Least Squares Geometry Fitting

Hi All,

Has anyone tried approximating arbitrary freeform surfaces with primitive geometry such as a cylinder or torus? I am interested in something flexible and probably code based.

A simpler example of this is the "Plane Fit" component - maybe David Rutten can shed some light on how fitting was achieved here?

I have briefly read over the following: Gradient descent, Gauss Newton, Monte Carlo. Any one have any advice on these?

I would very much appreciate any advice on this.

Cheers,

Sam