Neural networks applied to CubeSats

[Dylan_]

Hello everyone. First of all, I hope this is the correct subforum to post this, because I couldn't do it in another one.

I'm an Electronic Engineering student and I'm a beginner in the aerospace world, but last two weeks I've been reading a lot (specially from NASA (earthobservatory.nasa)) about this. I have to make the final project of a course called "Neural Networks" and I decided to do something related with micro-satellites but I don't know what.

I found two ideas:
- A complete control project, by Schreiner (or something like that). It's too complicated to do that as my first project
- An atmospheric density prediction, based on CHAMP measurements. I liked this project but I was told that it's not too much useful to predict the drag force for micro-satellites.

Does anyone have any idea?
Thanks!

[wumpus]

On 9/11/2016 at 7:06 PM, Dylan_ said:

Hello everyone. First of all, I hope this is the correct subforum to post this, because I couldn't do it in another one.

I'm an Electronic Engineering student and I'm a beginner in the aerospace world, but last two weeks I've been reading a lot (specially from NASA (earthobservatory.nasa)) about this. I have to make the final project of a course called "Neural Networks" and I decided to do something related with micro-satellites but I don't know what.

I found two ideas:
- A complete control project, by Schreiner (or something like that). It's too complicated to do that as my first project
- An atmospheric density prediction, based on CHAMP measurements. I liked this project but I was told that it's not too much useful to predict the drag force for micro-satellites.

Does anyone have any idea?
Thanks!

It sounds like any high-coverage data from such satellites would work, such that you could remove something like 10% of your data as a control system (and presumably break that up into sub-groups: training with less than the full group and comparing it to the the sub groups, then taking the best of those and comparing it to the "real" controls).  What I'd really want to do with aerospace and neural nets is to hack up a kOS routine that would land a booster spacex-style (or possibly simply get into orbit).  I strongly suspect that C# and kOS just aren't up to the challenge (of building/training/running a neural-net) and that trying to find another rocket simulator would be prohibitive.