Designing a Computer Simulation for Genetic Optimization of RC Aircraft

[philquiet]

Hello,
My project is to design a new type of flying RC toy, and since i am not an engineer and I don't know aerodynanic well, I am trying to craft some kind of aerodynamic simulator on my computer, then run some genetic programming to design from scratch any kind of small aircraft capable to fly in the simulated environnement.

Ok, that's not clear...
I mean:
-First, I would try to get any flight simulation software accurate enough to test a new virtual aircraft design (the kind than could have wings and propellers of any shape, in any configuration);
-Then run a search algorithm, certainly some form of genetic computing, to evolve a population of flying stuff. The simulation would be used to evaluate the candidates' flying abilities.
-Finally, make some blueprints from the best fit evolved critter.

Since I'm all but aerodynamic saavy, I'm stuck on the problem of the simulation. Any premade open-source stuff would be perfect, but if I'm going to do the simulator myself, what is the fields I'm the most likely to have to study?

Most important, do you have any suggestion for the design of the simulator or any part of my project?

 

[Cyrus]

err......my suggestion. Don't waste your time. Just build it and see how it goes. You do not have the know how to do what you propose, sorry.

It also just doesn't work as easily as you stated. Get a book on model airplane design as a starting point.

 

[philquiet]

I have at least some hints about how I would do this. I already own a starter book on plane modeling, and I have some useful hobyist knowledge in plane modeling, robotic, programming and AI.

Also, I know exactly how I would pull out the genetic algorithm part. That's not the hardest part. I also have some ideas for the blueprint.
My goal is not some king of conventionnal aircraft but something that is more like the Draganflyer helicopter or some hovering stuff like that.

And "wasting my time" is not quite a threat since the project is still somewhat hypotetical. I'm on the research part, and it's going well.

 

[Cyrus]

I have at least some hints about how I would do this. I already own a starter book on plane modeling, and I have some useful hobyist knowledge in plane modeling, robotic, programming and AI.

Having some hints is not enough to do what you want. You need to have a working understanding of aerodynamics, stability, and performance. A starter book on plane modeling isn't going to cut it.

My goal is not some king of conventionnal aircraft but something that is more like the Draganflyer helicopter or some hovering stuff like that.

And "wasting my time" is not quite a threat since the project is still somewhat hypotetical. I'm on the research part, and it's going well.

Again, just build one and see how it works. If its bad, build a new version. Eventually something will start to show promise.

 

[philquiet]

Okay, okay...
Anyway, thanks! Sorry of being some kind of stubborn smarta**, I understand the point. Though I'll maybe still try to evolve flying critter just for the fun of it, the whole point is that I'm in some kind of hobbyist fantasy here, with the illusion that I will be able to achieve this only with precise pratical DIY knowledge. Error and trial will do, I'm sure.
Thanks for replying! Anyway, are you in this stuff too? I'm wondering because of your avatar image.

 

[Cyrus]

Okay, okay...
Anyway, thanks! Sorry of being some kind of stubborn smarta**, I understand the point. Though I'll maybe still try to evolve flying critter just for the fun of it, the whole point is that I'm in some kind of hobbyist fantasy here, with the illusion that I will be able to achieve this only with precise pratical DIY knowledge. Error and trial will do, I'm sure.
Thanks for replying! Anyway, are you in this stuff too? I'm wondering because of your avatar image.

I do flight dynamics, simulation, and control. What you want to do, would take a team of PhDs. You could get better knowledge by just making it and recording your results.

 

[FredGarvin]

Just clarify, because this is a hot topic for me, computers don't design. People design. Computers are tools to help in that endeavor, but they are not there to help people who have no education in an area to design.

Educate yourself is the best way, trial and error is the next best.

 

[RolfRomeo]

A helpful bunch this was...

If I'm reading you correctly you want to do this mainly for fun and hobbyists interest? At least not bring the latest aerodynamic breakthrough to market?

Doing a full simulation of the aerodynamics of 1000's of (initially) randomly generated designs is going to take a while, and you are probably better off building and testing them, unless you have access to some serious digital muscle.

Now provided I was correct that this is a project for fun and personal satisfaction, how about hooking into a plane-simulator? Specifically X-Plane, as suggested http://www.jefflewis.net/x-plane_engineering.html".

If you do go along with this, keep us updated. I would love to know more about how you are going to define your "penalty"-function. (Basically what your design goals are.)

 

[Cyrus]

A helpful bunch this was...

If I'm reading you correctly you want to do this mainly for fun and hobbyists interest? At least not bring the latest aerodynamic breakthrough to market?

