'm betting your practical experience was before $50 GPS and $3
accelerometers? (say, 5 years ago). Am I right?




Ron Webb