[aschen]

This is a tougher thing to automate than it probably seems. It is multidicipline for mechanics, machine learning, and metrology

There are a lot of ways for a machine to physically measure a part, Photogrammetry, Laser scanning, direct probe. Hopefully the basis for the sanding or machining of the parts already exists for adaptation or it could get VERY expensive to develop.

I don't have any great useful advice and I don't understand the requirements well obviously but my experience is that this sort of thing is very expensive to do well, but the only way it can be economically feasible is if existing machine platforms can be adapted.

Another thought is that if you have a desired final shape a machine can be used with open loop tool passes programed to the right geometry, maybe on the low material condition side of the tolerance. That way you can knock down high spots to an acceptable geometry.

If a machine needs to recognize qualitatively unsightly defects and make decisions on how to blend them into the rest of the part, it could be quite an ambitious project.