Doing a full simulation of the aerodynamics of 1000's of (initially) randomly generated designs is going to take a while, and you are probably better off building and testing them, unless you have access to some serious digital muscle.

Now provided I was correct that this is a project for fun and personal satisfaction, how about hooking into a plane-simulator? Specifically X-Plane, as suggested http://www.jefflewis.net/x-plane_engineering.html".

If you do go along with this, keep us updated. I would love to know more about how you are going to define your "penalty"-function. (Basically what your design goals are.)

X-plane is not a design tool, it does not tell you about the stability derivatives, nor does it give you the aerodynamic lift/drag/moment/side force/etc curves. This is what he needs to generate, and is what X-plane (or any simulator) needs to work. The point is that simply messing around with a computer program and not knowing what you are doing is going to produce garbage results. There are many approximations being made that he (and you) don't even appreciate, which can be completely eliminated by simply building a design and seeing how it works.

If you want my professional answer go read:

[1] Aircraft Design: A conceptual Approach by D.P. Raymer
[2] see the DATCOM database
[3] see Jan Roskams books on performance

If you get past page one of these sources, I'll be impressed.

 

[minger]

Just clarify, because this is a hot topic for me, computers don't design. People design. Computers are tools to help in that endeavor, but they are not there to help people who have no education in an area to design.

Educate yourself is the best way, trial and error is the next best.

My adviser once told me about a really bright guy who developed a code that could parametrically "design" airfoils. In theory his program could design the perfect airfoil. He ran it and the program spat out some toroidial looking debacle. Needless to say I've always tried to follow a rule that

numerics (FEA,CFD,etc) can make a good engineer a great engineer, but can make a bad engineer dangerous.

 

[RolfRomeo]

I don't want to bite, but I can't resist :)

You design airplanes for a living - I get it. You're also correct that I don't, and hopefully neither does the OP.
But I believe you're completely missing the point.

What was asked for was a simulator to use as part of an Evolutionary Algorithm. Now, I'm almost certain this is not how engineers design aircraft, but I guarantee you this is an approach that can potentially perform better than garbage, given enough clock-cycles. (Assuming the simulator is relatively accurate, and the genome and fitness of the population has been defined in a sane way.)

This is not engineering, this is evolution, and as such can come up with effective solutions 1000's of engineers with 1000's of sliderules could never dream of! ;)

So, in your professional opinion, do you know of any open-source/free aerodynamics simulators?
I don't know of any, but a search revealed the link above. Now I understand that X-Plane is not an accurate simulator of aerodynamics, but apparently it does this sort of thing to a certain degree.

A search on forums.x-plane.org revealed this link: http://tetruss.larc.nasa.gov/" which may be what the OP is looking for, provided he is an US citizen. It's probably very advanced, but I guess that's inevitable.

May I even suggest to try and post the same question on that discussion board?

 

[Cyrus]

I don't want to bite, but I can't resist :)

You design airplanes for a living - I get it. You're also correct that I don't, and hopefully neither does the OP.
But I believe you're completely missing the point.

What was asked for was a simulator to use as part of an Evolutionary Algorithm. Now, I'm almost certain this is not how engineers design aircraft, but I guarantee you this is an approach that can potentially perform better than garbage, given enough clock-cycles. (Assuming the simulator is relatively accurate, and the genome and fitness of the population has been defined in a sane way.)

This is not engineering, this is evolution, and as such can come up with effective solutions 1000's of engineers with 1000's of sliderules could never dream of! ;)

So, in your professional opinion, do you know of any open-source/free aerodynamics simulators?
I don't know of any, but a search revealed the link above. Now I understand that X-Plane is not an accurate simulator of aerodynamics, but apparently it does this sort of thing to a certain degree.

A search on forums.x-plane.org revealed this link: http://tetruss.larc.nasa.gov/" which may be what the OP is looking for, provided he is an US citizen. It's probably very advanced, but I guess that's inevitable.

May I even suggest to try and post the same question on that discussion board?

X-planes is a simulator, it does not, in any way, predict aerodynamic performance. Running CFD is a non-trivial process. Again, just build a model and see how it performs. Then make a change, and see how that affects the performance. Tabulate the results that you find.

I'm not sure why you think no one uses evolutionary programs to design aircraft. AAI corp does: well, it optimizes the geometry based on some design constraints you predefine.

Your misunderstanding is that simulators simulate aerodynamics - they do not.

 

[RolfRomeo]

My adviser once told me about a really bright guy who developed a code that could parametrically "design" airfoils. In theory his program could design the perfect airfoil. He ran it and the program spat out some toroidial looking debacle. Needless to say I've always tried to follow a rule that

Did they test it in a wind-tunnel? ;)

For a time human engineers thought that to minimize drag a surface should be as smooth as possible. We now know this is not always correct, something natures evolutionary algorithm figured out 100's of millions of years ago, e.g. shark-skin.

