View Single Post
  #2  
Old 12-24-2018, 09:25 AM
crystallographic crystallographic is offline
MetalShaper of the Month October '14 , April '16, July 2020, Jan 2023
 
Join Date: Dec 2010
Location: Western Sierra Nevadas, Badger Hill, CA
Posts: 4,388
Default

Quote:
Originally Posted by Michael Moore View Post
I'll offer a few comments on DIY scanning. I have a 2nd gen (HD) NextEngine laser scanner

http://www.nextengine.com/products

which is designed for small/medium parts and is stationary on a table/tripod during a scan. It can capture a lot of detail, but getting clean mesh data is not something that always happens very quickly. The Artech that I've seen mentioned being used for scanning bodywork is slick, but the cheap one looks to be $19K vs $3K so it is the type of thing where you really need a commercial goal to justify a purchase.

I've been recently scanning a streamlined motorcycle seat back/bulkhead that I borrowed from a pal. Photos of the development of the Mark 2 version of the seat are here:

http://craigvetter.com/pages/2015-St...enges-p87.html

You can see there's a lot of shape in the part. The goal is to get a CAD model of it and also to get some practice with the scanner and software.

I originally scanned with the standard software provided with the scanner. Because of the line of sight issues (if the lasers can't see it, the surface isn't there) I had to stitch a lot (about 55) of scans (about legal typing paper size) together. Black/dark surfaces drop out and glossy surfaces often aren't seen well by the scanner either. Ideally you'd have a matte white surface. I recently got the OK to lightly scuff the gel coat to kill the shine but so far I've only scanned the unmodified part so there have been "issues".

55 scans and alignment of them was an all-day job. It made me decide to buy the "Pro Scan" software from NE to increase the depth/width of field and speed of scans. With that I've gotten down to about 20-25 scans. The scans go quicker if at a lower resolution and low crown areas don't need a lot of points to define the surface so I'm using "standard" density.

But there are still spots that didn't get picked up, and other spots that have multiple layers of scan data because I had to hunt for enough scan targets in the individual scans to get them aligned, and that might mean a lot of overlay. For example, the "wings" at the bottom of the seat -- you can get one side or the other or a 2" wide strip down the ridge that might include a bit of surface off to the side. So one of those sections might need 5-6 scans from different angles to get at least three common targets visible . All the inside corners cause issues too. Something like a typical car fender might not be too hard to scan, but a part with deep/sharp ridges/hollows and other line of sight issues may be a chore.

A surprise to me was that .004" thick scan target labels get picked up by the scanner, so there are bumps from those in the data. This later version of the seat was very Q&D and is lumpy and asymmetric and all of that gets picked up in the scan data too.

You spend a lot of time snipping stray bits of mesh off the scans, or bits of floor that got picked up in the scan.

Once the scans are aligned into a part-like shape you get to start the clean-up. This is repairing defects in the mesh, little holes, big holes, removing the bumps from the targets, smoothing the surfaces (but trying to not loose detail you want) etc etc. All of that takes time while the computer manipulates the data so you get to do something and then wait before doing the next thing.

If you are going to do a 3D print (which I'm not) you also need to make a watertight surface and sometimes the automatic features for that get carried away and make surfaces where you didn't want any, so you get to undo and start that process over again.

I got a seat of an older version of Rapid Works (reverse engineering software to generate parametric CAD models from the scan data) with the scanner and that's got a 700 page manual with it. I'm not sure how useful it is for things like this vs "mechanical" types of parts where it can see a clear cylinder, plane surface or other typical CAD feature. It does have an "autosurface" command which I've used to surface the mesh before pulling that surface into Rhino. It also has a "remove target" command that I'm getting ready to try. It presumes a round label and you tell it what diameter to look for and it searches for those bumps and takes them off the mesh. My first attempts were with cut up mailing labels as scan targets but I've now got some 3/4" round labels that I've printed targets on and I'll do another round of scanning and see how well RW removes them.

I've tried modeling the seat in Rhino from scratch as I would a regular part and that gets complicated quickly. I've also been trying a different way where a clean NURBS surface that looks similar to the part is generated from a network of curves abstracted from the Rapid Works surfaces or the STL mesh. That surface can then be manipulated with control points so it stays one fairly clean surface rather than a bunch of small pieces patched together.

I've gotten something that is getting close, using the Rapid Works generated surface as a "buck" underneath the Rhino surface, but as with metal it is easy to get one part close and then see a section nearby that was looking good has gotten pulled out of place.

To sum up, it can be done, but don't count on it being easy! I think the Next Engine is actually a pretty capable scanner for the price, and it can pick up very fine detail. But once you get a part bigger than a shoe box (or so) you are probably going to have to do a lot more work to get it scanned and stitched together. After that, cleaning up the scan data is probably going to largely depend on how powerful your software is and how skilled you are in using it. I often find that I have software so powerful it can only be used for good or evil, and I'm not sure how to make it give me the good results.

Whether scanning is something you need may well depend on what you want to do with it. If you need a +/- 1/8" model of an engine or other part to plug into another model as a place holder that is pretty doable. Getting a "good enough to run clean toolpaths for CNC" model may take some doing, and that will depend on the part too.

FYI, I got my scanner with the RW software from a friend who'd bought it thinking he was going to be making replica bodywork parts (side covers, tool boxes, etc) for some vintage Italian motorcycles. He found that for the things he was doing it was quicker for him to just model the parts from a clean screen in Solidworks than to scan, clean up the data, and try to extract a parametric model (to import into SW) with Rapid Works. So I got everything for about 1/6 what he paid to purchase the hardware/software new. At that price, it is quite affordable to mess with. At 6X the price (or maybe for those fancy commercial scanners 30-60X) your mileage may vary.

cheers,
Michael

Very useful info, Michael. Thank you !
Glad to see what you are up to.
Happy Christmas!
__________________
Kent

http://www.tinmantech.com

"All it takes is a little practical experience to blow the he!! out of a perfectly good theory." --- Lloyd Rosenquist, charter member AWS, 1919.
Reply With Quote