Maybe this idea ?

Before t=0, turn at w1 and w2 disk. We need to give energy for that. W1>W2. Ar t=0, walls of disk has friction with air only at red point, this give forces F1 and F2 and reduce w2, like w1>w2, the energy increase, no ?

It's possible to give a torque to disk with external green object. Each object is accelerating, its energy is increasing. w2 slow down, not w1. The energy is :

1/2md²w1²+1/2mr²(w1−w2)²

with :

d = lenght of black arm
r = radius of disk
m = mass of disk

The energy of disk increase. The system increas its energy.

Edit: use a ring instead of disk for have (w1-w2) term.

nobody for help me ?

There is no motor after t=0, only before for launch disk. At t=0, apply friction ONLY at red point, not all around the disk. Red points change all the time their positions, this need special wall if you use friction (w1>w2). I don't know if it's clear enough. I think the formula is good only for a ring it's a little more complex for a disk. The disk win energy because w2 decrease. It's logical if you think with some points around the disk, at external the velocity is dw1-rw2 but at internal it's dw1+rw2, like the energy is a square function the external disk "cancel" more energy than it add at internal. I tested with Algodoo, you can test if you want it's very easy, fixe w1 and change w2 like you want and look at the sum of energy you will see the energy increase when w2 decrease (w1>w2). It's strange to say the energy increase if you slow down something but it is. And you can slow down the disk only at red point where w1 is the same for 2 points.

QuoteWhat is it that you would like help with?

maybe if you understand it allow me to explain the system, after if you can test for watch the energy increase with Algodoo, if you agree with that it's a first step for me. Test in reality must be difficult due the dynamic wall needed.

Yeah, I think we're ok. When I said it's difficult to built, it's for a pure 100% mechanical system. 

What's number is wrong, the formula ? I think it's ok for a ring and maybe for a disk, a french guy on a forum said the formula is ok for a disk but I'm not sure, another guy on another forum said it's ok for a ring not for a disk. I looked for 10 hours on Internet and I don't find the formula.