This silly statement was meant to illustrate that perhaps the toroidal looking debacle actually was the perfect airfoil, but since it didn't look like one it was trashed. Of course it's more likely his program had some bugs, or that it managed to find some optima that originated from approximations done in the evaluating simulation. Or something. Good story though.

 

[Cyrus]

Did they test it in a wind-tunnel? ;)

For a long time human engineers thought that to minimize drag a surface should be as smooth as possible. We now know this is not correct, something natures evolutionary algorithm figured out 100's of millions of years ago, e.g. shark-skin.

Um, you might want to rethink that statement. This is generally not correct. I would ask that you please not make statements you are not well versed in. There are Reynolds numbers where smooth surfaces are better, and Reynolds numbers where rough is better.

 

[RolfRomeo]

X-planes is a simulator, it does not, in any way, predict aerodynamic performance. Running CFD is a non-trivial process. Again, just build a model and see how it performs. Then make a change, and see how that affects the performance. Tabulate the results that you find.

I'm not sure why you think no one uses evolutionary programs to design aircraft. AAI corp does: well, it optimizes the geometry based on some design constraints you predefine.

Your misunderstanding is that simulators simulate aerodynamics - they do not.

I'm glad to hear that evolutionary algorithms are being used by engineers. They are effective for optimization, and certain problems are pretty insurmountable without them. The comment was more based on my interpretation of your reaction to the topic, than any knowledge I (don't) have of the industry.

I am aware that the norm for flight-sims is that the 3d-model is simply a visual representation, but please skim through http://http://www.jefflewis.net/x-plane_engineering.html" . I get the impression that X-Plane does nothing near full CFD, but it seems to do enough to make it a valuable tool for testing new ideas. I may be reading it wrong though.

edit: Of course I'm not arguing that (s)he can use X-plane as an evaluator and then make the model and expect it to fly. I'm merely suggesting it as an alternative if making the algorithm is still interesting, even after realizing the problem was darpa-hard (Heh, watched an awful show with a bunch of darpas using darpa-hard every 3 seconds. It's friggin darpa-hard to get that darpin word out of my head.) It should at least accommodate testing different airfoils, weight distributions and placing, count, and size of propellers.
I suspect that making an algorithm that designs flying machines is the goal here, more so than making flying machines. If you can see the distinction. But now I'm speculating about the intentions of another person. I'll cease and desist.

 

[RolfRomeo]

Um, you might want to rethink that statement. This is generally not correct. I would ask that you please not make statements you are not well versed in. There are Reynolds numbers where smooth surfaces are better, and Reynolds numbers where rough is better.

Thanks for correcting me :) I changed it to "this is not always correct". Ok?

 

[Cyrus]

Thanks for correcting me :) I changed it to "this is not always correct". Ok?

I would not say it is 'correct' or 'incorrect', the answer is that it depends on the type of flow (which is based on the Reynolds number). For aircraft, you absolutely want smooth surfaces on the wings and body. Having rough shark like skin would be terrible. Conversely, having smooth surfaces on a golf ball would be terrible. It's all about the Reynolds number.

 

[Cyrus]

I'm glad to hear that evolutionary algorithms are being used by engineers. They are effective for optimization, and certain problems are pretty insurmountable without them. The comment was more based on my interpretation of your reaction to the topic, than any knowledge I (don't) have of the industry.

I am aware that the norm for flight-sims is that the 3d-model is simply a visual representation, but please skim through http://http://www.jefflewis.net/x-plane_engineering.html" . I get the impression that X-Plane does nothing near full CFD, but it seems to do enough to make it a valuable tool for testing new ideas. I may be reading it wrong though.

edit: Of course I'm not arguing that (s)he can use X-plane as an evaluator and then make the model and expect it to fly. I'm merely suggesting it as an alternative if making the algorithm is still interesting, even after realizing the problem was darpa-hard (Heh, watched an awful show with a bunch of darpas using darpa-hard every 3 seconds. It's friggin darpa-hard to get that darpin word out of my head.) It should at least accommodate testing different airfoils, weight distributions and placing, count, and size of propellers.

Ok, perhaps this will clear things up. Think of it this way: a flight simulator is a dynamics model - nothing more. You give it: thrust, weight, inertia, mass, and aerodynamics, and it produces the evolution of what the aircraft does in time and shows you the image.

On the other hand, CFD (or wind tunnel testing) gives you the aerodynamics, that goes *into* the flight simulator. THAT is where changing airfoil shape, wing span, etc, etc, tells you what is going to change performance -wise.

