3D Laser Scanner

[GimpLizard]

So I saw this scanner (http://www.dynamism.com/3d-printers/matter-and-form.shtml), last summer, on Indegogo and decided it might be worth a try. The machine will output STL files. My plan was scan some 1/64, 1/32, and if possible 1.25 scale model cars. Then scale them down to N scale and have them 3Dprinted. (That's the plan... anyway.) Well, it's been a long wait, but my scanner finally arrived last week. So I loaded up the software and gave it a sample go.

First off, let me say the software could not be any easier. Once you calibrate the machine (which is also very easy) it is literally a one-button process to scan. Just lay the object on the turntable, choose the resolution, and hit the scan button. Depending on the size of the object, and the chosen resolution, the scan can take anywhere from a few minutes, to an hour and a half.

I looked around for something small for my first sample scans, and picked up an old Aurora Postage Stamp covered hopper that was sitting on my desk. I scanned it 3 times. Once each at all 3 resolution levels. (low, med, high) The low res scan took about 7 minutes. The high about 15 minutes. Sadly, the initial results are less than stellar.

I set the car down on the trucks, and ran the scan. The lasers take a shot. Then the turntable rotates a few degrees and the laser pop off again. That repeats until the table has made a full, 360 degree rotation. The amount the table turns at each shot - and thus the number of shots per scan - depends on the chosen resolution. So more shots at the higher res. This scan was at the medium resolution. Unfortunately, one scan at high didn't look much better.

[ Guests cannot view attachments ] [ Guests cannot view attachments ] [ Guests cannot view attachments ] [ Guests cannot view attachments ] [ Guests cannot view attachments ]

If I get enough time this weekend, I'll play around with it some more. The software allows you to "combine" scans. So I'll try multiple scans, at high resolution, and then combine them together. But that will take some time to do. A couple hours, at least, I figure. We'll see.

[wcfn100]

Pretty much as should be expected.  Scanning always seems to lag behind printing.

We're probably a long way from being able to scan a scale model to produce a scale model.  Just like 2D scanning you need a higher resolution scanning in than what you're going to output.  So even for Shapeways FUD which only gives an output of .1 mm, you'd want to have a scanning resolution of .03 (or about 15x better than what that scanner will do) if you're starting with something of like size.  You're best bet is to scan the largest scale model you can and see what it looks like.

The other issue with scanning is the garbled mess of an object you get back.  In many cases it will take more time to clean it up than it would have taken to draw it in the first place.

No doubt it will catch up and spring forth a whole new class of IP lawyers.

Jason

[GimpLizard]

If I can fit a 1/25 scale model on the scanner bed, and get decent results, I'm hoping I'll get a "smooth enough" finish once it's scaled down to 1/160. But, at this point, it's all just pie-in-the-sky speculation.

The "sample" scans the scanner maker has at their web site sure look better. So maybe I just have me some learnin' to do.

[Sokramiketes]

The covered hopper is a tricky subject.  I actually suspect you'll have better luck with an organic shape, with no internal voids like the open end platforms, such as a car model.

Make sure you throw some primer on the windows so the laser isn't shining through to internal surfaces as well.

[peteski]

The other issue with scanning is the garbled mess of an object you get back.  In many cases it will take more time to clean it up than it would have taken to draw it in the first place.

A big +1.

[cv_acr]

And you're scanning something *really* small.

I know Rapido Trains took a laser scanner to an FPA-4 locomotive, but that's 160 times larger in all dimensions, and they probably still had a lot of cleanup and adjusting to do.

[peteski]

My dentist (and lots of others) use scanners specially designed for scanning the teeth (for making crowns). These scanners are specially designed for objects smaller than N scale trains (teeth).  But I suspect that even those scanners would have hard time dealing with things shaped like hopper cars (lots of flat surfaces with ribs).  They seem to be happier dealing with smoothly curved surfaces.

[Scottl]

These are pretty cool.  We use larger scanners like this for topographic mapping (we call them ground-based or airborne Lidar).  Precise units are expensive, you need to get the right balance of "hits" to get a point cloud that is accurate, and the perspective-shadow effect is a complex problem to deal with.  It is the same scaled down to models or small objects.  You need to scan from multiple perspectives to constrain a 3D object, and there are always issues with areas that are not visible due to protrusions and hollows.  Cleaning up the data from a small scene is a substantial task and there are uncertainties in the data that would probably result in irregularities that would be unacceptable for modellers.  In principle, these are pretty cool tools, but having spent a lot of time cleaning up point clouds and teaching people to do it, I would suggest that 3D modelling is a more efficient way to make precision models.

