Hi,
The mesh in your lower picture is probably already the ideal quad mesh for the Kangaroo simulation given your boundary situation, since it has a nice topology of roughly equal quads. An even better simulation could potentially be created with a neat and tidy tri-mesh.
The topology plays an important role in what emerges later from the simulation.
The result that this will give is in my opinion probably nearer to reality than what a more regular grid like mesh would produce in the same situation.
In terms of architecture, it is pretty clear that no matter how the output mesh from Kangaroo is tessellated, it hopefully won’t be the finished roof, right? I mean, it has zero thickness, no space for rafters, purlins, insulation, windows, etc.
I would simply look at the output mesh as a guide or base surface that you can now use to maybe project base curves for a roof structure onto. These curves could then be swept with different rafter profiles, and you could go on from there.
Now, if you want it all to be “one thing”, meaning the primary roof structure and every thing else emerging from the mesh tesselation, then you need to be way more strategic in how you design your initial mesh.
I’d even say that the wavy boundary is secondary to this, since the sinusoïdal waves could be introduced in the Kangaroo simulation, instead of “hard coding” them into a three-dimensional curve from the get-go.
By being strategic, I mean be aware of the limitations of (quad) meshes, the limitations of Kangaroo, and marry all that knowledge to what you have in mind for the roof structure.
Meshes are to be manipulated prudently and a low subdivision count is best at first since things remain humanly manageable.
It is also not good practice to simply carelessly cut and chip away from meshes, since you’ll need to remesh to clean up the tesselation after the fact, and you’ll loose your custom, defined topological features and details. This also entails that openings are not simply cut out from anywhere on the mesh, but you must remove single or groups of faces.
This means the input mesh needs to be simple - polygonal (hexagonal?) not round -, more subdivisions will add roundness at the end, before or even after the relaxation simulation.
Prominent features have to be reflected by mesh topology from the get-go.