This is why X-plane won't tell you anything about the airplane. You are putting the cart before the horse.

 

[daveb572]

Good morning Gentlemen! There are several great books on this subject that are currently available on the market through a local book seller. One that comes to mind immediately is written by Andy Lennon. It's entitled "Basics of R/C Model Aircraft Design: Practical Techniques for Building Better Models." Mr. Lennon's books do go somewhat into the engineering level but not so much as to be confusing to the average person. Another author is Carlos Reyes.

The internet is a wonderful place to search out the many aspects of RC aircraft design. Most all of the internet sites deal primarily with full-scale aircraft, but can be successfully used to design model aircraft. The differences will include Reynolds Numbers and a few others. The thrust to weight ratios will be somewhat absurd, too. Our models have tremendous thrust to weight ratios! This is evident in the latest type of RC flying called 3-D aerobatics. A good example is a maneuver called the Rolling Harrier Circle in which the model is hovering on it's prop vertically while torque-rolling AND also moving in a circular path! Quite a sight! Especially if an onboard smoke system is involved!

Also check out the local library. If they don't have one on the shelf, they can borrow one from another library that does. Also, check out the online hobby shops like Great Planes or Tower Hobbies. They post the specs of the planes they sell and this is great source of information regarding engine size vs. model type and span.

Some years ago I was into designing and building my own RC planes and almost exclusively used Mr. Lennon's book. Now, I'm not an engineer and do not claim to be but the models I have designed and built did in fact, fly very well. I would start off by deciding on a wingspan. Then I would "guess-timate" the weight based upon experience and other similar models of the same size. This would dictate the power plant needed to fly the model. Then the design work started.

In addition, I believe the original poster wants is a CAD program that includes a Virtual Windtunnel. Basically, a program that one can export a CAD design into and virtually "fly" the model and spit out a printed report on the aerodynamic properties of the proposed design. There was a DOS CAD program called "Computer Aircraft Design" a very long time ago. So long ago that it didn't work on anything above a 486DX2 computer. It contained a virtual windtunnel. I used that one to do my design work and windtunnel tests. Worked pretty well, too. There is one out there today called Model CAD 3000 but I don't believe it has a virtual windtunnel. It has a 15 day free trial but then you have to buy the program. Might be a possible solution to your issue. The virtual wind tunnel aspect may be solved with another program from RCAdvisor. com. They have a "calculator" program online for free. But again, full functionality is available through purchase only.

Wing planforms can be found on the internet also, but they are more in line with high-end engineering. They do work however for our application here. Try that in addition to Mr. Lennon's book and the graphs and charts become easier to understand.

Thanks! Dave Bryer

 

[daveb572]

Just a few updates on my previos post: I have purchased the full ModelCAD 3000 program along with an additional program called Wing Master. Both appear excellent for designing model aircraft. The wing Master program contains 1,250 wing planforms and design for you any wing shape, sweep, or multiple thereof for any wing you desire to construct. It will export to a printer or plotter your design either on one sheet of paper or a full size plan. The other program/website I mentioned is in fact a virtual windtunnel for testing your designs. That one is on the RCAdvisor website and is called "Calculator". Have a great day! Dave Bryer

 

[creyes123]

Thank you for mentioning my book and online calculator. The approach I took in my book was very different from the one that Lennon followed. Mine is not as number heavy - I put a bigger emphasis on the qualitative aspects. My goal was for the reader to understand enough about each subject so that they could take a reasonable guess without doing number crunching.

I used this philosophy in my second book, RCadvisor's ModiFly. There's not a single equation in that book. It's a very detailed and honest account of the process I took in coming up with that new design.

If you like to play with numbers, there's no better way than to try out my calculator. It's very friendly and very interactive. Online tutorials explain all the features, if you need them. The free version is surprisingly feature rich.

Carlos Reyes, AMA Scientific Leader member #4601
Author of RCadvisor's Model Airplane Design Made Easy and other books.
www.RCadvisor.com founder - Brainy calculator, so you don't have to be.

Good morning Gentlemen! There are several great books on this subject that are currently available on the market through a local book seller. ... Another author is Carlos Reyes.
...
The virtual wind tunnel aspect may be solved with another program from RCAdvisor. com. They have a "calculator" program online for free. But again, full functionality is available through purchase only.

 

[Mech_Engineer]

I've done some work in using genetic algorithms for optimization of engineering design, and it is a difficult, time-intensive process. Doing a genetic optimization can take thousands of iterations; meaning if your simulation takes a few hours to solve (like an accurate CFD might take) you're talking thousands of hours of solve time.