Having said that, I have recently been fooling with making rock exposure models from lidar data and I think this kind of scanning is quite useful for some modellers.  We worked a rock cut up last week with a lidar unit (six hours, four perspectives) and I am getting a 3D print of the resultant model soon.  It will basically be a 100% accurate model of a particular outcrop.  Kind of over the top, but the equipment was available and I like rocks!  If it works out, I might try to make models/molds of some rocks that are tricky to simulate otherwise.

[peteski]

Having said that, I have recently been fooling with making rock exposure models from lidar data and I think this kind of scanning is quite useful for some modellers.  We worked a rock cut up last week with a lidar unit (six hours, four perspectives) and I am getting a 3D print of the resultant model soon.  It will basically be a 100% accurate model of a particular outcrop.  Kind of over the top, but the equipment was available and I like rocks!  If it works out, I might try to make models/molds of some rocks that are tricky to simulate otherwise.

Nice.  I knew we had some members here who work in esoteric fields,
but I never knew that we had our own Rocket  Scientist! 

[GimpLizard]

A lot of good points (no pun intended, though gladly accepted) have been made here.

The hopper might not have been the best "first" subject. But it was readily at hand. But then I grab another object from my desk... an antacid bottle. The foggy, translucent plastic kind.

[ Guests cannot view attachments ]

I don't have it to show, but... I wasn't terribly impressed with the scan of it, either. It was better, but still had bumps - and gaps - that shouldn't be there. It might be a problem with the resolution of the machine.

As I mentioned before, as the table rotates, it pauses every so-many degrees, (5, 10, whatever), the vertical line lasers - one right of camera, one left - go off and the camera grabs an image. It then rotates to the next position and the lasers flash again. And so on until the table has rotated the full 360 degrees. If the object is tall, the laser/camera head raises up to the next level and the table rotates again. If need be, the head will raise again, and the rotation repeats. Until the whole object has been scanned.

A scan is made up of a series of vertical lines (Lines of pixels, actually) every 'X' number of degrees around the object. The problem is, there are gaps between those lines. Even at the high resolution. I'm hoping multiple scans will fill in those gaps. Though I have reservations. Seems to me the "stops" in the rotation are fixed, rather than random. If that's true, I suspect the lines will simply overlay one another. At which point I'd have to say... what's the point?

However, the good is... the scan was in full living color. (Though it did appear out of focus.  )

[Scottl]

Our scanners are essentially random point clouds, with sufficient density to give a good surface.  For instance, in airborne lidar, we want the flight velocity and altitude to give us a density of about 6-7 points per square meter.    Yours sounds different but I suspect there might be some settings to work with to improve things.  I would definitely decrease the scan angle for best detail.  The surface interpolation method might be adjustable as well.

[Lemosteam]

The scan on the left was made with a million dollar scanner.  The picture on the right is the cad model.  For the money, I would invest in a great modeling tool like Catia, pay for the training, and send all of the .stl files I want.  Scanning simply cannot capture all of the accuracy of a real, machined surface.  A castiing maybe.  The point cloud that Scottl refers to cannot pick up every infinie point on a real surface, so it must be retouched before using it.  CAD skips this step entirely and goes right to the dimensioned data.  You could use a scan to estimate dimensions I suppose, in order to model things you don't have dimensioned drawings for...

[Scottl]

I agree, and the example you give has many perspective artifacts remaining.  The standard error of the method ensures that a flat surface, even if scanned at nadir (right angles), will be irregular when interpolated.  There are ways to automate combing visible images with the scan data to do some grouping of points to develop surfaces in the model, but these are complex and need coordinated image and laser scanning. 

Where these systems shine is with relatively irregular surfaces like the rock face I did.  If I was doing a locomotive or car, I would scan sides, top and bottom only, and take point measurement data on selected features to build up a 3D CAD model. There are ways to reduce the point cloud to the principal surface features (each "depth" in the scan). Given how simple most railroad objects are (curves aside), this hybrid approach might actually be a way forward, especially if the modeller wants to scope the dimensions of the model envelop, and then manually add the details.

Still, a good CAD modeller can probably do the work faster and more accurately direct from drawings.