Yes, it is a general line detection method, so any point cloud input with lines in it can be inputted. For 3d scans, point clouds have some sensor noise and a robust implementation/version of this method would be better suited. But in case of a perfect cad model like the example, even the one that I provided in my prev comment should work fine. Do you only want to detect lines, or planes and lines? For planes, there are plane fitting approaches and then, one can find intersection of planes (as an alternative to Hough).

Hough Transform would work fine in this given case. It is based on setting up a hypothesis and evaluating the points in a voting scheme.

After lines are acquired, one can project on a 2d image from a given viewpoint.

You can check this repo which also provides visualization: https://github.com/cdalitz/hough-3d-lines