You can scan the surfaces without a reference model, but you must go in by hand(via CAD) and fix the areas that are bad. Once this is done you can then use the model to create a tool path for re-cutting. What you want to do is skip the hand part completely, and to do this you must have a solid model (perfect) of the desired shape.
What you want to do is achievable but its definitely next level.....get the right group of nerds on it and you may have something!

The surface material is highly relevant. Unless it's been highly polished a cast material is going to give different results than a stamped or forged. Those differing results may not matter in some applications.

And sure, you can paint it black but depending on the type of black paint you use again, slightly different results. Matte black sandable primer works best because it dries quick and even and has some texture to it.

Other stuff we use is Aesub Blue which is a product specifically designed for this. Before Aesub we used to use Magnaflux Spotcheck which is a product used to detect stress fractures in things like jet engine fan blades but it's not cheap and is basically asbestos in a can. Oh, and it displaces oxygen so if you use it in an unventilated space you die.

One of the most difficult concepts we have to convey is you have to approach this from the POV that you are blind.

The scanner, the software, none of it has a clue what it's looking at. It's using the data it collects to reconstruct. So like a blind person using their hands to touch your face to get an impression of "what you look like" that blind person can only know what it's touched.

We can't reconstruct what we have not seen and I know it sounds simple to say "well I'll scan it all' it's not that simple a thing to do.

Imagine trying to paint a car with invisible paint. Are you really sure you got every nook and cranny? Now laser scanners will give you visual feedback on a computer display so that helps.

But like painting a part you may need to elevate or suspend it so you can get to all sides without repositioning the part. This is where the skill of the scanner operator starts to come into play.

It's not always an easy task to get around an entire object without breaking registration which means you end up with multiple pieces of the entire object that now need to be aligned and combined into a single model. That's skilled man hours at a computer.

Laser scanning is either phase based or TOF, time of flight. Phase based is faster but generates more noise than TOF, which also generates noise.

It's the noise that is part of your challenge.

Look here, the face scan, that is raw laser data, note that the eyes did not reconstruct because they have transparency, the light energy passed through the surface.

https://rigsters.com/services/post-optimisation/

I know you are not doing transparent so that was just something to note.

Look at her skin, see how mottled it is. That is noise and every single solution out there will impart some noise that will need to be smoothed which effectively creates a 3D model that deviates from the original object.

Think of it as sanding or using body filler to fix the 3D model in software. May not be much but it's in the range of tolerances you are trying to hit.

Here's the frustrating part for your application, the less feature rich / smoother the original scanned part, the more noise you get, it's just a fact.

You are far more likely to end up with a raw scan that has more defects than the original part you scanned.

If you were scanning a piece for reproduction, generating a model, cleaning it up then using it to mill a new piece from blank stock your idea is sound and it's already being done.

But to take an original part, create a model from it and use that modified model as a base to modify the original part is asking for trouble because the reference 3D model is not the same as the original physical piece.

It's a problem mostly because of the tight tolerances you are trying to hit.

There is no point and shoot solution here, it all involves man hours and a skillset that is not cheap.

Monkey,
We have 2 types of scanners. A white light scanner which fits your description above perfectly. It is old technology. However our Blue light laser scanner does not seem to have these issues at all. It should be noted that the only thing we scan are metal parts, some with a very smooth surface some with not....

Which Blue Light devices are you using?

Faro

So, you are in Ohio.

Would you be open to Shaun, if he can, sending you one of the parts he is thinking about working with to scan and provide sample models for evaluation of what else would be required in the workflow?

Just to be transparent I've not been trying to give him a hard time, he is one of my best friends and I confide **** in him that I don't anyone else so I'm really motivated to have his back as I know he has mine.

We can do this.
Let me know.

One thing we cannot do and what I think Shawn really needs....is what we call reverse engineering. Start with a laser scan then import into the correct software to be able to create the surfaces he needs. In the past we have used this company to "reverse engineer"

https://www.sis-inspection.net/services/dimensional-inspection/

We can give him the point cloud, but we do not have the software for him to be able to actually use it to develop the tool paths.....

I have a contact at SIS (website above) that is excellent for this type of work.
Let me know.

Thank you, I really appreciate it! I will send you an email now.

I can work with the point cloud to generate a mesh that can be converted to a nurbs asset.