Here I started with a surface with the right topology, and pulled those 2 squares where it touches together with springs, while also using bending, length and shear springs.

Because the divisions of the initial surface are uneven, you can't use too stiff a bend force initially without it exploding (the current bending model works best when consecutive segments are the same length).

So I put in sliders for all the different stiffnesses, drag and timestep, and you can gradually adjust these as it gets more into shape - it's a little delicate, but does reach a solution.

I didn't try it yet, but it would probably be better to add a planarization force to the quads, and this would also give a better guarantee of ending up with a developable surface.

That sounds like two very good additions!

I'm planning for a large scale test of this double loop shape using double layers of 2 mm plywood joined to a length of 6 meters...